 Hello and welcome to the second season of Online Talks. It appears that worldwide we may have to get used to this format talk for a little while longer. So for the upcoming academic year we aim to organize bi-weekly seminars hosted either by us or the CIS exhibition group. Hello and welcome to the second season of Online Talks. It appears that worldwide ongoing talks are still part of the worldwide neuro-initiative. As you may have seen the number of participants increased quite a bit so there is usually more than one talk every day organized on this common platform which is for the most neuroscience field. I can only recommend you to take a look at the upcoming talks or subscribe to the mailing list so you will not miss any topic you might be interested in. You will find all relevant information in the description below. So back to us. My name is Maxime and I'm part of Combattance Lab at the University of CIS. Today I'm very glad to receive our colleague and neighbor as a lab that just separated by fly stairs, Leon Lagnado. Leon is the professor of neuroscience at the University of CIS. He did his PhD with Peter McNugton at the University of Cambridge and then worked as a postdoc in the laboratory of Denise Baylor at Stanford University. He then came back to Cambridge and worked for 20 years as a group leader at the MRC laboratory of molecular biology. Now he has moved south, Brighton, to the Center for Research and Perception and Pognition at the University of Sussex where his research focuses on understanding how the synaptic machinery of the retina contributes to the processing of visual signals. Good afternoon, Leon. Hi. Okay, so hello everybody. I've not done a Zoom seminar before so I hope this goes okay. Let me first of all get the right screen. Right, so as Maxime mentioned, we're interested in the function of the visual system in general and the retinal circuitry in particular and over a couple of decades or more we've been particularly interested in the contribution of synapses to the computations that the retina carries out and what I'd like to do in this talk is give three examples of how the plasticity of synapses contributes to the plasticity of the computations that the retinal circuitry as a whole carries out. Now I suspect most of you are very familiar with the landscape in which I'll be talking the retina but I'll just quickly run through some basics for those of you who might not be. Well, I would try to, right, here we go. Okay, so here's an example of a zebrafish retina and we can think of the signals flowing through the circuitry as straight through signals, the excitatory signals, photoreceptors to bipolar cells through glutamateric synapse and bipolar cell to gangrene cell. Most of the computations that we think of such as the generation of center surround receptive fields, contrast adaptation, detection of motion, signaling of object orientation and the variations in those computations are processes that go on in the inner retina, the inner plexiform layer where bipolar cells transmit to gangrene cells and where this transmission is modulated by inhibitory intranurons, the amicrine cells and we'll be looking at how the interactions between inhibitory neurons and the synapses of bipolar cells modulate signal flow through the circuitry. And I'm going to be talking about the plasticity of the retinal circuitry in three general contexts. Thinking of plasticity, first of all as a simple changing game and then secondly as changes in the tuning of neurons, that's to say what is, what feature of the stimulus are they responding to optimally? And I'm going to show an example of where gangrene scales can be retuned to optimally single signal objects of different orientations. And then lastly I'll be thinking about adaptation as retuning again but in this time retuning not of the neuron as a whole but of individual synapses and I'm going to show you how synapses of bipolar cells can be tuned between the on and off polarities, something that was very surprising to us and I'll be describing the mechanisms of that and potential context in which retuning of synapses between on and off channels might operate. Okay so let's first of all think about adaptation as changes in game. The visual system adapts the different features of the visual stimulus it adapts to the mean luminance for instance that's processes and photoreceptors but it also adapts to variations in luminance as we know contrast. So here's an example of a visual scene in the blue square and in the green square the mean luminance doesn't vary as a function of time as we scan those aspects of the scene but the contrast is higher in the green square compared to the blue square and it turns out that the visual system adapts to contrast and that adaptation begins in the retinal circuitry. And here's an example of how we set about looking at adaptation to contrast using an imaging approach. So most of the experiments I'll be describing use imaging as the fundamental experimental approach. So here we've got the retina of a larval zebrafish so we use zebrafish for our research because they've got beautiful big retinas you can see a larval zebrafish here they're relatively see-through we can put them under a multi-photo microscope and image different types of neurons that we can target genetically and here we're targeting the output neurons the retinal ganglion cells and we're looking at that activity simply with a calcium reporter a GCAP calcium reporter and here's some examples of outputs from the three different retinal ganglion cells initially when we just apply a step of light from darkness you can see a strong response and all three adapt to some degree so that's just adaptation to luminance we now apply high contrast stimulus in this case at 10 Hertz you can see this bottom ganglion cell doesn't much care about temporal contrast its response is pretty well sustained and at the top here we see what's been classically described over several decades a ganglion cell which initially responds strongly to the increase in contrast and then adapts as a decreasing gain so adaptation as a decreasing gain is something that we've all read about in the textbooks and the basic idea or thinking is that it's helpful to sensory systems because it prevents saturation it readjusts the gain so that if the contrast should increase again at some point in the future there's still some dynamic range with which to signal that increase in contrast but what we also found is a third type of ganglion cell which initially responds very weakly to temporal contrast but then gradually sensitizes so this is a form of adaptation where the gain change is now in the opposite direction the neuron is initially not terribly sensitive to contrast and then gradually sensitizes okay so we're particularly interested in getting into the circuitry within into the black box to understand how these input output relations arise from the neurons and the synapses are in the circuitry so i'll give you a quick run through now of how we disentangled what was happening here okay so we began to look at synapses in the retina using a reporter called sci-fi which is based on guero mesobox fluorine a gfp set of a ph sensitive gfp which is fused to synaptic vesicle protein an example of the kind of signals you can see with sci-fi as shown here so here we've targeted the report of the synapses of bipolar cells using a rib eye promoter and we can apply luminances different luminances here and you can see some synapses get brighter these are on synapses some get dimmer these are off synapses and so we're looking at vesicle release here using what is actually a very sub optimal reporter we could do much better to directly image a synaptic transmission and i'll get to that later in the talk but at the time that was the best we could do but we could push it along and when we looked at the output now from bipolar cells the neurons which drive provide the excitatory drive to the ganglion cells again we found we could find three basic types of bipolar cell output so in blue here is shown a bipolar cell which was relatively indifferent to a temporal contrast in black here we can see a bipolar cell where the output is initially strong and then it adapts as a decrease in gain it desensitizes and then in red here we've got a bipolar cell which initially responds to the increase in contrast very weakly and then sensitizes so we can see these opposing forms of plasticity now i'm not just in the ganglion cells that provide the output from the rest but also in the bipolar cells that drive them and what are the mechanisms what's going on here well it turns out i'll show you in a moment the depressing adaptation basically reflects a decrease in the number of vesicles available for release it's presynaptic depression as vesicle depletion the sensitizing adaptation is more subtle and what that involves is also synaptic depression but depression in the inhibitory synapses that control the gain of the bipolar cell output so let me show you some evidence for that okay so now again we're looking at the output from bipolar cells we're looking at the whole population but now our reporter is what we call Psi G-Camp this is a calcium reporter that we made by fusing a G-Camp to a vesicle protein synaptopheicin and this allows us to look with much improved signal to noise compared to cytoplasmic G-Camp specifically at synaptic signals so you can see on the right here a movie and now we're looking at synaptic changes in calcium in response to steps of light and again you can see on and off synapses in different layers of the inner retina okay now if you look at these synaptic calcium signals and you compare them in different layers of the inner retina that's shown here we can look at the output from bipolar cells and in parallel experiments not simultaneous parallel experiments we can also look at the output the synaptic activity of amicrine cells in the same layers and what we find is that the different forms of plasticity sensitization versus desensitization partition differentially in different layers of the inner retina so in the most superficial layers when we provide a high contrast of five hertz full field stimulus in this case you can see the majority of bipolar cells are sensitizing they respond weekly then the response gradually grows in that same layer amicrine cells are predominantly desensitizing they initially respond strongly and then gradually their response declines so it's very tempting looking at that correlation to think well what's happening here is that the amicrine cells are initially inhibiting the bipolar cell synapse strongly but then as the amicrine cell depress the bipolar cell is released from inhibition and one piece of evidence supporting that idea comes from directly manipulating inhibitory transmission using picrotoxin so here's an example where we're plotting the release rate from a population of bipolar cell synapses now and on average in the whole population shown by the black trace you can see the contribution of the desensitizing bipolar cells and then oh sorry and then the gradually sensitizing bipolar cells but now when we block inhibitory transmission with picrotoxin the sensitizing component disappears and we have as it were one way adaptation in the population that straight desensitization okay so here's the basic circuitry by which we can explain these two forms of plasticity depression in the retinal output reflecting a decrease in excitatory driving bipolar cells whilst sensitization shown in red here is reflecting the depression inhibitory inputs from the amicrine cell that control the gain of the bipolar cell synapse so why would the retina when we now look at populations of neurons why would it implement these two opposing game changes simultaneously in different neurons and that was looked at beautifully by Steve Bacchus who demonstrated the basic phenomenology here opposing forms of contrast adaptation but looked into how that affected the information that was transmitted out of the retina and basically what Steve demonstrated was that a mixed population of sensitizing and desensitizing ganglion cells did better at transmitting from the retina information about changes in contrast when those changes in contrasts were randomly fluctuating so in real life our visual system has to cope with increases in contrast decreases in contrast and they're unpredictable and it turns out that's a mixed population of ganglion cells which express these opposing forms of plasticity do a better job overall than a single desensitizing population at signaling both increases and decreases in contrast so we can basically think of depressing cells as encoding the higher contrast the strong signals and then the sense sensitizing cells as being getting ready to encode a decrease in contrast should that come along at some point in the unpredictable future and the basic circuitry I've described here which we analyzed in in zebrafish Mike Manukin has recently demonstrated also the basic ideas are still also operating up to the primate retina so evolution has held on to this principle of having mixed populations of neurons which adapt to contrast in opposite ways okay so here's an example of how we've been using imaging and reporters of synaptic activity to get into the retinal circuitry to understand the input output relation what we kind of generally term computations and this is an example where we've got adaptation as changes in game and what I want to now turn to is adaptations in the retinal circuitry but which now reflect not just simply changing how strongly a neuron responds it's game but changing what feature of the stimulus it is responding best to that's to say it's tuning okay and to look at that question we've been using a better reporter of synaptic activity than sci-fi intense glue sniffer so this is a reporter which is based on a bacterial glutamate transporter coupled to a circularly permitted GFP and when it binds glutamate the fluorescence increases it was actually initially the first prototype was developed in Roger Chen's lab but it was much improved in Lauren Luger's lab and we could target this reporter to different types of neuron in the retina so shown here are bipolar cells in dead rights up here receiving inputs from the photoreceptors outputs down here in different layers of the inner retina is there anything around here okay so and here's some examples of the kind of signals we can obtain with glue sniffer so the movie that's running here is showing the dendritic tree of a ganglin cell and we're actually stimulating it with moving bars and I hope if you look carefully you can see synapses on the dendrites being activated in sequence and you should be able to detect the direction of motion of the bar there at the bottom right here we've got a very narrow field amicron cell almost certainly glycinergic amicron cell and I hope you can see hot spots there on its dendritic tree reflecting individual synaptic outputs so this report is wonderful very high signal to noise great temporal resolution and it's opening up all kinds of questions for us the question that I'm going to be looking at next came from experiments with Jamie Johnston who was a postdoc in the lab at the time and we asked a very simple question which is how does the output of a ganglin cell the signal it delivers to the optic tecton in a zebrafish how does the output depend on the excitatory inputs that it's receiving so we used glue sniffer to image both the output of a ganglin cell in the optic tecton I think this is labeled the wrong way around here's the ganglin cell and here's the output in the optic tecton and in the b here what we are looking at is a raster plot of the glue sniffer signals delivered from a whole bunch of ganglin cells just over 100 and the stimulus here is one of them is a moving grating okay so classical stimulus and all we're doing is changing the orientation of the grating from vertical to horizontal vertical horizontal and you can see some of the ganglin cells at the bottom here for instance they don't much care about these changes in the orientation of the grating but many respond particularly strongly just when there's a transition when the orientation of the grating changes and that's shown more clearly in c there so here we're looking at the outputs from a bunch of different ganglin cells in green are examples of ganglin cells that don't care about orientation completely untuned in red here we've got examples of ganglin cells where the output is sensitive to orientation but it's so called statically tuned so when we go from horizontal to vertical in this example we've got a small well let's use this example we've got a response when we go from vertical to horizontal in both cases but we don't get a response with the transition from horizontal to vertical and you can contrast that with these examples in black these are ganglin cells which generate a response every time the orientation changes so yeah if we look at this guy here this guy signals um sorry this guy here is signaling both the vertical to horizontal transition and the horizontal and the opposite one okay so basically what we've got here is a ganglin cell which is first of all tuned to respond to the vertical that response then is snuffed out it adapts very strongly to that orientation of stimulus but then it becomes retuned to then signal the transition to the opposite orientation okay so as soon as we saw this we immediately thought of a wonderful paper from Marcus Meister's lab with Hesoya and with Bacchus where they demonstrated that the retina of actually a number of species um could adapt to changes in orientation that's to say they were ganglin cells that were tuned to orientation but then dynamically changing their orientation selectivity dynamically retuning to optimally signal a different orientation okay so this is this is a form of predictive coding in the sense that the neurons aren't interested in unchanging stimulus they're primarily signaling a change in the stimulus but not only are they signaling a change in the stimulus they're retuning so that they can signal changes between different orientations okay so called dynamic predictive tuning so as soon as we saw these signals at the outputs of ganglin cells we thought okay well we're in a position here potentially to try and understand the mechanism that generates dynamic predictive coding of orientation and a starting point was a model actually that was proposed within the same paper they actually proposed a couple of models one we think is the correct one and the basic idea was that you could make a ganglin cell that retunes to different orientations of activities by providing it with inputs from here they're called inch and neurons but we're talking about bipolar cells inputs which are intrinsically tuned to signal different orientations yeah but then changing the weighting of these different inputs so for instance if you apply vertical stimulus as shown here this input will adapt the gain of that synapse will reduce and the notion is then should the stimulus change to a horizontal orientation what will happen is that this other input which has not adapted will become dominant and so the ganglin cell will respond again to that transition in orientation and I'm going to provide evidence that that model is indeed correct and I'm also going to show you how it needs to be modified to account for some of the important aspects of the dynamics with which adaptation occurs and the dynamics with it the dynamic predictive aspect is readjusted okay so what's going on right okay so the first question is are there so-called pattern detectors in the retina do the inputs from to ganglin cells really encode the orientation of a stimulus you know we've we all know about orientation selectivity in the cortex you know from the work of Hubel and Wiesel and so on and Marcus and his group were the first to demonstrate that that kind of computation began already in the retina where in the retinas it began so what I'm going to show you is here direct evidence that it begins in the synapses of bipolar cells so here again is a raster plot and we're looking at the synaptic activity of a whole we're looking at the synaptic inputs I should say to different ganglin cells so I hope you can see the pink and blue dashes here that's showing you different ganglin cells and again it's the same stimulus we're transitioning from vertically orientated moving grating to a horizontally orientated drifting grating and you'll see some of the synaptic inputs the ganglin cells are indifferent to the changes in orientation but there are others such as those here which are very strongly selective for one orientation over the other and some of these examples are shown more clearly in gray here so at the top at the top here we've got a ganglin cell that doesn't care about orientation at the middle here we've got a ganglin cell which is select it responds to both orientations but it's clearly selective to one over the other here we've got a third example which only signals one orientation of the bars oh sorry I don't have full control over my tools here okay so if we now compare so we do indeed have orientation selective bipolar cell outputs how does that orientation selectivity arise and we consider two kind of basic mechanisms that might be at work one model one shown here is potentially that the receptive field of the bipolar cells are themselves are not circularly symmetric but are skewed such that the bipolar cell output becomes selected to the orientation of the stimulus okay so that's an intrinsic mechanism that would generate an orientation selective output the second model is one potentially in which the intrinsic receptive field of the bipolar cell is circularly symmetric but it receives inhibitory inputs which are determined by the orientation of the stimulus so here's an example which in which the inhibitory input would be activated by a horizontal stimulus yeah which would make this bipolar cell output signal vertical stimuli more strongly okay and it turns out that it's actually a combination of both these intrinsic mechanism and extrinsic inhibition that accounts for the orientation selectivity in the output of bipolar cells although to varying degrees okay so how do we get at that first of all we mapped the receptive fields of individual synapses again our assay for activity is glutamate release measured with glue sniffer and we reconstructed receptive fields using a we didn't use the kind of classical white noise checkerboard type of stimulus we used an approach based on filtered brat projection which we published a few years ago and allows us to use imaging to reconstruct receptive fields much more quickly than a flickering checkerboard and it turns out that if you look at the receptive fields of individual synapses they're very often not circular they're often elliptical and here are two examples and in this histogram we're looking at the distribution of ellipticity you know zero would be a perfectly circular receptive field very fewer like that they're all skewed to some extent basically but the degree of ellipticity that we measure in the receptive field of the synapses isn't sufficient to account for the degree of orientation selectivity that we measure when we compare the outputs to different orientated stimuli so that's shown in D here in D here we're using a linear model to predict the degree of orientation selectivity based on the receptive fields that we've reconstructed and in red we're showing the actually observed distribution of orientation selectivities in the output of bipolar cells and you can see the intrinsic mechanism just isn't strong enough to account for how selective bipolar cell outputs are okay so let's go to the second potential mechanism inhibition and what we're looking at now again is the activity in bipolar cell synapses but measured using the calcium reporter now and what we noticed was that in many bipolar cells they responded not just to their preferred orientation but they were also inhibited by the orthogonal orientation immediately telling us that these bipolar cells were being inhibited by interneurons and that this was determining the degree to which they became orientation selective so here are a couple of examples the orientation selectivity indexes are shown to the right here in some cases we can't see the net inhibition even though they're perfectly orientation selective but in these top two cases we can so to test this idea we did an obvious experiment we used pharmacology to block inhibition and what we find is that this strongly affects the distribution of orientation selectivities that we can measure in bipolar cell synapses so in black is the control distribution you can see that the orientation selectivity is actually falling to two clumps there's a subset of bipolar cells in which the output is very highly orientation selective and that subset is completely lost when we inhibit you know disrupt inhibition as shown by that red distribution another way to to look at the same thing is to look at how the distribution of preferred orientations is altered by this manipulation so in in black here we show the distribution preferred orientations between 0 and 180 degrees for bipolar cell synapses in blue is the same experiment we're looking at amocrine cell synapses so of course this is a key substrate here for this mechanism that the inhibition might itself be orientation selective and you can see that amocrine cells are to don't signal orientations equally and again we lose a large degree of the distribution preferred orientations now becomes flat in bipolar cells when we disrupt inhibition so natural inhibition it's not just a question the pattern detectors that we're talking about to generate this dynamic predictive code reflects a combination of intrinsic asymmetric receptive fields and then a contribution of inhibitory inputs from amocrine cells that more strongly tunes these pattern generators okay so that's important how are the pattern generators being generated but the second key kind of extra level of detail that we were able able to go into in terms of understanding this form of dynamic predictive coding was in terms of the kinetics okay so let me elaborate on what I mean so what we're showing here on the right is a comparison of the output of a ganglion cell in the optic tecton that's shown in black as we switch the orientation of the grating you can see this is an example which very strongly encodes both the vertical and horizontal switches in either direction it's very strongly dynamically retuning but what we've got in red now are the excitatory input so that's same ganglion cell and what you can see is a couple of features first of all most importantly we're looking first of all I should say just to the inputs tuned to the 90 degrees and although the output from the ganglion cell adapts very rapidly and completely these inputs only adapt partially and much more slowly okay so there's an important part of understanding this computation and that's its dynamics what is causing the net output from the ganglion cell to adapt so much more rapidly and more completely okay and it turns out that this is a feed-forward inhibition that there's a high pass filter operating within the inner retina and that that high pass filter is generated by amicron cells which feed forward onto the same dendrites that these excitatory inputs from bipolar cells connect to. Just before I show you the evidence for that let me just tell you what I'm showing you in the middle here what I'm showing you in the middle here is an I'm trying to give you an idea of the variety of different combinations of inputs that different ganglion cells can receive so what we're showing at the top here we've got four examples here's a ganglion cell in which the inputs are just completely selectively tuned to the vertical orientation yeah here we've and the average of all these inputs is shown in gray okay and you can see here very clearly also the time course of adaptation adaptation is a simple decrease in gain due to pre-synaptic depression here's a ganglion cell in which the inputs are untuned then the ganglion cells which will dynamically predictively code according to the pattern generator model are those that will receive mixed inputs tuned to different orientations and two different examples are shown at the bottom here so if we look at e here you can see different synaptic inputs tuned to different orientations and the average is shown in gray at the bottom you can see that this average of all excitatory inputs whatever the orientation is again far removed from the net output shown in black is delivered to detector okay so let's show you some how we got to the conclusion that feed forward inhibition was the basic mechanism that determined these kinetics and that was based on modeling so in a separate set of experiments that Jamie did electrophysiological experiments looking at inputs into ganglion cells in the goldfish retina we were able to look at the kinetics both of excitatory inputs to dendrites and inhibitory inputs to dendrites and what we're showing here in B is experimental based on experimental measurements where we're looking at how the kinetics of the net excitation into a ganglion cell varies as a function of the strength of inhibition that it also receives so in black is the input to the model based on the filters that we've measured experimentally and you can see the stronger the inhibition the more transient the net excitatory signal and the more complete the adaptation and so we considered two models one in which this form of feed forward inhibition modulated the excitatory inputs to ganglion cells that's model one shown here and we compared it with another possibility which neglected the notion of feed forward inhibition and simply postulated that the there was a thresholding non-linearity that acted on the excitatory signal transmitted from the bipolar cell to the ganglion cell and an example here from one ganglion cell we're comparing the data shown in gray here with the predictions of the model model one in red and model two in blue and it turns out that the model that incorporates feed forward inhibition does a much better job of accounting for the dynamics of the adaptive effect and the extent of the adaptive effect so putting it all together this is the circuitry that we believe can account and explains the data and can also account for the computation the dynamic changes in orientation selectivity it's one in which we have bipolar cells in in which the synapses show orientation selectivity so here's one tune to the vertical here's one tune to the horizontal these bipolar cells this is something we've not directly tested but we think it's most likely these bipolar cells drive amicron cells that then become orientation select if we've demonstrated amicron cells do indeed become orientation selected and then by a process of lateral inhibition the intrinsic tuning of the bipolar cell generated by its receptive field becomes sharper at the synapse at which the amicron cells act so that's the orientation tuning but then in addition to account for the kinetics we also believe that we have feed forward inhibition from the bipolar cell through an amicron cell to create a high pass filter okay so that's an example of plasticity by retuning the output neurons the ganglion cells and I've tried to explain to you what we understand of the synaptic basis of that retuning what I now want to get onto is a project that's been driven by Sophie Sable in the lab and this is again another example of predictive coding and based on dynamic changes in tuning but now we're talking about one of the most fundamental aspects of visual processing which is the splitting of the signal into on and off channels and what I want to show you is that these on and off channels actually don't run in parallel through bipolar cells but that bipolar cell synapses can become retuned from on to off okay so that's going to sound a bit crazy let me show you the evidence okay so we're using glue sniffer again and as I mentioned I love glue sniffer here's an example of a bipolar retina we've got a bunch of bipolar cells and we've got them sparsely labeled and this is actually important sparse labeling allows us to home in on individual synaptic compartments and it minimizes background so we can really improve signal to noise and we improve signal to noise also by collecting photos not just through the objective but also through a condenser and if you do these things I'm going to show you now a line scan through a synaptic compartment time goes upwards each frame is a is a we're scanning at a kilohertz each frame here is 100 milliseconds okay nothing much happening here across this line and now we're stimulating actually that was spontaneous activity yeah now we're stimulating with actually relatively high contrast stimulus here and you can see these flashes of light these are glutamate release if we now look at the intensity profile across this line as a function of time that's plotted at the top here what we can see actually there are two epicenters for these explosions of glutamate release which we think reflect two different active zones and we can use a very simple Gaussian fitting and demixing to these intensity profiles to independently measure the events of these two active zones so we've got the red active zone at the top here active zone two and the black active zone at the bottom here active zone one and what you can see are signals of very high resolution this is a a five hertz stimulus actually wasn't as high that was high contrast in the movie in this example it's only 20 contrast and there's a number of things we can see here first we can see when we look at these two active zones from the same synaptic compartment sometimes they both respond to the same cycle of the stimulus sometimes they don't synapses are noisy we're looking here at a fundamental property the stochastic nature of vesicle release but the other incredibly interesting thing is that the amplitudes of these events varies enormously and it turns out that the amplitudes of synaptic events changes because of so-called multi vesicular release okay so that's the kind of resolution that we can obtain with glue sniffer and that's why I love it but let's let me get back to tuning of synapses is between on and off here's a simple experiment now where what Sophie did is she scanned her line not through one synaptic compartment but through a bunch so here's a bipolar cell and you can see it's got a number of compartments in different layers of the inner retina and here I'm showing a a movie of activity in these different compartments this is at relatively lower resolution to a hundred hertz and what I'm plotting here now is the activities in these different synaptic compartments in response to a step increase in intensity actually the way we did this experiment is we're using a one hertz square wave and here I'm just looking at the average response to one cycle you can see this synaptic compartment generates an off response which is rather transient this one here releases glutamate both at the on phase when intensity increases and the off phase when intensity decreases this compartment is clearly an on an on response which is you know it's got a transient component as well as a sustained component and then this last compartment here again is purely on with a very large sustained component so we've got different outputs from one bipolar cell varying in their kinetics and also in their polarity so here's one example here's another example now where the axon of the bipolar cell isn't extending through the different layers of the ipr but is projecting laterally through one layer and again we can see in this example different outputs we've got an off output in compartment one on output in compartment four and so on and we actually see this mixture of outputs in in about 50 percent actually perhaps closer to 60 percent of the bipolar cells that we look at which have more than one recognizable compartment and so here are just some examples of the different morphologies of the axons in which we can see what we're calling synaptic multiplexing on and off channels on and off signals being transmitted through the same neuron multiplexing okay what the hell's going on here how how how can one bipolar cell transmit an on signal through one output and an off signal through another synaptic compartment let me show you the model and then I'll show you what it's based on okay so here's the notion here we've got a bipolar cell it's got two different synaptic compartments in different layers of the inner retina here's the ganglion cell onto which it's connected and let's just say this is a bipolar cell which is intrinsically on okay so on outputs to both dendrites of the ganglion cell what we believe is happening is that one of these synaptic compartments is receiving inhibitory input from a specific type of amicron cell glycinergic amicron cells these these have much narrower receptive fields than GABAergic amicron cells so when when this bipolar cell is excited so is this amicron cell and it inhibits this synapse and what that is doing is activating HCN channels okay hyperpolarizing hyperpolarization activated cation channels so for those of you who aren't familiar with these channels these are distributed in synaptic compartments primarily in the retina they're found in photoreceptor synaptic compartments but they're also an important feature of synaptic compartments of bipolar cells and these channels are cation selective but they're open when the membrane hyperpolarizes so whilst this amicron cell is active these channels open and then when this inhibition is removed they're still open and you get a rebound excitation they're cation selective okay and this whole thing is enabled by dopamine okay so that's the model let me run through some evidence okay so here's an example where we're looking at the actually it's the one at the role of inhibition okay so here we've got an axon here we've got a proximal synaptic compartment a distal synaptic compartment and what we're showing here is first of all the control response of both these compartments in black you can see these generate both on and off outputs okay we inhibit glycinergic inhibition with strict need and we block the off response maintaining the on response so we think this is an intrinsic on bipolar cell in which the gaba urgic in sorry the glycinergic inhibition glycinergic is switching some of the synapses from on to off you can compare this and I think this is a really interesting comparison with the effects of playing around with gaba urgic inhibition if we block gaba urgic inhibition with gabazine you can see the gain is increased we've got larger responses both to the on and off phases but it is to both the on and off we don't have a clear switch in polarity so kind of generalizing beyond this it's looking to us like the glycinergic amicron cells actually have very little role in gain control they're not really changing the game they're switching polarity whereas the gaba urgic inhibition is involved in basic gain control that first mode of adaptation I described at the beginning of my talking in the context of contrast adaptation okay so that's the role of inhibition time this is really important let's look here at how the rebound response varies as a function of duration of the on step this is a 0.25 second stimulus 0.5 one second you can see the longer light increment lasts the stronger the rebound response okay so this again is a hallmark of dynamically predictively coding you're retuning the synapse as a function of time it's accumulating information about the present state of luminance and it's becoming retuned now to signal a transition in the opposite direction okay what are these integrators how does this integration happen the integrators we believe are hcn channels and the the time course of the retuning we believe is primarily reflecting the time course of opening of these channels so here's an example where we're specifically blocking hcn channels with this block of zd7288 okay and if we look at the proximal synapse we can see here that the rebound of response is completely blocked there's something else changing here as well in this example we can see that when the light intensity increased there's actually also net inhibition in this synapse which is also removed again another piece of evidence for for the model that I propose at the start um finally let me show you how this mechanism can be modulated on a longer time scale um as I mentioned a lot of the actions going on in the inner retina and actually if we go from the inner retina and look at the bipolar cell terminal these synaptic compartments I think it's very clear a kind of one of the richest computational elements in the in the retina they integrate signals from different types of inhibitory into neuron they transform signals because of their intrinsic electrophysiological properties I didn't haven't gone into how these terminals can generate calcium spikes for instance but there are also sites of longer term control through neuromodulator so many bipolar cell terminals for instance have expressed dopamine receptors they actually express receptors variety of neuromodulators and one of the channels to which which dopamine receptors can modulator HCN channels basically they enable the activity of these channels by increasing cyclic Np levels and if we get back to the phenomena that I'm describing here uh if we antagonize D1 receptors um sorry activate D1 receptors here in this experiment we're activating rather inhibiting you can see examples of synaptic compartments in which the rebound off response is enabled by the activation of D1 receptors this this kind of puts this mechanism in a kind of really interesting context so on the short term we've got tuning between on and off channels in a sub-second time scale but then the actions of dopamine is now opening up the possibility that this mechanism can be modulated on longer time scales according to factors such as slower changes in luminance and the circadian cycle for instance and actually we have preliminary evidence that their circadian control of the number of on and off synapse is active in the zebrafish retina kind of more generally more generally placing this mechanism under control dopamine also allows it to be altered by what is kind of vaguely termed internal state which basically means uh uh situations in the case of a zebrafish such as hunger uh so for instance in another study a few years ago we demonstrated how um dopaminergic signaling was modulated by our factory inputs into uh in a zebrafish so here uh we've got to I think an entrant a fundamental mechanism that uh not only will allow for uh predictive coding that's emphasizing changes from a recent stimulus property but it's also placing that under longer term control and we can also I think think of this as a mechanism that helps overcome the bottleneck in the transmission of visual information imposed by the limited number of neurons and synapses effectively what's happening here is we've got all these on bipolar cells yeah they signal when the light goes on and what the retina is doing is when the light goes off instead of leaving those on synapses dormant it's repurposing them it's saying hold on I've got all these synapses not doing anything I'm going to switch their tuning so that when the light intensity goes down they can enhance my ability to signal that change so retuning of synapses this example thereof it'll be really interesting to look at how how that impacts on net information transmission through the circuitry and the output from ganglion cells and I think that's one thing that we're going to have to kind of look at in the future okay I've tried to show you different forms of retinal plasticity first of all and the circuit basis of different forms of plasticity and I've tried to show you how changes in gain enable adaptation and how those changes in gain can operate in different directions I've tried to show you how individual neurons can be retuned both by intrinsic mechanisms within synapses causing synaptic depression but then also by the local circuitry and finally I've shown you how new synapses themselves can become retuned and then how these different mechanisms can be put under a longer term control of a neuromodulator okay and I mean I think the final thing I have to conclude with is you know I've given you a kind of little smorgasbord here how do we kind of get these into a coherent overview and I think what we're going to have to do in the long term is to understand how these circuit mechanisms ultimately affect behavior the behavioral consequences potential advantages but that of course will itself depend on the external conditions in which the animal is operating okay so finally let me tell you about the people who are involved in this work I've got a wonderful group of folk that I work with and they're shown here they're two people I particularly want to highlight they are Sophie Helen Sable so Sophie uncovered the synaptic multiplexing through individual bipolar cells the retuning from Montauv and Jamie now he now he's running his own lab at the University of Leeds and he has contributed to much progress but particularly to the project looking at dynamically retuning ganglion cells the different orientation selectivities and yep thank you all very much thank you Liam that was a very nice talk I'm talking to our audiences I want to join us I can just click on the link shared on the on a chat so they can come and ask question or discuss this topic with us should I be looking at my YouTube now don't worry I'll do that for you so we have a very nice question from Anton who's asking if it's possible that the filter back projection algorithm artificially generate an elongated orientated receptive field in case of orientation selectivity cell so I think it depends on how many for people who don't know about this algorithm the basic approach is is kind of uh relatively we assume we have a receptive field of a particular size um Maxime I'm getting some feedback what do I do feedback on okay okay yeah okay okay so we probe there's a receptive field and we probe it with bars of different orientations and we look at the response to those bars of different orientation and just using a linear model we try to reconstruct the receptive field in which those bars were acting on and it's certainly true that you need to sample a fair number of different orientations we we used eight different orientations um uh to reconstruct these receptive fields and if you used fewer say four different orientations round the clock you you you might well kind of get dodgy outputs in terms of uh but I don't think it's a major concern I mean when we looked at the uh degrees of electricity um um they they were pretty uniformly distributed we didn't kind of get uh say peaks which might reflect the eight different orientations that we used for instance I I can't think of a reason why uh I didn't I didn't see anything to worry about that let's put it that way enough um I have a question from Charlson which is a bit controversial um what defines the cell subtype within the bipolar cell population um well that that would be the glutamate receptor type so so so you know if you measure signals in the soma with a with a sharp electrode uh you know there's there's no or a patch electrode there's no uh um problem uh an on cell is an on cell it's always an on cell it doesn't generate off responses an off cell is an off cell it doesn't generate on responses and that's just reflecting the two different glutamate receptors on their dendrites the action is happening the action is happening when the current flows into the synaptic compartments at the end of the axon and I should say I perhaps should have said that that this mechanism seems to basically only operate in on bipolar cells we can't completely rule out that we might have switching of the synapses from an intrinsically off bipolar cell to on but um uh we haven't convinced ourselves so let's put it this way we haven't convinced ourselves of examples of that and uh the vast majority of the ganglion cells the bipolar cells that we look at that do this switching are intrinsically on um I got another question from Anton our general is dynamic predictive coding did you try other stimuli like direction of motion or spatial frequency how general is it um we um I guess I don't know of course the way to know is to do the kind of experiments that Anton suggested um we haven't looked at how for instance frequency tuning might change um uh certainly there's uh uh um changing in receptive field sizes I mean that's uh that's a feature of uh adaptation to luminance but I also think that I'm trying to pin down the papers now that uh the distribution of spatial frequencies uh in a fluctuating stimulus that that uh uh also perhaps somebody can help me here uh also uh can retune receptive field sizes in ganglion cells I can't I've got to see the chat let me let me open the chat it's not that wrong I mean if Anton if you're listening to us please join in the meantime I can ask Marla to join us she has a couple of questions I see she's with her in a room so Marla if you want to ask yourself your hello hi Marla oh my god Leanne it's still dark here oh thank you it can be nice to see you Marla very nice to see you uh the imaging was spectacular the single bipolar cell terminals so um I had a question on the orientation tuning if in one of your slides it looked like that when you looked at the distribution of the bipolar cell tuning that there was an over like a dramatic dramatically more tuned to zero and 180 and very few that were tuned at 90 if I I might not have seen that plot correctly I lost you for a moment there Marla could you just repeat the question I lost you for a second sure that is there an over distribution is there an over representation of horizontal orientation been vertical in the bipolar cells and is that also true in the ganglion cells yeah um uh um so we so when you look in the ganglion cell output we have these experiments it's actually pretty small it's about 27 ganglion cells but uh Martin Mayer has much more comprehensively looked at the distribution of orientation selectivities and the outputs of ganglion cells uh in zebrafish and um uh they're actually the distribution has three peaks as I recall Marla um yeah the bipolar yeah I'm thinking of the one slide you showed of the bipolar cells it seems to be there are two peaks uh vertical and horizontal no hold on let me just let me get the data I thought it was zero and 180 but maybe uh um can you see my are you seeing my screen now I stop you have to reshare it okay I'll just quickly reshare it Marla uh share screen and there okay um so uh if we look at the preferred orientations Marla yeah um uh looks like zero and 180 there yes it's zero and 180 it's um it's vertical over that's vertical okay so they're preferred and is that so is that but and very few horizontal then so is that also true in the ganglion cells I guess no question no no so uh uh no in the ganglion cells as I recall there are three populations and the the the peaks are separated by roughly 120 degrees I should double check that but that's what I recall from Martin May's work so um so you're wondering how to get those different so he did this by measuring calcium signals in the ganglion cell axon that's how he surveyed the ganglion cell outputs you know who might be able to fill in here is Anton is Anton still on he's been looking at this he's enjoying this yes oh okay oh is this this is the breakout bit yes but you should be warned we're still being taped I have made this mistake already yeah right exactly okay but that's a really good question Marla how do you how do you create how do you create these different uh uh outputs um oh okay oh it must be a combination of different degrees someone has to turn off their uh yeah anyway all right you can take other questions lovely talk Leanne thank you thanks thanks for that um I see that Anna from Thibbingen is in a chat and she had a question for you if you want to ask it directly yeah sure um I'm interested in the second part of your talk um when you talked about uh the sparsely labeled bipolar cells that were showing both on and off responses in different parts of their kind of axon whether you've checked to make sure that the um signals you're seeing kind of further closer to the somas of these on bipolar cells is not just spillover from the off layer just having lots of glutamate in it during times when the light is off um um so spillover as a component of these signals is something that you know we obviously worried about and uh there's a few different pieces of evidence that um that tell us that these signals are very little affected by spillover so uh one of them um I showed in the talk which is if for instance we do a line scan through a single compartment yeah uh what we can find within there are foci uh uh which we believe are individual active zones which don't always respond in synchrony you that the the kind of stochastic aspect of transmitter release you can see by the fact that one of these foci might respond on one trial and the other one won't and they're they're often separated just by a couple of microphones so that's one piece of evidence um you know if we were looking at spillover I don't think you would get nearly that degree of local heterogeneity um another piece of evidence is um if we uh killed the bipolar cell that we're recording from at the cell body so we're recording you know we record the glucinophysignal in its synaptic compartment then we use the same layer and blow out its cell body uh so that the you know intrinsic activity of that bipolar cell disappears uh so do the signals that we measure uh so uh you know which should still be there if those signals were affected to any degree well you know we should get some kind of uh you know um uh um spillover signal so I'm not saying spillover doesn't happen but I think it must be much too slow or too small to uh infect what we're measuring on these timescales at least um yeah you could block the off the uh off bipolar cell response right with canate receptor antagonists and see if you still see that yeah that's another absolutely that's another thing we could do you know um I showed other kind of pharmacology manipulations which block HCN channels for instance and uh how these have differential effects on different compartments of the same uh bipolar cell uh again I think that's hard to account for if these signals are strongly affected by spillover um but one of the sorry I was just going to say presumably any pharmacology you do affects the whole retinal network and so if what you're measuring is an is a signal coming from a neighboring off bipolar cell that's releasing glutamate and then you're picking it up on your kind of glutamate antenna that you have running through the retina then you would manipulate that with any pharmacology that you do but I think Barlow's experiment is a good one because um that that should abolish the the any off signals that are coming from directly from off bipolar cells yeah that's absolutely uh yeah a nice experiment and a doable experiment but I mean yeah I mean uh I thought the ablation experiment was pretty good as well yeah it is pretty good as well although I guess I don't know do you affect the integrity of the eye glue sniffer if you blow out the soma that's good I guess I don't know yes I mean that I don't know but you're right the the more evidence or the more angles by which you can test this the better but you're right the k-nate is as uh is so fion I'm wondering if you say fion I'm sure we did experiments with ap4 but I can't remember there is some I'm afraid but you're right so you're particularly worried that there might be a form of spillover which is uh specifically enhancing off responses I think is that right Barlow I mean to kind of put it in a nutshell yeah okay so perhaps we should do the ap4 experiment well I'm sure we've done it actually I can't I don't want to report it because I can't remember what the traces look like um yeah either of those would be a great way to show that the on pathway is generating an off response right and it's not coming from the off pathway so ap4 would be great too I think it would be easier to just image calcium and then it's not spillover whatever happens yes yes it's amazing that expressing a calcium indicator is somehow easier than just putting on a drug an experiment that's already working a very different view of the world so so there's a bit of a problem with that I mean yes you can do that Tom one thing is that these rebound responses are really very fast and uh Sophie has done experiments with Sophie are you there yeah I can see you um hello Sophie talk to us I can't hear you can you get muted hi Sophie can you hear us yes now I can hear you sorry I just joined so Sophie with the calcium reporters which you've done some experiments with hello yeah just give me sorry I have still youtube running and I hear funny funny feedback and you're doing an experiment I can see yeah I try to multitasking it's not going well though uh so yeah I did some LAP4 experiments but that has been early days and the results were so diverse basically I couldn't I wasn't sure if the drug I can't uh I injected was strong enough um so I basically what I'm trying to say I cannot tell you anything about the outcome of these experiments um yeah okay so we should try them again Sophie probably I could yeah no uh Leo this is so Weili here great talk hi hi Weili quick question so uh hcn channel at least in the uh mammalian bipolar cell they often also expressed in the dendritic tip uh when you use zd to block them uh do you worry about the impact on the uh on the dendritic site um so I mean of course I mean it's something that we have to think about um we and yeah I think probably if we look closely at the traces there's changes in the kinetics uh uh um of the responses um but the the effects at least I'm reporting here Weili were very kind of uh uh qualitatively yes or no you know um uh the rebound responses were either there or not there so um uh so I think there's this qualitative change superimposed on more subtle kinetic changes in the responses uh when they are there is that is that what you mean that you it expects changes in the kinetics uh I think that that makes sense that uh if if there's modulation from the uh bipolar input side it will be more global that's what you said I guess yeah that makes sense yeah thanks I mean I've got to say the one of the experience that I found kind of most intriguing was when Sophie blocked lysinergic inhibition which had a very clear qualitative effect of blocking the rebound whereas playing with GABA-ergic inhibition had uh uh just increased the gain uh which I which I kind of found you know really intriguing I hadn't quite kind of thought about the relative roles of GABA-ergic versus lysinergic inhibition in gain control um because one one thing is that the lysinergic ones are vertical usually right and the GABA-ergic ones are horizontal so many of the GABA-ergic ones can't connect you on and off layer which is what you need yes thanks Tom that that so I hadn't quite thought of it but are they really most of the lysinergic amicron cells are quite uh that's sort of the view on the stage but then of course there's species differences and all of that yeah yeah yeah did you try to measure the receptor fields of different components are they different no Anton that would be great absolutely it would be really nice to not just compare the kinetics I mean but also polarities but also spatial properties absolutely and that's kind of on our list of things to do yeah yeah all right I will just finish with one question from Tim Gollish and then after I will close the session so Marla can speak openly lovely to see you Marla oh no she's leaving already okay so I have one last question from Tim Gollish who's asking is the retuning of hypercell synapses is layer specific and have you looked yet at what ganglion cells are doing for example under HCN channel block okay so okay so that's that's a really good question um there's an absolute rule there's an absolute rule in terms of layers um and I need to show this I think I really need to show this there's an absolute rule uh uh when you compare proximal synapses let's say proximal to the cell body versus distal synapses so let me show it with this six uh yes so I do share again don't I yeah share screen uh yeah uh okay so can people see that so this is this is this is the um uh oh no that's not a good example uh okay okay so so here's an example so um um we're comparing here the effect of the HCN channel blocker on the synapse that's proximal to the cell body versus the synapse that's distal and what you can see is that um uh the distal synapse generates both on and off responses the proximal synapse only generates the rebound off response the HCN channel blocker selectively uh uh on the distal synapse blocks the rebound um and this is this is a kind of general pattern um that's to say that um uh uh we think this mechanism is primarily acting on the synaptic compartment which are most proximal to the uh cell body does that make sense to people so you know one possibility one possibility is that the distal synapse is activating that glycinergic amicrine cell which then feeds back to the proximal synapse uh that's the route by which the inhibitory signal opens the HCN channels I think that basically chimes with your finding that the off the intrinsically off bipolar cells don't tend to do much in that in that realm right it's always the ons so the direction of signal flow is basically from up to from from the lower IPL to the higher IPL so from on to off so yeah that's right Tom so you know initially we thought this might be a mechanism that could in principle operate in either on or off but uh uh uh you know when you see mixed responses you have to have a good criteria for identifying the intrinsic polarity of the off of the neuron um and uh uh uh one of the best ways that we found to do that is is to look at uh is am I still sharing my screen guys uh no oh sorry um is to look at uh uh which response is varying as a function of time as a function of the duration of the stimulus so you know here's an example where I'm showing that the rebound response takes time to grow it grows on a sub second time scale and uh on that basis in this example we think this is an intrinsic on bipolar cell which generates a rebound uh off response um if we just go with that operational definition of the response to stimuli of different durations we haven't yet found any convincing examples of cells in which it's the on phase it's the on response that is growing as a as a function of time yeah so um and I think when I say no convincing examples I think that's from a sample of uh 25 bipolar cells 23 are clearly show this qualitatively this behavior uh uh uh we haven't found any convincing examples where it's the on phase that that grows as a function of the preceding off phase um yeah so there was one paper uh in uh so basically a red study where they um did antibody staining unfortunately there are no antibodies available for zebra fish but what they did show show in that uh in on bipolar cells you can find in all sub types basically HCM channels but in just one do you hear me okay do you hear me yes okay cool but just uh one off bipolar cell subtype is labeled with HCM channel the antibodies so they are rare in reds and I saw post probably the same situation that's happening in zebrafish that's interesting okay yeah I guess I guess the other thing is this weird thing that in some species and zebrafish is included you've got these bipolar cells that clearly stratify in both traditional on and off layers right and you don't really have that as much in for example mammals um but if you look at these weird bipolar cells the ones that stratify all over the place they tend to be M-glue R6 ones right so there's this old paper from from wikikonautom I think who basically puffed glutamate and and check that so all the long ones are basically actually on cells that happen to grow in off terminal um at some point in the evolutionary past and then I guess mechanisms got invented that turned it into an off terminal um okay yeah that is quite interesting uh thanks everyone I will just close the session now for the people who are still listening I hope to see you in two weeks and we'll have another talk thanks for organizing Maxine sorry over and thanks all guys Hannah and even Steve