 and I believe we are officially good to go. Hello everybody and welcome to another session of our Sussex Vision Seminar Series, Always Within the Worldwide Neuroinitiative. I'm George Cafetsis, a former master's student in Tomas Oilers Lab and relatively speaking, newly arrived PhD student with Tom Badden. As your host for today, I would like to once again begin by thanking Tim Vogels and Panos Bozellos for putting forward this very initiative towards a greener and much more accessible seminar world. And of course, having said that, allow me to get back to the reason we all gathered here for today and introduce our guest from University of Sussex, Dr. Silvia Schroeder. Following a bachelor degree in cognitive sciences from University of Osnabruk, she then moved for her master's and PhD in Neural Systems and Biology to Etecha in Zurich where she worked with Kivan Martin on functionally characterizing neural responses in the primary visual cortex of cats. In 2014, she relocated to UCL in London where she joined the labs of Matteo Carandini and Kenneth Harris. And there, successfully, successfully funded by Marie Curie and the BBSRC grants, she investigated the effect of behavioral modulation on visual processing by recording either optically with two photomicroscopy or electrically with the famous neuro pixels, the neural activity at different visual areas. At the beginning of this very year, she moved to Sussex where she is currently setting up her own lab. And without any further ado from my side, it is with great pleasure that I'm leaving the stage for her, Dr. Schroeder, for a talk entitled Arousal Modulates Retinal Activity. This stage is all yours. Thank you very much, George. And thanks also for the invitation to talk here. It's really a big honor to share my results here and to this big audience. And I'm also hoping to get lots of feedback and comments if people have those. So let me share my screen. Yes, I hope you can see that. Yeah, good to go. Good to go. So yeah, today I will talk about how responses in the early visual system, including the output of the retina, are modulated by behavioral or the internal state of the animal. And this is actually what I mean by the term arousal. So it's nothing else, but kind of an alerted state or an also a behaviorally active state. And at the center of my talk will be these green fibers that you can see here. These are actually retinal axons coming from the retina to the rest of the brain. Here, for example, the LGN, I believe. But before I go into the results of my studies, I wanna provide you with some context. And one of the bigger questions that these results and these findings address is, do we see the world differently when we experience different internal states? So if you look, for example, at these two guys, I hope you can believe me that they're in very different states. So he seems to be quite amazed, surprised by what he is seeing, whatever he is seeing, whereas he seems to be relatively bored. And even if they're looking at the very same thing, the question is, do they see it differently? And what do I mean by this? I do not mean that emotions or internal states can lead to different perceptions or make you attend to different things. That's, of course, also true. But what I mean is, does the eye or other early visual brain areas send different signals to the brain in these two different states? So this idea is not very new, that the idea that internal states influence neural responses, even in the early visual system. So about more than 10 years ago now, the following was found in the visual neurons. If you have a rodent here on a running wheel, you can see when it's running, like here, also usually the pupil dilates at the same time. And here you can see the activity of multiple units and you can see that this increases when the animal starts running and also when the pupil dilates, okay? So these findings were originally made in an area called the primary visual cortex. And here just as a reminder, I'm showing you the early visual system. You have here the eye with a retina which projects directly to the thalamus, to the LGN, also to the superior click loss and then the thalamus projects to the visual cortex. So the initial findings that visual responses are modulated by internal state or by local motion were seen in primary visual cortex. Later on, in the last 10 years, people also saw that visual neurons in the superior click loss are modulated by internal states and arousal as well as neurons in the LGN. So areas where the retina directly projects to. So having seen all this, we were asking, is this modulation already happening in the retina? Okay, so this is what this talk is mainly about. Now the first problem is actually measuring retinal activity in an awake animal, in an awake mouse. So we came up with a trick. Instead of recording the activity in the eye from the retinal ganglion cells directly, which is difficult because the eyes are moving around, we instead record the activity of the axon endings in the rest of the brain, in this case, in the superior click loss. So we can basically image this area and see the activity of those axons. You may wonder whether it is a good choice to go to a superior click loss. And that's why I'm showing you this diagram here. All you have to take from this is that there are lots of cell types in the retina shown down here. And the retina projects a lot of areas in the brain. But here, this one shows you the superior click loss, SC. And you can see that most of the cell types and actually also most of the retinal ganglion cells, I think the estimation is around 80 to 90% of retinal ganglion cells are projecting to the superior click loss. So we can actually see a very good portion of that the retinal output in the superior click loss. Right. So how do we do this exactly? So we do inject a calcium indicator into the eye. This is called Psi G-Camp 6F. It works like G-Camp 6F. So basically it gets fluorescent when the neurons are active, but it is co-located to the synapses only. So you get the output, the fluorescence, mainly at the synapses. And then we have a window on top of the brain, of the mouse, and can image the activity. As you can see in this picture here, showing you the retinal terminals in the superior click loss, you can actually see that superior click loss is covered by the cortex. And that's true for most of superior click loss except the very posterior part. And this is actually the part we accessed because we did not want to disturb any other brain area. So we put our recording window right at the back of the cortex. And this is what we can see once we have this window in place. So here, anteriorly, you can see the superior click loss. So it's a very specific portion of the superior click loss. Here you can actually already see the inferior click loss. So once we have this in place, we can then image using two photon imaging, the activity of the retinal axons. And we do this by imaging multiple planes, as you can see here. And these planes, we positioned them very close to each other, about two microns. Because the problem is that the brain is moving, especially when the animal is moving, is running, for example. And the retinal butyls are very small structures. So it could be that the brain and the retinal axons are moving in and out of the imaging plane. So in order to track single retinal axons, we have this volume here. And if the brain is moving, we can then post-talk track the single axons through this volume, okay? Just to show you what the data looks like afterwards. So here, what we presented to the mouse are these kinds of grating stimuli. They're presented on three monitors. So this is what you can see here. And these gratings are drifting. And I will start this movie in a bit. So you will see single trial data in this movie. Whenever the grating is on, you will see an arrow appearing here in this corner and it will show you in which direction the grating is moving, okay? So let me start the video. I hope you can see this okay. So you can see often quite strong off responses of the boutons. So they light up as soon as the gratings come off. But if you look closer or at other boutons, you can also see that they are tuned to very certain directions of the stimulus, okay? So this is data, this is sped up 10 times. And it is kind of smooth in time and it shows you the data after aligning an X, Y, and Z. Okay, so in addition to this neural data, we then also recorded the behavior of the mouse or the arousal state. And we do this with two measures. One is the running speed. So as you can see here in our setups, the mouse, the mice are always head fixed, but they are sitting on either this ball that is floating on air or on a treadmill that can rotate into one direction. And yeah, the animals also surrounded by these three monitors. So we are just recording how quickly the mouse is running. And it's free to run in these, in our studies whenever it wants to. And as a second measure, we are looking at the pupil size. So we have a video of the eyes. You can see here quite clearly that the pupil is changing in size quite a bit. And if the pupil is large, we're taking this as a measure of high arousal when it's small. It shows that the animal is in a state of low arousal. Right, and now we are coming to the data. So let me just switch on the laser pointer again. Yeah, so these data, this data set here was collected in darkness. So the mouse is in complete darkness, but it is free to run. And here you can see this running trace going up and down. In this case, we could actually only measure the running and not the pupil because in darkness, the pupil in our case was fully dilated. So we couldn't actually track any changes in size. And here's what the recordings from the retinal axons look like. So here is one data set and each little row, each line of this plot here shows you the activity of one single bouton. And so we have about 300 or so boutons and they are sorted according to whether, yeah, how positively and how negatively they're correlated with running. So on top you see they're positively correlated with running at the bottom, they're negatively correlated. And here you just see two boutons that occurred in this data set just as a single trace. Okay, so this shows you that in darkness, the activity of the retinal boutons are modulated by the behavior. And this cannot be driven by any visual input because the animal is in darkness. So then we recorded from the same retinal boutons but showed them visual stimuli. So these same gratings that you've seen before, now we can also measure the pupil and you can see that the results are quite similar. So actually I should tell you that now we have here the gratings. You can see the onsets of the grating is marked by these little tick marks and that induces the strong visual response. So that's why you can see this kind of stripy pattern. It's very thin stripes, okay? So that's the visual response. But even on top of the strong visual response is the modulation by behavior. So again, when the mouse is running, these boutons are up-modulated. So the responses are stronger whereas the boutons down here have a reduced response. And again, these are two examples, the same boutons when we showed the gratings. So what you can take away from this that even when the boutons are strongly driven by visual input, you still have this a behavioral modulation. And now comes kind of the quantification of this using all the data we had recorded. And the first plots show you that the effects are heterogeneous, meaning a rousal induces positive correlations in about half of the boutons and negative correlations in the other half of the boutons. And it doesn't really matter whether this is in darkness or whether you show gratings, okay? You can see this here in this histogram. The black bars show you when the correlations were significant and the white bars when the correlations were insignificant. And the whole, this whole histogram, this whole distribution was significantly different from what you would expect by chance which is shown here by this gray shade. And the same basically you can see here for the gratings. Now you can, we can also compare this more directly. We can plot the correlation for each bouton during darkness here in the X axis and during gratings on the Y axis. And you see there is a strong correlation. It is not exactly the same. So there are some things are different but the main message here is that the effects are very similar whether you have whether you do have visual stimulation or you don't have visual stimulation like in darkness. And you probably already realize that I'm kind of switching between running and pupil size. We kind of treat it as more or less one in the same measure of a rouser. This is because these two measures are highly correlated as shown here. So there is a very strong relation between pupil size here on the X axis and running speed on the Y axis. Now if you look at the correlation values of the single boutons like here on the X axis you can see correlation with running on the Y axis correlation with pupil. This relationship is seems even stronger. So the similarity of these correlation measures is very strong even stronger maybe than the similarity between pupil and running. Okay, so that's why we didn't try and figure out what the difference exactly is between running and pupil size. Right, so this was kind of a first result showing that there is some modulation and now we're looking a bit closer at what the effects are for different stimuli, particularly for different movements of directions for the gratings. So here we took the pupil size as one measure and we divided the data into those where the pupil was small and those where the pupil was large and here you can see tuning curves for one single retinal bouton. In black you can see the direction tuning curve for small pupil and in red for large pupil and you can see for this bouton that the response decreased when arousal was high or when the pupil was larger. The same or for this bouton it was similar so for the preferred direction the responses went down with arousal but for other stimuli, so for those that were non-preferred responses went actually up. So it shows you that there seems to be some kind of selectivity or difference between stimuli. So the effect of arousal can be different depending on the stimulus. But overall we also see that the preferred direction is not affected. So here we have preferred direction when the pupil is small, when the pupil is large on the y-axis and you see most of the dots following the diagonal showing that this is not changed. But if we then focus on the responses at the preferred direction and look at how the response is modulated during arousal which is shown here, you see the modulation is mostly negative meaning for most of the boutons, not all, but for most of them the response goes down, okay? So at the preferred direction mostly responses are suppressed. And this was very surprising to us mostly because in other visual areas specifically primary visual cortex people mostly saw that the responses go up with arousal or at least it was kind of 50-50. But we were not the only ones seeing this. There was another group who published this last year having a very similar setup but recording the axons of the retina in the LGN, okay? So they also had the mouse here head fixed and running on a ball showing them gratings but now they recorded activity in the LGN so they had to take off the cortex and then they could image this and this is kind of an example image that you can see when you express like we did cheek and 6F in the retinal axons. And kind of summarizing one of their main results is that if you change, well first of all if you compare the activity during low arousal states with those in high arousal states you see mostly the activity goes down. So this was similar to us, to our findings but also very interesting is if you change the spatial frequency so basically how thin or thick the stripes are of the grating you see that the suppression is different. So for low spatial frequencies the suppression is much stronger than for higher spatial frequencies, okay? So yeah, now we've seen, I've shown you that the activity of retinal boutons is modulated. Now we were asking ourselves what could be the possible mechanisms behind this, okay? And we came up basically with two main hypotheses. One is, and we hadn't considered this when we started the recordings but you have to think about what you're actually measuring, right? You measure or we measure the activity of retinal axons in the superior click loss. So you're not only seeing the firing rate of the retinal ganglion cells but this is modulated in addition by presynaptic neuromotulation on those axons, okay? So what we could have seen is that there are neuromotulators, for example, serotonin, acting on those retinal axons and this is actually what we are seeing, okay? This is how retinal axons get modulated. This is one hypothesis. The other one is of course that the modulation is happening in the eye so that somehow information about the behavioral state of the animal goes back to the eye and modulates the activity, the firing rate of the retinal ganglion cells in the retina more directly. So one way is by axons coming back from the brain, there are very few of those, but there are some. So this is one possibility. This information could reach the retina, okay? So then we went on and tried to figure out which hypothesis it is and we tried to test the second hypothesis and for this, we now switch to ETHIS. So we use neuro-pixels probes to record from retinal axons. So again, we couldn't figure out how to measure directly from the eye in an awake mouse. So we thought the best next thing is trying to record from the axons which we think should not be affected by neuromotulation happening much later at the pre-synaptic site in the superior clickers, for example, right? So this is our assumption that the activity in those axons is not modulated further. So basically what we can see is the firing rate of the retinal ganglion cells without any later neuromotulation. And these are results from two of these axons. So again, we recorded this in complete darkness and the mouse was free to run whenever it wanted to. So here in this first trace, you see the running speed of the mouse over I think 40 minutes or so. And at the same time here, you can see the firing rate recorded from one single axon in the optic tract. And you can see it's hard to see by eye but if you look at the cross correlogram of this data, you can see that there is a positive correlation between the running speed and the firing rate of this axon. And it is actually higher, more positive than expected by chance, which is shown by this red shade here. And here you see a second axon which was recorded. Again, you see the running speed of the animal and the firing rate. And again, this one shows a positive correlation which is significant, so more than you would expect by chance, okay? Now we managed to record from a total of 25 axons, which is really not a lot, given that we use the neuro-pixels probes. But yeah, you have to remember that the neural... Sorry. The optic tract is a very thin structure and it's also very deep in the brain. So it's quite difficult to hit it. And even if you get there with your probe, you will not record from lots of axons because again, axons are very thin and it's kind of hard to record from them. It's also very hard to keep a stable recording over time. So we had very stringent, how do you say? Measures to ensure that we have a very stable recording and that we are recording from retinal axons rather than from anything else. So that's why this number is relatively low, okay? But what you can see here on the x-axis is the correlation of each axon with running and on the y-axis is the p-value. So every dot that is above this, 0.5, a 0.05 p-value is significant, okay? And these were, I mean, like, I forgot, nine or so. Now I said it wrong, of course. So of course, you have to look at the dots below there. Yes, I get this plot is somewhat confusing. So we have swapped the y-axis here. So the smaller values are at the top of the y-axis. So whatever is on top of this dotted line has a very small p-value and is therefore significant. So we found that nine of 25 retinal axons have a significant correlation with running in darkness, okay? So this shows you that the activity of retinal ganglion cells is apparently some of the retinal ganglion cells are modulated by running, okay? So I want to just emphasize that this does support our second hypothesis that somehow information gets to the eye and therefore modulates the activity in the retina directly, but it does not exclude the other hypothesis. So in top of this, you can still have modulation at the pre-synaptic site in the superior colliculus. And I think this is very likely to happen anyway because there are receptors at the pre-synaptic site and there are neuromotivators projecting in the superior colliculus. I think it's very likely that both things are happening. Right, so after this, we thought, well, how does this affect the neurons that are listening to the retina so that get input from the retina? Namely, what is happening to the neurons in the superior colliculus post-synaptically? So now we did the same thing, but instead of imaging retinal vultures as in the beginning, we now image the activity from neurons in the superior colliculus, which you can see here. And here's again the results from one data set. You'll see pupil size in green and the running speed in yellow and several hundreds of neurons. And you clearly see that when we showed the gratings, there's visual input, but on top of this is the modulation by behavior. So again, these neurons were sorted by the strength of correlation with running speed. And this is basically the histogram or the cumulative histogram of correlation with pupil. In black, you can see those for the superior colliculus neurons in the superficial layer, so the visual neurons. And in brown, you see the correlation or the distribution for the boutons, but you can hardly see the brown line because they're basically on top of each other, meaning the distribution of correlations is very, very similar. Now, again, we looked at how the tuning curves of these neurons are affected. And you can see very similar things as to the retinal boutons, but now we also see quite a few neurons whose activity is up-modulated. So in red, you see the activity during higher levels of arousals. On this neuron, it is up-modulated. And here you can see the distribution of this response modulation to the preferred direction. And you see for the black line, which is for the neurons that about half of the neurons are up-modulated and half are down-modulated. Whereas for the boutons, there is a bias towards a decrease, as I have said before. So there are strong similarities, also some differences in how neurons and boutons, retinal boutons are affected. So it doesn't prove that the superior colliculus neurons inherit this modulation by the boutons, but it does show that it's very similar and probably the effect that we see in the neuron, that we see in the boutons, also affect the responses in the neurons, okay? One thing that might be slightly confusing is here, especially for the boutons we saw and that the correlations are kind of 50-50, positive-negative, whereas here, response modulation is mostly negative. So it's relatively simple here. When we look at the correlation, we really look at the activity along the whole time, right? And how is this correlated? Whereas here, this response modulation only looks at the preferred direction. So the response, so the preferred direction of motion, okay? So we do not look, we kind of ignore what's happening with the responses to other stimuli. That's why you can actually have this difference, okay? So overall, we think the modulation of retinal boutons and superior colliculus neurons are relatively similar, but the other input that the neurons in superior colliculus get is from primary visual cortex and we already know that these neurons are also heavily modulated by behavior. So now we wondered, is there also an effect from V1 in superior colliculus? Okay? That's why we, again, recorded now using electrophysiology from superior colliculus, but in one condition, we did nothing to primary visual cortex and in the other condition, we inactivated primary visual cortex and we did this using optogenetics. So we expressed an opsin in inhibitory neurons and PV neurons, and then when you shine the blue light on primary visual cortex, these inhibitory neurons get active and therefore decrease the activity in the projection neurons that project to the superior colliculus, okay? So here is now the activity of one example neuron in from the superior colliculus. Here is in the control condition. So you see it's nicely modulated by arousal. The activity goes up. Now, if we inactivate V1, generally the response goes down by quite a bit in this neuron, but the effect from arousal is still there. So still the response is up modulated, okay? And actually the relative amount of the up modulation is very similar in both cases. So therefore we concluded that this modulation is not inherited from V1. And here's just the quantification for all neurons that we recorded. Here again, I showed this response modulation showing what happens to the neurons, to the responses to the preferred direction. So here you see many neurons are up modulated, some are down modulated, and the same is true even if you inactivate V1. And you can also compare this directly. So here is response modulation during the control condition and here is when V1 is inactivated. Most of the dots kind of fall on the diagonals, so there's not much difference. Only the black dots actually show a significant difference. However, there are also many black dots when V1 is inactivated. So actually showing that the modulation was bigger when V1 was activated, okay? So overall I think it shows that V1, the input from V1 can not explain the modulation by arousal in superior clickers. All right, that's my time, we have 10 minutes. So at this point I've shown you that responses of the retina and also of superior click loads are modulated by these states, locomotion and arousal. And you may wonder, well, what else does the retina or its output know about? How much else does it know? And what I want to tell you now does, it does not seem to have access to spatial information which occurs first in primary visual cortex, okay? So this was a study led by Aman Salim. He had a mouse running through a virtual reality, basically through a corridor, which you can see here. And this corridor had four different landmarks, okay? Basically the first half of the corridor looked exactly the same as the second. So this landmark and the third one were exactly the same and the same one. Also the second and the fourth landmark looked exactly the same. So if a visual neuron purely cared about visual input, the responses of this neuron should look the same during the first and the second half, okay? Now what he found looking, recording from V1 is that some neurons, for some neurons, that is the case. So here the solid gray line shows you the response of one V1 neuron and the dotted line is just a copy of the response from the other corridor, basically. And you see for this neuron, the responses look very similar for first and second half. Whereas for this neuron, it's very different. It responds much more in the second half of the corridor. And in this neuron it's exactly the opposite. It's response much more in the first half. So somehow these two neurons care about the spatial location of the mouse, right? Whether it's early in the corridor or late in the corridor. Later on a study spearheaded by Mika Diamanti looked at LGN boutons that are coming from LGN two primary visual cortex and she imaged those boutons. And you can see here the activity of three different LGN boutons. And what you can see, hopefully, is that the responses here for all of these example boutons and also in the rest of the population, they were very similar in the first and the second half, okay? So it seems like any kind of spatial information, first is represented in primary visual cortex and not before. All right, so that just as a side story, now I want to summarize my results. I've shown you that arousal does modulate retinal output. It can or does differ across retinal terminals because some were positively, some were negatively influenced. It's also stimulus dependent, okay? So for some directions, you saw that the response can go down and for others it can go up even in the same retinal bouton. These effects are even observable in the firing rates of the retinal ganglion cells which speaks for a modulation at the site of the retina. And arousal also has very similar effects in the post-synaptic neurons, okay? So these are the findings as we have them now. I think, oops, sorry. The big open question for me is what is the computational advantage of this behavioral modulation as early as in the retinal output? And I just want to kind of show you two things I'm kind of thinking about. One is kind of along this idea of predictive coding and retina already does predictive coding amazingly. So what do I mean by this? So if you look at natural stimuli like this one, and you can also imagine how this moves along, there are lots of correlations in space and in time. And the retina is clever enough to compress this information by encoding only those features that are unexpected. So somehow it knows of this stimulus statistics of natural stimuli and kind of gets rid of everything that's boring and expected, okay? So that's one way of predictive coding that's happening in the retina. And then there's also even a dynamic version of this. So if you, you know, here is a, well, this is actually a blank stimulus. And if you adapt the retina to these kinds of stimuli and then measure the receptor fields, it might look like this, so typical centers around. But then if you show lots of vertical stimuli to the retina, you see that the receptor field changes so that it responds less to verticals of the expected stimulus and responds more to horizontal unexpected stimuli and the other way around, okay? So this whole predictive coding can happen also on a relatively short time scale, shorter than life. So I think it could also be good if, you know, if the, assume that the stimulus statistics changes depending on behavioral state, for example, running or not running. If the retina knew about this, it could already change its receptor fields or its response properties so that it adapts to those expected stimulus statistics. Okay? So this could be one way why this could be good thing to do to integrate or be modulated by behavioral state. And the other one is, well, being in different behavioral states could mean that different things are relevant to your behavior and you may want to enhance or suppress specific stimulus features depending on which state you're in. So for example, you might want to increase stimulus sensitivity during choir sense. So you are alerted to, for example, dangerous stimuli even if you're kind of dozing away. Whereas you might want to increase stimulus selectivity when you're alert and really looking for a very specific thing. Okay? So these are kind of the ideas of why this might be good. And in the future, I want to look at kind of three, three different questions or approaches I want to take is looking at the much larger variety of racial stimuli. I think that's pretty straightforward. So in this study, I only use gratings and vary them in their direction of movement. But of course, there's lots more parameters we can and should probe to see where which parameters are modulated or responses to which parameters are modulated and how is the modulation happening in which way basically. And also looking kind of similarly, looking at different functional cell types to see which cell type is modulated in which direction and seeing what different functions those cells have could give us a clue of which functions are enhanced and which ones are decreased. And of course, also the third point is looking at mechanisms. So how is this coming about? How do retinal boutons and superior click-flash neurons know about behavioral states? And I think as I mentioned before, neuromotulation is a big candidate. So looking into this, I think will give us a clue how this could be happening and also give us more details in what is possible to be modulated. All right, and at last, I want to thank co-authors on my paper and of course my collaborators these people here in my previous lab and also in different labs now. And I want to thank my funding bodies, the EU Welcome Trust, BBSSC. And yeah, it was mentioned at the beginning that I just started my lab last month in the University of Sussex. I'm still looking for a PhD student. So if you're interested in these kinds of questions, please get in touch with me. And yeah, also again, thanking my funding bodies here that make it possible for me to do future research on this topic. Yeah, that's it. Thank you very much. And I hope you have lots of questions. Thank you very much, Silvia. Indeed, a very interesting topic, UNTALK. Lots to unpack and definitely of crucial nature, the recordings you actually did of the RGC action tracks. There are already a couple of questions in the chat and just to remind our audience, like both regular and newcomers that there will be a short discussion like with the questions you post on the chat. And then we will invite you in the very Zoom room that we are currently sitting in for a post-talk informal conversation. So I will start with the questions like one is from Luisa Ramirez but you have already or maybe at least partially addressed I don't know if you want to add anything to that but the question was biologically is there a motivation for having inhibition directly in the eye? It was one of the first outlook points that you made like why we would already have this effect in the retina. I don't know if you want to add anything to that if not I will proceed with the next. Yeah, yeah, I can't add a lot because it's just guesswork at this point. But yeah, I agree that this is the most one of the most important questions why is this happening? And I'm very, very curious myself and specifically because we have the opposite mostly the opposite behavior later on in the hierarchy, right? In the primary visual cortex it seems like that the neurons are enhanced during arousal or the responses are enhanced. So yeah, it's a very big question. And I think we can only answer this with the kind of approaches that I showed by looking because maybe, yeah, maybe we were unlucky we are biased by the stimuli that we've shown and actually for other stimuli it could be different. So this has to be kept in mind and we should also not forget that some of those retinal butans there was an increase in activity and I think the most important step now is also looking at different cell types and seeing which ones were actually up modulated and which were down modulated. And yeah, then I think we have to look at the different stages along the hierarchy and seeing how is the effect of the retina actually inherited by superior clickers. So now I did these measurements in two different animal sets of animals basically. So I think, for example, recording at the same time and figuring out what all the different inputs to the neurons actually do is getting, will get us one step closer to the answer of this question. But yeah, for now I don't know. Yeah, thank you for that. Next one up is Grace Lindsay and she asks the following, the correlation between modulation from running and pupil size is surprising as you'd expect a change in pupil size to impact firing rates on its own. Is the impact of pupil size on activity small? I'm not sure I got the question correct. So she thinks that the running speed basically causes the fluctuations. So she claims that it's surprising that you have a correlation between the pupil size and the running speed. Does that imply that the effect from pupil size on the activity is small? No, it doesn't. I mean, it could be both of these things, right? So how I think about this is there is some brain state which we call arousal which is affecting the running speed and it shows up by the changing pupil diameter and also affects the neural activity, right? So there's, yeah, this vague or this third player called arousal and we just measure it by running speed and pupil. And yeah, it happens that these two measures, running speed and pupil are highly correlated which one affects activity more? I don't know. And maybe it's not one of them directly. I don't think pupil diameter itself causes retinal activity to change. I mean, it is important when you have light conditions. So when we show the gratings because of the changing pupil diameter there's more or less light coming into the eye and therefore this could trivially explain changes in responses. But that's why we did also the recordings during darkness where you don't have this confound basically. But yeah, as I also mentioned in the talk we didn't really try to separate between the effects of running and pupil dilation yet because they're so highly correlated. So it's an open question which whether there is a difference between the effect of running speed and pupil. Next one is from Michael Reiser. The balance of modulatory effects is very interesting. Could it be that the retinal modulation is mainly positive correlation while the effect of arousal modulation in the superior colliculus is mainly suppressive? So this is about comparing the E-Phys results versus imaging, right? In the E-Phys we had these nine axons and most of them were actually had a positive correlation only two had a negative. So I guess the question is, is there now a difference because from the E-Phys results we would expect more positive correlations whereas imaging we've got like a 50-50 correlation. I think the numbers are just too small to make any meaningful statement about this. So I think what we can conclude from the E-Phys results so from the recordings of the axons is simply that something is happening. So somehow the activity in the axons and presumably in the cell bodies of the retinal then get information about running speed, okay? But I wouldn't claim anything about numbers because it's just too, too little. And I think that if you do any kind of test significance tests, it wouldn't hold up to really compare differences in distribution from axons and imaging. So yeah, I wouldn't read too much into the exact distribution of correlations. Thank you very much. The last one appearing on the chat is from Hervik Bayer. Hi Sylvia, nice talk. You may want to look at Cortisol as a modulator. There is evidence from Zebraphys that glucocorticoid receptor activation can alter retinal physiology. It's more of a comment and less of a question. That's it, yeah, okay. So this is at the side of the retina and the cortisol would come from the bloodstream or? Unfortunately, I will be stepping on thin ground if I attempt to clarify that but I'm pretty sure that he can join us in the discussion, Maxim already posted this room link in the chat. So I will keep the live broadcasting for a couple of minutes more as people start already to join us here in the room. I have one question myself because you also mentioned simultaneously recording from the boutons or from the post synaptic cells and the axonal tracts. Couldn't you use a neuropixel probe properly tilted so you can simultaneously record both from the axons and the superior colliculus or that is impossible to assume? Well, okay, first of, I guess the answer to your question is yes, you could do this very easily with two probes, right? So the problem is not getting superior colliculus at the same time. It is much more of hitting the optic tract. So I even had two probes kind of trying to hit the left and the right optic tract at the same time just to increase my numbers, but I wasn't super successful even then. So that's not a problem. The problem is also even if you hit both that you don't know which of the superior colliculus cells is actually getting the retinal input, right? So you still don't have this one-to-one relationship. So I think this is really the big advantage of the imaging. So what we're actually planning to do is having dual color imaging of the retinal boutons at the same time as with the different colors of the neurons. Of course, it's still a problem to really know where the synaptic input, which axon is talking to which neuron, but I think it will give us a better clue of what is happening at least at the same time and in the same vicinity at least of the retinal input or output in the neurons that are listening to it. Thank you very much for that. I see there is a clarification from Grace. She says, sorry, I meant that when you change pupil size, you change the light that enters the retina which would change activity even without top-down modulation. Exactly. So yes, that's also what we thought. So we actually did those recordings first. So we recorded first under light conditions. So we recorded the responses to the ratings when this could be a problem, right? And therefore, we then decided to introduce the complete darkness condition, where you do not change the visual input anymore. So also I said during complete darkness, the pupil size does not change anymore. It's fully dilated. So you're getting around this and we still see even in darkness that the retinal boutons are modulated. So I don't think it's the whole... It's not the whole explanation. Now, yes, I think I would just repeat what I said before. I don't think there is running speed and pupil that are affecting these responses, but it's some other third component arousal which just shows up by these measures, running speed and pupil, yeah. Great. Yeah, I think that pretty much sums it up from the audience side. I would just like to remind everyone that you can join us in the Zoom room we are currently sitting in. And with that, I would like to thank Sylvia for honoring us and giving a talk about behavioral modulation of visual processing. And officially now, like before I terminate the live broadcasting, so I will give another couple of minutes so people can join. I would like to wave my rights. As moderator, so people who join can freely ask their questions without waiting for my green light, let's say. Yeah, thanks again for having me. It's really a great seminar series, yeah. Thanks. So if somebody doesn't want to go ahead with a question, I think I have another one, probably quite naive, I would say. Because we study, like you study, the effect of behavioral modulation on visual processing, but then again, you do not try to influence arousal per se, right? So I was wondering if you can play some more exciting stimuli for the behavioral arousal state to see if there is like an effect of vision on the arousal that comes back to vision again. Ooh. Like similar to the most exciting images approach that they have, like that they play different images and they try to figure which one triggers the higher responses and then they replay this again. So in a similar context, but behaviorally modulating, including. Maybe it's really naive and stupid, I don't know. Yeah, I think, I mean, it's definitely possible. I think the, so from my view, I think this would cause several problems though, because then it's kind of this loop and you don't really know anymore what's causing what. So is it the visual stimulus itself or is it kind of the excitement content in the stimulus that is acting on the neural response? So I'm always trying to separate visual stimulus from behavior. So having the same stimulus under different behavioral conditions or the other way around, right? So I think having these stimuli that always induce a specific behavioral state that gets you into problems. But yes, I am thinking about controlling the behavior better by simple things, like for example, having a motorized wheel. So yeah, you can kind of induce running or stop running in the animal. And yeah, I also want to look at kind of behaviorally relevant stimuli. For example, those that look more like a predator, the typical looming stimuli that everyone's using now or also stimuli that look like food for the mouse. So it would induce kind of approaching behavior. I think that's very interesting, but maybe has already similar problems that I just alluded to. So if you see, of course, if the mouse sees a predator, it probably gets very excited. So yeah, I don't know. Yeah, kind of interesting to see what would happen then. So yeah, I see Simon is here with his microphone on. I don't know Simon, do you want to go ahead with a question? Yeah, I've got a question for Sylvia. It was a really good talk. And I'm very excited to see that people are getting responses from ganglion cells in intact behaving animals. Because as far as I understand it, you might be able to correct me, but as far as I understand it, almost all of the work on the coding of natural images by ganglion cells is done on unnatural ganglion cells because they're in isolated retinas or in their isolated eyes. And so my question is, do you see from these very early recordings of yours that they're the first of their type, I think, are they? Are these? Well, there is one group that managed to record using a kind of electrode array, putting it behind the eye. They were able to record from retinal ganglion cells, but I don't think they, yeah, I don't know how stable this. So it's kind of a similar measurement with a different approach, which came up around the same time. So my question is, is there anything in your data or in your observations that suggests that the idea is about there being a very sparse code and very low firing rates when gang, average mean firing rates when ganglion cells are coding natural scenes? Is there anything that contradicts that? Because some of your spike rates seem quite high and sustained even in the dark. Yes, I was very, I mean, I'm not the retina person by training. So I didn't know much, but I was very surprised to see those high firing rates myself during darkness. Yeah.