 those of our vision talk series. For those who are maybe not a year away yet, we are part of the worldwide in Europe. It's a COVID-inspired seminar on the platform for neuroscience. Talks are organized there every day on a broad variety of topics. Most of them are already available as podcasts. So take a look at the website and subscribe to the newsletter if you don't want to miss any talk you or your colleague might be interested in. I will also tell you this opportunity to talk to you about a new platform that InvoGirls and Thanos Bozellos, the founders of Worldwide Neuro just created, is that Worldwide Universe. It's a platform which is actually an academic jobs board and to disseminate listening from both employers and job seekers. That means that in addition to a traditional board disseminating listening by employers, only curious fellows like in terms of the existence of Bosdok can advertise their ability on the market too. Those both job seekers and employers can benefit by this hybrid model in traditional plus reverse way. So to make it more appealing and useful, they have also implemented a smart matching functionality. So if you opt in for it, your listening will be displayed with your nearest scientific neighbors. Technically, your peers can be a postdocs or PIs who might work on topics very similar to your own interests. That way you can get more ideas on who to consider as your next employer or job advocate. So please check it out. Links are in the description below and just consider taking the time to submit your own listening. It only takes five minutes. Now for our today's talk, very glad to receive Martin Kermans from the Netherlands Institute for Neuroscience. Martin up in his PhD from the University of Amsterdam in 1985, sorry, 1989, sorry. Then moved to the US for a postdoctoral position at the University of California at Berkeley with Frank Rublin. And then he went back to the Netherlands. He received the Royal Netherlands Academy of Hearts and Science Fellowship. He then established the Retinal Signal Processing Lab at the Netherlands of Thermal Research Institute that he later moved to the Netherlands Institute for Neuroscience. He is now a full professor in neurophysiology, specialized in sensory physiology as the University of Amsterdam. His research group focuses among other topics on how are these cells inhibited photoreceptors and what consequences this has for vision. So hello, Martin. And thank you for accepting our invitation today. Okay. So I have to share my screen now, I think. Yeah. And I will send you. Okay. Okay, so first of all, I would like to thank the organizers for giving me the opportunity to present the work of my group on this platform. So we will take a long journey today and we will start discussing synaptic transmission in the outer retina. So where photoreceptors and bipolar cells and horizontal cells interact. We will look at detailed interactions between horizontal cells and photoreceptors. In the second part of the talk, we will discuss how these specifics of these synaptic connections influence vision and in special influence color vision. So let's start at the beginning. So basically the first indications that there was something interesting going on in the outer retina came from the paper by Werbelin and Dowling in 1968. That is showed that when you project a spot of light on the retina, photoreceptors hyperpolarize, horizontal cells hyperpolarize and bipolar cells hyperpolarize. However, when you present an analyst photoreceptors don't respond, horizontal cells still respond and bipolar cells depolarize. This experiment showed that there was a center surround organization already present in the outer retina. So in that sort we are going to discuss today the mechanism of how that comes about. So everybody agrees that for photoreceptors project to horizontal cells via a gluteometric pathway. However, how horizontal cells feed back to the photoreceptors is a matter of debate. In the 80s of the past century everybody was convinced that GABA was the neurotransmitter. In 2001, we proposed another mechanism, an aseptic mechanism. And in 2003, Akika Neko and colleagues proposed that horizontal cells used protons to feed back to horizontal cells. Since then basically there was a big fight which mechanism was the real feedback mechanism. What I would like to do today in this talk is to discuss a synthesis of those three mechanisms. What I will propose is that all three mechanisms are present and have specific functions. So the aseptic mechanism is one of the fastest inhibitory systems known and is most likely involved in spatial redundancy reduction. The pH buffer mechanism is relatively slow and is involved in temporal redundancy reduction. And the GABA-ergic mechanism is very slow and modulates feedback signal during light-dark adaptation. So, but let's start to look, have a closer look at this feedback signal from horizontal cells to cones. When you record from a cone in a voltage clamp condition, you see a standing inward current. And when you flash on a spot of light, then you see that this inward current is reduced and it remained reduced as long as the light is on. And then when the light is switched off, the response reappears, of the standing inward current reappears. However, when we keep the spot on and a full field stimulus in addition, we see this small inward current. This is a small current, look at the scale bar. This is 10 picofamps and this is 100 picofamps. So the first thing we did was look at the nature of this current, which channel generates this current. The result was that it is a modulation of the calcium current in the photoreceptor. So this is the calcium current of a photoreceptor when feedback is not active. When we activate feedback, the calcium current is shifted to negative potentials and there is a slight increase in amplitude. So the first results, this is completely different than what you would expect on my GABA logic mechanism. And indeed, this shift of the calcium current is independent of GABA, but we'll come back to that later. So this was found in goldfish. However, since then, many, many people found this similar behavior in different animals, in goldfish, in nude, in macaque, in mouse, and in salamander. So this modulation of the calcium current is a very conserved feature of negative feedback from horizontal cells to currents. Recently, we also found that this shift of the calcium current occurs in cultured human retinas. So when you look, have a very close look at the feedback response measured in photoreceptors, you can see that this feedback consists of two processes. A very fast one and a very slow one. The fast one has a time constant of about 30 milliseconds and the slow one, a time constant of let's say 200 milliseconds. So this was also found by other people. This is by the group of Wally Tronson in the salamander. So in summary, feedback from horizontal cells to cones shifts the activation potential of the calcium channels in photoreceptors. This is conserved over all vertebrate species tested so far. There's a fast one and a slow feedback component. However, how does this fast component work? The mechanism we proposed for this fast component is inspired from the work of Pesov and Sura Bura, which was published in 1986. So to understand this mechanism, we have to look at the cone photoreceptor synaptic terminal in detail. So in red, you see the cone synaptic terminal is the synaptic ribbons over here and vesicles alongside the ribbons. So there are many ribbons in the photoreceptor. These ribbons are aligned along the synaptic ribbon and the glutamate release occurs here, very close to the horizontal cell dendrites illustrated over here. And these processes are the bipolar cells. So one thing you can notice is that this synapse is a kind of enclosed compartment. The photoreceptor wraps around the dendrites of horizontal cells and bipolar cells. And that will become a very important label. So to understand how negative feedback from horizontal cells to cones works, we have to make a little sidestep and talk about connexins. Connexins are the proteins that form gap junctions. So gap junctions consist of two so-called hemichannels, which are docked and in that way, connect two cells and create an ion channel which connects the intercellular solution of both cells. Those are the gap junctions. However, in the rare occasions, hemichannels can also occur as a single hemichannels. And the key feature of those hemichannels is that there is only labeling or you can only find them in one membrane and not in the opposing membrane. Various in gap junctions, you find connexins in those membranes. So in a collaboration with Ray-to-Wirela, we looked at the localization of connexins in the Goldfish retina and we found that horizontal cell dendrites express connexins at the tips of the dendrites. But it was only present in one membrane, not in the membrane of the cone and not in the membrane of the bipolar cell. So there were hemichannels. So how could hemichannels be involved in negative feedback from horizontal cells to cones? So let's, I will talk you through this. So calcium channels are located at the near the synaptic ribbon in the cone photoreceptor. Calcium channels are voltage sensitive and regulate the calcium influx according to the membrane potential they sense. The calcium influx leads to a glutamate release. Basically, the calcium channels are a kind of voltage meter that measures the membrane potential and adjusts the release of glutamate. Hemichannels are expressed at the tips of the horizontal cell dendrites. Hemichannels are basically just holes in the cell, are not specific iron channels and therefore we can exchange them by a resistor. So in the dark, when horizontal cells rest at minus 40 millivolts, current will flow through the hemichannels into the horizontal cell. This current has to come from outside the complex convoluted synaptic structure. This convoluted synaptic structure, the extracellular space in this convoluted synaptic structure will not be zero. So current through a resistor will lead to a voltage drop. So just the current that enters the hemichannels will lead to a small negativity deep in the synaptic cleft, let's say minus 10 millivolts. This has major consequences for the calcium channels that we are sitting here. They will sense now the membrane potential of minus 40 minus minus 10 is only minus 30 millivolts. So when the hyperparadise horizontal cells by light, for instance, they will hyperparadise to minus 80 millivolts. That will lead to an increase in current and therefore the increase of the negativity deep in the synaptic cleft. So for the calcium channels, this is a huge difference. Now the voltage they sense is minus 40, minus minus 20 is only minus 20 millivolts. So hyperparadise horizontal cells leads to a local depolarisation of the photogenic cell. This is a negative feedback pathway. So we did a lot of pharmacological experiments to show whether this was indeed the case and we could confirm that in all those experiments. I'm not going to discuss these experiments in detail. What I will show you is the results of what happens with feedback when we make a knockout animal without the connection in horizontal cells. So this is the calcium current in control condition in green. When we depolarise horizontal cells, we scanate the calcium current shift to positive potentials. And when we hyperparadise horizontal cells with d and qx, the calcium current shift to negative potentials. This is what happens in the wild type animal. However, when we do the same experiment in the animals which don't have the connection at the dendrites of the horizontal cells, we see that the shift, the hyperpolarisation induced shift is absent. So directly showing that the connection hemichannels are mediating negative feedback from horizontal cells to curves. So the mechanism we propose, the septic mechanism, is purely electrical. And that would mean that there should not be a synaptic delay between the horizontal cell response and the feedback response. So we set out to determine the synaptic delay between the feedback response and the horizontal cell response. To do that, we used the following methods. So we modulated the horizontal cell membrane potential. This is sinusoidally. And then measured the feedback response in the cones. And as soon as there is a delay between the... And we determined the delay between the two sinusoidal responses. This delay would be the synaptic delay. And you can do this for many different frequencies. And we did that for many different frequencies simultaneously. So we made the stimulus consisting of many different frequencies. We added all those stimuli together. And this is the stimulus, complicated stimulus. This is the response of the horizontal cells and the feedback response of the feedback response. And using Fourier transform, you can calculate the gain function and the phase function. And the phase function basically gives you the delay. There are two things to notice. First of all, there's no additional filtering going on for higher frequencies. And secondly, there's no delay, but no change in phase at high frequencies. And then you do that. When we look at very closely at this, you can calculate that you can determine that the synaptic delay is not significantly different from zero millivolt. So in summary, the fast component of feedback is mediated by hemichannel. This might be one of the fastest inhibitory systems. What about slow component? For that, we have to go to an older paper of Barnes and Buie from 1991. They showed that the current of the calcium channels in photoreceptors are strongly influenced by the pH. In an alkaline condition, the calcium current is shifted negative. And in a acidic condition, the calcium current is shifted positive. So there is a shift of the activation potential, but there is also an increase in amplitude. This modulation of the calcium current can be due to so-called surface charge effects, or could be a direct action of protons on the calcium channel. So Aki Kaneko looked at this further in the 2003 paper. What they did, they measured feedback responses in the nude photoreceptors, and the same way as we did, but then they applied the high delts of hippies. That's a pH buffer. And what they found was that the feedback responses could be blocked by a high dose of hippies. And there was a recovery. So this effect of hippies is, again, a very general effect. Basically, everybody who tested it in vertebrate animals found that hippies blocked feedback. It's in nude, in a mouse, in goldfish, and in salamander. So the feedback seems to depend on pH. So in the Rich Kramer lab, it was the first to show that there was indeed a pH change in the synaptic left. So what they did, they made an animal, a zebrafish, which had a pH sensor hooked up to the calcium channel. So with that pH sensor, they could measure the pH in the synaptic left. And what they found was that the pH changed, indeed, in the synaptic left, and the time constant of this change was very slow, about 200 milliseconds. So could it be that the slow feedback component is the pH-dependent component? So one of the things that led us to believe that was that the time constant of the slow component is very similar to the time constant of the pH change Rich Kramer finds. So we looked in this a little bit further, and the first thing we did was we looked at how the slow and the fast component depended on the Konexin-55 HEMI channels. So when you remove the HEMI channels, what you see is that the fast component amplitude is reduced strongly, whereas the slow component amplitude is hardly affected. So the slow component is independent of HEMI channels. However, the slow component of all of feedback can be blocked by carbon oxalone. Carbon oxalone is a gap junction blocker. So you couldn't do this, you can block feedback completely with many different gap junction blockers. However, gap junction blockers are not that specific. For instance, gap junction blockers also block Ponexins. So what are Ponexins? Ponexins are a class of membrane proteins with sequence homology to Konexins. Contrary to Konexins, Ponexins only form HEMI channels. They don't form gap junctions. And furthermore, Ponexin channels are implicated in ATP release. So could it be that Ponexin channels are involved in the pH mechanism, the pH feedback from horizontal cells to cones? So the first thing we did was we labeled the zebrafish retina as an antibody against Ponexins, Ponexin-1 in this case. And what we found was nice labeling in the synaptic terminals of the photoreceptors. So this is the outer plexiform layer. And these dots are the synaptic terminals of the photoreceptor. When you look on the EM level, you see that the Ponexin-1 labeling is around horizontal cell bedrugs. The next thing we tested was whether horizontal cells are capable of releasing ATP, and whether this ATP release was mediated by Ponexin channels. So for that, we dissociated to horizontal cells and measured using fluorescent methods the ATP release. When we depolarize horizontal cells by AMPA, you see that there is an increase in ATP release. And when we block the Hemi-channels with Probenesit, which is a specific block of Ponexins, we saw a decrease in ATP release. So this indicates that horizontal cells are capable of releasing ATP. So I don't have time to discuss that, but we tested whether the ATP acted on Ponexin-1 receptors. But that in this case, the Ponexin-1 receptors are not involved. So could ATP release change the pH in the synaptic cleft? So this is the structure of ATP. And we all know that ATP can be hydrolyzed and that ATP can lose phosphate groups. And that's done by an enzyme called NTPDase. It's an extracellular enzyme which is present as an extracellular enzyme. And so ATP can be hydrolyzed to adenosine. So you end up with adenosine and you end up with phosphates, groups and protons. And basically this is a phosphate buffer. And this phosphate buffer has a pKa of 7.2. So that means that if you hydrolyze an adenosine of ATP, basically you create a phosphate buffer that pushes the pH to 7.2. And adenosine is further degraded to adenosine by the enzyme called ada. So the next thing we did was to look whether these extracellular enzymes are present in the synaptic cleft. And what we can show you here is the terminals of the photoreceptors in green. This is the localization of the Panexin channels. Here in green, the localization of the NTPDase that hydrolyzes ATP, which overlaps with the Clouard II labeling, which labels horizontal cell dendrites. The same holds for ada, which is also present in the synaptic terminal of the photoreceptors and also is close to the approximation of the Clouard II labeling. Remember, these two enzymes are extracellular enzymes. They are present in the synaptic cleft. So how does this ATP release lead to a negative feedback pulse? So this is the hypothesis we have. So we have Panexin-Hemmy channels, Panexin-Hemmy channels, and calcium channels. So in the dark horizontal cells rest at minus 40 millifold, and due to the faptic interaction, the potential in the synaptic cleft will be about minus 10 because of the current flowing through the connection channels. At the same time, ATP is released by Panexin channels because Panexin channels are open at the depolarized state. ATP is released in the synaptic cleft with the two enzymes, NTPDase and ada, the hydrolyzed ATP to ison and phosphate buffer. And this phosphate buffer makes the synaptic cleft acidic and inhibits the calcium channel. Furthermore, the phosphate buffer will diffuse out of the synaptic cleft. And so we will get a steady state of the creation of the phosphate buffer and the diffusion array of the phosphate buffer. All these processes together will set the activation potential of the calcium current in the dark. Now we hyperpolarize horizontal cells. The faptic current increases, makes the potential deep in the synaptic cleft slightly negative, and that leads to the shift of the calcium to negative potential. However, in due time, the buffer will start diffusing away from the synaptic cleft, and that will make that the pH in the synaptic cleft becomes more alkaline. And that leads to a further shift of the calcium current and an increase in the calcium current. So the fast feedback mechanism isn't the faptic mechanism and that shifts the calcium current to negative potential. The slow feedback mechanism is a pH buffer modulation and that leads to an amplitude increase and a shift of the calcium current to negative potential. Again, we did a lot of pharmacological experiments to prove this hypothesis, but the most convincing, the most direct way of showing this is making an animal that does not have the Panexin channels. And that's exactly what we did. However, CBERFISH has an additional gene duplication, and that meant that there were two Panexin-1 channels, Panexin-1a and Panexin-1b, and so we had to make two knockout animals. But as you can see, when you knock out R to Panexin-1a or 1b, the feedback response reduces. In black, we have the wild-type response. In red, we have the response in the mutant. The double mutant is also reduced and if we knock out Panexin-55.5 in the Panexin, almost all feedback is gone. So when we look a little bit closer to the feedback responses, you'll see that at least there is an indication that the reduction due to the deletion of the Panexin affects especially the slow component of feedback. Feedback becomes very fast in the knockout. Okay, so you might have noticed that I talked about changing in buffer capacity in the synaptic terminal of in the synaptic cleft. So let me explain why that is so important. So this is the synaptic terminal of a photoreceptor in the goldfish and a number of years ago, we set out to determine the exocellular space in the synapse. So we measured the exocellular space and calculated its volume and its volume is about 0.88 cubic micrometer. This volume contains about 38 protons. So based on the work of Steve Barnes, you can calculate that for every millivolt of shift in activation function of the calcium current, you need to have 0.1 pH unit change. That means that for a feedback response in the physiological range, 19 protons have to move. So if the system depends on the release of protons and the uptake of protons, it becomes a very noisy system which has only a very small dynamic range. As soon as you modulate the buffer capacity, those numbers of protons become statistical properties of the system and in that way you can move, you can change the number of protons nicely in a graded manner. So the only way this pH mechanism can work in a stable and low-noise condition is if you change the buffer capacity instead of releasing protons or taking up protons. So in summary, the slow feedback component is modulating the pH in the synaptic cleft. The pH change is due to a change in buffer capacity. ATP release via AT-1-HEMI channels creates the pH buffer in the synaptic cleft. So what about GABA? So there is general agreement that horizontal cells contain GABA. That's found in the fish here, in the goldfish. At least one type of horizontal cells contains GABA, the H1 cells. And it is found in guinea pig. But also there is evidence that mouse horizontal cells and red horizontal cells contain GABA. There's also consensus that cones and horizontal cells have GABA air receptors. So this is worked from Tachybana and Caneco, where they puffed GABA on the synaptic terminal of the photoreceptor and got a very nice response. We showed it in the goldfish. Goldfish cones have GABA-A conductance, which reverses nicely at the equilibrium potential of chloride. Furthermore, we showed that horizontal cells have GABA-A receptors. That's in salamander. Also, we're currently nicely reserved versus around ECL at the sparcotoxin sensitive. And Steve Barnes showed recently that also in mouse horizontal cells you have a GABA-A receptor. So horizontal cells release GABA and cones and or horizontal cells have GABA-A receptors. So what does this GABA-ergic mechanism do? So again, there is great consensus of the effect of GABA on feedback responses. So for Wayne Schnapp's showed in 2003 that when you apply GABA, the feedback responses measured in the photoreceptor of macaque monkey is reduced. When you apply pachrotoxin, the feedback responses increased. It showed the same in goldfish. So GABA inhibits negative feedback and pachrotoxin, blocking the GABA receptors, enhances negative feedback. And the same was found recently by the laboratory of Steve Barnes in the mouse rat and guinea pig. So in general terms, GABA seems always to enhance the feedback of macaque monkey. So in general terms GABA seems always to inhibit negative feedback and pachrotoxin seems to enhance negative feedback. So although there seems great consistency in this, there is one thing we have to keep in mind. The non-marmadian horizontal cells release GABA via GABA transporter. The mammalian horizontal cells release GABA via vesicles. So we looked a little bit further in this issue of GABA. So what GABA is doing, so we recorded from the cone photoreceptor and applied GABA. And what you see is that there is a current appearing with a reversal potential around ECL. When we apply pachrotoxin in the endogenous condition where we did not apply any additional GABA, you see that also the current is blocked which reverses around ECL. And finally, when we block the GABA transporter, which in fish is essential for the release of GABA by horizontal cells, we see that a conductance which reverses around ECL also closes. So this means that in the fish retina, horizontal cells release GABA and that this GABA opens agloride conductance in photoreceptors. So horizontal cells feed back to the cones via GABA allergic mechanisms. How then can it be that we did not see this in our light-induced feedback responses? So for that, we wanted to see whether we could find any evidence that horizontal cells modulate the GABA a conductance in photoreceptors in a light-driven manner. So we clamped the horizontal cell at minus 100 millivolts, set ECL very positive at around minus 10 millivolts, and then when you do this, you see this current. This is in the dark. Then we applied a long light stimulus. And as you can see, basically nothing happened with the GABA a conductance in the photoreceptors. From this panel, we know horizontal cells project to the photoreceptors via a GABA allergic mechanism. However, here we show that hyperparalization of horizontal cells by light does not lead to a change in GABA release and does to a change in the GABA conductance in the photoreceptor. So this led us to the idea that maybe the GABA release is modulated only very, very, very slowly. And indeed, there is a lot of evidence for that. So one part of the evidence is the following. So when I was in the Wirbelin lab, we looked at horizontal cells in the salamander. And we noticed that when you apply 100 micromolar GABA, the response shape of horizontal cell responses became exactly the same as the dark-adapted response shape. This is the fast horizontal cell response in the light-adapted condition. This is when you apply GABA or it could be exactly the same when you have a very dark-adapted retic. So it seems that GABA release is very high in the dark-adapted condition. So work of Steve Jezula also indicates that. When you take a retina and measure the GABA release and you put the retina in the dark, you see a slow increase in GABA release. When you put the retina back in the light, you see a very slow decrease in GABA release. So GABA seems to be modulated in a very slow manner. And finally, an old work of O'Brien and Dowling that show that the GABA release by horizontal cells is under dopaminergic control. GABA for dopamine inhibits GABA release via second messenger system in horizontal cells. So dopamine release is high during the day and therefore GABA release is inhibited during the day. Dopamine is low in the dark and therefore GABA release is high in the dark-adapted condition. So that would mean that negative feedback from horizontal cells to cones is inhibited during the dark-adapted condition because GABA inhibits negative feedback. And indeed, that's what has been found by a number of people. And here I show you an example of the work of Wagner and Jammelkos. So in the fish retina you have three types of horizontal cells. The monophasic horizontal cells, the bifasic horizontal cells and the triphasic horizontal cells. These are responses to different colors of light. And the bifasic horizontal cell, the H2 cell, depolarizes to red light stimuli. However, when you record from an H2 horizontal cell in the completely dark-adapted retina there's no depolarization. These depolarizations are purely driven by negative feedback from horizontal cells to cones. So in the dark-adapted condition, no depolarization and that means that there's no feedback. When we light-adapt the retina you'll see that the depolarizations become very prominent. So in the dark-adapted condition negative feedback is weak and in the light-adapted condition negative feedback is strong. So we came to the following hypothesis. So in the light-adapted condition we have the dopamine release is high, GABA release is inhibited and feedback from horizontal cells to cones is fully functional. And that functions partly via the effect mechanism and partly via the panaxin-mediated mechanism. In the dark-adapted condition however, GABA release is high although the dopamine release is low, GABA release is high and the GABA will open a GABA conductance. In this case in the photoreceptor. What happens now is that the photoreceptor GABA receptor will start supplying current to the synaptic cleft and this current will flow into the hemichannels and basically shunt the effect mechanism. So all the current needed to enter the horizontal cell is now supplied by the photoreceptor and there's no need anymore for to recruit current from outside the synaptic terminal. That means there's no modulation anymore of this the current through these resistors and no effective interaction. So basically the GABA receptor shunts the effect mechanism. Interestingly this it doesn't matter where the GABA receptor is whether it is on the cone photoreceptor or whether it is on the horizontal cell dendrites. In both cases the GABA receptor can do the same. So in summary there are three the three proposed feedback mechanisms working synergy a fast, effective feedback a slow pH buffer feedback and a very slow GABAergic modulatory mechanism. The GABAergic mechanism is presumably important for the reduction of feedback in the darker depth condition. The GABAergic mechanism works via shunting of the effect mechanism. But other mechanisms like H2O3 flux through GABA receptors may also be involved in the modulation of feedback. So what's the function of this complicated feedback synapse? So this is the view out of my window in the laboratory in Amsterdam. And basically it is the extremely boring view. Nothing is happening. You see occasionally a car coming by and there's a windmill spinning. But most of the time nothing happens both in space and in time. So the official system is basically built to remove redundancies. And there's a lot of redundant information in this sea both in space and in time. So what we propose is that the spatial redundancy reduction is mostly done by the effect feedback mechanism. Because it's very fast why should the spatial redundancy reduction need a very fast inhibitory system? So what the spatial redundancy reduction is it measures the global activity and subtracts that from what a certain cone sees. If this was very slow then the surroundings of this response would lack the center. So for spatial redundancy reduction you need a fast system. This system needs horizontal cell integration because it has to estimate the activity over space. Temporal redundancy reduction on the other hand is due to the Ponexin 1a ATP feedback mechanism. It's slow because it has to estimate the activity over time. Therefore it needs to be slow. And one interesting thing is that this ATP feedback, Ponexin feedback mechanism might even be relatively local in the tips of the dendrites of horizontal cell. So I told you that there are two components to feedback, an effective feedback mechanism and a Ponexin feedback mechanism. Are those components present to the same extent in all animals? So what we expect is that the animals with high visual acuity like zebrafish have a strong fast and effective feedback component because they are highly they are they have a big need for spatial redundancy reduction. However, animals with low visual acuity like mice have a strong, a stronger slow pH buffer dependent feedback component. So, however, this needs to be looked at very carefully. I think it's time to start comparative physiological experiments to see how the various feedback components depend on the environment the different animals live in. But what is the functional consequence of this feedback as a whole? So this is of course something you all know that they have gray dots here a gray dot in a light environment looks darker than a gray dot in the dark environment whereas all the dots are equally gray. So this is due to horizontal cell to cone negative feedback. So when there is a lot of activity the center will be inhibited when there is a little activity the center will not be inhibited. However, there's something weird to this. Now we have a flickering spot of light and you see it barely flickers in a dark environment. So, I will switch on the surrounds in a minute but what we expect is that when we activate the surrounds that will inhibit the center that will reduce the flickering. However, what we experience is that the flickering becomes more vivid. Now you barely see the flickering and now it's really obvious that the spot is flickering. So what's going on? To understand what's going on we have to look at the relation of cone photoreceptors and ganglion cells and the specifics of the feedback signups. So cone photoreceptors show a very nice sustained response. These are really DC neurons. The memory potential is completely dependent on the amount of light the cone sees. Ganglion cells on the other hand most of them respond with very transient responses. The consequence is that the ganglion cell is hardly interested in the sustained change in memory potential. They are very, very interested in the change in cone memory potential. So far we have discussed negative feedback from horizontal cells to cones only in relation to the sustained memory potential changes. Now let's see what happens if we consider what happens with a change in memory potential. So this is what I already showed you. This is the calcium current without feedback. When you activate feedback the calcium current shifts to negative potentials. This is an expanded version of this part of the calcium current. Cones hyperpolarize from minus 40 to minus let's say 50 that leads to reduction in the calcium current. When negative feedback shifts the calcium current you see an increase in the calcium current. So reduction in glutamate release increase in glutamate release. This is a negative feedback pathway when we consider the sustained memory potential. However if we modulate the memory potential of a cone slightly around its dark and resting memory potential you will see that that will lead to a big change in calcium current because the slope of the calcium current is very steep over here. When we hyperpolarize the photoreceptor to let's say minus 50, the slope reduces. So the same modulation of the memory potential here will only induce a small change in calcium current. So hyperpolarization of a cone reduces the synaptic gain. When we shift the calcium current to negative potential what happens is that synaptic gain, the slope of the calcium current increases again. Feedback from horizontal cell increases the synaptic gain. So can we see this? Can we measure this? So we did the following experiment. We modulated the memory potential of a cone in conditions where there was no feedback and when feedback was active. So this is the memory potential change in a voltage clamped cone and this is the resulting current when we clamp a cell at minus 40 mqt and this is basically the calcium current. This is in condition without a horizontal cell activation when we hyperpolarize horizontal cells you get this result and this is the calcium current with hyperpolarized horizontal cells and you can see that it has increased tremendously. Also you can see this effect at horizontal cell level. This is a horizontal cell response to a flash of one second flash of light of one second. Early on the feedback is low because only the effect feedback works whereas later on the response feedback is high because both the effect and the ph feedback are active. That would mean that in this condition the synaptic gain is low and in this condition the synaptic gain is high. What we did was we stimulated the horizontal cells now not with a flash of light but with a flash of sinusoidally modulated light and as you can see that here is that the amplitudes of the sinusoid of the response to the sinusoid increases directly showing that the synaptic gain increases. This is important and it will be important for the rest of the talk. What I would like to do is to consider what happens when we project some kind of stimulus onto the photoreceptors. So we have a kind of random stimulus over here let's say this is the response of the photoreceptors to this random stimulus. This is the response of the photoreceptors. That response is sent to the horizontal cells and the horizontal cells integrate this signal and basically come with this response to the right one which is basically the average signal of the photoreceptors. That's the predicted signal. Basically the horizontal cells send a signal back to the cones. This is what you would see if there was no detail in the stimulus. Horizontal cells there is this subtractive mechanism this negative feedback mechanism to the cones. So the signal that's sent to the bipolar cells will basically be just a smaller signal which is just a smaller signal. Now the mean activity has been subtracted. However I've shown you that also the synaptic gain changes. So we don't have only the subtractive activity for interaction but also this multiplicative interaction enhances the signal again such that it is scaled properly for the dynamic range of the bipolar cells. So this is a form of predictive coding. So horizontal cells estimate what the response of a photoreceptor would be if there was no detail in the scene. That signal is sent back to the photoreceptors and subtracted from the activity and the details that remain are amplified by this multiplicative interaction. Okay, so so far we have basically discussed cones only discussed cones. However there are three types of cones, red, green and blue cones. I told you that horizontal cells feedback to the cones and that this interaction is a kind of subtractive and multiplicative interaction. Now the question is do red cones receive a red feedback signal? Green cones, green feedback signal? And blue cones, a blue feedback signal? Or are the spectral inputs of the various photoreceptors already mixed at this level? So what we are going to for one of the major features of any color vision system is color constancy. And I will show you what color constancy is. So you'll see a red apple over here. And when I put the blue filter on that apple we get a really blue apple and nobody would consider this as a a good pick from that fruit basket. However when I expand this filter over the whole scene you will immediately recognize this apple as a red apple again. And this is instantaneous. It's not an adaptational stuff thing. It's immediately there. So this is color constancy. So color constancy is the ability of the visual system to perceive colors rather constant despite of considerable changes in the spectral composition of the illuminate. So how and where is color constancy generated? In primates the first color constant neurons are found in V4 and therefore people argued that color constancy was calculated in V4. However all animals with color vision tested so far on color constancy are color constant. So it seems to be a fundamental property of any color vision system. So the group of Kerstin Neumeyer tested this in goldfish. So they trained goldfish to swim to colored objects. So there was a training field and there were blue objects and yellow objects. And after a long training session the fish learned to swim to the training test field. This was when the fish, when the scene was illuminated with white light. Then they applied blue light and you can calculate that the blue light that the spectral output of the test field under white light is similar to the yellow test field. This yellow test field under blue light. So if the fish was not color constant it would swim to this test field. However goldfish were color constant. So this suggests that goldfish which don't have a V4 are color constant. This was not only true for blue light but also was true for yellow light. So how does this work? So we were intrigued by this finding and this mechanism how this could work. And so we started to record the spectral sensitivities of photoreceptors. So these are flashes of light of different colors and this is different intensities. When you look at this closely you see that the various responses that you can exchange intensity and wavelengths. So a certain intensity and wavelengths basically generates a response similar to another wavelengths and another intensity. And when you connect all the points of equal amplitude with each other you get a so-called spectral sensitivity curve. When you do that for many cones you can calculate those curves. Those are the spectral sensitivity curves of the L, M and S cones in the goldfish. The blue cones to green cones and the red cones. Now the question was do the various cones do the blue cones receive a blue feedback signal the green cones, the green feedback signal red cones or red feedback signal. So we did the same experiment but now for the feedback responses again you can connect the points with equal amplitude with each other and obtain the feedback spectra. And what we found was that the feedback spectra were completely different than from the the spectral sensitivities of the cones themselves. They were always very broad. So now we have to go back to what I told you before cone hyperpolarization reduces the synaptic gain whereas feedback from horizontal cells increase the synaptic gain. So this is important. So let's assume that we illuminate a scene with a lot of red light. For the blue cone the blue cone itself is not sensitive to red light. So its synaptic gain will not be reduced due to direct light stimulation. However it will get a lot of feedback from the red part of the spectrum. So that leads to an increase in synaptic gain. So with a red, a global red illumination the blue cones become more sensitive. The red cones on the other hand are strongly stimulated by the red light and therefore they reduce the synaptic gain. They also increase the synaptic gain slightly due to negative feedback but the effect of the hyperpolarization is much larger than the effect of feedback. So global red illumination reduces the synaptic gain of L cones and increases the synaptic gain of S cones. So that led us to think that this mechanism would lead to color constancy because there it compensates for the global color of the illumination. So what we did was we wait so we so what we did was we simulated now a color constancy experiment. So we present the white dots of light and illuminate that with different colors of light. And we will present the responses of the cones in color space. So the response of our red cone looking at the white dot plotted along this axis of the green cone along that axis and the blue cone along that axis. And that gives you a vector in color space. You can normalize that and you have a point in the color triangle. So we basically built the whole autoretta of the horizontal cell cone system in a model and looked at how this model behaved depending on the color of the global illumination. And it went from pure green from green to red. So this is pure green. This is pure red stimulation and it's pure blue. So the stimulus changed from green to red. The cone membrane potentials first of all they shifted to the center because of adaptation of the cones but then the green lights got green cone stronger than the red lights. So it followed basically nicely the stimulus. This is the cone membrane potential response. However, when you look at the cone output which is influenced by negative feedback you see that basically that was independent of the color of the global illumination. So it not only holds for the red green axis but also for the red axis. So when we change the color of the illumination from blue to red it's behaved similar. Okay, so why then are horizontal cells spectrally cone? So we have three cone types blue, green and red cones which strongly overlapping spectra. They project to horizontal cells and horizontal cells have those complicated spectral sensitivities hyperpolarizing over the whole spectrum, a biphasic cell which depolarizes in the red part and the triphasic one. And then the feedback signal those horizontal cells generate are again a spectrally broad signal. To understand why horizontal cells have this complicated spectral we have to go to an old paper of Buchsbaum and Gottschalk. They ask the question how can I code the information present in these three cone types most efficiently in three other cells? And basically their answer was, the medical solution was that you need a cell that basically sums all those spectra a broadband cell a motivational horizontal cell a biphasic cell and a triphasic cell. So these are basically the three principal components of the responses of the photoreceptor. So horizontal cells seem to store the principal components of the the global illumination. So we have an input that is activating the cone photoreceptors. They send it to the three horizontal cell types where the spectral composition is stored in three principal components. These three principal components are used to generate broad feedback spectra specific for each cone type presumably via kind of linear combinations. This feedback signal is sent back to the photoreceptor to modify the output of the cones such that they become independent for the global illumination. So the input is converted into its principal components and then fed back to normalize the output of the photoreceptor. And in that way generates a form of color consistency. So in fish horizontal cells are spectrally coded. One could argue oh well this is nice maybe this color consistency mechanism only works but this works nicely for fish but does it also work for primates because primate horizontal cells are not spectrally coded. So these are simulated horizontal cell responses and so we took the color consistency model and replaced the cone photoreceptors for the human photoreceptors and horizontal cells for primate horizontal cells and then calculated that this model could also generate color consistency. And indeed these are the responses of the cones due to the change in the global illumination and this is the output of the cones. So you don't need per se the opponents of the horizontal cell responses to generate color consistency. We can discuss later in discussion why fish use this scheme and primates use this scheme to do basically the same job. So to finally to finalize this argument let's see how we see color. So color is based on the ratio of the activity of the red, green and blue cones. So one thing we have to realize is that in the outside world we have wavelengths in our brain that's the only place where color exists. There's no color in the outside world. So light illuminates a scene and the reflected light enters our eye and there that activates cone photoreceptors and they send a signal to higher visual areas and the ratio of those signals determines the color. And then we perceive this apple as red. So now we change the illumination. Now that means that also the activation of the various cone types has different. Now there's a lot of blue light so that suggests that a big blue signal is sent to the brain and only a minor red signal and that's the apple we perceive we see is not red anymore. But that's if finally the apple is perceived red because there is a color correction and Zeke suggested that that happens in v4. However, I hope I've shown to you that presumably this is not happening in v4 but that we get a so-called wavelengths correction by feedback from or the cells to cones already in a retina such that the signals sent to v4 are already balanced and that we therefore see the apple as red. So in conclusion the feedback mechanism from horizontal cells to cones is highly conserved. The feedback from horizontal cells to cones consist of a fast effect and a slow pH buffer dependent mechanism. Feedback from horizontal cells to cones is modulated by a gyroergic mechanism which might function to shut down feedback from horizontal cells to cones in the darker state. The ratio of the two feedback components might vary between different animal species depending on their ecological niche. Negative feedback from horizontal cells to cones increases the synaptic gain of cones. The various horizontal cell layers function in concert to generate the first steps of color constancy. So this work has been done with many many people as collaborators and I'm very grateful to all those people. However there are a special group of people who have been working in my lab over the last years that made this work really possible. And I would like to acknowledge those people because those people have done the majority of the experiments in my lab and I'm very grateful that I was able to work together with those people. I would like to end here and answer some questions. Right, thanks a lot Martin, that was a very deep way through. First I would like to apologize to everybody that just joined us if you had the bad timing on your schedule for this talk and very sorry if you're just joining us. This was due especially if you're in America this was due to a time changing here in Europe so very sorry for that. Maybe you can rewind to the beginning of this talk but we're going to move this question now so if you want to join us join us and maybe watch the talk later again very sorry for that. Thanks a lot Martin. I have a question from Tom Badden. Do you think there is a possibility for different horizontal types to use different balance of his mechanism? I don't know where Tom posted this so I guess he's referring to the three mechanism you described at the beginning of the talk. So do you think there's different HC type to use different balance of his mechanism and Zeus setting up different properties in different spectral cone combination. So we looked at the expression of the various connections in the various horizontal cells for instance and we see that all horizontal cells express connection 55.5 so for the effect mechanism we don't have any evidence that the different horizontal cell types the H1 2 or 3 behave differently. Also for the pannexin staining we did not see a clear difference between we did not see any difference in the dendritic staining of the different types of horizontal cells. So in principle it could be but we didn't find it. Do you think there might be some chromatic selectivity between the horizontal cell or in cone? Sorry. I think you think there might be some chromatic selective center surround in cones or in horizontal cells. So in the fish retina and so basically in all animals most of the cones receive input from more than one horizontal cell. So there has been a long discussion whether there is the H1 horizontal cells and the H2 and 3 horizontal cells. One is a luminosity cell and the other chromatic cells. However in the end all the horizontal cells feed back to the cones again and the feedback signal measured in the cones becomes very broad. So we don't find so I spent a lot of time looking at horizontal cell responses. However I came finally to the conclusion that it's not that useful to look at horizontal cell responses. We have to look at the outputs of the horizontal cells and that's the cones and that's the feedback signal in the cones and the only signal that matters is what a cone receives from the horizontal cells and that's a spectrically broad signal. Sorry. Thank you for that. If you want to come ask a question yourself we shared the link to this room so if you want to join us please do so. I will continue with a question from Brent Young. Do you think that the color constancy mechanism will differ between fish and primates with more central processing in primates and peripheral processing in fish? So I think color constancy is so fundamental for color vision so basically if you are not color constant color is not very useful because it will continuous change when you move around. So an object does not have a fixed color so if you are not color constant the use of color to discriminate an object becomes less efficient. So my feeling is that color constancy is a fundamental property of the color vision system. So and given the organization of the auto retina it's likely that a major part of the color constancy is generated at that level. So the question whether it is different in primates or in fish our simulations at least show that the mechanism will function as well in primates as in fish. So the only difference is that in primates the L&M cones are very close together. So the spectra are very overlapping. So if you want to make a biphasic horizontal cell for instance in the primates basically you subtract two almost equal signals and that will result in a very very small signal and that's not very efficient. The final result of the three horizontal cells is a spectrically broad feedback signal. So you can also generate that by two spectrically broad horizontal cells. The oponency is not critical. The oponency is there because it is an efficient way of coding. It's not a fundamental property of the mechanism. Thank you for that. I have a question from I'm sorry if I spelled this wrong. Santanam abirami will there be any shift in this feedback mechanism during RP or any other photorespirant damage? Sorry, could you repeat it? Will there be any shift in this feedback mechanism during RP, or any other photorespirant damage? Yeah, so during those disease so the output of the cone and the horizontal cells keep each other balanced. So as soon as you are removing an input of the horizontal cell for instance, the whole system might get out of balance and you might move it to a range where you are not very efficiently transmitting information anymore. So I don't know what will happen in RP but it's a very delicate system which needs to be balanced. You just cannot remove one thing because you cannot remove horizontal cells for instance because then the calcium current will move all the way positive and you will lose synaptic transmission. So it's very sensitive to I think to degeneration of the horizontal cells. I have one from Gautam Awatramani. So intense talk, thanks Martin. What is there no role of IH in mediating slow feedback? I'm not trying to understand this question. What is there no role of IH? IH maybe? In mediating slow feedback. I'm not sure. Gautam Awatramani if you want to join? Inevitatory current, thank you. Someone just so where is there no role of inevitatory current in mediating slow feedback? It doesn't make sense. He's wondering if there's a role for IH in the feedback process. So in which cell? It means sorry. So there is one interesting thing. So when you horizontal cells are depolarized they are relatively high input resistance. However when you hyperpolarize them the potassium channel opens which makes them the input resistance lower and it allows for more current to flow through the hemichannels. So there was a whole issue about that the input resistance of this shaded horizontal cells was not low enough to generate all the current which is needed for an effect mechanism. But horizontal cells have only high input resistance if they are depolarized. So if you hyperpolarize horizontal cells channels open and they become more leaky such that you can generate more current. Maybe that's I don't know. I guess good time if you want some follow-up you can join us. I have a question from Simon which is with us in this room. So Simon if you want to ask your question yourself. Hello Simon. Hi Martin. Great talk. I've just given lectures to my students and you've put me to shame. So anyway my question is about the isolation of the different synaptic ribbons within the cone pedicel and also the isolation the so is there much crosstalk in this ifactic and ph feedback between the different synaptic ribbons. So I think that depends on which animal you're looking at. Right. So the synaptic structure of the primates is slightly different than from the zebrafish for instance. So in the zebrafish I think there might be a big chance that there is an effect of crosstalk. For the pH mechanism there is one thing I'm still very intrigued about so Panexin channels are gated by intercellar calcium. So you have the Panexin channel you have the dendrites. There's a glutamate receptor which is permeable from calcium and just very close to that you have the Panexin channel which can be modulated basically by opening the glutamate receptor on that dendrite so which is extremely local mechanism that could work in one dendrite there's no need for integration in the horizontal cells. So so it is kind of appealing to me so I don't have any evidence that is this works but it's kind of appealing thought to me that for the temporal redundancy reduction you can do this basically at individual dendrites of horizontal cells whereas for the effect mechanism you need the spatial integration. So what about the flat bipolar cells are they going to get the same feedback as the invaginating bipolar cells? So Steve Fevers looked at the difference between the timing of the invaginating and the bipolar so they are looking at the same glutamate release of the photoreceptor. But do they see the same pH signal and do they see? The question is is there a direct modulation of the glutamate receptors on the bipolar cells by the pH? Because in your circuit diagram you place your resistor in the neck around the invaginating bipolar. The critical point for modulation is the calcium current and then you get glutamate release and that affects the bipolar cells. However, I cannot exclude that there is a pH effect on the glutamate receptors on the bipolar cells. Alright people, I will just end the stream right now so if you want to join us in the zoom room do it now. We'll just pass a message from Christian Poehler who apparently really wants to organize an informal zoom meeting on a focus on the autorectina so if you're interested in that just contact Christian. We're going to receive really in early December if I remember right so maybe I will talk about it later on. Thanks everybody for this talk I will now close the live stream, join us on the zoom room if you want to continue talking about it. Thank you.