Tom Baden - Evolution of the Eye

Hey, hi everyone. So, it's my pleasure to introduce this second talk of the colloquium series of the priority program. So, today we have Tom Baden, professor of neuroscience at the University of Sussex, where his lab studies the function and evolution of neural circuits in the vertebrate retina. So, they work across many different species, from zebrafish to frogs to birds. I have to say that Tom has done an outstanding contribution to our understanding of visual circuits evolution, which have been of course recognized by multiple awards and grants. I can for instance highlight the recent ERC advanced grant. And I also want to say that in addition to the great science from the Baden's lab, it's worth to highlight the commitment to open science initiatives, including open science hardware and open data availability, something from which I myself have benefit in the past. So, with that I just want to thank Tom for being here today, and the stage is all yours. >> Great. So, yeah. So, thanks very much for the invite. I'm delighted to be able to show you all some of the work we've been doing recently, very much centered around the evolution of the eye. So, just before I jump in, um I wanted to first briefly acknowledge the key people that have been doing the work I'll be talking about, and also give you a bit of an outline of what I'll be talking about. So, there's going to be three sections. First, I'm going to talk about recently published work about the origin origins of the eye, and this was led by George in our lab in collaboration with Mike and Dan Eric in in Lund, and contribution from others as listed here. Then, I will next jump into sort of a topic that's sort of a new interest in the lab, where we're trying to do what I like to call real systems neuroscience in an evolutionary context, so many species trying to integrate data and see what the patterns are. Um, and then I will talk about a recent preprint, um, which was led by Shinwei, which talks about latent potential, the idea that circuits have the ability to do computations, um, in principle, it's just that those computations are normally suppressed, but they can be unlocked, um, both by evolutionary processes, but also, um, throughout the life of an animal as computational needs shift. So, um, with that being said, let's jump back here and, um, let me introduce you to your ancestor. So, this guy here is the artist's idea of, uh, the Uurbilaterian. So, this is the last common ancestor of all animals with bilateral symmetry. So, this includes us, but it includes flies, it could include octopus, spiders, all kinds of interesting critters, anything that's bilaterally symmetric. Um, and this guy lived about 600 million years ago. It's about the size of a fingernail, and it's a benthic grazer. And what that means is it moves around the seafloor, um, and it's looking for nutrients that don't run away. Um, and it is already a fairly sophisticated little critter here. Um, so, the idea is that there's already a clear cephalic region as distinct from the rest of the body. Inside that cephalic region, some sort of brainy kind of thing, whether it's a proper brain or some sort of nerve net, who knows. Um, and clearly this animal was able to do interesting things like navigate, um, and it has it was equipped with all kinds of sensory apparatus that would have allowed us to do, uh, allowed it to do that, uh, including not not only vision, but, um, some sort of light sensitivity would have been in this animal as illustrated here by this guy sort of pointing his his his head towards the light. Um, and the reason I'm bringing this guy up is because I think we need to understand this sort of situation, or we need to at least think about it, um, if we're going to try to understand where our eyes come from. Now, why would I say that? So, um where do our eyes come from? If If you look at pretty much any textbook that that discusses this, you're going to get some sort of version that sort of is illustrated over here. So, the idea is that we have a sort of classical sequence that goes from simple to not so simple. Um in a sort of stepwise manner. So, you've got isolated neurons in some sort of un for stage that sort of a little bit cone and ganglion cell. Um they come together form a two-layer circuit. Then that two-layer circuit turns into a three-layer circuit probably by duplication of one of the elements, maybe the top one. Um that diversifies gives you on and off pathways. And then this is kind of a retina already. In the sense you've got cones, bipolar cells, ganglion cells, possibly some other stuff that I'm not plotting here. Um but the rods are not yet there. And then we get the rods later. And but because this circuit is already kind of cooked and it's thing, you can't just unplug it and plug the rods in. So, the rods need to invent this new sort of roundabout way of of piggybacking onto the circuit involving an extra bipolar cell and a weird amacrine cell and all kinds of other stuff, which then feeds the rod signals into the cone circuit and ultimately all feeds into the brain. So, that's kind of how we for a long time have been thought to thought to think about how how retina evolved. Um and this picture was uh built um largely by anatomical and physiological evidence that sort of compare the cross species. A little bit of molecular stuff, but it really didn't have the power of modern molecular comparative um techniques at hand. And turns out that if you use those techniques, so this is like detailed genomics, transcriptomics, that sort of level, um this picture just doesn't hold it up. Um and let me just give you a few reasons why not. So, the idea that um we have a two-layer retina leading into a three retina layer retina comes from hagfish mainly. And hagfish is a vertebrate. It's a It's a very early diverging vertebrate. It has eyes, you can see them here, but the eyes look a bit weird. Right? So, there's there's skin over the eyes. This is not a proper cornea. There's no lens, so the the optics are questionable. And uh if you look inside, there is a retina. Um but that retina, if you zoom in, it looks two-layered, not three-layered. So, the eyes The idea is that this is sort of a remnant of this this stage. However, the problem is that that's incorrect because A, you can stain uh using molecular tools for interneuron markers. So, things that we that should be in interneurons, and for sure you get hits. Um so, it sort of looks like these two-layer retinas um include a third layer. [clears throat] It's just somehow collapsed. And perhaps more importantly, um we've revised the position of hagfish in a phylogenetic tree. So, we used to think that hagfish might be sort of an ancestral vertebrate slightly before the lamprey, so a different lineage. However, it turns out that's not true because um hagfish and lampreys share a whole genome duplication, which is distinct from the whole genome duplication of all the other vertebrates. And that means that hagfish and lampreys must be monophyletic. They're part of the same branch. And lampreys have three-layer retinas. And that means that whatever hagfish have um is not necessarily the ancestral state, but it is just a regressed version of what lampreys have. And that means that really this kind of whole chunk goes away. There's really no evidence for it. Um next. Um with uh you've probably come across this idea of a simplex versus a duplex retina. Um and the idea is a simplex retina has just cones, and a duplex retina has cones and the rods as well. Um turns out that's um and and the idea is that this guy sort of precedes this guy uh in an evolutionary sense. And much of this was built on the idea that lampreys, one of the earliest diverging vertebrates, don't have rods, but everybody else does. And therefore, this is sort of the original. But it turns out that's wrong because lampreys do have rods. The lamprey people have been saying this for years, um but now the molecular evidence has sort of really nailed this. It's completely clear that if you look through single cell transcriptomics on lamprey eyes, um, you get rods and cones and they're homologous to the rods and cones of all the other vertebrates. So, the rods clearly are part of the ancestral complement of the vertebrate eye. Um, and we've taken this a little bit further uh, in this collaborative work here where we've sort of tried to under use comparative transcriptomics across a bunch of species to sort of try the lineages of photoreceptors. Um, and this is sort of our most likely scenario. So, we start with some sort of ancestral photoreceptor, probably a cone-like thing, that a very early on, before the last common ancestor of the vertebrates diverged into a cone and a rod. Um, and then the rod just stays the rod. And the rod is always the rod. So, the rod of a lamprey is homologous to the rod of a human. Um, and then the cones, that's also true. For example, the original rod is probably homologous to the LM cones in humans, the the dominant ones. There's also this guy, which would be the blue cone in humans. And all kinds of diversification and loss has happened over lineages, but at the end of the day, really, it's just two diversification steps. Um, and much of this is pretty much ancestral. So, photoreceptor complement, including rods and cones, are both ancestral in vertebrates. I think it's quite difficult to dispute that now. Um, it's not just the rods that are ancestral, though, it's also all the schlock that comes afterwards. So, for example, the rod bipolar cells, that was also shown in this paper, they also exist in lampreys. So, that's also an ancestral trait. Um, and then Takeshi, when he was, uh, Takeshi Yoshimatsu, when he was in the lab, he showed, um, actually using data that previ- that came from from previous lab, from Rachel Wong's lab, that the zebrafish has a rod bipolar cell pathway that is essentially the same as the one of the mouse, including the A2 cell. Um, we also see this now in catsharks. So, this is George Kafitzis here. He's done a volumetric EM, and he's he finds the rod bipolar cells and all the circuits that, uh, we expect there to be there. So, catsharks also have this. And, in fact, uh, recent transcriptomics work, uh, this guy here, um, showed that transcriptomically the A2 and the A17 are pretty much present in all the vertebrates. So, all of that stuff is just O H basically. Um and really what that means is that this what we have here is the sort of the end of the evolutionary sequence. That's the thing that's present in all the vertebrates as far as we can tell. Unless they've lost it. Um and that also means that whatever this state is, um we've got if it even existed, the link to whatever this ends up being is completely unclear. And um while it is exciting in some way that, you know, all of that stuff that we know to exist, for example, in a mouse is probably also present pretty much all the other vertebrates. Um it kind of leaves us a little bit in the lurch in terms of understanding where all of this comes from. Right? And it really does mean that to understand where this comes from, we can't just look in the vertebrates because all the vertebrates have the same thing. You need to look before the vertebrates. Um and so this is where then Dan-Eric Nilsson and Mike Bock came in with a massive survey of animal eyes. By animal eyes in this case, I don't mean all the animals. In this case, I mean bilaterian animals, which is just the majority of the animals. There are of course non-bilaterian animals out there as well. So, if what what they what they've done is you basically look across species, what kind of eyes they have, where on the head they are, and what the lineage identity of the neurons and sight those eyes probably is. And the long story short, the the the big picture really is that eyes only really exist in two places. They're either on top of the head in the middle or they're on the side of the head and they're paired. And more than that, the ones in the top of the head, they come in two flavors. They come in the flavor of pigmented, so directional, and unpigmented, non-directional. Um whereas the ones on the side of the head always come in pigmented, so directional. Now, we need to think a little bit about what that means. So, for example, if you stick a simple photoreceptor system on top of your head and you don't pigment it, that system isn't directional. It doesn't know where the light comes from. But it's still it's it's useless, right? It still tells you that there is light and it can tell you, for example, about the wavelength composition of the light, about the timing of the light, intensity of light. So, these kind of cues are essential for various tasks. Circadian rhythms is the obvious one, but of course there's others. For example, if you sit in the water, you might be able to measure water depth. Or you might be able to guess what the what what the weather is, or if you happen to be in the shade. So, all of these kind of things you can do with a non-directional photoreceptor system in the middle of the head. But the second you stick a bit of pigment around it, that thing becomes directional, and that almost by definition it turns it into an optical statocyst, a system that tells you where's up. And you know, if you're bilaterian, one of the most fundamental things that you need to know in the world is where's up, because unless you've got your head straight, all that left-right symmetry that you sort of evolved to exploit to do interesting things goes away, because if it's sort of tilts left and right, then left and right is no longer left and right. Left and right becomes up and down. So, you need to keep your head straight in order to use your left-right system for anything useful. Then, so basically this disambiguates the vast majority of things that an animal needs to do in order to sort of set itself up for some success in in terms of being able to navigate with with a much of, you know, like image-forming vision yet. And then you've got your lateral eyes. And the one thing that the lateral eyes can do that the median eye can't do by virtue of its position, by virtue of that it's paired, is that it can resolve left-right decisions. So, basically the yaw dimension. But it makes no sense to make left-right decisions unless you already know where's up. First need to know where's up, and then you need to know this. Now, so this is what these eyes are fundamentally for. And of course, eventually, once you've got these eyes, you can make them bigger, you can make them image-forming, then you can sort of start developing circuits for detecting optic flow, translation optic flow, that's very useful. And eventually, you know, you make it more sophisticated, you go for prey capture, and one day you go for appreciating sunsets. But that really is quite far along the line in in terms of where the evolution has started from here. Okay, so those are the kind of eyes that you see in the eyes of animals, but to make it slightly more concrete, uh let me just give you an example. So, here's a fly. Um and flies have eyes. Um these guys here, these are big compound eyes, these are the lateral eyes, so these guys here. Um but they've also got ocelli, so these are median eyes, pigmented median eyes. So, these are these guys. So, these are in the fly, we know they are optical status, they help flight control and all kinds of other interesting things. Um plus the fly has also unpigmented distributed photoreceptors, which sit in the middle of the head. Now, um what we've done here, we've annotated the position and identities of these, but more than that, we have also annotated the kind of lineages of the neurons that are inside of those eyes. And that's the color code. And without going into any detail, just appreciate that the basically everything here is blue. And blue in this case means rhabdomeric, the R opsin. So, this is we we know, of course, flies use a rhabdomeric system for pretty much all of the vision. The compound eyes are full of it, the ocelli are full of it, and some of these unpigmented ones are also full of it. In addition, there's also a different a color here, there's green. Green stands in this case for a C opsin, so a ciliary like the rods and cones, but it's not the one of the rods and cones. It's a different lineage of C opsins that acts through different kinds of G proteins. So, it's just a more ancient version of the phototransduction system that ends up being in our eyes. Now, that's a fly, but the fly is not special in this regard. If you look, for example, across the arthropods, so basically anything with an exoskeleton, it's always the same thing. Yes, the eyes move around a bit and they're doing interesting things in different lineages, but basically it's always all blue and you always have all these systems present. And that means that what you can do is you can take these systems and you can reconstruct the most likely ancestor of all the arthropods. And that would have looked like this, already with big paired eyes and all of the slaughter in the middle. It's a little bit analogous to what modern Limulus still has today. Now, this is the arthropods, but of course, there's other animals out there. For example, here we've got the lophotrochozoans, so this is basically this is your octopus and your your your your polychaete worms and scallops and snails and fan worms, and you know, all of these kind of guys that I'm showing here. Um, and more there's a bit more variety in terms of what they do with their eyes. Um, but big picture is it really ends up being kind of the same just sort of developments of the same theme. So, again, you can sort of reconstruct the ancestor, and it would have looked like this. Very similar to this guy. Main difference being, I guess, that there's another c-opsin system present. It's just a different version of ancient ciliary opsins, but again, not the ones in the rods and cones. So, out of these guys, so but these guys also have an ancestor. So, in their last common ancestor, it gets you pretty close to the last common ancestor of bilaterians. And then, if you sort of consider these guys here, plus all the stuff that based on molecular evidence, we know predates bilaterians. Um, so this becomes the by far most parsimonious solution to what the bilaterian ancestor would have looked like. Already with some sort of lateralized rhabdomeric pigmented system. Probably not an image-forming eye, more like an ocellus. Um, already a medium pigmented system, also rhabdomeric, and then some non-pigmented uh, ciliary um, system along the midline. Okay. So, if that's what the bilaterian looks the obilaterian looks like, then this is also our ancestor. So, if we're going to try to understand where our eyes come from, we're going to have to somehow relate it to whatever this guy's got going on. Now, if we look at what vertebrates look like, and this here is a lamprey, so one of the earliest diverging ones, it also has a lateral system and a median system. The lateral system is just what we call eyes, and the median system is what we call the pineal gland. Now, a lot of animals, um, a lot of vertebrates out there that are aquatic, and especially the ones that sort of swim close to the surface, they will still have a pineal gland that's actually sensitive to light, um, and guides behavior. In mammals and birds, specifically, it hasn't been lost, but it has been internalized under the skull. in most cases the light sensitivity has been lost, but the sort of the core function of pineal for example regulating melatonin release are still there. So, we we still have this, it's just under the skull. Um but it is originally a median eye. Now, um if we have this as we can that means we can also look at the lineage identity of the neurons that are sitting inside those eyes. And this is where it's completely different from the rest of the tree. In this case now, we're getting rainbows, but we're getting rainbows both in the lateral eyes and in the median eye. And that should immediately tell us that there's a very good chance that there's some sort of relation between this median system and the lateral system. Now, how can we go from the bilaterian to us? Uh the first clue comes from looking at all the other guys on our our branch of the tree. So, these are deuterostomes and there's not a lot of them left other than the vertebrates, um but there's a handful. For example, lancelets or as cephalochordates, but also sea squirt and salp which are actually a little bit closer to us. Uh these guys um well, they're round and if you look at them they have a head, but they don't have lateral eyes. They only have median eyes. And if you look at the lineage identities of these median eyes, it's all of the ancient stuff. It's the stuff that was already present in the beginning. They don't have the lateral systems. Right? And that what that kind of tells us is the most parsimonious solution to getting therefore from here to here taking into account what's happening here is a two-step process. Step one is we go through a phase of loss. We keep the median stuff, most of it anyway, and we lose the lateral stuff. And then the second step, specifically in the vertebrate, we gain an extra opsin identity. This is where there's red here and red is the C opsin that's sitting in rods and cones, it which doesn't exist in any form outside of the vertebrates. Um that is present in the median eye and then that median eye gives rise um de novo to a uh a pair of paired eyes that are lateral. Now, why would I say this makes any sense at all? Let me try to make this slightly less abstract. So, here is a cartoon of the obeliterian. It's moving around the seafloor. It's got the lateral and the median system. And it's the small benthic grazer. Now, if you going to want to go from here to pretty much any protostomes or arthropods or whatever goes on, um the main the the main thing really you have to do in terms of vision is you take these little ocelli and you make them better. You make them bigger. Turn them into eyes. So, it's a fairly sort of straightforward conceptual jump from here. Um but you can't do that for us because for us um the idea is that we've lost these eyes. Now, why would we lose these eyes? Well, turns out that there's a lot of fossil evidence that suggests that actually in our lineage um there was this short but important stage where instead of moving around um we became sessile filter feeders. Um so, basically you stick your bum in the ground, you stick your mouth into the water, and you hope for nice things to float by. Um it's a it's a it's a pretty good lifestyle. Um and it's a lifestyle that means that you don't need to navigate anymore. So, if the original left-right ocelli, so the lateral eyes are for navigation to basically tell about left and right, uh but you're an animal that doesn't move left or right because you just sort of uh stick your bum in the ground, then you don't need it, so you lose it. So, um just a little bit of evidence. So, the fossil evidence suggests that this happened, but actually if you look at all of the deuterostomes um that are not vertebrates, that's there around, they still do that. So, here's your uh your your lancelets. They are filter feeders like this. The sea squirts also like this. And even the lampreys, the larval form is a sessile filter feeder. And you can see the eye isn't quite there yet. So, I think there's a pretty good evidence that this is what happened in our lineage. A sort of cyclops moment in our evolutionary history. Um but then eventually these guys that became our ancestors must have come out of the ground again, started free-swimming uh filter feeding. It's a more efficient way of filter feeding. Um but when you do that, um well, then you need to start to navigate again, because if you can't navigate, you're basically planked on. You're just being moved around by the water currents. That's not a very powerful uh strategy in the long term. Um, so you need left and right eyes, but the left and right eyes are gone. So, where do you get your left eyes left and right eyes from? It's where you take your median system and you lateralize it. And I think this was probably helped by the invention of this new lineage of opsin and phototransduction systems. It's now sits in the rods and cones, which is only present in our the lineage. So, something must have happened. You must have have this system present. This guy adds in. Somehow everything merges into a common eye cup. So, this is sort of mega eye kind of situation, and that starts to lateralize. And this the second it starts to lateralize even a tiny bit, you're starting unlocking vertical light gradients, which A give you more information in this looking up. Um, and also they start to unlock left-right information. So, the second you do that, the evolutionary selection to make the eyes properly lateralized, to really do uh resolve left and right will be high. So, the selection will be a very high evolutionary rate to lateralize those eyes. Um, so we lateralize them. Um, these become our dominant eyes, and what is in the middle becomes eventually the pineal. Um, but note that there's no optics yet, because you don't need optics really to do any left-right decision. You just need the pigment. Um, but the optics, of course, came in eventually, and they came in plugged from the outside, from the surface ectoderm. So, this is your lens and your cornea. Um, so this then gives rise to the modern vertebrate eye. Now, if this is right, it kind of implies that this is really the eye, but this guy here already, that's the retina. So, the retina by that logic would predate the eye, which I think is is a cute tagline. Now, if anything of this is right, we should strongly expect to be able to look in the pineal of animals, of vertebrates, and look for the molecular signatures and relate them to molecular signatures that we know exist in the retinas. Um so here's a pineal data set um from from Inbal that she kindly shared and just just to annotate what's in there, this is zebrafish. Um first off, you get a bunch of clusters and if you annotate the clusters and make educated guesses what they are, it looks like these guys seem to be the rods, quite a lot of them. And these guys seem to be cones. The rods and cones, there's clear molecular signatures for that. There's also molecular signatures for retinal pigment epithelium-like cells, so the things that keep the rods and cones happy in the retina. There's also a big cluster of Müller glia-like cells, so these are of course key cells in the retina that keep the retina as a whole happy uh and help development of the retina in the first place. So all of that stuff seems to be there in the pineal, so that's a good start. But of course you might say, "Okay, great. So you've got a bunch of molecular markers for these cells. Do we know that these cells do the things that we know they do in the retina?" Um so we wanted to know that and so Tessa here in the lab uh tested this directly. So she went uh in into our database of baby zebrafish transgenic lines and turns out that well, if the if the retina and the pineal are related in an evolutionary sense, then usually when you express something in the retina, you're going to get expression in the pineal as well and that's true. And so for example, in this case, this is a pan cone retina GCaMP line. Lots of cones. Uh you point the two-photon microscope at the pineal and sure enough, you're getting expression. And then what she did is she just pointed the at the two-photon microscope, played a bunch of stimuli, the sort of stimuli that we use to characterize retinal photoreceptors and sure enough, you're getting very clear, very fast, very high amplitude responses. These are not averaged. This is just what comes out of these cells. So these are proper cones. And really, if you compare the sort of response properties of these photoreceptors to what we've previously imaged to be uh the photoreceptors in the eye, it's not all that easy to tell them apart. So these are really just cones that sit in the pineal of the zebrafish. Um what about rods? Same story. So here's a rod line retina. Here's the same rod line in the pineal. There's clear expression um and you can image them. The signal-to-noise is and that's actually an interesting feature because what that tells you is that these rods are ridiculously light sensitive. Um, so basically the second you switch on the two-photon laser, they just go voom to zero, and you get nothing because they're off cells, because they respond to the laser. So, you need to take the laser to really, really low levels, which gives terrible signal to noise, and then go really, really slow. Uh, notice the time scale. Um, so that you minimize the exposure of laser to the cells. And if you do that, you can get them to escape saturation and image these uh, rod responses in the pineal. So, I think it's quite clear that both molecularly speaking and functionally speaking, um, rods and cones that behave like rods and cones do sit in the median eye of the vertebrate even today. Um, there's other things in the pineal. So, transcriptomically, you can also find what looks like ganglion cells. It's a very small cluster. It has this ganglion cell marker and plus a a bunch of other things. Um, and you can find these guys. These are parietopsin photoreceptors. They're near to the cones, but they're not cones. They're transcriptomically distinct, and they express a different opsin. They express parietopsin. Parietopsin is an opsin that you do find outside of vertebrates. You find them all over the bilaterian tree. It's one of these ancient ones. It's the green lineage that I had previously. They exist in the pineal. Um, so, what can we say about those? Well, it turns out that if you can image rods and cones in the pineal, you can also image all the other cells. So, if you take one of these pan-neuronal lines, um, which in the retina, uh, look like this, um, yes, sure enough, you get lots of cell bodies in the pineal. And then what Tessa did is she used in situs to identify the ganglion cells, in this case by staining melanopsin. Melanopsin is a raptomeric opsin, so the opsin that originally sits in the ganglion cell lineage. Um, and sure enough, you find them. You get two, uh, two per side, one, two, one, and I I don't know where the last one is, but you tend to get two per side in the pineal of the baby fish. And then she specifically records using GCaMP from these cells. Uh, before that, actually, obviously, in situ comes later. Um, and these are the response properties of these cells. Now, they're off for UV and on for green, blue, red. Now, you might be able to explain this response here with melanopsin because we know that's roughly where it's sensitive, and melanopsin is an on-acting opsin when coupled through its normal phototransduction cascade. So, this is sort of fine, but you can't explain the UV off based on melanopsin because melanopsin doesn't do that. However, parietopsin does because parietopsin is an ancient bistable C opsin. Uh it's off for UV and it's on for green. So, basically, we know we're recording from ganglion cell-like neurons because we've identified them, but the response properties that they have are those of parietopsin, which isn't expressed in the ganglion cells, but it is expressed in these guys. So, what that implies is that this circuit exists in the pineal, ciliary ancient ciliary feeding into ancient rhabdomeric projecting to the brain. There's also other stuff lastly in the in the pineal. There's some more neurons. So, there's a big cluster of neurons. No one knows what they are. There's also this cluster of neurons. Who knows what they are? And possibly some immature ones. But, if you look at the molecular signatures of these, you do find the odd bipolar cell-like marker. So, maybe there's some sort of relation there. Um, and of course, if we can record those handful of ganglion cells that are in the pineal, you can also record from all the other cells, and these are all the other cells you can catch, and they're basically all off cells, just off. Sometimes very broad spectrally, sometimes slightly more narrow, but never super narrow. And the dominant energy in these cells is in the red-green range. And the red-green range in this case is the opsin spectral sensitivity that you would expect from the from the ancestral rods and cones. And what that really implies is, again, we're recording from these kind of cells, but we're getting which don't have opsin as far as we can tell from the um uh from the transcriptome, and they're responding to light, and they're responding to light in a way that's consistent with responses from rods and cones. So, it kind of implies that this circuit somehow exists as well in the pineal. Okay. So, it looks like we've got these two circuits. We've got the ancient part, ancient opsin, ancient vertebrate opsin in the ciliary, and then we've got the modern part. We know these two are modern ciliary opsins, and these look transcriptomically speaking more modern than ancient. So, two sides, possibly not talking to each other. What can we do about this? Well, what we can do is we can look at lamprey, and if you look in the old literature on anatomical molecular characterization of the lamprey retina, you do find a lot of interesting hits. First of all, the lamprey in general has a very nicely organized pineal gland, much more nicely organized than the zebrafish. Um and it kind of looks like a big median composite eye that's sort of the result of multiple ancient midline systemics being smashed into each other into a common eye cup, but without completely interconnecting all the neurons. So, you've got on the one side, you've got all the ancient stuff, and on the other side, you've got all the modern stuff. And you can actually anatomically speaking find a circuit that looks like this. The paratopsin system feeding into ganglion cell like system projecting to the brain, and on the other side, you can find rods and cone-like cells using ribbon synapses to connect onto these weird neurons which have lentil clubs, and look they look odd, but they're not photoreceptors, and they project neurons that go to the brain. Now, how do you build a retina from that? With nothing it easier than this. So, if this here is the original outer retina, and this here is the original inner retina, then you sort of just plug and play the circuits like indicated. So, which means that you end up building rods and cones out here. You're getting bipolar cells in two flavors. This is a bipolar cell, but so is this. And then you get your ganglion cells feeding out. Now, what that really strongly implies is that bipolar cells, if this is right, have two origins, not one. And that's been one confusing thing. But actually, if you look transcriptomically at the molecular hierarchies that we see in the bipolar cells across vertebrates, it's completely obvious that this is by far the most obvious solution. So, for example, here in the lamprey, the rod bipolar cell tree and the cone bipolar cell tree are really quite molecularly distant. It's but it's certainly the most distant one. The same story in zebrafish, same story in mouse. Um and it really looks like you've got basically this rod bipolar cell tree, which is on, it has non-cascade driven by MGLUR6, which is incidentally also seven-transmembrane like the parietopsin, and it acts through GO like the parietopsin. So, it sort of feels like this is the ancient lineage here. Uh and then you've got all the the cone bipolar cells. And actually in lamprey, most of the cone bipolar cells are off and transcriptomically a single one is on, and that one expresses MGLUR6 and GO. So, it kind of looks like we've got a co-option of the on system onto the off system that gives rise to the on bipolar cell. Um same thing in mouse, just the on system is more elaborate, and actually the most convincing example of this is zebrafish, where you've got this massive schlot of nerve tree, and you've got multiple incidences of onness emerging on the off tree, as if the sort of co-option has happened more than once. Now, um okay, so this is what this might look slightly more anatomically correct. And then you can sort of imagine that you take this proto-retina, and whether or not amacrines and horizontal cells were already there, who knows, this is a question for the future, but basically you boil this for a bit, not very long at all, and you can sort of imagine how it would turn into this ancestral-like retina. So, this is sort of a schematized retina of what the retina of a lamprey is uh we know what the retina of a lamprey looks like. So, it has all the right bits, it has the rods and cones and stuff. Um it's got some oddities, so for example, the axons tend to run on the wrong side. Um it has these bipolar-cell-like neurons, some of which project what we would call correctly from what we know in mouse, but some of them actually go to the brain straight away, a little bit like these guys probably would have done originally. So, it kind of it feels like it makes sense. Um and then you can boil it for a tad longer and before you know it uh we've reached today. And we've got a mouse. Um and the mouse retina is is is a bit of a engineering [clears throat] marvel, right? It does all these really interesting computations and it's it's extremely well understood. And I mean the reason it's well understood, of course, is because so many people have put so much time into trying to understand this this retina. And I would I would argue that almost the mouse retina is probably the best understood complex part of the vertebrate brain. Um which is quite the achievement, I think. Um um but but but then what really have we learned? Let's say we understand the mouse retina, and that's not true. We don't understand the mouse retina. It's just we understand the mouse retina better than most other bits of brains. Well, maybe not the flies, but I mean not invertebrates at least. Um but then but what does it really mean to be a mouse? Right? What's the context? How's the mouse related to the human retina or the fish retina or any other retina? And I think for for these kind of questions we're really lacking um we just don't know. Um and this this is where I just want to take you on a sort of almost philosophical little detour. So, the mouse really is just one solution to vision in a tree of the vertebrates which looks like this. There's 67,000 of them. Almost all of the vertebrates have some sort of retina. Um these retinas are all a little bit different from each other. So, what's what's the context? Um but imagine you could measure mouse retina, what it means to be a mouse retina architecturally speaking, functionally speaking, molecularly speaking, and sort of boil it down to to a thing. And then you do the same thing for a bunch of other species and then you have all these data and you throw it into some sort of algorithm that sort of maps this data into a space, the space of the evolutionary possibility space of retinas. Right? What if if we could do that, then immediately it produces a bunch of things. One, it positions deep model species like the mouse in the broader evolutionary context. Right? Is the mouse representative? Is it an outlier? What is it representative for what not? Right? Um but it's actually better than this because it also gives us a a a hypothesis-driven framework for testing how evolution might even work in the first place. Um So, here's my logic. So, imagine you measure a a bit of the retina in a certain way. Um and then you go to a slightly different part of a mouse retina or to a different mouse. You and you measure the same thing again. You're not going to get exactly the same data. You're going to get slightly different data. And if you keep doing that, you're going to populate a a a region in this evolutionary possibility space which makes up the mouse. So, these are all the bits of possibilities that a mouse can achieve. So, it can achieve something like this, but it can't achieve something like this. But if this is true, then other species will have this, too. Other species will also occupy some region in this possibility space. Um and that means, fundamentally, and I think this this has to be true, but it also fundamentally means that within retina variation over space, over development with developmental time, over circadian days, any sort of variation you get within a species must exist in a common space that also exists across species. Right? So, within and across species variation must exist on the same hyper plane in this weird data space. Um and if that's true, we should be able to use this kind of space, the understanding of this space, in order to try to understand how evolution works. So, you can, for example, um imagine that there is this thing like I like to call it the latent possibility state. So, latent possibility state means it's a state of a retina that is easily achievable, but not normally used. Right? So, imagine you've got um let's say you disinhibit the retina, right? So, you can imagine how you could take out a couple of amacrine cells, there's less inhibition of the retina, certain circuits pop out. Um that retina will functionally change, but it might not break altogether, right? Um so, this is would be a something that evolution can use in order to sort of shuffle the shuffle the possibility space of one species along within this overall possibility space. And you just can keep doing that a few times, and you sort of can imagine [clears throat] how you sort of move species around within this possibility space, right? So, um I just want to argue that this is maybe a useful way of thinking about it. I'm just going to illustrate that now with a little bit of preliminary data. Now, so if we were able to do this, well, fundamentally means we need some sort of data that's relatable across uh retinas. And of course, we can't get the sort of depth that we have today in a mouse and sort of just get the same thing for all the other species. That's not realistic. But what we can do is we can choose aspects that are fundamental to how a retina works, which are really easy to measure. Therefore, example, the by far easiest thing to measure in a retina is the architectural the fundamental blueprint. How many cells are there, and where are they in the layer? So, for example, this is a histological section in a mouse, and it's completely obvious that all of that stuff up here is rods, right? So, I can quantify how many rods per unit area the mouse retina has. And I can do it for the internal retina, and I can do it for the projection neurons down here, and I can see how thick this layer is, and I can see how big the outer segments are. So, this kind of really basic level information is extremely accessible in pretty much any vertebrate the second you've got a bit of tissue, right? So, and just to illustrate this, so here's sort of the same thing for a hook, and it looks completely different in one way because the densities are all over the place and everything. But on the other hand, I can still go and like, yeah, here, these are the rods, right? Easy peasy. These are mostly bipolar cells. These are my projection neurons. Yes, there's going to be some amacrine cells thrown in, but that really is a very small level of variation within the grand scheme of things because we've really got lots of different density differences here. So, we can sort of quantify this and yes, sort of quantification of what it means to be a mouse versus a hawk in the sort of density space. And for example, you can see that mouse has lots of rods. We can see that by eye. Um the hawk has a lot of interneurons. We can see that by eye. Right? So, we can do that for the hawk and the mouse and a bunch of other species. We can cluster them and we get sort of these hierarchical trees. And fair enough, the mouse sits here, the hawk sits there. That's nice. Um we can do the mouse more than once and yes, we're going to get slight jiggle in where the mouse sits. This is different parts of the mouse retina, but also different mice. Um so, there's some jiggle. And then we can look at all the other species that we've measured here and this is about 100 species at this point. And you can see there's very obvious patterns, right? So, there's sort of this pattern here, then there's the rod pattern, then there's some that have no particular density hotspots and so on. And then we can take this data and map it into this possibility space. And of course, this is much more reduced than what I sort of showed you in my concept, but it is a measured possibility space of retinal architecture. And then we can ask, where do my animals sit? Or we can first ask, what is even the thing that defines the space? So, if you do the PCA, you can ask the PCA, what are the features that you used? Right? And for example, one thing that pops out quite clearly is that there's a rod-cone axis that really obviously pops out. Um and there's also a cellular density axis. For example, if you plot inner nuclear layer cells, you get lots over here and not so many over here. So, these are two of the main axes that it uses, but they're not perfect correlations with the PCA because of course, there's more features in there. Now, so where are my animals? Well, here are some animals. The hawk is a massive outlier up here, right? It's pretty sort of in its own league. Um opposite to the hawk in the space is something like a lungfish, which has this extremely low density, fairly homogeneous retina in terms of the density of the different layers. The mouse sits on the edge of the space. You can see that the different data points of the mouse have a bit of jiggle room, right? So, this might be sort of starting to approximate the possibility of the possibility space of the mouse specifically here. So, it can sort of jiggle around a little bit, but the mouse is never going to reach the hawk. Here's the same sort of experiment done for a bunch of other usual suspect model species, right? So, for example, here's Xenopus, here's the chicken, here's a bit of catshark. Zebrafish Zebrafish sort of spreads quite nicely, as it turns out. Um and here's a human. And you can see the human periphery sits somehow here, but the human fovea, which is obviously architecturally very different, sits over here, different part of possibility space, right? So, with all this in mind, we can then ask, so instead of looking at single species, let's just look at whole taxons. Let's look at the big picture. What what is in this space? So, for example, down here, um I'm just highlighting the very early [clears throat] diverging things. So, jawless fish, your lampreys, then cartilaginous fish, this is your sharks, rays, and skates, and then your lobe-finned fish. So, this is lungfish and coelacanths. So, these are a very rare today uh version of fish, but they are the most direct ancestors to the tetrapods. Um and you can see how they spread through the space. So, the jawless fish start here nicely in the middle, incidentally, right smack in the middle you've got the human. Um and then your cartilaginous fish, they sort of take a certain corner of this possibility space. And yes, they they use it quite broadly, but they for example, you're never going to get a shark that sits here or that sits here. And then your lobe-finned fish, well, there's only a handful left, so you can't get a lot of data points. They sit in their own little corner here, right? And whether or not this is ancestral or derived is going to be impossible to work out, I think. Um but it's an interesting notion that sort of forces you into this corner. And then what happens? Well, we've got our tetrapods. Tetrapods are things with four legs. Um or two legs and two wings eventually. So, for example, you've got our amphibians, and you've got the reptiles, and you've got the birds. And I think it's it's very obvious that they seem to form some sort of arc in this possibility space that sort of feels very intuitive. So, the amphibians are a closer to your lungfish, and then your reptiles are a little bit closer there, and then the birds are sort of have started to explore these crazy parts of possibility space. Um I haven't shown you the mammals yet because the mammals sit here. So, the mammals have taken a bit of a detour compared to all the other tetrapods in this space. Um but they're not without overlap. So, you can see there's a little bit of overlap with interesting things. And then if we're just going to highlight this. So, there's there's a few cute notions. So, for example, the human sits sort of fairly nicely within mammal territory, but the fovea is a complete outlier and it really sits in reptiles territory. Um then there's some additional outliers. So, if you just look at who these guys are. This guy here is a ground squirrel. It's a famously It's It's a completely diurnal mammal. It's It's It's a mammal that's useless at night and very good during the day. And it's a completely cone-dominated animal. Um this guy here is platypus, by the way. Um and this one I think is macaque. Um and then here on the other side, uh you've got three species of nocturnal snake, uh which are nicely hugging the mice. So, it kind of really feels you've got a day night axis going on here. And then of course, this is intuitive because we know that mammals have passed through this nocturnal bottleneck um during the age of the dinosaurs. So, there's a there's a very long nocturnal history, and of course, that's going to have shaped how mammalian retinas look. But it looks like you can break out of this space, right? Because you can turn a mammal a little bit reptile-like, or you can take a reptile and make it a little bit mammal-like um if you give it the right kind of environment. Um cartilaginous fish and mammals also have a decent amount of overlap. And I I I like to show this because it just shows how the human doesn't just sit in the middle of the jawless fish, it also sits on the boundary of your mammals and your sharks, which I think is just cute. Um and then you've got this guy. Um these are your ray-finned fish. So, these are your fish fish. All the fish that I haven't yet mentioned. Basically, the vast majority [clears throat] of fish are ray-finned. And what's really interesting here is that they really have massive adaptive radiation across the the They're basically You can find a fish for pretty much any other non-fish out there, right? So, for example, here you've got yourself a frog-like fish. This guy is shark-like. This guy is seal-like. And this guy here is bird-like. It's a mudskipper, but it's an amphibious fish. It spends a lot of time outside of the water. And just to sort of hammer this home, so the mudskipper is sort of best friends with a hummingbird and the crow. So, this is really quite the extreme adaptation, but in a fish. Okay? Um now, all of this might be the a sort of a bit entertaining to think about where the different animals sit, but really without sort of linking this to something a little bit deeper than how many cells an animal has, um we really need to start thinking about how can we do this at the level of function. And here I think I would argue that retina really is an absolute sweet spot because retina is cells cells cells cells cells cable brain. Okay? And that means that cable, which is optic nerve, is the absolute intuitive a bottleneck that if we can measure the signal there, we really describe the retinal output in its completeness. Um and we can do that across species. And the obvious way to do that is multi-electrode array recordings. Now, people have established this and and done lots of data sets for all kinds of mammals and a handful of non-mammals, but it really there there there there exists a massive a bottleneck in in the non-mammalian space. And so, this has annoyed us a little bit. So, a few years ago and then um Marvin uh heroically responded to a PhD advert to say, "Should we should we see if we can record from a bird?" Uh and it took a while to record from this bird, but uh he he did manage to do it. And so, he got the first uh nice light responses that I know of um ever measured from the retina of a bird directly from the ganglion cells using multi-electrode arrays. Here's just an illustration of what that looks like if you computationally process those signals. These are just spikes flying across the array. Um triggered by light. And this is from the chicken. But then sort of as soon as he sort of unlocked the chicken possibilities, we started to think, "Well, maybe we can sort of scale this." And so, by now we've established Xenopus, zebrafish, and catshark, plus as it happens a handful of other species as well. And so, we can now really start doing these kind of experiments across the full vertebrate phylogeny, at least for some example species that sort of sit within different branches of this tree. Um and just to show you how what this looks like. So, I call this the comparative retinal physiology. And just to show what that looks like. So, here is sort of the simplest possible stimulus that you can think of, I think, which is a step of light. We switch on the light and then we switch off the light. It's the same amount of light, but the same wavelength composition. So, we're literally showing these animals the same thing and we're asking, "Given that I've shown you a flash of light, what is the population response of the ganglion cells?" And so, if you do that for the mouse, uh it looks a bit like this. You're getting your on response, there's an off response. This is very decorrelated, so different cells do different things. There's a lot of sustained stuff going on. Um in the shark, you kind of have a similar situation. You get your nice uh on responses and off responses, very nicely segregated. Of course, it looks different from the mammal, but it's sort of reminiscent conceptually. And then you've got all these guys in the middle. And the guys in the middle, they really love a bit of on-off. So, there's very few on cells or off cells in these animals. The most cells are on-off. They do both. Um and there's also massive variation in terms of the kinetics, as you can sort of see across these species, but also in terms of just how many spikes they invest per ganglion cell per unit time to convey a message, right? So, for example, if you compare yourself to a chicken and a shark, you've got more than a log unit difference in spike rates. So, that that that must tell us something. Um so, what we can do is we can now take every single ganglion cell here and like before we've done for retinal architecture, we can again map that into a possibility space, but this is not a computational possibility space. And in this case, it looks like this. Each dot now is a ganglion cell and of course, we haven't told the PCA which dot is which species, right? So, now we can ask, "Where do the different species sit?" But before we do that, let's just see what are the sort of general properties that PCA has picked up on. And for example, uh unsurprisingly, of course, in this case, it picks up one of the key signatures of sort of on-off mix. So, these are sort of the pure on cells in various kinetic flavors. These are the pure off cells in various kinetic flavors. And here's all the sort of the intermediate stuff, right? So, they sort of nicely divide the space up. And then we can see where the different animals sit. So, for example, now, if we sort of project these guys in the middle here, yes, they're all on-offy, but it's not like they're the same kind of code, right? The chicken sits here, the fish sits here, the frog sits here. There's a little bit of overlap, but I would argue it's surprisingly little overlap. So, I would almost call this three different computational solutions to the problem of encoding the stimulus. All right? Um what about sharks? Well, the sharks are sort of extreme in their They're very good at on, and they're very good at off, and they really don't like to mix it up, right? So, they're extremely clean. Um mammals look like this in this case, this data set anyway, right? So, you get your on and off cells. There's a bit of non-bias in this data set, that's just how it is. Um and of course, there's a bunch of mixed cells. And you know what what really sort of strikes if you compare your mammals versus your with your sharks, they sort of occupy similar bits of space, but with very different sort of strategy. The shark is ultra clean, and the the mammal is ultra decorrelated. Different ganglion cells just do different things. And what that tells you is that the mouse invests computational finesse in telling apart all kinds of interesting aspects of this really basic stimulus, whereas the shark just goes, "Well, this was on, this was off." Right? It doesn't really invest a lot of computational finesse in decoding that. Does that mean the shark is useless in terms of more sophisticated things? Maybe, but maybe it just says that the shark isn't investing in that stimulus because it's investing in other stimuli that we haven't tested. Um Now, human is a little bit mousy, but with different biases, but you can hopefully appreciate that they're sort of fairly similar. Um and if you put your mouse and your cat shark and your human on top of each other, you can see that they do share the space, right? So, this sort of is a little bit reminiscent what I showed you from the architectural space, that architecturally there's sort of a very nice similarity, but also functionally. So, that implies then, if this sort of scales to across species, is that there's a predictability that goes between function and architecture, right? Which sort of goes in the idea that that that there's maybe some sort of evolutionary trajectories that we can start to think about when we try to link these spaces. Now, um however, um there's some bigger patterns here that I think are important. So, for example, if you sort of just consider where these different species sit and what species they are, phylogeny really doesn't explain the space, right? Because for example, the distance between a shark and a human is pretty vast. Um and you've got your chicken and your your fish your zebrafish over here. So, this phylogeny doesn't explain the space. But, what could explain the space is just the rod-cone dominance. Because these are extremely diurnal cone-dominated animals and these are mostly night rod-dominated animals, right? So, there's a sort of night-day axis going on again. Um and what that implies, if this is what it is, is that there is some sort of reason that if you're cone-dominated, you choose this mixed space, whereas if you're extremely decorrelated nocturnal, and the shark is sort of an extreme version of that, you go for extreme decorrelated. And maybe that makes sense, right? If imagine if you sort of sit in the night, you're really fighting for every photon. Signal-to-noise is limiting. Um and so, what you need to do is you need to make sure that whatever you're telling the brain um isn't so noisy that it's useless, right? So, maybe um keeping things clean is a is a strategy to keep things uh to to to fight against noise. But then, if you're getting on the other end, if you have loads of light, light is no longer limiting, signal-to-noise is sitting in a completely different regime, then you can maybe start to exploit the signal to start to multiplex. Basically, put multiple signals into your message at the same time because the signal-to-noise is so good that you can work with that. Um so, the idea um therefore is that the chicken sort of just packs more information into its code than the shark for something like this. And we sort of tested this in the first paper um in a very sort of conceptual way. Um we just played on and off steps, but we played them at different intensities and in different wavelengths. And what is I think quite striking is that if you You this for off step, if you look at the off and remember most cells are on off. And if you look at the off responses, if you change wavelength or if you change intensity, we just get scaled version of the same response, right? So basically, it's really difficult using the off step alone to tell apart wavelength from intensity. So this is not a color system. It's just an intensity system basically. Whereas if you do it for the on steps, hopefully you can sort of see that the the the black and white lines, which are sort of just intensity in white, they give you similar shape and kinetically, but actually the different wavelengths, they give you very different shapes. So it looks like sort of in in kinetic space, wavelength gets mapped onto the timing, whereas intensity gets mapped onto just the number of spikes. And you can sort of show this more systematically. You do PCA through the space. And if you look at how the colored dots versus the black and white dots aligned in the on steps, they're almost orthogonal. So it sort of looks like they're simultaneously encoded here. Whereas in the off step, that's not the case, right? So there's sort of one course interpretation from this, and this is of course superficial compared to what we've done in this paper. Sort of argues that it can now multiplexing intensity and color information, basically popping the intensity information into its off system, popping the color information into its on system, and then sort of using single axon to convey more information per per axon. Now, um if that is so, that really then begs the question. So if you've got yourself something like a nocturnal-ish animal in the mammals, like the mouse for example, um what would happen if you took a mouse and you made it more diurnal? And of course that's happened. I talked about the ground squirrel earlier, but there's others. There's even closer versions of mice that are more nocturnal. For example, there's this thing called the grass mouse. And as it happens, I met Alan recently published a paper where they recorded from grass mouse. And look what that guy looks like. It looks like a zebrafish, right? It doesn't look like a mouse at all. Um and if you look at where the grass mouse sits compared to the mouse mouse, it really it pushes deeply into this mixed space and this really is the zebrafish territory but it's not completely in zebrafish territory. It keeps its tail in mouse territory and pushes into zebrafish territory. So it kind of implies that the grass mouse over this very short evolutionary distance that separates the mouse mouse and the grass mouse this has happened. So it can't be a major revamping of the retina because that just takes time, right? So something fairly straightforward must have happened in that circuit in order to unlock this. And more more sort of generally speaking it implies that you should be able to take a cell in any given animal and then over relatively short evolutionary distances give the cell a different functional identity a predictable different functional identity. So maybe this then comes back to the idea that you sort of have within species variation and across species variation existing in the same plane. So if you imagine that this ganglion cell during the course of the day in light adaptation all kinds of other interesting things that happen can sort of move around a little bit in this possibility space then maybe as the mouse grows or whatever you can push it a bit further and then over evolutionary time you can sort of keep pushing it a little bit further still. Now there might also be bits that are just inaccessible. Maybe you just lack components. You no matter what you do you can't reach there unless those components convergently appear. So so how can we sort of test this and this sort of the last bit I want to talk about I just want to quickly check how I'm doing for time. Um poorly poorly. Okay, my apologies. I'll try to be quick. So on off and mixed here. How does this happen in the retina? Um well we've got cancer off we know this. Bipolar cells in the mouse come in on and off flavors. So you've got the sign inversion here. So you start here you can move it into the on space and then on off is thought to mainly arise when you take your on and your off channels and you combine it one level down in the ganglion cells. Yeah? And then after that everything is just inherited in the brain more or less. So the brain basically inherits the on off and on off polarity of the retina. Um but a evolutionary speaking, that's probably not the original, right? Because first of all, if this happened, then the original ganglion cell probably was on-off already. Um and also more so, if the originally on cone bipolar cell um comes from an off-on co-option, then it would have also been on-off to begin with. So, it kind of feels like on-off is the original and separated on-off um is is derived. Um so, how's this built? Uh well, we've got different glutamate receptors. We've got these guys and we've got these guys and this is now very briefly talking about the preprint um that I mentioned. Um so, depending which ones you pop into your dendrites of a bipolar cell, you're an on or an off cell. That's kind of the idea. But actually, if you look transcriptomically, um and this is mouse now, these are off cells, these are on cells, it's not super clean. So, yes, there's more off stuff in the on and there's more on stuff in the ons, but it's not exclusive, right? So, for example, here this is um on stuff in off cells and this is off stuff in on cells. So, they should be dirty. Um so, again, what we can do is we can use comparative transcriptomics uh to see um to basically take bipolar cells across species and map them out and here's where the mouse sits. So, these are the mouse bipolar cells and they sort of sit nicely in the middle. Zebrafish bipolar cells sit completely elsewhere. They use a different molecular regime for their um on-off system. Um we can look what's inside the space. You've got on versus off separation, but there's a hell of a lot of overlap where for example, this off cell and this is an on cell very similar to each other. Um and you can look what correlates with that and what happens is that you've got um in this off space, the kainate receptor, which is one of the off receptors, really nicely maps onto that. And then mGluR6, which is one of the on receptors, really nicely maps onto the on space. But the other two don't. AMPA receptors actually map onto an orthogonal receptors. These are meant to be off, but they're not off. They're orthogonal to the off and on axis. And EAT, which is the other on thing, is not on. It's orthogonal to on in the other direction. And what's that direction? Of course, it's again rods-cones, right? So, it kind of feels like this is the original on-off version and this is actually a rod cone choosing kind of regime that these guys use. Um so, just to test this in zebrafish, um so, first of all, zebrafish have a lot of cross expression. So, bipolar cells should be on-off. Yet, when you stick them in the photo microscope and you record and you flash the lights, they're not. They're on or they're off. They're not on-off. So, what's up with that? Well, turns out Well, you can quantify this. This is off and this is on. Um but, turns out that if you actually block inhibition, just block the amacrine cells, you get basically the same responses, but now really clear on-off responses as for example highlighted here. Um you can quantify but they sit in certain specific layers of the retina. So, on-off is actually intrinsic mode of quite a lot of bipolar cells. It's just you don't see it because inhibition suppresses it. So, this is what I would call the latent possibility space of this bipolar cell. It's not normally used, but it's really easy to get there. You just take out some amacrine cells and the cells become on-off. Um we can also do this by taking out the individual receptors. For example, if you do an EAAT mutants, the on thing in zebrafish, um yes, you lose the on, but you don't just lose the on, you gain offs, implying that you've sort of unlocked a latent ability to be off. The reverse also works. If you take out kainate, you lose the off, but you gain on in the places where you've lost the offs. Um and if you look where in the IPL it sits, where you get the latent and on-offs and where you get the on and the offs are the same layers. So, it's it's just types of bipolar cells that sit in these layers that do this. Um in the bipolar cells, it's about 30%, but then we asked, what happens downstream? Turns out if you do the same experiment for ganglion cells, it's 75% and if you do the same experiment, now disinhibiting the brain, not the eye, um you get 90 plus percent of brain neurons that are capable of representing on-off information. They just don't do it under normal conditions. Now, of course, this is a pathological fish. This is not a happy fish, right? But, it just shows that these bipolar that these central circuits, the vast majority of these cells have access to both on and off information. Now, um why would you do this? Well, one idea is if it is ambition, then you should be able to on the fly regulate the amount of inhibition and sort of shift your on versus your on off and your on off regime uh on the fly. Um and one reason that might happen, for example, is with growth. Here is a baby fish, small eye. Here is an adult fish, big eye. Um so, this guy is kind of more diurnal than this guy because it just gets more light. So, we should get more on off here than here. And we already know that's true. I showed you that the zebrafish dominated in adult in on and off, which larva isn't. You need to disinhibit the larva to get there. Um but actually, what you can then do is you can take an adult zebrafish, give it light, get these on off responses, and then give it less light. And what you can see is that you start to lose the on off responses and they just become monopolar. This is just one example done here by Laura, but you can actually look at this in this in this computation space. So, this here is the uh the ganglion cells uh in the light, and this is the same ganglion cells with less light. And hopefully, you can sort of see how sort of stuff starts to project the the same bump ganglion cells systematically move into the monopolar spaces increasingly with lower light. Um so, this is dynamic. Um what about the mouse? So, we teamed up um with Thomas Euler's lab, and Dominik von Schorlemer here did this experiment. Here is ganglion cells of the mouse um in control condition, same thing with inhibition block. And sure enough, you're unlocking on off ganglion cells, quite a few of them in the mouse. Fewer than in zebrafish, makes sense. The mouse starts in a different part of possibility space. Um so, you don't need to push it all the way to uh 75% that the zebrafish achieves, but the same concept applies. And then lastly, the shark. So, this the shark is sort of this extreme uh in this world. It really only does on and off under control condition. But, if in shark you block the on pathway, first of all, you lose all the ons, but you don't just lose them, they become off cells. So, these on cells can do off stuff. They just don't do it under control condition. And if you disinhibit the shark, well, sure enough, they end up in the middle. Okay? So, it's even for shark, it's quite easy to turn the shark into an on off coding regime. Um and where does the shark land? Well, the shark actually lands in frog land in this case. So, you could sort of argue that with this disinhibition trick, you can take a cat shark and turn it a little bit more into a frog. You can't turn it into a chicken. It's just a different regime. It's too far away. But now, of course, you might ask, "Well, maybe if you take a frog and you start to disinhibit it, does that become like a chicken?" So, this is sort of the start of of of a journey that we're hoping to get onto. And this is where I'm going to close. I just want to briefly acknowledge again the people that have whose work I've talked about here over the arrows and the funding and of course, all of you for your attention. And I do apologize. I believe I'm a little bit over. Yes. So, sorry about that. >> Great. So, thank you so much for this great talk, Tom.