 Okay, so now we are officially live. Hello, everyone, and welcome to a new seminar of our seminar series, Sussex Visions. I'm Antonio Nujosa, I'm a postdoc in Leon Lanyado Lab, and I'm working in Moz Cortex. And this series of seminars are part of the Worldwide Neuro Initiative, which is a, sorry, I have the, sorry. So as I was saying, this is part of Worldwide Neuro, which is an initiative to watch a greener and more accessible kind of talks. So this one, this initiative was started by Tim Boggels and Panos Bocellos, and I would like to thank them for this initiative. And so basically just for the newies, the format is we have a talk of 45 minutes plus the questions. And so for everyone who has some questions, please post them in the chat. And by the end of the talk, I'll be very happy to ask the questions to the speaker. So, and finally, before I introduce the speaker today, I would like to encourage you to join the channel if you want to, if you don't want to miss any of the talks that we have every week. Okay, so today, we are very happy to have with us Professor David Berson. And so I will make a little introduction about his career. So Berson graduated from Brown University in psychology. And after that, he joined the lab of professor Anne Gravelle at MIT with here, his PhD studying salamocortical connections. Berson began his postdoctoral research at Brown University with Professor James T. McCallway. And then he did another post-doc with Professor Peter Hardline at the iResearch Institute of Retina Foundation in Boston. During this period, he continued studying various visual circuits, like for example, the connection from retina to superior colliculus. So Professor Berson accepted in 1985, he had a faculty position in the department of physiology at Brown University. And as a PI, he has focused on studying the retinal amplitude ganglion cells. And one of the main discoveries that I'm sure most of you know has been the intrinsically photosensitive melanoxin cells. These are ganglion cells that they are able to set the circadian rhythm from the eye by being photosensitive, being sensitive to light. So at present, he's current as the chair of the department of neuroscience at Brown University. And he's also a Sydney A. Fox and Dorothea Dr. Fox professor of ophthalmology and visual science. And just to finish, I would like to highlight that he has received many professional honors throughout his career. And that includes being a fellow of the American Association for the Advancement of Science. And also he received a boycott prize for outstanding career contributions to retinal research. He's also received from Brown University, the Presidential Faculty Award. And finally, the Freedom World Award from the Association for Research in Vision of Thalmology. So today we're very happy to have you here, David. And we are looking forward for your talk. Thank you. Thank you, Antonio. I have one small correction and that is I no longer chair of the department and I couldn't be happier about that because I'm back doing science again. So that's a great relief for me. But thank you so much for that nice introduction and thank you so much to the organizers for the opportunity to speak today. I'm gonna try to share my screen here and hope this works well. How does that look, Antonio? Yeah, it looks great. Beautiful. Super. So what I wanna talk about today is our efforts over the last decade or so to try to work towards a connectome for the interplexiform layer of the mouse retina. And in particular today, I'm gonna be focusing mainly on connections between bipolar cells, which of course link the photoreceptors to the ganglion cells and then the ganglion cell targets. We're working on many other aspects of the connectivity of the IPL. I gave a talk at Arvo recently on the roles of these circuits in visual reflex functions of various kinds. And at the upcoming FASSEB meeting in June, I'll be talking much more about amortarine cells. So today's focus really is much more on really kind of the bipolar cell to ganglion cell connectome. Before I get into the heart of the talk, I want to credit three unbelievably sharp and productive young scientists. Megan Lair at the left is a recently graduated PhD student in the lab. She has been working very hard. Oops, I'm sorry. I meant to, okay. I meant to get a better pointer here. Okay, I'll just suspense with that. So Megan recently graduated. She's done a ton of work on connectomics in the mouse retina. Shai was a postdoc for several years in my lab, now has his own lab at Hebrew University in Israel and was heavily involved in physiology of various kinds I won't be talking about today. But he also ran a team of very dedicated undergraduates which you see listed at the right who helped him with connectomic analysis related to the intrinsically photosensitive retinal ganglion cells. And finally, at the right, we have Drew Linsley who's my current collaborator, he's a research assistant professor at Brown in machine vision. And we've been working together on methods for automated segmentation and synapse identification which we are now harnessing for this project. I'll show you a little bit of what's coming out of that at the end of the talk. And then of course I'd like to acknowledge the funding for the work which without which we really couldn't have done it. So... Sorry, David. Did you want to put on a different cursor? Yeah, I was just going to try that, but... I think it's just if you right-click, maybe you can have some options. I mean, just if it improves. Right-click. If you just right-click in the presentation. Yeah, pointer options, okay, pen. Yeah, that's the one that's best, pen, let's try that. That's not so good. No. Okay, we can just stick with the original cursor. Can you see that well enough? Yeah, I can see, but I just saw that you tried, but if there is no other option, I thought there was more options. Highlighter, maybe. Highlight. Is that any better? Yeah, it's better, but yeah. Normally there is one that is a circle, but I can see that now. Let's see it. Well, we'll use this for now. Okay. So, we all know that ganglion cells, shown here, many of these are ganglion cells at the bottom in this work from the turn of the 19th century, from the work of Santiago Ramon E. Cajal. We've known these cells are very diverse morphologically and Cajal himself inferred that they were making contacts with other presynaptic cells and that this was underlying the computations that the retina did to allow us to see clearly. Of course, the feed into these cells from the outer retina comes from bipolar cells and we now know that there are a great diversity of these cells as well, but not nearly as diverse as the ganglion cells. And so really at the core of what I'm trying to get at today is the question of to what extent are the very obvious functional differences among different ganglion cell types, traceable at least in part to differences in their connectivity with these bipolar cells, which we have glimpses of differences in functional properties of, although there's much more to be learned in that realm. So there's a longstanding notion traced back to Peters who suggested that connectivity among neuronal types is really based upon the positioning of presynaptic elements. If they're arbors, these synaptic arbors of the axonal arbors of the bipolar cells are stratifying at a particular level of the IPL, which you certainly saw in this last slide here, very systematic differences from type to type in different layers. Two different flavors of cells, we know the off types are stratifying, they're arbors near the top of the IPL, the on types shown in these green or cooler colors down in the lower part of the IPL, the on sublayer. That's gonna dictate which types of ganglion of cells with their arbors at different levels in the IPL they might make contact with. But that ultimately, that you can really read out what the likely connectivity is simply from their stratification essentially. And what I wanna do today is probe that at some level of detail with new anatomical evidence based on this serial EM reconstruction that we've been involved in. The volume we're using is the K0725 dataset. This is from Kevin Briegelman's lab, back when he was at NIH. And the dimensions of this thing are roughly 260 by 200 microns. It's really pretty much restricted just to the IPL. You get a little bit of the ganglion cell layer, a little bit of the inner nuclear layer. It's adult mouse. Loxals are about 13 by 13 by 26 nanometers across. This block has been stained for synaptic detail using heavy metals, unlike some of the earlier volumes that came out of the work with Winford Dank. And this side kind of shows you what we're dealing with here. So you have to understand that we are blind to what's happening in the outer part of the retina. And that's a bit of a limitation for us but we can work around it. Here's a reference to the original paper down at the bottom left. Just for reference, this is the volume from Helmstetter at all. This is one of those original Dank group papers, first looking at the mouse retina. And that of course was not stained for synaptic detail and is much smaller in extent than the current one that we're working with that gives you much better access to cells with more just widely dispersed dendritic arbore just like many of the ganglion cell types. So this is valuable from that point of view. And of course for identifying actual synapses. This gives you a little bit of a look at the EM in this material is serial block face which already is a bit of a limitation. It's not as good as the transmitted EM that you might get. It's also a little bit of a legacy volume by now. Not nearly as good as some of the stuff that comes from serial block face from current instrumentation. But this is really not that bad a limitation if you take the time with the volume, you can see the synapses. Here's a nice synapse here, for example, onto a bipolar shaft. And you can do a lot with this volume. So we're looking to other volumes going forward but we have a lot invested in this volume and you'll see how much you can learn from it. It's also publicly available by the way. So most of our work has been done with a process we call skeletonization that is dropping nodes into the interior of cells at various points connected by little connectors. And this creates a skeleton for an individual cell and you can do this over and over again for different cell types and get a cartoon, a stick figure version of what the cell looks like in three dimensions. Overall, we have now 10,000 or so of these skeletons. I wouldn't say these are necessarily all unique cells because in many cases you have processes coming in from the outside of the volume and some of those may be coming from the same cell but we wouldn't know that. So we call that a separate skeleton and we have a total of about two and a half million nodes making these things up. We have about 160 ganglion cells in the volume. Here you see a map of the somas of the ganglion cell layer in this volume and all these little annotations we know what all these types are. The smaller ones, some of which have been completely reconstructed are the amherst cells but the general rule rule here is that the bluer and greener colors are ganglion cells and the rusty colors are the amherst cells and has the right kind of breakdown. These are the blood vessels of the ganglion cell layer. We now have pretty complete mosaics for all of the major bipolar types. I didn't have quite time to update all of these. This is more typical of what we have. These are type six and seven bipolar cells. Here's type nine down here. We filled in many of the gaps that you see here. So we have almost complete mosaics for the great majority of the bipolar cell types although we've pretty much ignored the rod bipolar cells because of their lack of direct connectivity to the ganglion cells. So I don't know that I can see the number up there but it's about 1500 bipolar cells all together. This just shows you a breakdown of the numbers of bipolar cells in each of the types defined by their axonal arborization patterns. And you can see that there are radical difference in numbers. Type six, for example, there are 220 cells in the volume roughly speaking whereas the type nines are much fewer in number as are the XBCs, for example, here. But all of the major types are represented and as I say, we have mosaics for those and we can identify ribbon snap to context between each of these types and whichever ganglion cell type we've skeletonized and are trying to work out the connectivity of. This is a side view of all of the ganglion cell skeletons. You've seen that before and what I really wanted to show you here is that you can learn a lot just from looking at the skeletons themselves even without any synaptic data. And of course we've seen this already from the Eyewire Museum where synaptic detail is lacking. But what I wanted to point out here was that you can already begin to see the glimpse of a non-uniform density of ganglion cell dendritic processes at different layers within the IPL and this is much more easily seen in the next movie which I realize may not play very well but I think it'll play well enough that you can see that there's an extra dense band here which is the off chat band where the off starburst amicron cells sit and another one in the on chat band here. And then in the middle of the IPL probably in just the beginning of the on sub layer between the chat bands, there's another dense zone and then zones that have sparser dendritic arborizations of ganglion cells overall. So things are complicated. So what I'm gonna try to do today is to give you a little bit of a tour through the different sub layers of the IPL relating bipolar that are terminating at that level to ganglion cells that are receiving those inputs to try to give you an overview of the bipolar to ganglion cell connectome in the mouse retina. And I'll start just with this division between the off and on sub layers of the IPL with this black line dividing the two. And I'm gonna be using a slightly different sub lamination scheme than the typical five layered scheme that just divides it into equal 20% increment. So we're gonna start with the chat bands. Everybody knows the chat bands. This is where the starburst cells terminate the off starburst terminate here in the off chat band the ons in the on chat band. And so these two layers will be considering independently outside toward the margins of the IPL we have two other layers one in the off one in the on and I call these outer off for the purposes of this talk and inner on. And then finally, we have on and off sub layers that are actually in between the chat bands. So I call these off inter chat that is between the chat bands and on inter chat. Okay, so these are the six main layers but then I have to add one more. And this is a little peculiar it's an on sub layer that is sort of patchy and lives in the wrong layer. And we're gonna have a lot more to say about this shortly but these are little bits of on channel input to ganglion cells and amaranth cells that sit in what seems to be a wrong layer the off sub layer. So these are the seven compartments I wanna be talking about. So the whole thing, this whole study really started with an interest in trying to understand what the nature of bipolar inputs to the intrinsically photosensitive retinal ganglion cells or IPRGCs were and Antonio already introduced these as melanopsin expressing cells that are capable of direct photosensitivity. Another thing you should know about them is that they're really the only ganglion cells in the retina that are actually keeping a representation of the ongoing integrated light intensity in the environment. Other ganglion cells pretty much adapt this out through temporal filtering. We know these are present throughout mammals probably to some extent throughout vertebrates but they're certainly present in our own eyes. And we know now that there are many sub types of these IPRGCs and I'm just summarizing those for you here. So here you see the six known types of IPRGCs and they're dendritic stratifications within the inner plexiform layer off sub layer at the top on at the bottom. And these can be, and you can see that some are monostratified like this one here the M1 cell like this one here the M4 cell also known as the on alpha cell. And some are in the so-called off layer some in the on layer. The ones off to the left are the ones that have been implicated in what are sometimes called the non-image forming visual functions. These are non-conscious functions that relate to visual reflexes like the pupillary light reflex, the entrainment of the circadian clock to the rising and setting of the sun and so forth. And they include these three types the M1 through the M3 cells. This collection here is more like other ganglion cells in the sense that they have projections into the lateral geniculate nucleus and they've been implicated in image forming or conscious perception. But all of them express at least some melanopsin and can respond in the absence of any input from the bipolar cells. So what we're really interested in initially is this population here the non-image forming varieties but we're interested in the full range. So let me start with this inner of these two bands containing melanopsin denerases in the innermost IPL we call that inner on sub-layer. And what I'm showing you here in this little cartoon and I'm gonna have cartoons like this for all of the sub-layers. So let me get you oriented here. We're gonna see the major bipolar cell types that have axonal arborizations in a position to make contact with ganglion cell stratifying in this region. The yellow indicates the location of the layer. So this is inner on. So it's in the inner part of the on sub-layer here shown in green. And over to the left here I'm listing the ganglion cells that have dendrites stratifying within this tier of the IPL. And the plot here, the sort of the table here is showing you that the, again, the same bipolar cell types that we show over here to the right. And this is in the cells here, you get a representation of the percentage of all of the ribbon synaptic input to the ganglion cell type listed at the left here from each of these types of bipolar cells. Now in some cases there are other elements that are missing here. I'm trying to keep things as simple as possible. This is a big sprawling story and tough to get through it all. So this is just the way I've simplified things. And I've tried to highlight the major inputs by bolding them and putting them in red. Relatively strong inputs are bolded but they're still black. And then relatively weak inputs are in the smaller font and in black. And then I've separately shown cells that are monostratified and bistratified. So you saw before that the, I've lost the M1 cell here somehow but that's fine, we'll come back to it. So the monostratified texts include the M2 here, the M4, the M5, whereas the M6 and the M3 are monostratified, right? I'm sorry, bistratified. They're stratified in both the inner and outer sublayers. But here we'd be talking about this inner part of their arbor here in the inner on sublayer. So what can we infer from the connectivity that we've reconstructed from this KO725 volume? The first thing is to say that the IPRGC types that are all shown marked here by the purple arrows here, these are dominated by type six. Every one of these has a dominant input as shown in this bright red percentage here from type six, we do not have a clear M3 cell in the volume. So we really can't say anything about those cells but type six is clearly playing a huge role in driving these cells. One of the things that fell out of this was that cells that look to be of the M2 type, those are stratifying, they're monostratified only in the inner on sublayer. They seem to fall into two groups. And you can see that in the plot here that one of these types which are called M2 subtype eight have a substantial amount of type eight input along with the type six input and relatively modest type nine input. It's like two or three times as much, three times at least as much eight as nine. Another type shown here is the subtype nine variant. And here you see that there's virtually no type eight input at all. It's very heavily from nine but even more heavily from six, okay? So this is true across all of the cells stratifying at this level that we can find that have this sparse architecture. The color scheme here just shows you the differences in connectivity here. I won't get into the details here but more red is a type nine and more purple is type eight. So the question is which of these is the real IPRGC? And the answers we don't really know at this point. My first expectation was that this M2 subtype nine might be the more obvious one because it has a lot of input from type nine which is true also the M1 cell. I'll show you that a bit later. And other IPRGC's often get a lot of type nine input but I now believe that of the two of these the one that I'm more confident really must be a true IPRGC is this one at the top, it's subtype eight variety and let me show you why I think that's true. We think this other cell is very tightly anatomically linked to a type of wide field of polyaxonal amicron cell that has sustained on responses and it turns out to be selectively labeled in the RBP for Cree reporter line. So Shah Sabah and some of my other students published paper on this some years ago there's some interesting ganglion cell types labeled by that Cree line but there's this amicron cell which has a relatively small dendritic arbor a few hundred microns across and extremely widespreading polyaxonal arbor and this is what they look like in the K0725 volume seen on faucet the bottom left and in vertical view here so it's stratifying below on chat here. Now the movie here is going to show you the wide field amicron cell here. See if I can get this to play, it's not playing. Okay, there we go. So this is the wide field amicron cell it's a spiny cell you see a couple of purple spines coming off the yellow cell is the M2 subtype 8 and you can see that the two are glommed on top of each and there's a lot of contact here but also elsewhere. And what I'm also showing you are the sites of synaptic ribbon synaptic contact from two types of bipolar cells. The type eight is shown in sort of bluer and the type six in greener axon terminals here. So the red marks are the bipolar contacts ribbon contacts. So we know that this rack one amicron cell is gap junctionally coupled to the M2 cell. If you inject dye into the neurobiotin dye into the wide field amicron cell you get coupling tracer coupling into the M2 melanopsin positive ganglion cell. So you suspect, although of course we can't really see the gap junctions in our material here. This is serial block phase that this one is in a position to make that kind of gap junctional connectivity. By contrast, the other type the type nine variety of the M2 cell really doesn't seem to care about this wide field amicron cell at all. It might occasionally bump up against it but there's really no extended contact. There's no shared dyad synapses the way you have from the type six and eight at those points of contact. So we're pretty confident going back a slide that this type here is a true IPRTC. The identity of this one remains unclear. One possibility is that it's a variant of the M2 cell that has a different bipolar input and maybe lacks the gap junctional connectivity with the wide field amicron cell. Another possibility is that it's just a lookalike. It's an imposter that's masquerading as an M2 cell but it's actually something else entirely. And we won't be able to resolve that until we get more targeted electron microscopic evidence, I think. Okay, so this brings us to the M1 cell and that reminds me why I didn't show the M1 cell in the previous slide. And that's because of course the M1 cell isn't arborizing in that inner-on cell layer which I showed you before. It's only arborizing in the outermost part of the IPL and that's shown here. And what we showed some years back as did Steve Massey and Steve Mills and their collaborators was that the reason the M1 cell is actually an on-cell like all the other IPRTCs that it's actually getting on-person input from bipolar cells at this level. This is the origin of what I'm calling the accessory on-sub layer. So let's talk about this layer. And so I'm showing it here as these little bits of the on-sub layer that are sort of shoved into the outermost part of the off-sub layer. Here you see the type six cell we were talking about. It's descending axon. It's making synaptic contacts here and driving on responses in certain post-synaptic targets. So what cells are stratifying at this level and are making contact with these on-person synapses, the M1 cells for sure, fully three-quarters of their synaptic input is coming from these on-person synapses. The M6 IPRTC, that's the bi-stratified type that has both the inner and the outer arbor. Its outer arbor is also getting input from these two populations, the type six and type nine. And the reason that I've highlighted these two cell types as particularly bright here is there really seem to be the ones that are making the most contacts onto post-synaptic anglian cells in the accessory on-sub layer. We certainly do see others on-person synapses and other on-types as have Ryan Jones and his group in the rabbit, but at least in our hands and in the mouse, type six and nine seem to be the most important. I'm gonna come back to this in a minute. I'm asserting their type nine, you won't have a reason to believe it until I show you the evidence. And then finally, we suspect that the M3 cells may also be picking up these inputs, but we'll just have to wait for a bigger block to get to that rare type. So a couple of points here, these synapses are unusual. They tend to be monad synapses. There's a single post-synaptic target. I'll show you what I mean by that in just a minute. So they also involve more than a single ribbon in most cases. We have evidence now that type nine bipolar arbors are bisradiified. And I'm gonna show you that in an upcoming slide. Overall, we believe IPRGCs are a preferred target of the type nine bipolar cell. The reason you might be interested in this particular cell type is it's the only one makes selective contact. That is, it receives its excitatory drive selectively from the true short wavelength cone bipolar cell. Makes selective context with the true UV cone in the mouse retina. And the other thing that I wanna add here about these on-synapses, they're not limited just to ganglion cells and particularly the IPRGCs. They're also targeting other amricone cell types that seem to be related to this luminance coding network, including dopaminergic amricone cells and NOS-1 amricone cells and a bisradiified type that we believe corresponds to one of the VIP positive amricone cell types. So here's a side view of an M1 IPRGC. This is a skeleton now. These are descending type six axon shafts, right? So the main arbor here, the type six cell is gonna be in the inner on-sub layer, which we talked about before. But as these axons go by, they're making these on-synapses onto type one, I mean, M1 IPRGCs as shown by the red dots here, almost all of the input to the cell comes from those. There's a little sprinkling here in the inner part of the on-sub layer as well. But that's really a minor part of the input to the type one cell. Now, the distribution of these shafts is of course much less dense than the axonal arbor. And this dictates the density of type six inputs onto the M1, which is really remarkably sparse compared, for example, to the bipolar inputs to inner stratifying IPRGC types like the ones shown here. So that makes the M1 cell pretty distinctive in terms of its bipolar drive. So this is just a schematic view showing you these multi-ribbon monad synapses. So these ribbons here, they're making a contact on a single post-synaptic cell. In this case, I've listed it as an IPRGC. But the same axonal shaft will give rise to a big arbor down here at the base, which will make classic dyad synapses with two post-synaptic partners and the potential for both feedback and feed-forward inhibition of the transmission from bipolar to ganglion cell. So there's something special about this accessory on sublayer, as we call it, where amicron cells don't really get their fingers into the mix here as much as they do elsewhere in the IPL. This just gives you a look at a couple of the synapses in the actual EM. This is a dopaminergic amicron cell here. This is a plaque of up to 13 distinct ribbons that are gonna be hard to see in this material, but the dark elements are the synaptic vesicles and the gray fuzz is actually the ribbon here. These are various examples of similar contacts onto the M1 IPRGC. They tend to be a little bit smaller, but often six or seven separate ribbons. They can involve these kinds of imaginations in some cases, but they're always moments. Now there's another player at the same level. And I'm gonna see if I can get this movie to play here. Here we go. Okay, so the yellow shaft here is a type six axonal shaft. As you get transparent here, you can see it's monad ribbon synaptic contact onto the blue M1 dendrite here, but there's another player here that looks kind of similar. It makes monad synapses onto M1 dendrites and onto many of the other cell types that I just described getting input from the type six, but it's stratifying only in the outermost IPL. And this had a stump for the longest time. They come into the volume and they sprawl across the volume much more than any other bipolar axon terminal stratifying at this level. Here are the T1 and T2 type one type two bipolar axon terminal shown on FOSF scale. And you can see these things are much, much more broadly spreading. You can see they're costratifying with the M1 IPRGC. These dots are the points of ribbon synaptic contact, the guemplas are again, monad synapses. And some of these send a few little branches up into the inner-on sublayer as well. And we couldn't really tell what these things were, but they looked for all the world to us like type nine bipolar cells, but kind of inverted. So the type nine stratifies exclusively below on chat here, but occasionally you find one with widespreading arbor at exactly the same level where you see these funny sort of novel bipolar types. So given their connectivity onto IPRGCs, which are on cells and they're similar morphology, we kind of were thinking they had to be an on and maybe type nine-ish somehow, but not really sure. And then we had a conversation with Wei Li, who said that he had a new reporter line for the type nine bipolar cell. He said that he could see occasional collaterals at this level in the outer off sublayer, the accessory on sublayer, I guess you might call it in this case, but that the branch points were pretty proximal, pretty near the cell body in the inner nuclear layer. And of course we only have the IPL, we couldn't see that branch. And so we began to infer that these cells, much these funny off stratifying type nine-ish looking cells must in fact be the outer arbor of by stratified type nine cells. We see a lot more of that than Wei sees generally, and that may be because of topographic location. We're not absolutely sure, but there was an exciting talk from Yao Shui from Jimmy Joe's lab at Arvo saying that they had recorded from cells that had this kind of property and in the off sublayer and found on responses. So all of this is converging, I think on a recognition now that at least in some cases, type nine bipolar cells in the mouse can be by stratified and they're exclusively on. So they're participating in the accessory on sublayer. Okay, now we've got a lot to cover in a very short creative time. So this is gonna be a bit of a whirlwind, let's take a tour of some of these other layers. We're still in the on sublayer, but now we're talking about the on chat band, right? This is where the on starbursts terminate. This is closely linked to the direction selectivity network. In fact, the major types that you find stratifying there are of course the on off and on type direct and selective ganglion cells. There's an overlapping, I call it a by stratified type. It's the F mini on cell. I'll have much more to say about that in just a minute. And there is a cell that Greg Schwartz calls the on transient medium receptive field cell that's kind of straddling the on chat band gets in there a little bit. And here you can see the bipolar input connectivity patterns for these cells. And what I first wanna do is to make the point that most of the cells that are stratifying in the on chat band really are by stratified, right? This is this one type that isn't. The thing that really has us a little bit surprised is that the pattern of connectivity to the on and on off direction selective ganglion cells seems to be somewhat different. I'd always assumed that the bipolar drive would be the same. These are both DS cells. We know that the on direction selective type like slow velocities, but some work from Adam Monty in my lab building awful work from Ben Sivier suggests that that's largely due to feed for glycinergic inhibition from V-glute three amicron cells. So we didn't really think that the bipolar inputs would be all that different, but you can see that the on off cells seem to be getting a lot more input from type 50 and type five I than does the on directional selective type. And the on type is getting its dominant on input from type five T. You get a different pattern for the F mini on here you see a lot of selectivity for type five I. Okay, so let's turn to on interchat. This is the uppermost part of the on sub layer. Here you have a lot of the type five family terminating at this level, including the XBC. And there are many, many ganglion cells that you can find stratifying at this level. And this is partly, I think why you see that density of ganglion cell dendrites at this level. And let me just make a couple of points here. The first is to call your attention to the XBC. And really for the most part, this bipolar cell type doesn't seem to be making a lot of contact onto ganglion cells. The one obvious exception for this is a cell we call OIA, but this is called the on transient alpha by others, including friend Ricky's group. So it's stratifying between the chat bands that has an alpha like morphology. And that one seems to be getting some substantial amount of XBC input. I've already pointed out that the F mini ons down here, the by stratified type is getting a tremendous amount of input from type five I will have more to show you about that in just a minute. Now, what I want to really point out here in this on interchat zone is that there are four bushy types that are really quite narrowly stratified at this level. One of them extends a bit above off chat. This is the W3B or UHD cell in great Schwartz's terminology. The others are pretty much limited to the interchat zone and mostly biased towards the on half of interchat. There's really not much to distinguish these in terms of stratification, but if you look at their bipolar inputs, they're really the four types are really quite distinct from one another. They differ in their on off balance. There's really only one that gets really strongly biased on versus off input and this one over here. And you can see that the waiting of the different on and off bipolar cells varies from type to type. And what's very satisfying about this is this maps very nicely onto, just to show you that the mosaics are very lawful, that respect the tiling rule very well, but these map really nicely onto the four types that Jason Jacoby and Greg Schwartz identified based on physiological and followed by di-filling data. We don't yet know how to map these two on top of each other other than to say this is the equivalent of our UHD, the type W3B is the smallest field cell of all, but we'll need more correlative information, I think before we can really assign one to one the remaining three types with the three connectivity types. Okay, so let's move to the off sub layer and here we're dealing with the outer off sub layer not to be confused with the accessory on sub layers. So this is the matrix within which that funny bit of the on sub layer sits, the major bipolar cell types terminating here are types one and two. And of course, this funny one that doesn't have an outer dendritic arbor, the gloomy cell first described by Rachel Wong's group that looks in other respects a lot like a type one cell. There are at least four major monostratified types that are stratifying at this level. And you can see that all of them are getting a ton of type two input here, whereas only one of these is sort of distributing its input more widely across types one and gloomy, okay? And that's the jam B cell, that's an interesting cell type but it seems to differ from other cells stratifying at this light by accessing these two types of bipolar cells. Here's some vice stratified cell types as well. Again, the dominant input from the bipolar cell seems to be mainly from type two. This suggests that a lot of type one and gloomy output is getting to the american cells, which is something we're gonna be looking at going forward. Okay, what about the off chat band? This is kind of a boring layer. There are no monostratified ganglion cell type stratifying in there, these two directional types we know about. There may be some subtle differences in here but I'm not really convinced the on cells really don't have much arbor in the off sub layer. And so the numbers are gonna be very small here. One last point I wanted to make is that we know that the off starboard cells which are stratifying in exactly this layer are getting quite a bit of type one and type two input but the ganglion cells don't really seem to get a lot of that, the direction selected ganglion cells. So that's an interesting difference between the bipolar drive to the starboard cells and the bipolar cells driving the ganglion cells. Okay, let's move into the off inter chat zone here. This is right up against the on sub layer boundary. And there are two interesting cell types that stratify at this level. They're both considered alpha cells. One is the off transient yet there is the mini version of that and you can see that there are differences between these two cell types, especially with respect to the weighting of input from types three A and three B but otherwise they are blending the same bipolar inputs to some extent. The other thing that I wanted to point out here is that the F mini on ganglion cell type which spans this layer really doesn't get any off input which makes sense because physiologically they're on cells. So this actually represents a pretty clear violation of Peter's rule in the sense that here are the chat bands. Here is the F mini off is clearly symmetrical around the middle of the IPL and spans this off inter chat zone. And yet I just showed you it gets virtually no input at all from the off system. So almost three quarters of it is coming from one type of on bipolar cell. So just to sort of give you an overview here, you know what we sort of wanna know is to what extent do ganglion cells just listen to one channel or are they blending lots of different channels? And there's a spectrum and at one end of the spectrum you have what you might call specialist ganglion cell types which tend to get most of their input from just a couple of types. The M1 cell is a good example here. He's getting a lot of input from type six and from type nine now that we understand that that's what that funny extra outer bipolar cell type really is the Delta cell is another good example here gets almost exclusively type two input. Same thing is true for the F mini off cell not to be confused with the F mini on I just described a minute ago. This is a cell that is called type 25 in the eye wire museum and it's a bistratified cell that it really only gets input from one type of on five T and from one type of off type two. So it's really quite a specialist it's just in both on and off sub layers. And then you can compare that with general lists like the on off directions like the ganglion cell which sort of distributes its bipolar input across a larger number of cell types and the ultra high definition or W3B small field ganglion cell I showed you before is really picking up from a lot of types as well. So it's not really quite either full extreme you don't get any ganglion cell type that draws from all types of bipolar cells nor do you have very many ganglion cell types that only listen to one there's somewhere away from the extremes of that spectrum but there are variations from cell type to cell type. So let me just give you some takeaways here. Each ganglion cell type exhibits a unique blend of bipolar input. Some are specialists with few input types others are generalists I just said that the cost stratification of bipolar cells and ganglion cell arbors largely does dictate the connectivity that's Peter's rule except sometimes it doesn't. And good examples of this are the M1 cell which is stratifying in the off sub layer but it's getting input only from the shafts of type six bipolar cells and the outer arbors of the type nines and the F mini-ons I just showed you that example where it's stratifying with both on and off bipolar cells but it only listens to the odds. To some extent we see the same discrepancy with the on and on off DS cells they're stratifying in the same layers they're in the chat bands and yet they get somewhat different bipolar inputs. And finally, type six and nine bipolar cells which are by stratified appear to dominate IPRGC and therefore luminous coating type one and glumis make relatively few ganglion cell synapses but they do onto the GMB type. So I'm just gonna close now with what's coming next and this is the segmentation work I was telling you about before. So what you're looking at here is a skeletonization skeletonization of lots of on starburst americans cells. So this is the kind of data we've had up to now and we're beginning now with automated segmentation and what you see here is the beginning of this little sample zone here where Drew has started to segment. I'm not sure this is gonna play here let's just hope that it does. Okay, so you're zooming in now on the little segments that have been pulled out by artificial intelligence convolutional neural network that Drew generated. And this is in KO725 the very volume we've been talking about. Drew has a draft segmentation of the entire volume and we're in the middle now of beginning to curate that. We hope to have this whole thing all of our skeletons and the segmentation posted for everybody's use within a few months I'm hoping but we're not that pleased with the EM in KO725 anymore and we're moving now to other volumes and here what you're looking at is what Drew's software can do in material that was provided by Rachel Wong's lab watching you and what you're looking at here is here at higher resolution just the EM and here's Drew segmentation which you see is pretty faithful to the underlying architecture. And then you can see that the segments span really quite a large part of the volume and I can give you a better sense of that by showing you a side view here that these meshed up and you can actually see their shapes and this is the full width of the volume and Drew's software with no input for me has segmented a full 50 microns of these ganglion self-dendrites across the full volume. So we're hoping to segment more volumes like this and make them available to folks which I think is gonna really supercharge our ability to do the kind of connectomic analysis that I've just been describing and in KO725 of course we have all those skeletons generated and we can use those nodes and skeletons to do the duration to agglomerate over segmented volumes to try to get to full cellular volumes much more rapidly than we could do otherwise. So that's the current state of the art and any questions you have about what might be possible please let me know. We would be delighted to share what we have with anybody who's trying to crack a particular circuit and needs information about the connectivity of bipolar cells with amyloid cells or ganglion cells or amyloid cells with ganglion cells. So thanks very much. Great. Great. Well, thank you. David it was a great talk, very, very good work very exhaustive as a kind of certain beautiful images for sure. So now I have, there are a couple of questions in the chat so I'm gonna formulate them. So we have several questions from Professor Marla Feller and so the first one is it thought that you have all the bipolar cells in the volume and therefore these percentages are not biased by numbers of different bipolar cells that were reconstructed. Right, so the mosaics that I showed were not as complete as what we actually have. There are certainly some missing bipolar cells still and there's a little bit of a bias in which types are complete and which ones are not quite complete. So I wouldn't bet the farm down to the plus or minus 5% but it's pretty close to a complete mosaic for all of the types. So those numbers are pretty good. And I think they follow what would have been inferred for example from the published at or at all analysis the skeletonization of the volume without any synaptic content. So I think those numbers are pretty accurate but if you had a specific question where you really wanted it down to the level of precision that you may be after you probably wanna ask me about that but the other complication here is that you have elements that are coming in from the outside of the volume. So if I've listed as a skeleton it might be two pieces of the same bipolar cell coming in from outside the volume and I call those separate skeletons those get listed as separate skeletons that's gonna bias the numbers as well. But you can avoid those edge effects you needed to we could crop the limit and come up with pretty decent numbers. Okay, thank you. So we have a second question from Marla Feller. Is where all this connectivity associated with ribbons or other synaptic structures? Great question. And the point is very well taken because several labs in the past have identified what appear to be synaptic contacts between bipolar cells and post-synaptic partners both ganglion cells and amyloid cells that don't have obvious ribbons, ribbonless synapses watching you, Rachel's lab has just done a nice analysis of this herself. And so those definitely exist we did not include those in our analysis. The EM is really not great for those kinds of cellular synapses but maybe we'll have a better shot at it when we start segmenting Rachel's volumes. Thanks for this question. Yeah, okay, thank you. And have the last question from her as well. Is generalist might help for robust, generalist cells might help for robust feature detection. For example, on-off DSGCs are directly selected for a variety of stimuli, velocity, et cetera. That's what I'm more a suggestion than a question. So I missed the very beginning of the question. What was the first? Yeah, so it's not, it's more a comment that so meaning that generally cells might help for robust feature detection. Right, yeah, I think that's right. I mean, I think that, you know if you imagine the number of different sort of temporal kinetic streams that are working their way, you know to influence the response properties of the postsynaptic ganglion cell, you know a cell that's drawing from many of those different channels might be in a position to approach feature detection in a way that's much more sophisticated than one that's just really kind of peeling off one of those channels. And of course, all of these bipolar cells are under American cell regulation as well. So you might get contextual effects that would dial up or dial down certain bipolar drives according to the context, right? You have more opportunity to mix and match in a cell like the W3B than you would with the Delta cell, for example. Okay, okay, thank you. So there is a comment from Wei Li. He says that the accessory on processes of type nine often stem from the soma. We call those bestratified type nine, Captain Morgan cells. And we didn't get the joke, but I guess he was gonna explain later in the chat that who he said. It's like Spice Rome or something. I don't know what the reference is there, Wei, but... Yeah, we'll know later for sure. So we have another question from Rachana Deven Somayan. So for the bipolar axons that are bestratified, are there differences between their inputs in different layers in terms of their strengths? Maybe counting active zones in electron microscopy data would help. Right, yeah, exactly. The kinds of questions that we've been wondering about as well. I mean, I showed you that very sparse distribution of type six inputs to the M1 IPRGC, you might say, well, then the input's gonna be relatively weaker than you'd find with inputs to other gangmate cells where you have very dense bipolar drive. But if those synapses are more secure, that might compensate to some extent. For example, if you counted the total number of ribbons onto the M1 cell, it would be much less different than if you'd discount the number of active zones, essentially, or contacts. So we would suspect that those synapses might be unusually secure. They might also be well adapted to high rates of steady state release over long periods of time, which seems to be critical for encoding luminance at high light levels. So that might be part of the reason why you have this specialized accessory on channel with these monad multi-ribbon synapses. But there's really very little work done at this point on the physiology of those synapses. A pair of recordings would be fantastic. Shai Sabah has a paper under review right now, she's been able to image the on responses using iglysnifer in the accessory on sublayer. And they don't seem particularly sustained, but I mean, they're sustained enough to carry in luminance signal, but they're not unusually sustained compared to the inner-on sublayer. In fact, maybe less so. So I think we just need more direct physiological studies to really sort out what that really looks like. Okay. Now we have another question from Michael Reiser. He asked, what fraction of RGC's inputs are from cells other than bipolar cells? Does this vary substantially for different cell types? Right. Yeah, so the other cells that are gonna be providing synaptic input to the ganglion cells are amicron cells. And those are even more diverse than the bipolar ganglion cells that you've seen today. Or maybe on the order of 80, 100 types of amicron cells. We haven't really worked out numbers yet, but what I can tell you is that there are big differences among ganglion cell types in terms of the density of conventional amicron cell synapses, just as there are differences in the density of ribbon synapses, there are certainly huge differences in the identity of those amicron cells. And that's gonna be a big part of what needs to be fleshed out trying to get to a complete connectome of the ganglion cell population. But that's a heavy lift. There's a lot of cell types and many of them are wide field and it takes a long time to reconstruct those. And many of them have somas outside the volume. So the amicron cells are definitely going to keep us busy for decades to come. So related to that, I was wondering if there are specific amicron cells that connect only to the photosensitive, injusically photosensitive retinal ganglion cells? Yeah, there probably are, but I don't really have great data on that right now. There seems to be a bit of a network linking IPRGCs to specific bipolar inputs and to specific amicron cells. One of the cells that seems to be in the mix here is the N-MOS1 type. It's a bi-stratified type that earlier described, along with those weird outer stratifying things we now take to be the type 9, outer arbor, by John Demme and Josh Sainer's group some years back. So the N-MOS1 seems to be tied into the same kind of circuits in a major way and that VIP or VIP amicron look-alike cell that I mentioned briefly also seems to be heavily involved with the IPRGCs. They seem to be making intense contacts onto the type 6 shaft where those monad synapses are, but also onto the IPRGCs themselves. So it's a complicated mix, but it looks like they're going to be networks of interconnected amicron and amicron cell types within the luminance system. Probably true. Their justice would be for the directional system. OK. OK, that makes sense. So I have a last question from Jeff Diamond. Are the PSDs opposite the multi-ribbon contacts continuous, or do they appear to be distinct postsynaptic regions opposite each ribbon? Well, the EM is not that great. It looks like a big plaque of ribbons smashed as close together as they possibly can be, sort of a palisade of ribbons right up against each other. And then you see sort of one continuous postsynaptic density on the IPRGC, for example, or on an N-MOS1 cell or VIP amicron cell. But this is not the material to be asking those fine-grained questions, Jeff. What we really need are to re-examine the same synapses first with material, serial blockage material, like Rachel has been providing me, but ultimately with transmission EM, and maybe even eventually with electron tomographic methods that will let you see the structure of those synapses with much better resolution. I mean, we're groping around in the dark with the quality of the volume at this point. OK, great. So there's no any more questions in the chat. So we will finish the talk here. I'm still going to leave the transmission on for a while because there is a lack, as I told you before. But thank you so much to everyone for coming to the talk. And please join us in the Zoom link that I posted just in the chat so you can join us through that link. Actually, there are three people that have been waiting for a while. I just let them enter. Thanks so much, Antonio. Amber, thank you. OK, hello. Hi. Sorry for that flurry of questions. There had been some Zoom bombing that I found very unnerving. And so I just was trying to fill the chat with my questions. So good to see you. What a beautiful talk. Thank you. I was like trying to wrestle a sea monster. There had been some Zoom bombing that I found very unnerving. And so I just was trying to fill the chat with my questions. So good to see you. What a beautiful talk. Thank you. It was like trying to wrestle a sea monster. There had been some Zoom bombing that I found very unnerving. And so I just was trying to fill the chat with my questions. Good. Everyone turn off their links to the chat. Thank you. You know, I was like trying to wrestle a sea monster. There had been some Zoom bombing that I found very unnerving. And so I just was trying to fill the chat with my questions. People, you must have had your, whatever, Safari. That thing open. OK, Wei, explain the Captain Morgan thing. OK. Let me, if I can share this screen real quick. Can you see it? Yeah. So this is type nine. And this is the kind of a very bottom. And this is the accessory on that day I was talking about it from the Soma. And it coming down like a foot stepping on that barrel of, I guess, whiskey. So that's going to get from. I'm going to go with rum on whatever. Captain Morgan. Yeah, rum. Anyway, wow, that is random. Isn't that random? Am I the only one who thinks that's random? Nice way. All right. All right, maybe I share. No, I don't share. Gordon. Yeah. So are you trying to unshare way? Oh, there you go. Michael, nice to see you all. Thanks for coming. Maybe I can ask you. Everyone. So I was wondering when you have said these cells that are that that direct different layers is always the same proportion of our organization that you find many in one layer and in another one is very few. Or it's you find some that are equally distributed. So the first thing to understand is that for most ganglion cell types, you're going to be getting a chunk. You're not going to get the whole arbor. So I actually brushed over some real complexity in the DS story because most of what we have are fragments from outside the volume. So if you have a fragment coming in from outside the volume, it's in the on chat band. How do you know whether that's an on or an on off type? So what I ended up doing is looking at the cells, the few cells that we do have some is in the volume. I really know all of those processes belong to one cell and a real on off DS cell just looks like it has a different blend of bipolar inputs than a real on DS cell. And then I use that signature to try to sort out the fragments. So sometimes you get a cell that it's only a fragment of a cell that's only an off chat. So your first guess would be it's going to be an on off type because the ons don't really have that much in the off layer. But some of those actually have input patterns that look more like on types. So things like that happen. I'm sort of halfway to an understanding here. And it's also possible that the difference between the types that I'm seeing is not really so much on versus on off but directional subtype within let's say on off. So I haven't completely ruled that out and we're not really going to be able to tell until we have a bigger volume where we get all of quite a few of these cells of each of the known directional subtypes. I can tell you what the directional preference is based on starburst asymmetry for most of the fragments that I have. So that's going to help sort the whole thing out. But it's going to take more data to really be sure about that off piece. But to your general point of, in general, do you find the same stratification pattern for the same type? If it has a little bit in the off layer but mostly in the on layer, is that generally true? And the answer is within the limits of this really restricted space.