 Okay. Okay. Now we are live. Hello and welcome everybody to a new seminar of Sussex Visions. I am Antonio Inajosa, a postdoc in Leon Leñado Lab, and I'm working in mouse cortex. These series of seminars are part of the World Wide Neuro, which is an initiative towards a greener and more accessible kind of talk. So I first I would like to thank the organizers of this initiative, Tim Bogos and Panos Bocellos for for starting it. And just for you to know the format of the talk, it will be around 45, 50 minutes of the talk. Then we will have some questions that you can type in the in the chat. And I will ask just at the end of the talk, the questions. And, and also after we finish the talk, we will have a discussion over Zoom. And so feel free to to join to that to that discussion. Also, I would like to to remind you that if you want to to see our our Sussex Vision Talks and don't miss any of them, you can join the channel in YouTube and and and that and this way you will get all of it. So today we have the pleasure of having with us Joshua Sainz. Hello, Jos. And and he he graduated from Yale in biochemistry and psychology. Then he joined the neurobiology lab of John Hildebrand, where he earned his PSD studying the development of moth neurons. And then he took a year off from experimental science to advise members of Congress or in the USA on scientific and technical issues. But then he went back to experimental science. And he began his doctoral research at Harvard with neurobiologists who are Jackson MacMahon, and later on with neuroscientists at University of California, San Francisco. During this period, he started his research on the mechanisms and molecules that drive the formation of neural connections within the neural junction. Sainz accepted a faculty position in the Department of Physiology at the University of the Washington University School of Medicine in San Luis. And then he continued his work identifying and characterizing molecular factors involved in neural muscular junction formation. It was also then when he contributed to generating a technique for tagging neurons with color coded fluorescent proteins combinations. So this is the famous rainbow mouse, which has been widely used around the world. Then in 2004, Sainz returned to Harvard University, and where he became the founding director of the Center for Brain Sciences. There, he researched transition to the analysis of genetic specificity in the retina. And he's currently a professor of molecular and cell biology at Harvard University. To finish, Sainz just has received many professional honors for his career, including he's a member of the National Academy of Sciences and also the American Academy of Art and Sciences. He's also been part of numerous advisory panels like the Wellcome Trust or the Hogwarts Huge Medical Institute. And he's also received numerous prizes like the Gruber Neuroscience Prize in 2017, or recently the Edward M. S. Colnick Prize in Neuroscience. So it's a pleasure to have you with us today, Josh, and we're looking forward to hear your talk. Okay, thank you very much. Let's see are we, you tell me whether you're seeing what you should see? Yes, I can see your presentation. Okay. All right. Well, it's a pleasure to almost be with you. I wish I were there in person, but there you have it. So I'm going to talk today about recent work from my lab that relies heavily on single cell high throughput transcriptomics. And because that makes people's eyes glaze over frequently, let me just start by telling you how we got into this. And it started with the interest that Antonio mentioned in synaptic specificity, how it is that axons entering an area of the developing nervous system choose among many potential target neurons, the ones on which to make synapses. In some cases, they even choose a particular part of a neuronal distal dendrite, a cell body, an axon, hillock, and so forth. And so close to 20 years ago, actually, we decided to investigate this issue in the retina. This is obviously not an audience that needs me to tell you what the retina is or why it's so wonderful. But very briefly, it is a part of the brain, embryologically and structurally. It's about as complicated as other parts of the brain. And I'll try to give you some reason to believe me on that in a few minutes. But it has a lot of advantages that make it easy to ask detailed questions about development structure and function that one might like to ask in the cerebral cortex, but one can't because of technical difficulties. It's obviously been pulled out of the skull. So it's very accessible, comes with its own lens. It has clear borders. Its neurons and synapses are arranged in beautiful layers, as you can see in this picture from Josh Morgan and Rachel Wong. Importantly, there are no long range inputs. So it is the whole circuit from light at the outset to optic nerve at the end. We were interested in it because there are many types of interneurons that synapse on many types of retinal ganglion cells. The numbers here are the ones that were thought to be correct at the time we started this work. Again, I'll update that soon. But in very brief, you probably know this, photoreceptors make synapses on interneurons in an outer synaptic layer called the outer plexiform layer. The interneurons, they're three classes, horizontal bipolar and amicron cells, and two of them bipolar and amicron cells make synapses on retinal ganglion cells in an inner synaptic or plexiform layer. And then the ganglion cells tell the rest of the brain about what they've learned. And because there are many types of interneurons making synapses on many types of ganglion cells, each type becomes particularly tuned to some visual feature, like motion in one direction or a color contrast or an edge or whatnot. And therefore, what the retina does is to send multiple parallel representations of the world that have been sparsified and tuned to a feature onto the rest of the brain. So for us, the interest was in this inner plexiform layer, where in a really very narrow space, especially when synapses are being made, very specific cohorts of interneurons make synapses on very specific cohorts of retinal ganglion cells. And clearly, it's a short distance phenomenon. So it really is synaptic specificity, not, for example, exon guidance. And so over a period of really mostly about eight years, we use the retina mostly based on mouse genetics to label particular cell types, monitor them, profile them, and then find out what they did in loss or gain of function experiments as measured either morphologically or physiologically. And a group of really wonderful graduate students and postdoctoral fellows whose names I've listed on the right, this is not all of them, but I would say the main players, came up with evidence that a set of adhesion molecules, recognition molecules like the catherines and proto catherines, and transcription factors like TBR1, SATB1, Phezef1 play key roles in setting up this specificity. So this was all rewarding, but it became clear as we went on that if the field was going to get a really satisfying explanation for how this worked, one was going to need another approach, not the retail approach of getting a transgenic line by hook or by crook and finding out something about the cells that are labeled, but rather getting a really comprehensive and unbiased classification of all of the interneurons and all of the retinal ganglion cells and characterizing what recognition molecules and transcription factors they expressed. And this seemed like an impossible dream for a long time, but in 2015 it became possible thanks to the invention of high throughput single cell RNA sequencing methods. It's such a great idea that it was invented three times simultaneously. I put in green the invention by Evan McCosco and Steve McCarroll's lab because we ended up collaborating with Evan in the development of this method or not really the development he developed it, but in using the retina to validate it and perfect it. And that provided the way to profile thousands, tens of thousands, now people do hundreds of thousands of individual cells at high enough speed and low enough cost to make it possible to conceive of having really a comprehensive and unbiased categorization and characterization. So I'm not really sure a single cell RNA seek has kind of become an extremely popular and much used method over the past six years, but I don't really know whether some of the people in this audience might need an explanation. So I'll try to split the difference and give you a very quick one. So essentially what one does, this is drop seek, but all three methods use the same basic sort of platforms. We actually now use commercial one called 10X, probably most of you do single cell transcriptomics do the same. So in any of them you dissociate a tissue into individual cells and then isolate the cells. Nowadays you can isolate the nuclei and then pass them through a fairly simple microfluidic device in which a single cell is encapsulated with a single micro particle in a nanoliter sized oil droplet. And inside that oil droplet the cell lices and messenger RNAs are captured by DNA oligonucleotides that are derivatized to the micro particle. They end in oligo DT and therefore they capture polyadenylated messenger RNA. Now then the whole emulsion is broken and the RNA is turned into DNA amplified and then sequenced. And the secret of the method is that you can now capture, amplify, sequence the messenger RNA transcriptome of literally thousands of cells in a single tube rather than needing thousands of tubes and dozens of graduate students and that's really the key to saving time and money. Now the DNA oligonucleotides on the micro particle are elaborately barcoded and so that means that computationally once you have all the sequence you can divide the multiple sequences into those that came from cell one those that came from cell two and all the way up to cell 5000. So here is a display of the results from Evan's first big experiment on retina in which he profiled almost 45 000 cells and then computationally grouped them into 39 clusters and displayed them on what's called a tisny plot. And so this display has 44 808 dots and their colors and their grouping represent what the computational algorithm thought were cells that were sufficiently similar to each other in terms of their transcriptome that is their catalog of messenger RNAs to be potentially all the same type. It's nominally an unsupervised method but as we can discuss later we twirl the dial so it isn't completely unsupervised but mostly. So at any rate because it was the retina and we had so much ground truth we could look at genes expressed by cells in each of these clusters and ask what they represented. And the good news was that each cluster represented a type or a small group of types that we knew about. Amicron cells, photoreceptors, bipolar, horizontal, scangling cells, those are the five canonical neuronal classes in the retina as well as non-neuronal cells including Euler glia. And in addition their transcriptomic relationships to each other which you can see in the dendrogram on the left made sense. So rods and cones were each other's closest relatives all the bipolar cells were each other's closest relatives and so forth. So that's the good news. The bad news is that there were only 39 clusters and we already knew that there were more than 39 cell types in the retina. And the problem was that even though this seemed to be a ton of cells and it was most of them were rods and therefore there weren't enough cells in the most heterogeneous classes bipolar's amicrons and ganglion cells to have the computational power to divide them into individual types in some cases to even capture the sparse types. And so what we went on and did then was to use methods to enrich ganglion cells, amicron cells, and bipolar cells separately so that we'd have enough of them to take the job through to completion. And over the next few years we did that finding 15 bipolar 46 ganglion cell and believe it or not 63 amicron cell types. And in each case we could look for genes that were selectively expressed in individual clusters putative cell types and then use morphological methods immunohistochemistry or in situ hybridization to see whether the types defined molecularly corresponded to types defined by criteria that neurobiologists would care about which were morphology and physiology. And the good news is that at least for the retina we don't know if this is going to be true throughout the nervous system but at least for the retina. They did we did a comprehensive match physiology to morphology to molecules for bipolar's and spot check the ganglion cells and amicron cells. So we ended up then by 2020 year ago with what we think is a nearly complete cell atlas of the mouse retina. I say nearly because of course there'll be no shock anybody if wanted to more show up but from everything we've seen so far it's pretty complete 130 neuronal cell types and 10 or so probably more than that non-neuronal cell types and that is what makes me really believe that the retina is about as complicated as other parts of the brain. This is a number very similar to what people have found for example in visual cortex, motor cortex and so forth. Interesting the number doubled over the decade between 2010 and 2020 and there's lots of reasons for that but I think it's safe to say that the biggest contributor to that increase is the single cell transcriptomaic method. So as this atlas near completion we of course wanted to use it to ask interesting questions about biology the atlas is an atlas it's not interesting in and of itself and so over the past few years we've been trying to use that atlas as a foundation to learn things about the development of retinal cells their responses to injury retinas and humans and non-human primates cell types in which disease associated genes are expressed and finally evolution of cell types and so what I'm going to do today is focused on the first second and fifth of these development selective vulnerability and evolution and tell you fairly briefly three stories that are preliminary at present unpublished although I'd like to think they're not in the long run going to be unpublishable. So let me start with development. Lots of questions there including the one that set us off in the first place about what recognition molecules and transcription factors the types express but I'm going to confine myself to how the cell types within the retinal ganglion cell class diversify and by way of background let me review some key points that have been discovered about the lineage of cell classes and by classes I mean the ganglion cells bipolar cells amicron cells and so forth and there's sort of three I would say big tenets that have been worked out over the years first is that all of these classes arise from multi-potential progenitors that divide and give rise progressively let's say to a retinal ganglion cell maybe later a amicron cell maybe later a rod and so forth and that was discovered in three labs more or less at the same time Christine Holt and Bill Harris your colleagues Scott Fraser and Connie Setgo and her colleagues. Second big idea is that as the progenitor goes through its cell divisions its propensity to generate different types of cells first retinal ganglion cells last rods and so forth changes because the intrinsic competence of those progenitors to generate a particular cell type changes it's not entirely clear how much that change is driven by intrinsic timekeeping mechanisms or by extrinsic factors but Connie Setgo I'd say preeminently has got a lot of evidence for the idea and then the third big idea is Bill Harris's which is that the competence is not a determinative one but rather a stochastic one that is at any given point the probability that a progenitor will spit off a let's say ganglion cell or amicron cell is a matter of probabilities it's not an all or none phenomenon. Now I'm going through all this because when we get the cell types that is how a retinal ganglion cell diversifies into its 46 types and newborn amicron cell into 63 types and so forth we know nothing about any of these that's largely because when Setgo Harris and all were doing their work the number of types really wasn't appreciated and the markers to distinguish multiple types were not available so we thought we'd have a crack at it and what we did to begin with and the main we here is Karthik Shekhar a computational biologist and Irene Whitney a neurobiologist Karthik now has his own lab at UC Berkeley and so we began by asking about two extreme models one extreme is that there are 46 different types of progenitors each one of which is fated to give one type of retinal ganglion cells and the other is that it all happens post-mythotically with some kind of stepwise determination in which a multi-potential precursor post-mythotically gradually becomes committed to a particular fate and so to distinguish that we repeated our isolation of ganglion cells and transcriptomic characterizations at earlier and earlier times all the way back to when the first ganglion cells were being generated and what we found was that the number of clusters increases and although I'm not going to talk about it so much the clusters became more clustering they became better separated from each other and so at face value with a lot of controls that I'm not going to go into that kind of argues against a strictly predetermined scheme that is there's no reason at all to believe that when ganglion cells are newly post-mythotic there are 46 types of them so then we asked about the stepwise mechanism and again you can imagine two extremes one is that each post-mythotic ganglion cell I call it a precursor is committed to give rise to a small number of types and that plays out the other is that they remain uncommitted early on and so any given precursor can give rise to some of the same types that a different type of precursor gives rise to and so the computational way to have a first crack at this is to organize what's called a confusion matrix in which one takes an early type and asks for its transcriptomic similarity to later more numerous types and so in a specific model on the left a given early type would be transcriptomically similar to a discrete set of later types whereas in a completely non-specific model the transcriptomic relationship would be more or less random because anything can become anything and the data are here and you can see it's sort of somewhere in between but clearly it's not the case that early clusters seem to be faded by this criterion which is admittedly indirect to become specific groups of adult clusters so we feel that we can tentatively rule out a completely specified model but that of course is a completely specified model at the level of these clusters or types and so it could be different from the point of view of individual precursor cells and I'll give you again two extremes one would be that the precursors in an individual cluster are actually each one faded to give rise to a particular type but they're so transcriptomically similar at this early time that we can't tell them apart the alternative is that within a cluster individual ganglion cells truly are multi-potential truly uncommitted so there is a way to distinguish these alternatives and that is to visualize a precursor follow it through to adulthood see what type of cell it becomes and then run back time and repeat the process over and over but that won't work that violates the laws of physics and so Karthik decided to use an indirect computational method called optimal transport which predicts how a cell at a given time distributes its so-called transcriptomic mass to cells at a later time and again I don't have time to explain this and to be honest I'm not sure I could do a good job of it anyways but the key point here is that it is dealing with cells one cell at a time independent of the fact that they're clusters so they're not really looking at the possibility it won't get caught by the possibility that we're mistaking types within a cluster for being homogeneous and the upshot is that indeed by this criterion the cells are very multi-potential each one contributes its transcriptomic mass which we view as a proxy for faith to an average of 10 to 12 adult types at the earliest stages and then that multi-potentiality decreases until in adulthood of course it's one because every cell has become what it's going to become and the umaps which are like tisneys on the right are just to show that this multi-potentiality is shared by many many different clusters or cell types it's not just a property of a few so that then makes us believe that these precursors are truly multi-potential early on each one having different probabilities of turning into different sets of adult types and those types are of course overlapping one precursor to another so of course now the big deal and we haven't done enough here is to ask what are the types that an individual precursor can give rise to and i'll give you just one small result these are these optimal transport results displayed in the form of what's called a sankey plot in which one looks at the weight in which cells at a given early time give to the 46 adult types which are shown at the right with every intermediate point in between and what you can see is that some groups of cells for example those that are characterized by expressing the transcription factor tbr2 in adulthood or expressing neuro d2 in adulthood seem to have some coherence in which groups of early cells they derive from whereas others like those expressing the transcription factor fox p2 don't and so what we think overall and i should point out for those of you into the details tbr2 marks intrinsically photosensitive retinal ganglion cells but not only them some other types that also turn out to be pretty close relatives so what we come up with then is a tentative model of diversification in which the post-mycotic cells are initially multi-potential they're biased towards generating subclasses and they separate it was kind of inherent in the sankey plots but i didn't point it out by an asynchronous process of fake decoupling where some become committed to a particular fate earlier than others and i put these words in blue because it's kind of interesting to me at least that they resemble in many ways the big ideas that harris phraser and sepco came up with for the diversification of cell classes it's just in this case the multi-potentiality stochastic bias and a synchrony are occurring post-mythotically rather than during the process of mitosis so let me turn now to the second story and this is about vulnerability of ganglion cells and it's a it's known very well that in many neurodegenerative diseases the insult is shared by many cell types think of al s arkinson's Alzheimer's but only certain cell types are vulnerable and end up dying and so we thought we might be able to look at this for retinal ganglion cells because in mouse after you crush the optic nerve just outside the orbit 80 percent of the ganglion cells die in two weeks 90 and four weeks and then i'll come back to it in a minute almost none regenerate axons but a few can be induced to do so and so in a collaboration which you're going he a children's hospital we initially ask which 80 die is it completely stochastic or maybe the cells that were active when the nerve were crushed or were inactive or were near a blood vessel or had a little bit more trophic factor or could it be that at the limit 80 percent of cell types are completely vulnerable and they all die whereas 20 percent of cell types are completely resistant and they all survive and so we asked that initially using a small set of transgenic lines and the answer was that there is a big cell type bias alpha retinal ganglion cells and one of the intrinsically photosensitive ganglion cells for example survive extremely well whereas many other types fail tuj1 is the marker in this case for all ganglion cells and you're seeing the the overall loss of 80 to 90 percent so once we had the atlas we thought we could go back and do a more comprehensive look at resilience and vulnerability and also characterize the resilient and vulnerable types the idea being that if we found some genes expressed by resilient types would it be possible to stick them into vulnerable types and make them more resilient or could we knock out genes and vulnerable types and make them more resilient so what we did was look at the frequency of the 46 types in the atlas and then crush the optic nerve wait various times thereafter isolate the ganglion cells and do the same thing over again and then divide one by the other to give us a measurement of resilience and what you can see are a few things first there are really dramatic differences the most resilient types survive to the tune of 80 or 90 percent the most vulnerable types die to the tune of more than 95 percent the other thing we found out is that it's a continuum not a dichotomy of resilient and vulnerable types which was a point that was sort of left unsettled when we only could look at a few types so now we could go into these types and ask whether their genes selectively expressed by resilient or vulnerable ones and we could do that in intact retina or at various times after performing optic nerve crush between a half day and three weeks and the bad news is that we couldn't find any genes that were either expressed by all and only resilient types or all and only susceptible types or whose expression was graded with resilience and vulnerability but clearly there were some interesting possibilities and so we went on and tested a few of them and the we here I should say is Nick Tran who now has his own man at Baylor, Karthik Scheker again, Ann Jacoby and Irene Whitney and so here two that were particularly interesting, urocortin is a peptide that's upregulated in some resilient cells but no susceptible cells and it's secreted intrinsic inhibitor, CRHBP a binding protein is selectively expressed in many susceptible but no resilient cells and that's true before after crush so the idea is urocortin is a peptide in a family with CRH and CRH binding protein is secreted binds urocortin and CRH and prevents both of them from activating the receptor so this seemed like an interesting conjunction and so we went on and either overexpressed urocortin or injected urocortin protein or knocked out CRHBP using a CRISPR based approach and in all those cases we significantly increased the survival of cells at two weeks after nerve crush the control value is shown by the red horizontal line with pink being the error bars and so this is not a huge effect to be sure but it's very encouraging of the approach I'm going to skip this and then go on to what we're doing now and in the new step that Ann Jacoby is leading with Nick and not Wen Zhen Yan we're trying to look harder at how resilience and vulnerability come about by looking at interventions that can save some of the ganglion cells and actually promote external regeneration from some and the three interventions we're using are ones that you're going still our collaborator had discovered many years ago and so they've been validated in multiple studies in his lab and other labs and they are deleting a phosphatase called p10 which is an endogenous inhibitor of mTOR signaling to leading a transcription factor called SOX3 which is an intrinsic inhibitor of jack stat signaling and overexpressing the neurotrophic factors CNTF and we did it in combinations either p10 conditional knockout alone it's being knocked out only in ganglion cells to avoid indirect effects and for p10 lethality p10 knockout with CNTF or all three interventions and you can see here really confirming what you're going and others have shown that they substantially in an integrated fashion increase survival of ganglion cells and the regeneration of axons from a minority of those cells that survive so the idea now is again at various times after performing these interventions an optic nerve crush to have a look at which cell types survive your cell types regenerate and what genes they're expressing and so for survival the answer is a fairly similar simple one it turns out that all of these interventions with with only a few exceptions promote the survival of cells in proportion to the fraction that survived initially so they're still resilient and vulnerable types but whatever the vulnerability was after crush alone superimposed on that as an overall increase in survival for regeneration is a little more complicated because what we needed to do was separate those cells that regenerate in a given retina from the much larger number that survived but didn't regenerate and so to take that on and Jacobi devised a really beautiful and extremely difficult surgical method in which she injects a fluorescent dye micro ruby into the optic nerve far enough away from the nerve crush that it won't be taken up by stumps and close enough to the nerve crush that it can intercept a fair number of the regenerating accents and she can then go back and look in the retina at the cells that are micro ruby positive and you can see them on the right and a whole amount of the retina those those little orange dots and then all the cells in this experiment are ganglion cells are labeled with yfp and so she can then use fact sorting to collect the cells that are yellow that survived but didn't regenerate or yellow plus red that survived and regenerated and obviously difficult method lots of controls that I won't go into so here's the result and it was a little more complicated when we only knock out p10 although many types of cells survive the regenerators are actually predominantly a single small subclass called alpha retinal ganglion cells this turns out not to be a surprise because we actually shown that already in an earlier study but what's a little more of a surprise is that when we look at all the interventions the cells are regenerating in rough proportion to their survival so what seems to be the case is when you activate jack stat signaling which both socks three knockout and cntf over expression do you overcome the type specific limitations on regeneration imposed by p10 and so of course it's particularly interesting to find out what's going on downstream how are these interventions managing to cause regeneration I should point out by the way that p10 and socks three are not great therapeutic candidates on their own because they're both tumor suppressors so that would be bad news and cntf has been tried but isn't terribly effective all on its own so Anne could now go in and ask what are the genes that are expressed selectively by regenerating cells compared to those that survived but didn't regenerate and using the same sort of strategy we used before and that I described briefly she's been able to test them by over expression of three genes the transcription factor wt1 and two neuropeptides crh and galinin by over expressing them and indeed all three of them promote regeneration of axons as you can see in examples on the left bottom and the the the group statistically analyzed data on the right and again this is not enough to restore vision so far no intervention that anybody's used has been enough to restore useful vision and amounts but it certainly again encourages our thought that by figuring out how these interventions work we may be able to move towards getting more useful ones now what we'd really like to do though I mean that's the long-term goal but in the short term we'd like to have a deeper look into the gene expression programs that are regulated by p10sox3 and cntf so that we can sort of have a more comprehensive idea of what they're doing and how they're doing it and so what Anne, Nick and Wenjin have done is to take all the cells from all the interventions at all the times and essentially recluster them and then use a variety of computational methods to figure out what genes are expressed and how those genes vary by intervention by time by survival by regeneration and so forth and the key point here is that four different methods all end up giving us three gene expression modules that is genes expressed by sets of cells they're not identical and from one point of view it's not a big shock because of course all four of the methods begin with the same data set but they all make very different assumptions and use very different algorithms to come up with their results so one is to cluster the cells by type of intervention one is to cluster them all by transcriptomic similarity without any regard to what type they are one emphasizes the states over the types and one perhaps the most interesting is called scenic which digs into the data to come up with what are called regulons that is groups of express transcription factors and co-express genes that are predicted to be transcriptional targets of those factors and again we don't know enough for me to give you sort of a comprehensive view of the result but i'll just give you one example here of three different modules in this case from surat no sorry in this case from scenic and what you can see on the right are in blue the cells that end up being in each module and you can see that they're not completely but largely non-overlapping cells and on the right although you can't really read it is some indication of what some of the key genes are but basically basically there's one module that seems to be expressed largely by vulnerable cells that either destined to die or already dying and the genes expressed in those cells include many that have been already implicated in apoptosis including transcription factor atf4 that's been implicated apoptosis another module is expressed by cells that are surviving but not necessarily regenerating and that has interestingly a lot of genes in it that have been implicated in neural development one can sort of make up the story in which these cells are trying to survive they're sort of regressing to an early developmental state but they're not quite making it and then the third module selectively expressed by the regenerating cells that include the candidates that I already showed you have a lot of genes that are already known to be implicated in axonal regeneration so called rag genes a regeneration associated genes in other in other systems and so you know we're now digging in I'm not going to stop and talk about this to really look at what some of the genes are what they're expressed where they're expressed what the regulatory networks are about so let me come to the last and shortest story shortest because we frankly don't have a lot of results but I'm going to tell you a little bit about it in hopes of provoking discussion just to be honest we're in need of ideas and so the idea here is to see if we can use the retina to find out something about the evolution of cell types and classes kind of by analogy to what people already know about the evolution of genes and gene families that's a very well developed field and there should be but there isn't a corresponding field for cells for cells and we think the retina is great for this enterprise because as you probably know its overall structure its layers and cell classes are beautifully conserved amongst all vertebrates and you can't say that for other structures for the cortex it's a little hard to look at the evolution of cell types in the cortex because their whole parts of the cortex that are missing for example in birds or fish and in some of them the whole cortex is missing frankly so what i've told you about so far is a cell atlas from the mouse now a few years ago iran peng began expanding this search to look at primates first macaques and then marmosets and humans and her reason at first was that although the mouse retina is quite a wonderful model for studying the development and function of neural circuits it actually is a horrible model for looking at vision and that's because our vision is dominated by a central fovea or macula that's only a small fraction of the whole retina but accounts for essentially all of our high acuity vision and really most of our color vision in daylight illumination it's clinically critically important because diseases like age-related macular degeneration which is probably the number one cause of blindness in the in the first world selectively affect the macula and among mammals only primates have a macula so these are structures and functions and diseases that really cannot be adequately studied in mice so what iran did starting with macaque was separate the fovea which is the central bit of the macula from peripheral retina and look at the transcriptomes of the cells coming up with atlases of as i said first macaque then human and then marmoset separately for fovea and periphery and of course when she had those atlases she could then ask what is it about the fovea that makes it different from the periphery different types of cells different portions of shared types similar proportions of shared types differentially express genes that's published i'm not going to talk about it but the short answer is that all three things are true but the vast majority of types are conserved so that's really not it but every single type has a substantial number of differentially expressed genes between foveal and peripheral cohorts and so we expect that in those differentially expressed genes are going to be the ones that give foveal cells their distinct structures their distinct functions and distinct connectivity at any rate we could then begin comparing the atlases of the primates to the mouse and in the past few years we've expanded that to look at a number of other mammals rabbits paramiscus which is a wild mouse squirrels pigs ferrets basically anything that somebody is euthanizing in a lab for another purpose but can spare the eyes and we've also done some work some of it published on chick and zebrafish and they're beginning to work on lizards and here's just sort of a not very but fairly recent census of the number of types that we're finding in these species and so what we're really going to focus on now are the bipolar and the retinal ganglion cells so i'll finish with just two snapshots one is about cell classes and this is a comparison that Wen Zhen Yan made of two primates human and macaque mouse and chicks um looking at classes and what she did uh kind of what i showed you about the interventions is take all the cells together and recluster them and you can see at the top the clusters by class in this case we're not going deeper to the types and on the bottom if you look the four colors represent the four species that contributed to this map and and what i want you to see is that for every class or subclass cells from all four species contributed so in other words we knew that at a structural and functional level the retina was highly conserved amongst rodents primates and birds but we didn't know whether that was also true at a molecular level and the answer is it is transcriptomically all of the cones are each other's relatives all of the rods are each other's relatives and closest relatives to the cones glycinergic and the aburgic amicrons and so forth one surprise out of this for aficionados in the audience is that a really fascinating type of amicron cells the starburst amicrons are actually more closely related to retinal ganglion cells than they are to the other amicron cells sort of driving home their special status that people have already commented on by other criteria for types we know less but here's just two examples so when Wenjin went on now and reclustered amicron cells by type or reclustered retinal ganglion cells by type she got quite different answers and i think you can see that intuitively on the left where the individual amicron cell clusters almost all contain cells by the particular criterion from all four species and the average number of species that contributed to each cluster was three again by this criterion similar for bipolar and amicron cells whereas for retinal ganglion cells they seem to cluster separately and again intuitively you can see that there's some purple clusters some red clusters some green clusters not so many rainbow clusters and so on average only about one species contributed to each of these clusters and so this kind of supports the idea that the outer retina may be a beautifully conserved and optimally wired graphics card that is conserved through a lot of vertebrate evolution whereas a main site at which evolution acts to essentially suit the visual capabilities of each species to its ecological niche is at the level of retinal ganglion cells where the information is being confided from the eye to the rest of the brain and just last month um Karthik Shekhar came up with another intriguing piece of evidence that supports this idea and this is I'm going to be the last slide I show you so if you look at the species this is not all of them but most of the species we profiled on the left you see a conventional evolutionary map of their relationships human and macaque and marmoset of course are each other's closest relatives mouse and perimiscus squirrels are pretty closely related chicks and fish or outgroups if we look now at the transcriptomic relationship of the bipolar cells it fits that evolutionary relationship perfectly well on the other hand everything goes haywire when you look at ganglion cells when you look at the transcriptomic relationship of ganglion cells zebrafish are the closest relative of the primates pigs and squirrels are close relatives who they shouldn't be chickens are the closest relatives of the runes and we're not really sure what this means but one simple idea that certainly intrigues us is that maybe a lot of evolution is occurring more or less by neutral drift so the separation among species is in some sense a clock of how long it's been since they diverged and that may be true for bipolar cells they're evolving by neutral drift just as is the whole species whereas for retinoganglion cells there are selective pressures whether they be purifying selection or whatever that those selective pressures are acting and they're overriding the evolutionary clock for reasons of their own again the idea would be that it's at the level of ganglion cells the evolutionary pressures are being felt the strongest to make sure that the right kind of information is being transmitted from the eye to the rest of the brain and of course this is evolution so we're never going to be able to test it but we're trying to get more species to see if this holds true and see if we can do more computational tests and look at what types there are that for example make zebrafish look like macaque in this respect so this has been the most collaborative project I've ever been involved in it's been a real pleasure I mentioned Steve McCarroll and Jagon he as I went along Mike Doe was a huge help in the initial studies of primates Aviv Ragev got us going on all the computation in fact Karthik Shekhar began as a postdoc in her laboratory and the zebrafish work was a wonderful collaboration with Yvonne Kouch a graduate student in Herbert Byers lab so I'll stop now thank you very much okay great thank you just it was a great talk really interesting and now we have a couple of questions in the chat that I will read to you so first we have from Professor Marla B. Feller and she asked how do you think neuropeptide over expression could enhance survival are there auto receptors on the cells and it somehow promotes signaling in the same cells that secret them yeah so we we don't know first of all I should say thank you Marla for getting up so early in some cases in in other cell types DRGs and so forth it looks like one main way they act is by enhancing increasing cyclic amp levels and activating the whole pathway downstream of cyclic amp D we don't know at all whether that's the case here in terms of receptors yes for urocortin and CRH for example the CRH receptor is expressed by pretty much all ganglion cell types we know less about the the element receptor okay thank you then we have George Caffetzis in what light conditions where mice kept post O&C could neuronal activity play a role in the survival of regeneration processes yeah that's a great question there certainly has been some evidence not from our lab but from others that activity can promote survival of ganglion cells that started first I think with Ben Barris demonstration in cell culture and there's some evidence in vivo we kept them in a normal daylight a normal daylight cycle we haven't tried changing activity in this in this context we did in another context if if you give me another 30 seconds we have been interested in whether the maturation of ganglion cells is affected by activity and so we've done a bunch of dark rearing experiments and they have to me shockingly modest effects on the transcriptome of adult retinal ganglion cells okay thank you then I have a I have a question so I think I observed that there are so many different types of amacrine cells in the clustering user and I know that they have different morphologies so I was wondering if you found many more pathways that are different among them or very different genes among them more than in in retinal ganglion cell or in vivo ourselves um yeah there's more types um and certainly when you look at them transcriptomically the GABAergic amacrines and the glycinergic amacrines sort into different clades um our attempts to try to come up with further subclasses the way we've done with ganglion cells have not really been very successful so far so my guess is they're there and we just haven't done a good enough job of looking for them okay um now we have a a question from Leon like that and Yadu how far can we relate transcriptomic overlap divergence of cell types between species to different in visual processing within some species within those species yeah that is the question that is what Karthik is really focused on and um we just don't know there's nothing I could say that wouldn't be stupid to be honest so I won't say anything yeah okay so we have another question from Simon Lafling great talk slightly surprised that Owen um people our cell cluster with off giving ideas that they evolved from photoreceptors the bipolar cells are are related to photoreceptors you're saying um yes I think I think they are pretty closely related to photoreceptors and I think I mean that seems reasonable to me they both have ribbons synapses um and they have a lot of physiological features in common um so you know I'm not sure what to say I don't I don't think anything is expected but I think that transcriptomically um that amicrons and ganglion cells are more closely related to each other and photoreceptors and bipolar cells are more closely related to each other and we've tried to look a little bit at what's known about the most primitive retinas before there were a lot of cell classes and I guess the best I can say is what we've seen is consistent with that but not really strong evidence okay um let me I think there is a last question is there some some way to identify the genes that are more susceptible to evolution yeah again in this very very preliminary little story that I told you that has not been done yet but certainly um there are methods that people hi Marla there are methods that people have worked out to track the evolution of genes and in fact that's a pretty well established field and so Karthik does believe that he'll be able to dig in and do that but again hasn't been done yet um you know whether it whether it can be done with enough certainty to be useful I think remains unclear okay okay thank you just so yeah I think we will finish the the talk now thank you Josh and thank you everyone for for being here and now we will uh I will leave the the transmission on for a while so that people can join uh but just please join the the discussion uh now thank you okay so we have more people joining Josh so good to see you it's been a long time to you I only hear about you from Karthik these days I know I talk to Karthik all the time he's doing he's doing great I hope he's doing great he's doing great he was in Boston a couple weeks ago for some chemical engineering method you got to get him out of that department yeah it's a good department I think isn't it well anyway we could talk about it no I think it's a great department they just have to teach a bunch yeah yes to teach a bunch but he really kind of likes to teach or so he told me but anyway anyway I like to keep my grab but not that much of it yeah yeah well it's a little bit you know those hard money positions you get you gotta you gotta get paid for doing something so uh but um so I have but I do talk to him a lot about this so um I have I have a list of questions I'm gonna go run and get then a lot of people ask questions so I'll be right back Josh could I ask a sort of general question um how do you I'm thinking a lot about homology versus analogy in evolution and where you see this uh finding that the that no ganglion cells are um somehow more related between uh fish and primates for instance than to other species is that because they're more analogous or um yeah this is the downside of talking about data that we don't understand I mean I like the idea and I credit Karthik for this that the wacky transcriptomic relationships are a sign that there was selective pressure exerted on the types of ganglion cells that overrode the dominant uh feature of the other cell classes which is essentially a timing mechanism dominated by drift yeah but then what you're asking is why what is that pressure can we learn something about that what the pressure is from say the fact that fish and primates are related to each other and I just have no idea I really have no idea my first thought was that it must have something to do with color vision