 And I believe we are officially live. Hello everybody and welcome to another session of our Sussex Vision Seminar Series, Always Within the Worldwide Neuroinitiative. I'm George Caffèdiz, a former master's student in Thomas Euler's lab and currently a PhD student with Tom Badden. And as your host for today, I would like once again to 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 LMU, Munich, Professor Laura Bush. Following her masters in Neural and Behavioral Sciences from the University of Chibigan, she moved to get again on the lab of Stefan Troje for her Dr. Rare, not in biology. After highly productive years with both Stefan and Matteo Carendini, first in San Francisco and then in UCL, she started in 2010, her own group back in Chibigan at the Center for Integrative Neuroscience. And in 2016, she moved to Munich and LMU where she's located ever since and nowadays holds the title of Professor of Organismic Neurobiology. With research interest ranging from natural stimuli and vision during action to how specific neurons form the basis of visual behavior, today we'll have the pleasure of hearing about the dorsolateral geniculate nucleus and how feed forward and feedback signals shape its activity. So without any further ado from my side, please all welcome Professor Bush. Laura, the stage is officially all yours. Thanks George for the introduction. Thanks for having me also. Thanks to Tom for the invitation. It's really a pleasure to talk here. I think like you put it in the beginning, this initiative is fantastic to bring neuroscience to a large audience without you having to travel anywhere. It's very accessible and I'm very proud that I can contribute to this. So let me share my screen. Can you see it okay? Yes, everything looks proper. Perfect. So, sorry, I wanted to go here. I wanted to introduce my title. So I will talk to you today about visual processing of feed forward and feedback signals in mouse thalamus. So my lab investigates visual processing along the retinogeniculo cortical pathway. And this pathway consists of retinal ganglion cells that form the output of the retina. They project via the optic tract to many places among them the dorsolateral geniculate nucleus of the thalamus and from there relay cells then send their axons to primary visual cortex. We think this pathway is an important one because it's what we consider the image forming pathway. So it's the pathway that underlies object recognition and visual perception. And it is also the most direct route for visual information to reach cortex. In the scheme between the retina with its powerful mechanisms for feature extraction and maybe primary visual cortex on the other side with its that is famous for being selective orientation and where this feature selectivity emerges. LGN at least according to the classic textbook notion is really considered a relay station. And I want to show you today that LGN does much more than just simply relaying this information that it receives from the retina to the primary visual cortex. So when you look at the circuits evolving around LGN this already suggests that LGN is much more than a simple relay. And of course, this scheme here on the left side is still a simplification but it shows you already the complexity of the circuit. So LGN itself, if you look here already has its own interesting circuits. So we have the relay cells that project the axons to primary visual cortex. And in addition, we have local inhibitory neurons. These are the red ones here. They are very few in the mouse, less than 5% but they have a large span. So they target many, many regions and subdivisions of LGN and they have an interesting dendritic signaling mechanism. In addition, then LGN is embedded in a complex network. So for example, we have the feed forward inputs from the retina that go to the relay cells and also to the local inhibitory neurons. We have an inhibitory loop via the thalamic reticular nucleus here. And we have feedback inputs from the primary visual cortex. So these would be layer six corticothalamic neurons and neuromodulatory inputs from the brain stem. Okay, so this already suggests LGN is more than a simple relay. And I want to point out that in addition it's really the final bottleneck before cortex. So if you look at the numbers at least from macaque LGN versus V1, you see that the numbers of neurons in primary visual cortex are about two orders of magnitude higher than those in LGN. And so LGN is really a place where you could modulate visual information processing very efficiently. In the past, my lab has contributed to investigating parts of these circuits from different perspectives. So together with Thomas Euler and Philip Behrens in tubing and we have looked at how retinal inputs converge onto the relay cells in LGN. But since this work is published today I want to focus on the other aspects of the circuit. I want to talk to you about modulations of LGN by corticothalamic feedback and also modulations of LGN by behavioral state. And for this I want to present research that we are trying to publish right now and you find it all on bioarchive. All right, so let's go to feedback first. So like in many other areas of the brain also at the level of LGN, feedback is very prominent. So if you look at the synaptic inputs to the relay cells in LGN of the thalamus you see that the synapses from retinal ganglion cells only make up about five to 10% versus you get about 30% that come from these feedback connections from cortex. So these, before like saying that this is now a really important question I think to understand what this feedback does I want to point out of course that these numbers do not mean everything. So of course these retinal inputs are very, very strong and driving inputs whereas the 30% that come from cortex considered modulatory. But since there is this big number of modulatory inputs we think it's really an important question to understand how feedback affects responses in LGN. And I also want to point out that the answer to this question is probably very difficult and very complex. So on the one hand, we have both direct excitatory feedback. So this layer six corticothalamic neurons project both directly to the relay cells in LGN. And we also have indirect inhibitory circuits either to the local inhibitory neurons in LGN or via the thalamic reticular noctis. What is interesting is that these components of the feedback they have different synaptic short-term dynamics and the balance with which these two branches of the feedback are engaged is basically unknown and probably context dependent. Okay, and maybe not so surprisingly given this complex circuit, the literature that has dealt with the question what does feedback do to LGN has suggested to that this feedback modulates really diverse aspects of LGN responses from spatial integration, temporal integration, changes in firing mode and so on. And this literature is really diverse sometimes inconsistent but what has struck us is that basically most of this knowledge that has been acquired about the role of feedback comes from studies using artificial stimuli such as gratings and bars. This we found interesting because early experiments already suggested that maybe complex visual patterns might better engage this feedback and then might maybe elicit more robust responses. And that's why in this first study that I want to show you we thought we should probe feedback, corticothalamic feedback with naturalistic movies. And the idea is that maybe we can engage this feedback more efficiently if the stimuli match an internal model that's built from the statistics of the world. So what we do is recordings in head fixed mice. We show them naturalistic movies. These are movies that we recorded with a handheld camera. You can see one frame here. The movies contain foliage and there are some shifts in the movies to simulate the cuts that the mice do with moving their gaze so their head and their eyes together. You can see a movie playing now here. You can see the grass and the leaves. So we present these movies to movie clips. They last about five seconds to head fixed mice. And in these head fixed mice we manipulate corticothalamic feedback by suppressing primary visual cortex. We do this by transducing parvalbumin-positive inhibitory interneurons in primary visual cortex with general drops in and then using blue light to activate these inhibitory, local inhibitory interneurons in V1. So we recruit the intracortical inhibition and thereby suppress activity in primary visual cortex including in the neurons, the layer six corticothalamic neurons that then provide the feedback to LGN. So we have also repeated these experiments with more specific suppression of layer six corticothalamic neurons directly. But the results are very similar. So we are going to focus for this talk now on the global PV activation method where we suppress V1 in its, like across all layers. Good. So let's look at an example neuron and its responses to these movie clips with primary visual cortex intact versus suppressed. So here you see 200 trials of an example neuron one in response to this naturalistic movie. It's five seconds long and it's repeated 200 times. Each of these black dots is an action potentially elicited by this neuron. And in red you see burst spikes. So these are spikes that come in very close succession to each other, less than four millisecond inter spike intervals and they are preceded by about 100 milliseconds of silence. Bursting is a response mode of thalamus in general and it's a signature of hyperpolarization because it is driven by a low special calcium current that it is elicited after prolonged hyperpolarization. So these are the responses of this example neuron when feedback is intact. So how does it look when we now use our optogenetic suppression method and suppress activity in primary visual cortex? So these are the responses that you obtain from the same neuron, same movie with one suppression. If I go back and forth, you can see that we basically keep the major response events in this response. So for example, around two seconds is something interesting in the movie that the neuron responds to and it did so before but you can also see very traumatic changes in the response. So when you look at these what you realize is that we have a much reduced firing rate in this neuron and we have many more of these burst spikes, the red spikes here. So let me go back and forth one more time. Okay, so from this it seems that with feedback intact maybe the neuron, this example neuron here is responding more in tonic mode. So these regular black spikes maybe with more graded responses. So we have either large responses, intermediate responses, small responses, no responses and there's a higher dynamic range versus with V1 suppression we see much more burst mode activity and the responses seem much more rectified and at the same time more sparse and reliable across trials. This is not only this particular neuron that responds in this way but you can also see this across the population of neurons that we recorded under these conditions. So with feedback intact, this is plotted on the Y-axis we have higher firing rate like in this example neuron we have lower bursting like in the example neuron but this comes at the expense of sparseness and reliability. So with feedback intact responses are less sparse and less reliable across trials. So what we've seen here with the movies is that the effects of feedback on these responses in LGN seem consistent with a net depolarization hence the enhanced firing rate and the reduced burst ratio. And this switched to a tonic mode firing. Okay, so one thing that was very prominent to us was this reduction in firing rate when we suppressed primary visual cortex so we wanted to go one step further and understand a bit more how like what is the mechanism behind this suppression and in particular, is it an addition or is it a multiplication that or a division basically that happens when we go from feedback intact versus suppression. So can we explain basically we ask can we explain the response during V1 suppression by an additive or by a subtractive and or a divisive mechanism based on the response during feedback intact. For that we took the two responses in the two conditions this is another example neuron this is now a PSDH where we plot the firing rate across the time during the movie presentation. And we thought we can ask basically can we model the response during feedback suppression with a simple threshold linear model where we allow a certain offset and slope and we have this small rectification space here place here because firing rates of course cannot be negative. So the question is can we now send into this model the response with feedback intact and predict the gray response here during V1 suppression. For this particular example neurons when you compare basically the model during feedback suppression or model responses during feedback suppression in blue against the actual measured responses in gray you can see that it's the simple threshold linear model performed really well. This is cross-validated we use 50% of the trials to make this model and the other half of the trials in order to evaluate this model and it is cross-validated evaluation we get an R square of about 90% which is really great. This is an example one example neuron that I should point out versus very little so you can see across time there's almost no bursting in this neuron this is the pink line. What about the other example neuron that we looked at before so this is called example neuron one where we looked at the raster plots you saw lots of bursting. Here you can see that the overall performance of the model is much worse. So it's only an R square about 30% and you can see that the models if you compare the blue versus the gray lines here does an okay job in many places but also fails considerably in others. So it does okay in areas like here or here but it fails considerably in areas or in time points where we have lots of bursting. So when we have this higher pink peaks we can see that there's a massive difference between what we predict with the model based on the response intact with feedback intact and the actual response during feedback suppression. So the actual response is much higher than what we predict based on our scaling model. And this is one example here. Here's another peak where the model completely misses the high firing rate during suppression and another one is maybe here and up here in the front. So it seems like at least in this example neuron whenever we have lots of bursting of course our simple threshold linear model fails very much so this is what we observe across the population. So overall when we have very little suppression in the neuron or when we have very little bursting in the neurons during suppression we see that the threshold linear model does a good job. This is our example neuron one and green when we have lots of bursting the simple model fails and it cannot explain the data well. So this regarding maybe this effect of bursting however what we can conclude from this modeling exercise is that in many neurons we have a slope that is when you fit this model we have a slope that is smaller than one. This is the red triangle here the average this means that we have a divisive mechanism and very little change in threshold. So from this we would conclude that maybe except for this periods when we have lots of bursting we want suppression seems to influence the responses in LGN to naturalistic movies predominantly by a divisive mechanism. So we get slopes that are smaller than one and little shift in this model. Okay, so now maybe you can think why is this useful? So why do we have these kinds of responses these tonic responses, many spikes and less little bursting with feedback and these more rectified bursting responses with V1 suppression. Maybe in order to interpret the results one we thought we should look back at maybe general explanations that have been proposed for the role of burst mode in Thalamus. And one very prominent one here in the literature is the so-called wakeup call hypothesis. So it's known that the first action potential of a burst is much more likely in this case more than twice as likely to evoke the V1 action potential than is an action potential during tonic mode. So if you have these action potentials during bursts you get larger EPSPs in the cortical cells and then you drive firing much more reliably than with these tonic spikes. So we thought maybe one idea could be that during this reduced feedback the burst mode could serve to signal salient events with high reliability. So these events that the neuron was supposed to signal basically it signals now in burst mode and this triggers V1 responses very reliably. And then I mean this is pure speculation but this wakeup call then this kind of wakeup call could trigger feedback and shift LGN to tonic firing mode for more linear encoding and you would have this high dynamic range responses in tonic mode here. Okay, good. So far I've shown you that this feedback to LGN can have effects on firing rate and on the firing mode of LGN neurons but these were very global modulations multiplicative modulations. So the next question we asked now was does the feedback from cortex on Thalamus maybe have more specific functions beyond this global firing mode modulations that I've showed you before. One candidate mechanism or one feature that is very salient about these feedback projections is that they are topographic and the topography of these feedback connections already suggests the role in spatial processing. So previous research in primates and cuts has shown that there's a margin retinotopy between the layer six corticothalamic pyramidal cells and then the terminal fields in the genicular target region. And so we next set out to see if this is also the case in mice. What we did was we've now performed triple color viral tracing studies where we injected into primary visual cortex three small portions of AAV viruses that would code for three different fluorophores. One for EGFP, one for M-scarlet and one for M-chloroquine. And these viruses express in a pre-dependent way and we injected these viruses in the NTSR1 crea mice. These are mice that express crea recombinase in to 90% specificity in layer six corticothalamic neurons. And what we now saw to do was look then in histological slices post-mortem at the terminal fields in the LGN and see if there's topographic arrangement. So what you can see here is the V1 injection site. So we performed experiments either along the azimuth direction of retinotopy and primary visual cortex or along the axis for elevation. You can see that what we, I mean, we were successful in these injections. We ended up transducing small populations of primary visual cortex neurons with these different viruses, the different colors and those were restrained to primary visual cortex basically. Okay, so now the question is what happens in LGN and here you see the result. So when we looked at the level of LGN at where these layer six corticothalamic neurons project to, you can see also at the level of LGN a very nice ordered topography, both for the azimuth direction as well as for the elevation direction. And if you know the literature on LGN retinotopy, you will realize that these projection patterns match quite nicely the known retinotopy of LGN. So this already suggests that maybe beyond these global modulations that I've shown you before LGN or this corticothalamic feedback to LGN is really in a prime position to modulate spatial processing given the spatial specificity and the retinotopic arrangement of these corticothalamic projections. So to test this also more specifically in terms of function, we next performed recordings after injecting the NTSR cream mice or them transducing layer six corticothalamic neurons not only with a fluorophore but also with general rhodopsin. So we could do photoactivation studies in order to functionally map these connections. So now we have local populations of primary visual cortex transduced with green and also with EGFP and also with general rhodopsin. After letting express the virus, we perform recordings first and primary visual cortex in the first session in order to map the receptive field locations at the injection site or around the injection site and determine where about the receptive field center is on average at this V1 injection site. You can see an example session here where we plot you three receptive fields that we found and also the outlines of all the receptive fields that we encountered in that recording session around the place where we did this virus injection. So presumably in the middle of this transduced volume in primary visual cortex. In the subsequent recording sessions, we then went to LGN and just positioned our electrode wherever we found LGN2B and wherever we found receptive fields. On the right side, you see two different sessions that we recorded in LGN. You can see that these differ in terms of azimuth and you can also see that along the electrode, we have receptive fields that go from the upper visual field to the lower visual field as is known for the retinotopy in LGN. So what this allows us now is locate the receptive fields in LGN of the neurons that we recorded in space and express their location relative to this mean or receptive field center at the V1 injection site. Okay, you see this here on this plot. And then the final step, we can now switch on the blue line because we have expressed not only EGSP in these neurons, but also channel rotopsin and use this, the opsin basically, photo-activate the opsin in order to map functionally the specificity of these connections. Okay, so what you see here is the result of all these experiments. On the x-axis, you see the distance of the LGN receptive fields to the mean V1 receptive field at the injection site. And on the y-axis, you see fall change. So how much the neurons activity change as a function of either activating the layer six corticothalamic cells or control condition without optogenetic activation. I've also plotted you three different example neurons. These correspond to the three example neurons up here. And I want to use them in order to illustrate the effects that we've seen. So when we look at first these portion of the part of the neurons, whose receptive field is close to the V1 receptive field at the injection site, we see that there's lots of diversity of the effects, including some neurons that are enhanced by this optogenetic activation or photo stimulation of the layer six corticothalamic neurons. You see the orientation tuning or direction tuning curve of that neuron. You can clearly see that we're driving this neuron with what we're activating the neuron with the blue light stimulation of the layer six corticothalamic neurons. When we move further out or further away, so we focus on neurons whose receptive field is about 30 to 60 degrees or 50 degrees away from the V1 receptive field at the injection site. We see that most of the neurons actually are suppressed. And an example is this example neuron two. So photo activation of layer six corticothalamic neurons reduces the responses of these neurons. And this is also evident in the mean that is significantly negative here in the population. And then when we go really far out, so about 50, 60 degrees, we have neurons like this example neuron three, where activation of layer six corticothalamic neurons does not have a very strong effect. So what this shows is that activation of layer V1 layer six corticothalamic feedback can exert distant suppressive influences. And it's consistent basically with this topographic arrangements. We have distant dependence effect, distance dependent effects. And in particular, we have this distant suppressive influences. Good. Next we tested more directly the role of V1 suppression and LGN spatial integration having seen this suppressive region here. What we did again is the first map receptive fields in recordings of LGN and then turned again to our PV activation model in order to suppress primary visual cortex in order to basically take out the influence of corticothalamic feedback. But what you see here is now first the control condition. So what we do is we center now the stimuli on the receptive field of the recorded neurons in LGN and we increase systematically the pseudo random way, the diameter of our gratings in order to map so-called size tuning curves. Here you see two example neurons in LGN that we measured with these different stimulus diameters of grating diameters. And you see the typical size tuning curves of visual neurons. The response first increases up to a certain size. This is the preferred size after which the response then doesn't increase again but rather gets suppressed. This is an interesting effect. It's an effect of context and it shows basically how local information is integrated into a global context and is important or we think that it's underlying things like objects or image segmentation and pop out and visual saliency. So in these size tuning curves we focus on two different aspects. The first one is the peak of this curve. So the preferred size, the mono size and the second one is the amount of surround suppression. So the suppressive influences for this big stimuli. So the question now is what happens if we take out primary visual cortex, if you suppress primary visual cortex with these size tuning curves. And here you see the result for these two example neurons. What you can see is done is a very interesting effect. On the one hand, we have a reduction of responses to small stimuli. This effect then reverses for the biggest stimuli and we have an enhancement with V1 suppression for the big stimuli. Together we have a shift of preferred receptive fields to more larger sizes and a reduction in this surround suppression. You can see this in both of these example neurons. So these effects are really interesting. We think they are not only there for the two example neurons but also for the population where we see that V1 suppression modulates LGN size tuning. In particular, you see recapitulated the effects that I explained for the two example neurons. So during V1 suppression, we have smaller responses for the small size stimuli. We have an enhancement of responses for this large size stimuli. We have a shift of the preferred receptive fields towards larger sizes and we have a reduction in surround suppression. So this seems to be the case that feedback from V1 to LGN boosts responses to the small stimuli and enhances this contextual modulation so it enhances spatial integration. Okay. So the next question we asked was via which circuit could this could feedback modulate spatial integration in this interesting way. So what we've seen again is that with feedback we have these enhanced small size responses and they seem to us really consistent with feedback activating through, acting through excitation. So this would be very consistent with this direct excitatory corticothalamic feedback circuit via which it could enhance basically the responses to these small stimuli. On the other hand, the second effect that I've shown you where the responses cross and we have an increased surround suppression with corticothalamic feedback is maybe easier to explain if you think that the feedback could also act via inhibition basically. So either via the local inhibitory neurons or via neurons in the TRN. In order to find out which of these two possibilities might be more likely to be involved, we next turn to some modeling in order to systematically manipulate the properties of these inhibitory feedback connections and see which one would be most consistent with the data. So what we turn to is simulations within the extended difference of Gaussian's model framework to examine the properties of corticothalamic feedback. And we did this in collaboration with Gautier Einefeld who originally developed this model and also provided us with simulations or published simulation tool in order to adapt it basically to the mouse visual system and systematically test the influence of this corticothalamic feedback coupling kernels. So what the model consists of is three cell types. It has a retinal ganglion cell layer, a relay cell layer and a V1 cortex cell layer. This is a mechanistic model. Basically, the retina is modeled as a difference of Gaussian and from there on the different stages are connected via coupling kernels, these coupling kernels K. What you see is what is important in our context is that we have an inhibitory and an excitatory coupling kernel from the cortex stage to the relay cell stage. And the model is of course now agnostic to the source of inhibition so it doesn't contain local inhibitory neurons in LGN or TRN neurons. But what we can do is vary the special properties of this inhibitory CT feedback projections in order to test with which stage the coupling kernel would be most consistent. And in particular, what we can do is either use a narrow projection from cortex back to thalamus or rely on wide projections from the cortex to thalamus. Okay, so we took the model, adapted it from the cat for which was originally developed to the mouse and then played around with the systematically varied basically this inhibitory corticothalamic feedback coupling kernel. So here you can see again the arrangement we have inhibitory and then inhibitory in red and excitatory coupling kernel. The weight of these are given by the model or as set as parameters in the model. What we do now is vary the sigma of the inhibitory feedback coupling kernel. In this condition here, we make it the same as the sigma or the width of the excitatory feedback coupling kernel. Here we make it three times as wide as the excitatory feedback coupling kernel here nine times and here 40 times as wide. When we now look at give the model our different stimulus diameters and look at the resulting size tuning curves in the absence of feedback. So any of this doesn't really matter. Of course we get four times the same response. But now the very interesting thing is if we turn on this in the model then corticothalamic feedback, we see very interesting and different patterns emerging depending on the size of these or the relative size of these inhibitory and excitatory feedback coupling kernels. And in particular only if we use coupling kernels that are wide enough, we can see that we can replicate at least qualitatively our experimental findings. So with the feedback coupling kernel inhibitory feedback coupling kernel that is nine times as big as the excitatory feedback coupling kernel, we see this enhancement of small size responses with feedback intact. We see this stronger responses for large stimuli with feedback suppressed. So there's a crossover between the two lines. We see the receptive fields getting bigger when we suppress feedback and we see the reduction of surround suppression. So this condition here at least qualitatively recapitulates our experimental findings. And we took from this exercise that the feedback-mediate enhancement of LGN surround suppression and the reduction of receptive field size requires a wide inhibitory feedback coupling kernel. So the question now is which of these circuits fulfills these requirements? And then we thought basically guided by these simulations that the ideal candidate circuit might be this indirect inhibition via the visual teramic reticular nucleus. Like I've introduced in the beginning that TRN is the sheets of inhibitory neurons between the cortex and the thalamus. It's considered something like the gate of the guardian of the gate to cortex and it has in the past already been implicated in gain control, behavioral state modulations, attentional selection. So it has very interesting roles. And if I say TRN, it's of course a very big area. What I mean specifically is the visual TRN now from now on, it's just the visual sector of the TRN that projects to LGN. So the hypothesis that we have is if this feedback via TRN indeed potentiated surround suppression in LGN, the neurons in the visual TRN should have large receptive fields because they should provide this suppression for these large sizes. They should themselves experience very little surround suppression and their responses should decrease during the one suppression also for these large sizes. And now we are going to test these hypothesis. So first, what we did is recordings in TRN and map receptive fields. And here you can see typical receptive fields that we obtain in LGN. So these are small receptive fields. Sometimes they look like the classical textbook receptive field center surround organization. And on the right side and red you see or labeled and red you see example receptive fields that be recorded from the visual sector of the telamic reticular nucleus. And what is striking here is that you can find lots and lots of different sizes, some that look like center surround like the LGN neurons but also others that have big, big receptive fields. And if we map this quantitatively and put everything together and look at the distributions, we see that the receptive field area in the visual part of the TRN is about 7.5 times larger than the area that we recorded under similar conditions from LGN, okay? So this means the first hypothesis holds true. Receptive fields in visual TRN can be large. Then we repeated our size tuning experiments. So positioned the stimuli in the center of these receptive fields and measured size tuning curves. And you can see an example response here of a visual TRN neuron. We sorted the trials here by a similar size and you can see the resulting size tuning curve here. And like what is evident from this example neuron that has very, very little surrounds the question itself. This is something we see also in the population. So in the recorded population where we perform the size tuning curves, we see similar to the example neuron that many neurons do not have significant or very strong suppression for the large similar sizes but there's also some of course that have. So there's a variety of responses but the predominant mode basically is very little surround suppression strength. We quantify this by an index. You can see that about half of the visual TRN population has a suppression index less than 0.05 which is very, very little surround suppression. And in contrast to that, you see the responses that we have or the suppression strength that we measured in LGN and you can see the striking difference between these two distributions. So here also the second aspect holds true a second hypothesis holds true the neurons in visual TRN have very little surround suppression themselves and are thus ideally poised to provide lots of inhibition during these times when we present big stimuli. Okay, so how about the last prediction? What happens when we suppress corticothalamic feedback? Before I go to the functional responses I want to show you briefly that we also measured the terminal fields in TRN with our triple injection paradigm from the beginning of the talk. You can also see that the corticothalamic feedback provides topographic feedback to TRN. So it's also able to do very interesting modulations at this processing stage. But now if I go and show you the other approach where we now instead of activating or just labeling these projections to TRN we're suppressing primary visual cortex again to see how feedback impacts on the responses of visual TRN neurons. So this is again the recording configuration. We have general rhodopsin in PV interneurons and primary visual cortex. We use the intracortical inhibition to suppress the entire suppress activity across all layers in primary visual cortex including the layer six corticothalamic neurons. And we're recording with our electrode in the visual part of the TRN. Here you can see example responses across time for different sizes of grating during the control condition in black and during the optogenetic suppression condition of primary visual cortex in blue and the corresponding size tuning curves. And maybe if you remember the corresponding plots that I showed you from LGN what is really very striking here is that the response suppression that the suppression of the responses in this in visual part of the TRN is really massive. So we had a small modulatory effect in LGN but here we are almost reducing the amplitude of the magnitude of the response by half. This is also the case for the population where we have a substantial reduction in firing rate when we suppress primary visual cortex. So this seems to be the case that here the inputs this corticothalamic feedback to TRN at least makes up almost half of the response of visual TRN. This is also quantified down here. In addition, I would like to point out that during the one suppression we have a small increase in preferred size for the visual TRN neurons and this might fit nicely to the increase in receptive field that we see in LGN. So maybe the inhibition just kicks in a bit later in LGN and we have no change in suppression index here. It's low in both cases but we have a massive reduction in responses here when we have suppressed primary visual cortex also for the big sizes. So with that, we can conclude this part and say we want suppression reduces visual TRN responses substantially and increases visual TRN receptive fields and this response reduction that I've shown you here is also active at these large sizes where we see this release from inhibition in LGN. Okay, let me conclude this part and this was the main part of the talk. So I have talked about the role of corticothalamic feedback. I showed you in the beginning that V1 corticothalamic feedback promotes LGN tonic firing mode. These V1 corticothalamic feedback projections beyond providing this global modulations have spatial anatomical and functional specificity. They are ideally suited to modulate LGN spatial integration and indeed when we probe this specifically we see that corticothalamic feedback enhances responses to small gratings and it sharpens receptive fields and increases surround suppression. We're very happy to see this in the mouse because these effects are consistent with some observations in primates and cats and it really points towards like maybe a conserved mechanism where we can see the same effects across different animal species. Also the effect of feedback has similar effects on spatial integration and other sensory modalities. So in the somatric sensory system similar effects have been reported and also in the echolocation system of that. So again, I think this is interesting because this points to maybe similar mechanisms being in place in these different modalities and speaks to the relevance of these findings. What our research further showed is that the response properties in visual tear and are consistent with the role in mediating this feedback enhanced surround suppression in LGN because they have these big receptive fields with visual tear and neurons themselves a little suppressed by big stimuli and they're strongly modulated by the feedback. And overall, I think this shows that cortical feedback can dynamically modulate the processing of visual information on route to cortex. So before concluding here, I want to show you just a few more slides on the other aspect that I wanted to talk about and this is behavioral state. So we've seen like beautiful effects of feedback on relay cells, but I also introduced in the beginning that there is a brainstorm modulatory influences on relay cells. And indeed also the relay cells are influenced by the behavioral state of the animal. So from research in primary visual cortex, mainly we've seen that locomotion and pupil size can be used as maybe partially overlapping proxies for behavioral state. And then in V1, if the animal is running you have an increased gain of responses at least on average, we have changes in receptive field size and surround suppression also we have reductions in noise correlations and many other things that happened. So there's a huge literature on that. But like we said, the LGN is also part of this circuit embedded in this circuit and the feedback circuit and also it receives these modulatory influences from brainstorm. So we think it's also an important question to see how behavioral state shapes neural activity in the stage before primary visual cortex, so in LGN. In the past, we've shown that modulations by locomotion occur already at the level of the LGN. So here you see two example neurons, time lock to locomotion onset. This is during spontaneous activity, during visual activity and in both cases the neurons respond with an increase in firing to this locomotion onset. We also have sustained responses. So when we look at LGN neurons response to drifting gratings and these split the trials according to whether the animal is sitting most of the time in orange or running most of the time in green, you can see that the firing seems to change. So we have higher firing rates during locomotion and this is also present in the population at least on average. So we replicated these findings now in the new work also by looking at more naturalistic stimuli. So now I've shown, I show you again the responses to the natural movies that you've seen in the beginning of the talk. This is the example neuron one. This is when the animal is mostly running throughout these five seconds of movie presentation and here the animal is mostly sitting during the five second movie presentations. And you see that there's maybe subtle changes but there are changes, the firing rate here is lower. There's more of these red birth spikes again and maybe the responses seem a bit more sparse in this case. And this is also what you see not only for this example neuron but for the population. So running in LGN like we would expect from the grating experiments enhances firing rates. It reduces bursting. This is also what Christine has shown long time before us and there is reduction in sparseness and reduction in reliability. So now maybe you think, okay, wait a minute this looks very much like what I've shown you for feedback where I also said that feedback enhances firing rates reduces burst ratio at the expense of sparseness and reliability. So is this all just inherited maybe from cortex? So the effects of locomotion here really resemble those of feedback and the question is do they just come from there? Are they inherited via feedback from the cortex? Okay, so in order to answer that question what we did is compute an index for each quantify basically effects of running and feedback for each neuron. We computed a run modulation index that we call RMI which is the difference in firing rate between running and sitting condition over the sum of the two firing rates. And we have made an analogous feedback modulation in the XFMI where we look at the responses during feedback intact minus the responses during suppression divided over their sum. And we made two predictions. So if the LGN run modulations which simply mediated by feedback then the strength of the run modulation and feedback should be positively correlated. So maybe if you have a very strong projection to one particular neuron that mediates both then both of these effects should be very high. What we find is that this is not the case so we have rather no correlation between running modulations and feedback modulations this line is pretty flat here. Maybe more powerfully what we can do is look at the run modulation during suppression of feedback. So the prediction would be if we take away feedback and everything would be modulated or would be mediated by this feedback connection then we should also get rid of the run modulation. But this is not the case. So even when we suppress primary visual cortex we see strong run modulation and in fact they are very correlated between neurons. So no matter if feedback is intact or suppressed the running modulations are very much similar between these two conditions. So we think that this was out the idea that these run modulations are largely inherited from primary visual cortex and rather points to independent effects of running and corticosalamic feedback and these are probably mediated by these neuromodulatory influences. Okay, last few slides. What about more fine-grained modulations of LGN spiking? So we looked at running now which is we classified it as either sitting or running so we binarized it even. How about more fine-grained modulation? So we think that the pupil signal so the pupil area is maybe a more multifaceted signal to characterize the behavioral state modulations in LGN spiking. Can see an example pupil trace here where we have different slow modulations that are also accompanied by modulations at a bit higher frequencies. In order to characterize this modulation of this multifaceted signal of the pupil area what we did is exploited a method based on the Hilbert-Rang transform. So what we did is we decomposed this pupil-sized signal into different components called intrinsic mode functions. So you see the pupil-sized signal all repeated in gray across these components and in color you see the different components that we extracted. Can see that the components span a wide range of frequencies across several orders of magnitude and they occur with different strength basically in this composite pupil area signal. What is nice about having these intrinsic mode functions is that these are very amenable to Hilbert spectral analysis so for each moment in time and each of these intrinsic mode functions we can now compute the instantaneous phase and the instantaneous frequency and amplitude. And this allows us now to relate basically the phase of these pupil signal components to the spiking activity in LGN. And again what we do is we because firing mode matters in LGN and because it's an indicator of behavioral state what we do is we separate the spikes that occur in birth. So these are in red versus those that occur during tonic firing mode in blue. And we can now relate when these birth events or the tonic spikes occur relative to the phases of the IMFs. And here you can see an example of a single LGN neuron and a single IMF and it's in this particular IMF here. And you see three really striking features. The first is that both the burst events and the tonic spikes they are significantly tuned to the phase of the pupil signal component. So both of these phase histograms are not circular but they have a deviation from circular. The second one is that the tuning for the burst events is stronger than for the tonic events. This is the strength of the tuning in the R square. And the third one is that the burst events preferentially happen during contraction of the pupil and the tonic spikes seem to occur preferentially during dilation of the pupil. When we look at now all the neurons that we recorded and all the IMFs that we decompose the pupil signal into you can see this particular pattern here. So on the x-axis you have the frequency of this pupil size component, the center frequency of this pupil size component on the y-axis you have the tuning phase and up here is contraction and down here is dilation. You can see basically in solid dots the significantly tuned neuron IMF pairs. And yes, that's it. That's what you see. What is evident from this picture is basically that you have this anti-phase relationship across many different frequencies of the IMFs of the pupil size components. You have significant tuning across many frequencies of the IMF. So here you see this quantified one more time. We have frequency on the x-axis, tuning strength on the y-axis. And you can see that the tuning for burst spikes is stronger like in the example neuron than the tuning for the tonic spikes. But this tuning is there for a huge range of frequencies and deviation from basically uniform distribution. And down here you see quantified the difference between tonic and burst phase tuning. You can see that for many neurons this is really anti-phase. So then burst spikes tend to happen during contraction and the tonic spikes tend to happen during dilation. Good. With this I'm ending. I want to say that spiking activity in LGN is significantly modulated across a broad range of time scales in the pupil signal and this modulation occurs in anti-phase for both for the burst and tonic spikes. And this is my overall summary. I've shown you that LGN is robustly modulated by behavioral state. And these modulations are largely independent from corticothalamic feedback. So they are not just inherited from cortex. We think likely sources of modulation are the neuromodulatory inputs from the brain state and beyond changing LGN firing rates and receptive fields. The behavioral state seems to modulate burst and tonic spiking across a broad range of time scales that we determined by this decomposition of the pupil signal. With that I want to finish and say I've hopefully convinced you that LGN is more than just a relay. It offers opportunities for feed forward convergence of retinal inputs and this modulates by feedback and behavioral state. And finally I want to thank the people involved that are working in my lab and they are involved in their projects. In particular, this was work that was done by Gregory Bourne, Yannick Bauer, Davide Cromby, Sinema Arisken, Felix Schneider and Martin Spadczyk. Also want to thank everybody else who's working on other projects, our collaborators and our funding. And thanks again for listening and having me here. This was impressive. Thank you very much, Laura, for your presentation. A lot of evidence that the Thalamus is not just a relay and I'm personally like really, really impressed about how systematically and meticulously you try to dissect the contributions of feed forward and feedback pathways. So there are already a couple of questions appearing in the chat. Before I start conveying them to you I would like to remind to the audience that they can either ask a question there or wait for the post-doc informal session where I will invite you and let you in. In the room we are currently sitting in in maybe like five or 10 minutes from now. So I will start with a couple of questions. First I will go with Tom. So the first question from Kim is in addition to the size selectivity effects under cortical suppression, do you also see changes in the shape of receptive fields? So thanks for the question, Tom. I think it's a very interesting one. We've never so far looked at beyond the size tuning how to quantify receptive fields with V1 suppression. So we map them of course, but then we've never looked at more complex changes in shape. It would be something interesting to do because I think there's also, you could hypothesize that maybe some of the directional orientation selectivity comes also via modulations from feedback. And for that it would be nice to see if the receptive fields are maybe more elongated during feedback intact, things like that, but we haven't done it. The second one again from Tom, for size tuning curves in example zone, the spikes seem to often follow it by phasic profile transient and sustained. Does size selectivity based on these two components differ? So you mean the temple, if I understand the question correctly, you mean there's a temple evolution of responses where maybe you have a transient response first and then the more sustained region. And the question is if there's differences between the effect of feedback on this, maybe size tuning for different components of the response, is that the question? So until we wait for Tom to reply, maybe we can also address this at the zoom call like soonies. Yeah, I can say maybe quickly that we never tried so far to look at temporal aspects of this modulation. So this is of course a really interesting question because these effects should evolve over time. For now what we did is just look at the average effects during the stimulus presentation, but since this is a loop and it has different frequencies of communication, I think this would be really fascinating to look at. I see that Tom is already in the room, like if he wants to elaborate, Tom, please go. No, no, he interpreted it right. Thanks, that was what I was asking, thanks. Great, then I will continue with conveying the rest. Next one up is Antónico live. I might have missed this, but how does size tuning care looks like for a V1 neuron? Is it similar to LZN? No, it's a good question. I think in principle, yes, we've never quantitatively compared things like suppression index or the receptive field size is a bit, the preferred size is a bit larger. For V1, you have to know that the size tuning itself is depending on layer. So the lower layers have generally larger receptive field and less surround suppression than the upper layers. So I think if you wanted to do that quantitative comparison, you would need to split V1 by layer, basically. But we've never done it. But it's a good question. Next one is from Stefan Denay. And I'm sorry if I'm mispronouncing names. Fantastic work. I'm curious as to why the cortex would have the function of linearizing the response of thalamic neurons. It seems that the linear response would be easy to implement without feedback. Yeah, so I think this is a puzzling aspect to us as well, this particular response under feedback. So for many aspects, you would think, maybe you want to have a sparse response. You want to have a response that is reliable across trials instead of this more tonic response that is less sparse and less reliable. So what could be the function of this? I think this is one interpretation that we came up with that maybe, I mean, we see that feedback is doing it. Like going along this wake up call hypothesis is one function that we could interpret. But you're absolutely right. So I think we need to understand better what it is and why it's using this circuit and not something else. Thank you very much. Along these lines, before I move on to the next question, I actually wanted to ask you, because you saw for different trials, all these trials are with the same image, right? Yes. And you always play these five seconds of this image. So what you observe is that the bursting part, like when you suppress V1, always appears I think at 0.5 seconds within the image. Do you have any speculation or idea why that might be the case? Yeah. And so I think the bursts are also, I mean, maybe by suppressing primary visual cortex, we take away this excitatory feedback. So the edge end is maybe more hyperpolarized. And then if there is something in the movie that is driving this particular neuron well, like it's depolarizing the movie, then it will respond with this burst. So the burst need like 100 millisecond of hyperpolarization and then something excitatory. So that will elicit a calcium spike and then this is crested by the sodium potassium action potential is the regular one. So we think in this movie is something that neuron likes at that moment of time and because it has been hyperpolarized or there was something like non-preferred basically before it was silent, it can respond with this burst. So it's something salient for the neuron that is happening in the movie. And because, I mean, we think we take away primary visual cortex, the neuron is more hyperpolarized in general, it will respond with this burst. I see. Yeah, I was just wondering because, you know, like the increase in bursting activity only happens there. And like it's quite regular, but like among trials, but not within its irregular throughout the trial. And have you tried like with other videos, like five-second clips or? Yeah. So you will see this bursting in other neurons in other places basically depending on their feature selectivity. We haven't systematically tried to find out like what is the thing in the movie that's driving this particular neuron at that time. But yes, you can see that in different parts. Or if you take a different movie, like you were asking, you'll find it in different time points. Right. So next one is from Jorge Julián Palacios Venegas. Thank you very much. If it's modulated by behavior, could the initial process of the visual attention, wait, it's phrased quite weirdly. So if it's modulated by behavioral state, could be the initial process of the visual attention organization responses? I'm not gonna touch. So maybe this, the question is about if there's behavioral state modulations, if there's also maybe more cognitive influences like attention that could use that mechanism. I assume so. Unfortunately, I'm not entirely sure. Yes. So I think, I mean, there is some pioneering work in Macaulon that looked at how attention would modulate TRN and LGN. And there's also some human imaging, neuroimaging studies that would indicate both of these regions and some attentional processes. So maybe one can speculate that like also this higher order cognitive top-down modulations could exploit the system, but we haven't tested it and we don't know. I can't say anything in the mouth about this. Okay. And the last one appearing on the chat and I would like to remind to the audience that if they want, they can join us in the Zoom room that we are seated following the link I posted in the chat. The last question is from Patricia Orlovska. Excellent talk. Thank you very much. It is known that pupil oscillates in the infra slow range and so do firing in the LGN, OPN and SCN. Have you seen correlation between pupil oscillations and firing in dorsolateral geniculate nucleus? So if you talk in particular about these infra slow fluctuations, I think we are just limited by our recording duration here because there's, I mean, if you have X minutes of recording, then you can only extract pupil fluctuations according up to a certain frequency, up to a lower frequency. And I think we have evidence to say that beyond what we can extract that should probably also be slower fluctuations that we just can't compute basically because our duration is too short, our recording durations are too short. So if we're particularly thinking about these infra slow fluctuations, they might be there, but we just can't resolve them with what we have now. Great, so as I do not see any more questions appearing in the chat, I would like to thank you once again, Laura, for this excellent talk. I would like to thank the audience for tuning in for another of our Sussex Vision Talks and would like to remind you that we should join soonies if you want to continue in the post-talk offline session as I will be stopping live transmission in a couple of minutes from now. So before people start jumping in, I would like to ask you another question because you also like mentioned limitations, like can you actually inhibit, like suppress directly or excite the really, like the phalamic reticular nucleus directly or this is impossible with current? So you can do it. What you have to do is implant a fiber basically and then manipulate these neurons directly. We haven't done it so far. We have just manipulated from outside the cortical neurons, but yeah, that would also be, of course, interesting then to see more interactions in this loop basically if you want to target it for multiple areas at the same time, maybe and manipulate in different places. Right, yes, to try to further dissect apart the components that come right. And like you mentioned that you know the feedback, like that the feedback is quite topographic. What do we know about the feed forward, like from DLGN to relay to reticular story? I don't know very well, but I think it should also be organized, of course, in some way topographically. It's a shorter feedback loop. I mean, it's also feedback loop. It's a shorter feedback loop. It contributes probably the remaining part of the response that we could still see when we inhibit the feedback. But I don't know if there's very specific, at least I don't know any very specific walls of that connection. So there's lots of things to investigate, I think. Yeah, and quite complicated circuitry in the end. Okay, so with that I will end the live stream. So once again, thank you very much, Laura, for your talk today with us. Thanks for the questions and thanks for listening and your attention and thanks for having me.