 I guess we're live. Hello everyone and thanks for being with us today. This is already the seventh talk for the Sussex Vision series. As you may know, hosting these events will not be possible without the logistics provided by the worldwide number of organizers, so once again I'd like to thank them for their continuous work and support. So for today I'll be a host, I'll be a host of this session. My name is Maxine and I'm currently finishing up a PhD in Tom Baden's lab. Today I'm particularly pleased to receive Catherine Franke. She, my senior in the sense, has Catherine completed her PhD in 2016 under Tom Baden and Thomas Solot's provision at the University of Tübingen. In 2017 she received from SFN's NEMCO prize in cellular and molecular engineering science for the work covered in her PhD thesis, functional characterization of the excitatory pathways in the mouse retinal. Since January 2017, Catherine had been a general research group leader at the Benstein Center for Computational Research Science and the Institute for Optimic Research in Tübingen. Her work focuses on how cells in a early visual system act together in complex terminal circuits to extract relevant visual information. And today we're looking forward to a talk about retinal cycle functional diversity and the role in chromatic processing. So feel free to use the YouTube touch to comment or to ask questions to our guests. They will be answered at the end of the talk. So hello Catherine and thank you for accepting our invitation today. Yeah, hi Maxine and hi everyone. Thank you very much for the invitation and giving me the opportunity to present some of my work that I did during my PhD and since I have my own group here at Tübingen. So I will now start sharing my screen. Okay. All right. So what I want to talk today is what the eye tells the brain, visual feature extraction in a mouse retinal. All right. So like all sensory systems, the visual system builds an internal representation of the outside world by extracting relevant information from the environment. And here it is important to consider that relevant depends on the species specific needs of the animals. So this is nicely illustrated by this picture here where, for example, the owl needs to have a very high spatial resolution to detect prey like the mouse from a large distance in the sky. And this is especially important when the mouse is located in a more natural environment. In contrast, the mouse might not care so much about high resolution, but it needs to detect large approaching predators from above. And this kind of visual feature extraction already starts in the retina. So the retina is a layered neural tissue located at the back of the eyeball, and the visual image is projected out to the retina by the optics of the eye. In the retina, the photoreceptors first transform the light into electrical signal, and this electrical signal is then forwarded to, is then transmitted from the bipolar cells to the retina's output neurons, the ganglion cells. And this vertical excitatory pathway is modulated by two classes of inhibitory neurons, the horizontal cell in the outer retina, and the armor green cells in the inner retina. So in general, this retinal blueprint is highly conserved across vertebrates with some species specific modifications, of course, like, for example, the fovea in primates, and that's present in primates, and also some bird species. In contrast to most other sensory systems, the sensory information is highly processed in the retina before it is sent to the brain. And this is why the retina is also considered mini brain in a dish. So here the question is, how does the retina process this information? Or in other words, what does the eye send to the brain? So the classical view is that the receptive fields of retinal neurons have a center-surround receptive field structure, meaning that these neurons are excited by a light spot within their receptive field center, increasing their spiking activity while they are inhibited by light in their surround, decreasing their spiking activity. And then each pixel of the incoming image would be filtered by these centers-surround kernels before transmission to the brain. However, already very early work has shown that receptive fields of retinal neurons might be more complex than that. So for example, in the study, what the frog eye tells the frog's brain, Latvin and colleagues identified retinal output neurons that responded specifically to a small dark spot moving in its receptive field, while independent on the illumination intensity and also background motion. So this corresponds to rather a complex feature detectors, which might be ideal to detect bugs. So since then, a lot of work has shown that the retina is not just decomposing the image into many pixels, which are then relayed to the brain one by one. But instead, the retina extracts different visual features like contrast, edges, or motion, and these different representations of the same scene are then sent to the brain in parallel. And since the first recordings of retinal neurons more than 50 years ago, there has been the question of how many feature channels exist from the eye to the brain and what do they encode. So during my PhD, we addressed this problem using the mouse as a model organism. So in general, for mammals, there are very few feedback projections from the eye to the from the brain to the eye, which is why the retina on the X-bevo retina can be considered like intact system. So here you can see the whole mounted X-planted mouse retina that was electroporated with a calcium indicator on Oregon green vector one, resulting in nearly complete labeling of this one cell layer, which is the Gamnion cell layer, the output layer of the retina. So then we projected different visual stimuli onto the photoreceptor cell layer of this tissue, where we recorded the light correlated activity of these neurons using two photon calcium imaging. And here in this movie, you can see the raw calcium changes of the cells, and we recorded different kinds of stimuli. So for each of the cells in the end, we extracted the response profile in response to these different stimuli. So for example, we recorded the full field chirp stimulus, which consisted of a step of light, and then frequency and contrast modulations, moving bar stimulus to test for direction selectivity, and also wide noise stimulus that allowed us to characterize the spatial and temporal receptive fields of the neurons. So they are preferred stimulus in space and time. And for this specific cell that is also marked here in this movie, we can see that this is an off cell, meaning that it shows an increase in calcium and activity when the light is switched off. It also nicely follows the frequency and contrast modulations of this full field chirp, is not direction selective, so shows a similar response to a bar moving in eight directions and has a relatively large receptive field. So in this study, we recorded these response profile of more than 10,000 cells and then used an unsupervised clustering algorithm to group them into functional types. And here I just want to mention that, so the result was that we obtained more than approximately 32 different functional ganglion cell groups in the mouse, broadly classified as off cells, on off cells, fast and slow on cells. And this number of 32 was approximately twice the number of previous estimates, suggesting that the information channels from the mouse's eye to the mouse's brain are substantially more diverse than previously thought. And since then, in the last four years, this result of ours has been confirmed both morphologically based on EM reconstruction data and also genetically by single cell transcriptomics. So now we can go back to the single neurons I have showed you on the previous slide and color code them according to the functional group they were assigned to. And now in this movie, you will see the responses of these cells to the moving bar stimulus and also to the chirp stimulus. And so these movies nicely illustrate that already a very simple stimulus like a moving bar or chirp stimulus, differentially activates these different cell types indicated by the fact that these different colors light up at different times of this visual stimulus. Okay. So at that point, we might have to question, so why are there so many different written output channels to the brain? And so when we plot here the on or off preference of the different functional groups versus the preference for local or global stimuli, we can see that our functional groups cover this relatively simple feature space like more or less homogeneously, meaning that each of these functional groups responds only to a highly specific combination of these two visual features. So this suggests that what the retina does it decomposes the visual input into many output channels and each is selective for one or very few visual features. And so computational studies suggest that this pathway splitting reduces redundancy and therefore provides an efficient code to the retina. So in the following, I want to I will focus on the mouse as a model system, but I just want to briefly mention that the number of retina output channels to the brain varies across species. And so for example, in mouse with more than 40 different output channels compared to, for example, primate where in the central retina, we only have four to five different four to five dominant cell types. And there has been since a long time this concept, the dam of the brain is smarter than retina, suggesting that lower vertebrate species have a smarter retina, while like higher vertebrate species like primates, for example, might have a more linear retina and relocate more complex computations like direction selectivity to the cortex. And I just want to highlight this recent paper from Stefan Denizlab, where they investigated this concept computationally, which is I think a very nice example how deep neural networks can help us to test like concept, conceptional ideas we have in visual neuroscience. All right, so so far I've shown you that there are approximately 40 different ganglion cell types in the mouse retina. So the question is, how does the retina generate such a functionally diverse output, especially considering that only two center cells upstream, we only have three types of photoreceptors. So in the following I want to show you three different studies that I worked on that address these questions from different angles. And the first study addressed the functional diversity of bipolar cells, which transmit the information from the photoreceptors to the ganglion cells. So here you can see a morphologically reconstructed bipolar cells from the mouse retina and the different types mainly differ with respect to the stratification of the axon terminals in the synaptic layer called inner plexiform layer or IPL. And to record the output of all these neurons, we express the glutamate indicator, I glue sniffer throughout the whole inner plexiform layer, which you can see here, and then recorded their responses to visual stimulus like a noise stimulus. So we use these responses to the stimulus and a local image correlation to then estimate regions of interest in this exemplary scan field. So here you can see a region of interest mark with different regions highlighted in colors. And in this study, we were able to show that these different regions correspond to the output of single bipolar cell axon terminals, and that within one scan field, we can record the output of different bipolar cell types. So here you can see now the response of a single axon terminal to two versions of the trap stimulus you have seen before. So the local trap stimulus is spatially limited and mainly activates the excitatory center provided mediated by excitatory input from food receptor at the dendrites of bipolar cells. In contrast, the full field trap stimulus is larger stimulus and in addition also activates the inhibitory surround mediated mainly by inhibitory input to the axon terminals of the cells. And this additional inhibition results in a different response profile of the cells. So in the study, what we found is that if you look at the output of different bipolar cell types of the same response polarity, like for two on cells type six and type nine, you can see that their output is highly correlated indicated here by a high correlation coefficient. So and they have the question, so why would you invest into many of these different cell types if they actually respond the same to visual stimulus? However, we found that upon this full field stimulation, these responses get decorrelated and in this specific case, for example, one type gets much more transient whereas the other one is more sustained. And this finding that full field stimulation decorrelates bipolar cell was also the case for an exemplary pair of off bipolar cells and in general across all bipolar cell types. So bipolar cell functional diversity relies on full field stimulation. And here the next question, so what is the neural correlate of this decorrelation? I already on the previous slide mentioned to you that the full field stimulus in addition to activating the center mediated by photoreceptor input, it activates the surround of the bipolar cells. And previous studies have demonstrated that this bipolar cell surround is mainly mediated by wide field GABA-ergic amicron cells. So to test whether these cells provide the decorrelating input to the cells, we performed pharmacological experiments where we blocked these cells specifically using a GABA receptor antagonist. Here you can see the response of one exemplary region of interest in response to the local stimulus and the full field stimulus. And you can see that you have very different response profiles. So up on blocking these GABA-ergic cells, however, the cells, the responses of the cells get much more correlated and this was again true across all different bipolar cell types. So what this showed is that wide field GABA-ergic amicron cells provide a decorrelating inhibition to bipolar cells. And on a more general level, this study highlighted the fact that bipolar cell functional diversity requires an interplay of excitatory and inhibitory inputs. All right, so this first study showed, so when we go back to the question, how does the retina generate a functionally diverse output, this first study showed that actually inhibitory inputs from amicron cells play a major role in diversifying the signal in the inner retina. So next we asked the question whether there are also cell type specific intrinsic properties that might help to increase the diversity of retina signals. So from cortical and retinal work, we know that the output of a neuron is critically dependent on how the inputs are integrated in the dendrites of the cells. And this is why we next, however, it's so far really not clear whether there are cell type specific dendritic integration profiles that help diversify signals across neuron types. And so in this next study, we focused on two cell types, the transient of mini cell and the transient of alpha cell, which receive overlapping excitatory inputs and ask the question whether they have cell type specific dendritic processing in these two cell types. And this study was performed by two PhD students, Yanli Ran, who was, who was close supervised by Thomas Euler and me, and Xiwei, who was a PhD student of Philip Barron's lab. So Yanli filled single ganglion cells using sharp electrodes with the calcium indicator, Oregon green peptide one, and then recorded the responses of different dendritic region of the same cell to a noise stimulus. This allowed us to estimate receptive fields of different dendritic regions. So here, relatively close to the soma and here a distal part of the dendrite. And then using a threshold, we were able to draw a contour of these receptive fields and overlay them with the cell's morphology. So Yanli have repeated this procedure many times for the same cell at different dendritic regions, resulting in a large number of receptive fields, which allowed us to systematically investigate how visual information is integrated in these dendrites. So to quantify dendritic integration, we looked at different parameters like receptive field size. Here you can see exemplary cells in different regions recorded across the dendrites with their respective receptive field contour. And we found that the transient of alpha cells showed a decrease in receptive field size when you go from the soma further to the distal dendrites. In contrast, this was not present for the transit of mini cells, which did not show a systematic difference in receptive field size. Another thing we quantified was a receptive field overlap. And we found that if you record from dendritic regions at different branches of these alpha cells, that they don't really have any overlap, meaning that they are spatially independent. However, receptive fields always overlapped for this transient of mini cell, even if we recorded at opposite sides of the cells dendritic arbor. And these differences are now highlighted in this movie here, where the red dot indicates the position that was recorded and the red outline corresponds to the respective receptive field contour. And in these movies, we go from one side of the receptive field dendrite to the other side. All right. And so you can see that when you look at the alpha cell, that the receptive field position and also the size changes systematically in the time of throughout this movie, while for the transit of mini cell, receptive field size and position remains relatively similar close to the soma. So these experiments showed that the cells have different dendritic integration profiles. So the transient of alpha cells have isolated and spatially and temporarily independent dendritic segments. In contrast, the transient of mini cells have highly synchronized spatial and temporal dendrites. So what could be the mechanism underlying these differences? So again, from previous work, we know that back propagation of somatic signals critically influences dendritic signaling. So to quantify the signal spread from the soma to the dendrites, so Yanli next patched these cells and injected somatic current, while at the same time recording the dendritic calcium signal of these cells either close to the cell or at distal dendrites. And what you see here is the current injection protocol, the spikes recorded at the soma and the calcium recorded either close to the soma or at the distal dendrite. And for this transient of alpha cell, we found that dendritic signals evoke very small, no somatic signals like the spikes evoke very small dendritic calcium signals, especially for dendrites that are further away from the soma. And she also performed this kind of experiment in transient of mini cells and found that here the somatic signals evoke much stronger dendritic signals, even for dendrites that are further away from the soma. And again, we quantified this by estimating the mean calcium signal for a specific number of somatic spikes and found that these calcium signals are much weaker for the transient of alpha cell than for the transient of mini cells. And that this is statistically significant. So this means that the stronger back propagation in the transient of mini cells might synchronize the dendrites resulting in these different dendritic integration profiles we have seen on the previous slide. And these different integration profiles in the end might help the cells which receive overlapping excitatory input to process it differentially to result in different output profiles. All right. Okay, so so far we have covered, we have found that inhibition is decorrelating retinal responses, which is a general feature across all bipolar cell types. And we have focused on two specific ganglion cell types and found that dendritic integration can help to diversify retinal signals. But ultimately what we want to understand is how does feature selectivity arise across all retinal layers. And we have recently tried to address this question by focusing on the retinal circuits that extract color information. So color information here or color in general is like a good test case, because it can easily be addressed by changing the wavelengths of the input stimulus. In general, the prerequisite for color vision is that different cone foot receptor types, or that you have different cone foot receptor types sensitive to different wavelength, and that the signals from these photoreceptors are somehow compared along the visual pathway. Primates, including humans have trichromatic vision, meaning that they have three different cone foot receptor types, expressing the blue sensitive S option, the green sensitive M option, or the red sensitive L option. In contrast, mice like most mammals have dichromatic vision, so they only have two foot receptor types, the S cone and the M cone. And interestingly, the S cone in the mice or rodents is shifted towards UV sensitivity. And this UV shift has been, or it has been proposed that this UV shift facilitates the detection of dark objects in the sky, like approaching predators. However, as we don't see UV light because our lens filters it, and also most cameras actually have a UV filter, we don't really know how the chromatic input to the mouse looks like, or in other words, how the mouse sees the colored world. And this is why a PhD student in Thomas Euler's lab, he recorded these natural scenes with a camera that was adjusted to the spectral sensitivity of mice, allowing to also record the UV spectrum of these scenes. And he also took into account other features of the mouse visual system, like the wide opening angle on the large field of view. And if you're interested in these natural movies, you should watch Thomas' talk next Friday on World Wide Neural. So in these natural scenes, what we found is that chromatic contrast is actually enriched in the upper visual fields of these scenes. So meaning that the contrast in green versus UV band is more different in these upper visual scenes. And so what this would suggest is that mice might be able to better discriminate color in this upper visual field, because why would the visual system invest into expensive circuits to extract chromatic information if this is not homogeneously distributed across the visual field. And indeed, this is what a recent paper found. So the authors, they looked at or investigated color discrimination in mice, tested three different positions along the vertical axis of the visual field, and found that mice were able to discriminate colors at these three locations in the upper visual field, but not in the lower visual field. However, the neural correlates for this behavior and the retina are still unclear. And especially, and so to the most part, this is because the retina, the mouse retina has actually quite a atypical asymmetric distribution of its, of its obscene types. So most animals have, for most animals, the different cone types and therefore the spectral sensitivity is more or less homogeneously distributed across the retina. And while this is the case for UV sensitive S cones in the mouse retina here indicated by the purple dots, the green sensitive M cones, they co-express the S option with an increasing expression, a co-expression ratio towards the ventral retina. In the end, what this results in is a ventral retina that is mainly UV dominant, and a dorsal retina that is mainly green dominant. And now, so this asymmetric distribution is not specific for mice, but can also be found in other species like rabbits and many insects. And however, since color vision requires that chromatic information is locally compared by neural circuits, it's unclear how these different animals can use this asymmetric distribution to see colors, especially in the ventral retina where you may mainly have one obscene type, which is S obscene. So therefore, for this next project, we wanted to look at the regional correlates of color vision in mice. And specifically, we aimed at the systematic characterization of chromatic processing along the vertical signaling pathway of the mouse retina. And this is the food receptors, the bipolar cells and the ganglion cells. So for chromatic stimulation, we used a visual stimulator that we recently described that is a DLP-based projector without any LEDs, but with a light gate port. And that allows you to couple in an arbitrary array of LEDs. And we use here a UV and a green LED matched to the mouse's spectral sensitivity. And this is important because conventional display devices have the already LEDs in conventional display devices. For example, here from a TFT monitor really fails to activate UV sensitive S obscene mice. So solutions and calibration notebooks for other species like zebrafish or mouse and vivo experiments can be found online. All right. So as a first step, we wanted to address the question, so how do cone food receptors process color information? And so previous studies focusing on this have mainly been performed in retinal slices, where the inhibitory network is not completely functional. And so here what we want, so to record the cone output, we express the glutamate indicator idlusnifer throughout the whole retina. And here you can see one scan field that was located in the dorsal retina in the outer plexiform layer where the cone food receptors have their axon terminals. So now if we apply flashes of light, we can see that there are some activity hotspots modulating their glutamate release upon this light. And based on these responses, we can estimate a correlation image where each pixel here is color coded according to the correlation of neighboring pixels. And so you can already see by eye that there are different activity hotspots. And we found that these glutamate activity hotspots, they align with anatomical cone axon terminals visualized using the disulfurodamine 101. And also form regular mosaics suggesting that these glutamate hotspots correspond to single cone axon terminals. So now to investigate the chromatic processing, we use green and UV center and surround stimulus. So here you can see the mean response of a single cone terminal to the stimulus. And so as vertebrate food receptors are off cell, you see a decrease in glutamate release upon the center stimulus. And because this scan field was recorded in the green dominant dorsal retina, this cell shows a stronger response to green than to UV. Now the advantage of our whole mount approach is that we can now also look at the chromatic preference of the spatially extended surround. And for this cone, you can see an antagonistic surround response that's also green dominant, showing that the cell has a center and a surround chromatic preference of green. So now we can repeat this experiment in the venture retina. And we find that in line with the obscene expression that the cell has a UV dominant center response. However, if the cones have an antagonistic surround, this is always green dominant, resulting in color opponent centers around receptor fields of these neurons. Here you can see the center and so the all the cells that Marie Lee, when she was a master student in the lab, recorded across the retina. And each cone terminal is color coded by its spectral preference, estimated as spectral contrast, which is minus one, if the cell only responds to UV, and one, if it only responds to green, and zero, if it responds to both colors the same way. And we can see that the center responses of cones actually nicely follow the obscene gradient across the retina with UV sensitive responses in ventral and green sensitive responses in dorsal retina. However, the surround distribution looks very different, especially in the ventral retina, the preference is highly shifted towards green preference. And there's a strong difference between center and surround chromatic preference of these ventral neurons, then results in color opponent field full field responses, meaning that they are single cone response with a different polarity to a UV or green full field spot. So the question, however, is, where does this green surround response in the ventral retina come from? So previously, they have has been a paper suggesting that or showing that one specific type of retinal ganglion cell shows such centers around color opponent responses in the ventral retina using rods signals that are relayed to cones via horizontal cells. So the inhibitory neurons of the outer retina. To test whether this was also the case in our experiments, we pharmacologically blocked horizontal cells with NBQX, resulting in an increase in baseline of cone responses, but also this green antagonistic surround response was then abolished, suggesting that color opponent C and ventral cones is mediated by horizontal cells that are likely driven by green sensitive rods. So to inform behavior, this signal in the end has to reach the brain. So the next question is, how is the chromatic information from cones processed by downstream circuits? So we next focused on the bipolar cells. I have already said that they transmit the information from the food receptors to the ganglion cells. And in mice, we know that there are 14 different bipolar cell types that have their axon terminals in different parts of the synaptic layer. And ideally, we want to record the glutamatergic output of these cells at the same time. So to achieve this, C.C. a postdoc in Thomas' lab, he equipped our setup with an electrical tunable lens that allows us to shift the focal plane of the laser very rapidly and record the different layers in the IPL more or less at the same time. So here you can see the response of an IPL to the chirp stimulus. And you can nicely see the segregation of the synaptic layer into on cells that respond to light increment and off cells that respond to light decrement, now responding in anti-phase to the flicker. Again, we used local image correlation to draw regions of interest and confirmed previous results that responses at different position in the IPL have very different response profiles, ranging from sustained off cells to more transient off cells and transient on cells to more sustained on cells. Showing that with this method, we can use this method to record the output of nearly all bipolar cells at the same time. So now we combined this method with chromatic stimulation to test how bipolar cells encode chromatic information. Here you can see the response of an exemplary scan field in the ventral region to UV and green centers around flicker stimulus. And again, we used image correlation to draw regions of interest where every region corresponds to the output of a single bipolar cell axon terminal. So now we can look at the response of single terminals to the stimulus, which as I said, a green and UV centers around flicker stimulus. So we can use these responses to estimate stimulus kernels for UV and green center and surround and also the mean glutamate response in response to fulfilled flashes, the onset and the offset of these flashes. So I'm happy to go through this in detail if you have questions later, but so we used these center surround kernels to estimate the preference of the cells. And we used these responses, the mean glutamate responses to fulfilled onset and offset spots as a measure of color oponency. So for example, for this specific scan field, we found that most of these regions of interest are UV sensitive in line with the position of this scan field in the ventral retina. However, if we now look at the surround responses, they are systematically shifted towards green and this difference was often more pronounced for the off layer here than the onset layer here. And this very strong difference between center and surround chromatic preference in the end resulted in color opponent responses of these neurons, which is here indicated in red and which can be seen, for example, here. So this antagonistic responses of these cells to UV and green can be nicely illustrated when we look at the response of this off layer in response to UV and green sign modulation of a full field stimulus, where you can see that this responses is in perfect anti phase. In contrast to the ventral retina, the dorsal retina showed mainly green dominant center responses and a slightly UV shifted surround. However, as this difference in chromatic preference was much weaker, there were rarely color opponent cells in the dorsal retina, which would be indicated in red here. Okay, so we performed these experiments across different regions on the retina. And here every dot corresponds to one, bipolar cell axon terminal. And again, these are color coded according to center and surround chromatic preference and full field opponent. So these results at the level of the bipolar cells are very similar to the cones, where the center responses mainly matched the option gradient across the retina. However, the surround chromatic preference seems to be nearly inverted with a green dominance around response in the ventral retina. And the strong difference again then results in full field color opponent see here indicated in red for most of these ventral neurons. So as in, as a set next step, we wanted to test whether they're, as I said, they are different bipolar cell types. And they could process this information differentially. So we therefore we plotted the difference in chromatic preference versus IPL depth, which is the synaptic layer. So here a value larger than one zero indicates that the surround is green shifted, and a value smaller than zero indicates that the surround is UV shifted relative to the center. And as we already know, the ventral retina has a green shifted surround. And interestingly, this was significantly more pronounced for the off layer and the on layer. And in contrast, the dorsal retina shows a much weaker difference between center and surround, and therefore little color opponent cells. All right. So these results show that mouse bipolar cells relay the chromatic signals from cone foot receptors to the inner retina. And they do so differently for on and off bipolar cells. So however, is this chromatic information done finally sent to the brain? And this is what Claudia, a PhD student in the lab looked at. So she again, used the bulk electroporation of Oregon Greenbub Davon to label the cells located in ganglion cell layer, and then recorded the calcium responses of these cells to the similar colored flicker stimulus. So first we recorded again the acromatic stimuli like full field chirp and moving bars. And similar to the bipolar cells, the responses of these cells to a UV and green center and surround flicker stimulus that gives us the stimulus kernels and also the full field events for estimating chromatic preference and full field opponents of the cells. Again, I'm happy to go into detail later if you have questions. So then she repeated this experiment across many different regions on the retina. And here each dot corresponds to one cell located in the ganglion cell layer color coded by center and surround chromatic preference and full field opponents. So basically we can reproduce the finding from the upstream processing layers that the center responses more or less follow the obscene gradient, although there's more variability now at the level of the original output. And that the surround looks very different with a strong green shift in the venture retina resulting in the fact that most color opponent retina ganglion cells are actually located in this part of the retina. So in the beginning of the talk, I mentioned that there are more than 40 different ganglion cell types in the mouse retina. So which are the types that then send this color opponent to the brain for driving any color vision behavior? So to address this question, we assigned the cells that we recorded here to the functional groups I have shown you in the beginning of the talk using the acromatic stimuli like the TREP and the moving bars. And now here you can see the blue bar indicate cells that are not opponent and the red bars indicate cells that are opponent across these different ganglion cell groups. So what we can see is that most ganglion cell groups actually contain a few color opponent cells, suggesting that this is a general feature of ventral retinal neurons that is likely inherited by the presynaptic circuits. However, there are also differences. For example, type 12 or type 18 have very few color opponent cells, while for example, type 27 and 31 have more than 50% of the cells are color opponent. So where do these differences originate from? So they could come from the fact that these cells have differences in center and surround chromatic preference. For example, let's consider the cells with few color opponent cells. They might have a very similar spectral preference for center and surround resulting in very few color opponent cells. In contrast, the cells with many color opponent cells could have a very strong difference in center and surround chromatic preference than resulting in many full field color opponent cells. So to test that, we use the permutation test and I will briefly explain it, but we can also go into detail later. So we generated a distribution of expected percentages of color opponent cells, given the center and surround chromatic preference, but shuffling the group labels. And then we compared this estimation to the raw data. So for example, group number 12, the box plot shows the distribution of expected percentages for cells that have very similar centers around preferences as the cell. And you can see that the black dots of the real data lies within this range. However, for example, group number 18 has significantly less color opponent cells than other cells with the same center and surround spectral preference. Similarly, we can find a lot of different groups here indicated by the red arrows that have more color opponent cells than expected from their center and surround chromatic preference. So what this means, or the take home message is that type specific, that there is type specific and nonlinear processing of chromatic centers around information at the level of the original output in contrast to the previous processing layers, and that this in the end results in a more diverse chromatic output to the brain. And before I sum up, I briefly want to mention that we find that color opponent of single original gamut cells varies across the retina. So for example, when we look at group 27, which has many color opponent cells, we recorded the cell type across many positions on the retina. And we can then look at the color opponent of the cell, which is here indicated by this red cells. And we find that only cells in the ventral retina are color opponent. So suggesting that the functional feature of a cell like color opponent might vary in one cell type depending on which part of the visual space it samples from. And this raises the question of how variable functional properties of single cells types are and how we should actually define a single cell. All right, to sum up this color part. So we have found at the level of the retina, that most color opponent retinal neurons are located here in this ventral part. And this actually nicely fits to behavioral data, showing that mice are best at discriminating different colors here in this upper visual field, which is encoded by the ventral retina. And this difference across the visual field might be an adaptation to the input statistics, which ensures that the chromatic information that is enriched in the upper visual field is efficiently encoded. And so far, we have mainly used artificial stimuli to address chromatic processing in these cells. But in the future, we also want to, we now use natural movies to investigate chromatic processing because there's increasing evidence that natural stimuli drives neural circuits differently. And therefore, this highlights the importance of using more natural stimuli in understanding circuit function. All right. So what I showed you today is that how does the retina generate a functionally diverse output? So the first study I showed you highlighted the importance of inhibitory circuits in generating diverse retinal output channels, and that this is done by transforming a change in stimulus size into a change in temporal response profile. So the second study showed that cell type specific dendritic integration profiles contribute to diversifying the signals in the inner retina. And finally, we tried with the example of color to see how the selectivity of single visual features arises and that it progressively increases from the photoreceptors to the original output and that it finally involves nonlinear integration of centers around information. And I think I'm nearly done. So I will skip the last two slides and just continue with thanking people. So I want to thank everyone who has been involved in the study, mainly. So all of that was performed in close collaboration with Thomas Euler and Philip Behrens and all the people who have helped with the experiment like Claudia, my PhD student and Marie Lee, who has been a master's student in the lab. And so currently we are working on chromatic processing in visual cortex and this is in close collaboration with Andreas Tullius and Fabian Sinsen also, analysis lab. And yeah, thank you for your attention and the funding sources here. Thank you, Catherine. That was a very nice overview. I guess I will give you a second here. So thanks for that. We actually have a couple of questions. There's a very popular one regarding the bipolar cell chromatic responses. So this question is from Mahler-Fehler. This seems to mean the center of the cones and surrounding the road will have very different dependence on light intensity. So those color or tendency depend on intensity. Yes. So yeah, that's a very good question. So the light intensity that we have used is in the low photopic range. And so it's not so for our experimental paradigm, it's not so easy to increase the light like significantly because that causes like bleed through in the channels to like in the two photon detection channels and also decreasing it a lot might not work because we have this background illumination due to the two photon laser and directed indirect activation. So what we, that's a very good question. And we performed some electrical recordings to make sure that this color opponent sees also present, let's say without the presence of the two photon laser. And in these electrical recordings, we could now change light intensities, like increase the light intensities or decrease and check whether this changes with light dependence. And so I think that's a very important point also because I think it's still there are like controversial results about in what light regimes rods are active in mice. So at least in the low photopic regime that we are using, there's now a lot of evidence that they are active. But it's really unclear or it's not really tested so far in which conditions they are active across the different lock units. We have another question from Greg Schwartz. Yes. VCs regarding the dendritic back propagation. Yes. Is the input resistance different in the d of mini and d of alpha? Does that explain the difference in back propagation of action controls? Okay. I'm not exactly sure about this answer. We would have to ask the PhD student. But what I can say is that we not only did this experimental, we not only tested the mechanisms of then different dendritic integration profiles experimentally, but we also together with a student in Philippines that we built like a simple biophysical model testing the contribution of morphology of the cells versus different conductances. And I'm happy to go into detail there. But for the specific answer about the input resistance or not, we would have to ask the student. Thanks for that. I have another one from Vikrant. Do you observe attractor like dynamics due to the correlation by a governed cells, something on the lines of upfield network? Can you, sorry, can you repeat that again? If you observe something similar to a upfield network, I'm not sure I get the question. Let me let me open. So who was the person who asked it? I can check it in the Google docs. This one is the first one. Oh, yeah. Okay. I see. Well, I'm actually not really sure what is meant by this attractor network. So maybe can we, can this person like specify or is that not possible? Yeah, or I can ask this person to join in the chat and we can postpone these particular questions. Yeah. Yeah. Yeah. Good idea. Yeah. So I have one from Brent Young. With non-specific calcium imaging in a ganglion cell layer, how do you distinguish between retinal ganglion cells and displaced amigurine cells? Yes. So yeah, that's a good question. So approximately 50% of the somata that are located in a ganglion cell layer are displaced amigurine cells. So in the study that I showed you in the beginning for a subset of experiments, we performed subsequent immunolabelling of the cells. So we labeled the amigurine cells using antibodies and then used these immunolabels also for clustering to distinguish between ganglion cells and displaced amigurine cells. Brett Schwartz is also asking, does a mouse have cone-cone co-openancy or is it just all road cone? Yes. So most of the evidence that we have points towards the direction that most of the oponency is really cone-rod oponency. And however, so we have specifically looked for ganglion cells that, for example, show color oponents receptive field center. And we find that there is a very low percentage of cells that does have a center oponent receptive field. But since we don't really center the stimulus on the receptive field of the neurons, I can't really tell you or I can't be sure whether that's really cone-cone color oponency or not. And there are some other studies. I think there's one from the Burson lab that shows that there's, but that's also a centers around mechanism, not really a center mechanism like in primates. So yeah, so so far we have relatively little evidence for that. Okay, a quick one from Christian Poehler. I can oponency be a general feature across ganglion cell types on versus off when oponency is strongly dominating in off bipolar cells. How is oponency biased toward off bipolar cell in the first place? Yes, that's a good question and I don't know. So yeah, so why off bipolar cells have a stronger difference in center and surround chromatic preference, I'm not sure. But then when you look at the percentage of color oponent bipolar cells, also the on bipolar cells are a full field opponent, although their chromatic preference is not as extremely different as the off bipolar cells. And so the question is, so why how can it be a general feature? So I think it is a general feature of bipolar cells. So it makes sense that it is a general feature of ganglion cells. But what is weird, for example, is that at the level of the ganglion cells, we find a lot of on ganglion cells have many color opponent neurons, while a lot of off bipolar cells are color opponents. So how this transformation is happening from bipolar cells to ganglion cells is not really clear. But we would be very interested to study in single types, for example, to do something similar with the dendrites, I mean, perform dendritic imaging of these cells and see how their output is generated by the inputs from the bipolar cells. But I can't really answer this question. I will just finish this session by a quick question from Dambaden. How does it have to get before you don't see nothing anymore? Yeah, that's a good question. I'm not sure. So I think that's something that we should test at some point because so there's also some evidence that we have suggesting that UV cones and I think Tom also finds this in zebrafish that UV cones are much more sensitive than green cones, for example. And so we have recently done a test experiment where we just kept the green intensity the same, but lowered the UV intensity by a factor of 10. And it seems like the overall chromatic preference doesn't change very much at all. So it would be interesting to like decrease the UV intensity even further. But as I said before Tomala's question, at some point we have to switch to electrical recordings because the background intensity or the activation direct and indirect from the laser might be stronger than our stimulus in the end. But that's an interesting question. We do observe that in zebrafish. So thanks a lot, Katrin. Yeah, thank you very much. I will encourage everybody to join us on the Zoom room that was shared on the chat. So if you want to further discuss with our guests, if you want to ask other questions or just talk about chromatic processing in a radio. So we're waiting for you. And I can see that Jeff Diamond is already here. Hello, Jeff. You had a comment. Well, I had a comment. I thought, I mean, oh, well, I was, my comment was that Tom was looking good, but his appearance degraded substantially in the last few seconds. I do apologize for that. Is that your sheep, Tom? Yes, it's one of my many sheep. It's what we do here. So we do sheep now. Hello, you. So I just saw this passing. Jeff, you had a comment about, I mean, I cannot find it in the chat. Oh, no, I mean, I was interested. Katrin, there was some discussion going on that Marla started actually about the intensity dependence of the surround. And I mean, you're focusing on the potential implications of chromatic surround. But it's sort of, you also might have a situation where in mesopic or photopic conditions, the surround is a constant, and the center is being modulated by relatively small changes in contrast. I have no idea what that would be good for. So you mean if the intensity is not changing at the same rate for center and surround, but if you have different changes in luminance, for example? Yeah, let's say you're up in a region where you're at the top of the rod activation level, and then you're modulating the contrast there. You'll get more modulation of the center than the surround. And is that useful? Yeah, I think the question about how part of that is useful, like in the end for the animal, or in what specific lightning conditions and scene conditions it would be useful is very interesting. But I'm not really sure how to easily address that, except for like really performing behavioral experiments in the end. I mean we can of course like quantify the dependence of center and surround on luminance and so on, but in the end to see for what specific conditions it would be helpful, it's like a really difficult question I think to address. Yeah, I mean is there any reason to think that that's any more less useful than the chromatic surround? So what do you mean exactly with that? Well, I mean I'm agreeing with your point that it's hard to know which feature of this would be most useful to the animal. Yeah, I'm not, yeah, hi. Hi, great talk, Katrin. Quick question. So how is surround changing from dorsal to ventral? So what do you think the mechanism could be? How is the surround changing from dorsal to ventral? If I understood correctly, you say you show that the surround in the dorsal is more towards UV, is that correct? The ventral is more towards green. So is it the cellular component change or how does that work? Yeah, yeah, especially for the bipolar cell data we see that the surround preference is like nearly inverted to the center preference, although in the dorsal region it's like quite balanced, meaning that you have a response to both colors but a little bit stronger to UV, but how this is really modulated. So we have started to do some pharmacology experiments where we specifically block different kinds of amacrine cell subgroups and then see how this affects the surround. And so for example in the ventral retina we find that the bipolar cell, so the green component of the surround is not mediated by amacrine cells but is only abolished when you also block horizontal cells. Why the UV component of the surround for these cells is blocked when you block amacrine cells? So it seems like they receive input or inhibitory inputs from where the surround is mediated partly by horizontal cells which is green, dominant in the ventral retina and partly by amacrine cells which is UV dominant in the ventral retina. But I mean how exactly this is in the dorsal retina? We didn't systematically address these questions yet. Just a quick comment. One cell to play with there in terms of the color potency of the surround might be the on delayed cell that we worked on. So it's not published in our paper but we've since messed around with that cell a bit and it appears that the far surround disinhibition, that GABA B disinhibition component appears to be UV shifted. So that may be a place to look for a UV. How does the cell look? I mean morphologically? It's the type 7-3 in eye wire. It's the recursively bi-stratified one. Oh yeah so it's a bi-stratified one. It's bi-stratified and yeah it's not exactly in the bottom but yeah it's bi-stratified in layer 7 and 3 and it's um there's a pretty good match to it in your chirp data I believe. I'll have to look at exactly which one it is but because it's so delayed but that's the key like for an on step it's got almost a half second latency so that it's really slow. Okay so I'm intrigued so I think in the 2017 paper we showed something was it the on's or the off's have a slightly bigger receptive field, one of them, the bipolar cells? So given that the dendrites of those bipolar cells between on and off are not all that different wouldn't then the assumption be that that's because one of the group gets a bigger surround than the other or stronger surround or something which then makes the receptive field different size. So can the same thing possibly be linked to the fact that you're seeing more opponents in yours? Yeah so I get your point so I think that's actually the case that the off bipolar cells I mean they have a stronger green component of this around right and so but your argument doesn't make sense because it's exactly the other way around you know. So what we found in the paper functionally is that the off bipolar cells have a larger receptive field than the on bipolar cells but that would mean that off bipolar cells receive let's say relatively speaking less surround but now for the color data we see the opposite. So one hypothesis might be that I mean as I said the bipolar cells they seem to receive a part of this around comes from horizontal cells and apart from the armor green cells and the relative contribution of these two surround pathways might differ for on and off cells. Okay are there any clues to that in the new gap junction data between cone bipolar cells from from Brian Jones or Robert Mark? I'm not sure I would have to check. That might contribute to the difference in the receptive field sizes. Yeah or gap junctions in the inner retina right. I mean part of the receptive field of on bipolar cells is their gap junctions to A2s. Sorry I will just intervene to say I'm going to hand the YouTube stream. So for YouTube people thanks for watching and hope to see you next week for Thomas Alert. Thank you. That's why we would probably be between cone bipolar because