 I just for putting together this amazing meeting and for inviting me, I'm really happy to be here and really excited to meet and discuss with people at the meeting. So to sort of position the interests of my lab as an introduction, us like most everybody in the room is interested in figuring out how the brain works, how it generates sensory perception, motor control, cognition, and then perhaps try and understand how those psychiatric and neurodegenerative disease make cripple in a brain function. Now we tackle these questions at the level of nervous circuits because nervous circuits are sort of the basic computational units the brain uses. So for example here, motor neurons in the spinal cord are essentially components of the sensory motor circuit that controls the movement of our limbs. Now I'm particularly fascinated in the diversity of the different types of neurons that constitute such neural circuits. This is evident at the cellular molecular levels. So for here, for example, again, motor neurons, but now distinct types of motor neurons labeled by distinct types or distinct patterns of gene expression. This diversity of cell types is very obvious at the morphological level. It's obvious at the, or it's very prominent at the level of local long range connectivity and perhaps most importantly at the level of functional and intrinsic properties. And I would argue that understanding neural diversity is one of the key challenges in trying to understand neural circuit function. So we chose the olfactory cortex. Pyrofoam cortex is our model system for the following reasons. So first, as you've heard throughout the meeting, olfaction is of outstanding ethological importance. Animals mice really care about what they smell. And this tight link allows us to establish causal relationships between circuit function and behavior. Olfactory cortex is a sophisticated thing. So it integrates segregated input channels from the olfactory bulb to create all the objects or all the perception. It also links all the perception to experience and learning and memory. So it has essential functions in associative olfactory learning and memory. Now, both in humans and in mouse models, the olfactory system is tightly linked to social and emotional behaviors aging and all degenerative disease. So there's a considerable interest in understanding olfactory circuit function from a translational perspective. However, for us, the most important reason we study olfactory cortex because it's a comparatively simple cortical circuit. And I'll go into that in a minute. As has been pointed out throughout the meeting, the odorants are detected in the olfactory betelion by odorant receptors expressed on olfactory sensory nons. These nons project to the olfactory bulb and nons expressing the same receptor can coalesce into individual glomerular olfactory bulbs such that others elicit these discrete segregated patterns of glomerular activity. Now, what I would like to point out the stress is that in order for us to perceive odors to generate these odor objects we can describe and that animals can report to us in terms of behavior, this segregated information has to be integrated. And this integration happens in olfactory cortex. So how is information transmitted from the bulb to the cortex? Michael and Tufted says, as Florin explained very nicely in the previous talk, project to large areas in the brain. This is the waveform cortex here. And in contrast to this segregated tight organization of projections to the olfactory bulb, these projections are widespread and diffuse and as such allow for the integration of different glomerular input channels. And consistent with this anatomy, both imaging data and electrophysiology have shown quite clearly that odors activate these very distributed patterns of nodal activity in waveform cortex. Now, olfactory cortex is close to the periphery as also Charlie introduced is only two synapses away from the sensory neuron and it's also relatively simple in organization. So olfactory cortex sits here at the bottom of the brain. This is the coronal section for the mouse brain below and the neocortex here is composed of three layers, layer one containing the main input axons from micron tufted cells synapses onto dendrites of principal cells in layers two and three which then form a large recurrent network and also accommodate associative top-down inputs from other cortical and sub-cortical areas. However, how these circuit functions emerge from the concerted activity of different types of nods in waveform cortex is fully understood and that's what I will try to address a little bit in during my talk. So I'll be discussing two sets of experiments. First, a set of positive imaging experiments which allow us to better understand how information about odors is encoded in olfactory neural circuits in the cortex. These data also highlights the functional diversity of different neurons in cortex. A second set of experiments where we use molecular genetics to identify markers and molecular signatures that identify the linear different types of nods and their connectivity patterns. And then I'll spend a little bit of time to illustrate where we're taking these experiments next. So to explore, and this is following up nicely on Florence talk with a lot of sort of overlapping interests and techniques. So to explore sensory processing in olfactory cortex, we use two photon imaging. So we stair-toxically inject the calcium indicator six, a G-Camp six into BFM cortex. The experiments I'll describe are performed in anesthetized animals which is a big caveat on one hand. On the other hand, we have now both with Kevin Franks, electrophysiology in awake and anesthetized animals and preliminary new data, imaging data in awake animals which really demonstrate that the kinds of parameters we use for the analysis I'll describe are fairly conserved across brain states and we're fairly confident that they're generally relevant. But I'll be happy to discuss that later. So this is what it looks like when we infect period from neurons. These are all the responses which you may have seen. So color coded from blue to red. These are very densely labeled areas. So we have to develop an algorithm to actually segment these cells in an automated fashion. So this algorithm uses correlation over time like many other algorithms but also takes into account spatial constraints and much works very well in particular for such densely labeled areas in the brain. And these are just the next slide are just some example responses of such experiments. So these are 12 different norms I believe in lines and 13 different order stimuli and there's basically two things we observe. First of all, the response properties we observe are very diverse. So we see neurons that respond with an increase in activity indicated in red, with a decrease in activity indicated in blue. There's neurons that are very broadly tuned that will respond to pretty much every order you give them. There's neurons that are very selective for particular odorants. There's neurons that are very reproducibly responded across different trials. Others that show a lot of trial to trial variability, et cetera. And so we can map these responses now back onto the imaging site and that's shown here. And we see these dispersed order responses that are consistent with previously published data. Now what's different here with this increased resolution both temporal and spatial or higher signal or noise is that we can use single trial instead of averaging. And that sort of highlights two important features of these responses. One is that there's a lot of trial to trial variability. So you see, for example, these neurons here responding to acetate and trial number one, they respond again, one disappears, the other one is active, the other one comes back, one disappears, et cetera. So across trials, about 50, 60% of neurons kind of come and go and there's a lot of variability that works. It's coming out of this, but it's very serious. Right, so we did this experiment, it's very low. We see about 10 to 15% active neurons spontaneous activities in the range of one to three percent. So over the same, yeah, if we do the same types of windows. The other aspect is that a lot of neurons show very overlapping response properties. So you see that these two neurons here respond happily to all three neurons without much of a preference. And if you go through these patterns, you find a lot of them. So what that tells you is that looking at individual neurons really doesn't tell you anything about stimulus information. And so we have to reformat these data in such ways that we can quantitatively get at understanding what they may represent. So this is such a matrix or vector where the cells of the particular imaging site, about 280, 300 cells are lined up on the y-axis. There's 13 different orders and four files each so that each data point represents the response average change in fluorescence over a short period of time after stimulus onset. And with these matrices, we can now begin to quantitatively compare and calculate similarities and differences in the response patterns to different orders. So in such a simple way by calculating the correlations. And what you see here is that so correlation, I mean similarities between one and the same stimulus per definition one is on the diagonal here. And this makes two points really that so four trials of hexanol, for example, are fairly correlated to each other and different from other order and that's true for some. But if you, for example, look up here at citronol, then you see that there's a lot of heterogeneity or sort of dissimilarity between different citronol trials and the similarities fairly equal across the different sets of orders, which reflects the sort of overlapping and trial to trial variability. Yes. So the orders are pseudo randomized such that we never use the same order twice in a row and the intertrial intervals are two minutes. We've done this with longer intertrial intervals, et cetera. And I think we're fairly confident that none of that reflects a bit duration, I think like four in this case, four to seven. So about 50 to 60% of neurons will respond on consecutive trials and the other half of neurons kind of come in and out. And so, no, come yet, why is it about the same? Half of the neurons will respond. Will respond all four times. No, will respond in consecutive trials and then they come back at some point later and we haven't done that in extensively enough. A man as such that we can ask how, what's the percentage of neurons total that would respond to a particular order and I cannot tell you that? Oh, yes. Absolutely. Right, right. Yes, I'll have to think if there's some logic to that. I can remember that. All cells have a different image in this case. Okay, so to more directly ask if these patterns are different such that they can be discriminated, we use the classifiers similar to what had been discussed before and this is now based on again a single imaging site and works pretty well so using a linear classifier we can predict based on single trial response patterns with an accuracy of above 70% the identity of an order within this order set of certain different orderants and if we combine our windows either within the mouse or across mice then we can very quickly reach accuracies that go beyond 80, 90% which is consistent with the physiology data. So which layer and waveform cortex? So we've played a lot with layers. I'll come back to that later. From my perspective, it's not always flat and so guessing as to where you are within a layer is not a good reference but I'll come back to that in a minute. So basically this tells us that in the response patterns we image contain sufficient information to predict the identity of an order. We can then ask how is this information distributed across peer-reformed cortex? Is it organized? Is it clustered? Or is it distributed as previously suggested? So these are 250. This is one imaging site on average 220, remember? If you generate pseudo populations by adding imaging sites then that's what you get. So we need about 200, 150 to 300 neurons to come to an accuracy of above 70%. The green is fluorescence in fact because so we're using very slow ways of dealing with that and there's a delay in classification which is entirely explained by the delay in fluorescence. 13, it's the same order set that is that order set. It's 13 orders and so the tens level is here, 7%. So very little time to be the trial correlation itself, right? So what, right, if you look at the, yeah, we'll get something to tie in. So the number is 0.6 on average, yeah. 0.6 correlation on average with it, 0.6. Right. And across this order set, the number is 0.45. The linear claspy results are a different order set. No, no, that's the same order set using a linear clasifier but just looking at correlations that I don't think we've, these are Pearson correlation but we've done, basically they tell you that there's some consistency and some noise we didn't really look into that. So maybe last quick question. Yes, yes, yes. And again, we've done several ways of classifying and it basically comes out in very similar way. This is Euclidean. Yes. So I'll come, maybe I'll come back to that in a minute. Yes, in fact we've done it with microcells and I'll show you the results. Okay, so we're looking at how this information is organized in space and we did the following experiments. So if information is closed, okay. So we calculated the amount of information contained in ensembles of increasing size and we built these ensembles from starter cells and then either built them with cells that around the starter cells, especially constrained clusters of increasing size around the starter cell or increasing size ensembles that are randomly put together. And the prediction is that if information is clustered in some way, then ensembles built from specially constrained starter cells should differ in their information content and should be different from randomly assembled clusters of known weight. And so without going into detail the data suggests that there's no spatial organization, no clustering such that whether or not you built your ensembles from specially constrained or random populations we don't find any statistical measure that can differentiate between the efficiency or the accuracy of classifications built from these different types of ensembles, right? So now we use orders, we use order paths that are very controlled, very reproducible concentrations in the environment, in the wild, and as Andreas and others have pointed out, orders come in order plumes and one aspect of these order plumes is that the concentration of a given order and at a certain point in time massively varies, right? And as Linda has pointed out last time interactions of order and receptors and order and very strongly depend on the concentration of order and as such information on all activity that goes into the system will vary wildly depending on these fluctuations and concentrations within order plumes. Now on the other hand, perception of the identity of an order, identity, perception of an order must remain stable in these order plumes otherwise we'd constantly get confused as to what we're smelling and so the prediction has been that or one of the models has been that order representations in clear form cortex are concentration invariant that they allow us to decode the identity of an order independent of concentration, right? So we tested that by and this is just showing in fact that we can measure that with PID so these are order paths which most people use. If we take the PID a meter away and we open the window then we get these massive fluctuations that Andreas described in much more detail. So we tested how order representations in clear form in particular how order coding, identity coding in clear form depends on fluctuations in order concentration by using three orders at 10 fold and a hundred fold changing concentrations so 10,000, 1,000 and one in a hundred and basically going through the same sets of experiments and what you see is that overall the patterns of order evoked activity don't massively change with increasing concentration so we get few more norms that are active, we also get more norms that are suppressed. If we look at tuning at the specificity with which orders respond we can use lifetime sparseness here as a measure for that and we see that while individual norms change the tuning overall the distribution of tuning does it remain stable across this 100 fold range and concentration. It's liquid dilution but it's verified with PID measurement so it looks very reliable. Now if we start looking at individual norms however they do care about concentration and they do care quite a bit. So if you look for it so this is again one imaging side this is now clustered just to make it more visible. If we look at this cluster of norms here you see quite clearly that they are much more active in response to high concentrations of a set of inon whereas this cluster on top here basically shuts down at high concentrations of hexanone. So individual norms care about concentration so how does that affect the encoding of odd identity in pair from cortex? If we go back to simply calculating correlations then you see the following two things. So one is if we look at an individual order at last state at increasing concentrations the representations become more reliable and more similar and that likely simply reflects the fact that there's more signal over noise so we're using high concentrations, less noisy. More interesting however, if you look across concentrations now for it lasted you see that the response patterns gradually de-correlate so they turn from orange to blue and they're about as blue as all the other organs I showed you before which would suggest that as concentrations increase the patterns become very much indiscriminable amongst each other and very much sort of similar or as dissimilar as we observe for different order and that obviously poses a problem to the olfactory system and its need to stably encode odd identity. Second? So this looks exactly the same in the wake brain. So Kevin has done that in extra cell recordings we've done a little bit but this looks very much the same. There are differences which I'll be happy to discuss. So in order to solve a sort of get at this problem we wanted to test the idea that order identity is not encoded in the entire ensemble but that the order identity independent on concentration may be in fact encoded in a subnetwork or subpopulations of norms. And so the question then becomes can we isolate concentration and variance subpopulation of pure form norms and can these subpopulations provide any advantage for the system to generalize, to identify an order across different concentrations. And so we did that using linear regression where we first select norms that are significant and modulated by the identity and intensity of an order but we eliminate norms with mixed selectivity where there's an interaction between identity and intensity for simplicity at this point. And then we take norms that are significantly modulated by the order. So they care about which order they smell and we ask whether within that population there's a subpopulation of norms which doesn't care at which concentration they smell this order. So the question is do these norms exist? They're about 12% of all order responsive norms, about 30% of those that do care about the identity of an order and some example traces are shown here. And to verify this type of selection we compute the correlations. And what you see I think quite clearly is that now with this using the subnetwork or subpopulation of norms we generate fairly homogeneous correlations across now a hundred fold range of concentration, okay? And so we can now go ahead and compare the coding properties of this subnetwork with the entire rest of the norms by for example projecting these patterns in PCA space and looking at its structure. And what you see is that while generic norms so all the other norms cluster primarily by concentration so all the low concentration orders are here independent of identity, concentration in their networks now cluster by the identity of an order but within that cluster they don't care about concentration and we can ask more directly does this subnetwork enhance the capability of the system to generalize that is to predict the identity of an order independent of its concentration. For that we went back to linear classifier and now these are a little bit difficult to read but what we did is we trained the classifier on a set of orders and concentrations and then tested it with an order at the concentration that it was not trained with. So that's asking to identify an order at the concentration that it's never encountered before which we think more accurately so reflects the challenges of the olfactory system in real life. And what you see is that for 10 fold range in concentration so these are here anywhere in yellow for example for this order the both generic norms and the concentration and band subnetwork does reasonably okay with an accuracy of about 70% but if you go to a more drastic change in concentration as a hundred fold then you see that the subnetwork very significantly outperforms all the other norms in predicting the identity of the order. So yeah so this is in fact a fairly complicated question because it's very much dependent on our data set. So we've done that by so that the criteria is that the norms care about which order. All right, so that's the first selection criteria and the second one is asking whether the concentration in there. So if you expand or shrink the order set then that changes the subpopulations we regenerate by this procedure. I did, how that relates to being able to discriminate between different sets of orders across different ranges of concentration in real life I think is difficult to predict but within our data sets these populations are fairly stable. Next slide, two after. So I've argued before that in order to represent the identity of an order you have to integrate segregated input channels from the olfactory bulk and that may suggest that the way we observe order coding in peer-reformed cortex is not observed as such in olfactory bulk output in micro and crafted cells. So we went back to a previously published data set which we looked at micro cells and did the same kinds of analysis to see whether there's concentration in there and subnetworks already at the input level of olfactory cortex and what you see is two things. So first of all we see fewer concentration in variant norms involved than we see in peer-reformed but more relevant is what Renke just suggested that when we do a shuffle control that is we randomize the cell identities so as a reference for the numbers of concentration in variant norms you would predict simply from the statistical distribution of the data set then what we observe in the olfactory bulk is that the observed value is within that range of noise whereas in peer-reformed cortex it's very significantly shifted to the right suggesting that the segregation into concentration in variant norms is something that is if not generated at least strongly enhanced in olfactory cortex with respect to its olfactory bulk inputs. So to summarize this set of experiments I've shown you that the identity of orders is encoded in distributed ensembles of peer-reformed norms that identity independent of intensity is encoded in subnetwork on a subpopulation of peer-reformed norms and that this concentration in variant subnetwork may be an emergent property of olfactory cortex and not as such observed in olfactory bulk. And I think more generally moving forward we would like to propose that distinct subpopulations of peer-reformed norms can be identified or encode different features of an olfactory stimulus that may be order identity, intensity, perhaps valence and other things. So these experiments like I said highlight some of the functional diversity of peer-reformed norms in the second part of my talk I'll come to the molecular diversity of cell type and diversity in cell type identities we can now assign to peer-reformed norms and this work is very much based on a lot of previous work and perhaps first defined by the just lab in the spinal cord where they identified in particular transcription factors which define the identity of a norm and its connectivity with downstream targets. So these are again motor norms in the spinal cord and as you can see here that this distinct types of motor norms which innovate different muscles in our limbs can be delineated by their distinct patterns of gene expression and often combinatorial patterns of transcription factor gene expression. In the neocortex similar experiments have been described by Maklis lab, Ninazistan, Rubenstein lab and others such that genes can identify or sort of reflect based signal circuit organization for example the laminar organization of the neocortex so Cox one is a marker for the superficial layer to three cells while TLE four is a marker for deep layer five, six cells in neocortex and moreover more recent experiments have shown that again combinations of transcription factors can not only tell you about the organization of neurons into the different laminar of neocortex but also tell you about their connectivity so if you have Cox one neuron you are likely to project to the controlateral in motor cortex, you're likely to project to controlateral motor cortex whereas if you have TLE four neuron you project to sub-survival targets including the thalamus and the corticospinal tract. So we figured that having these types of molecular markers that can identify the connectivity of different piriform neurons would be a big advantage to have in piriform so what we did is we cut out the layers with deep sequence and from these deep sequencing experiments we identified candidate genes which we tested by in-situ and antibody staining so these are just a few examples of genes that are selectively expressed in a particular layer so for example PRDM-8, PDPD-3 are exclusively expressed in layer 2 not in 1 and 3 I should mention this is a nisselstain just as a reference for layer 1, 2 and 3 and we find genes that are expressed are specific for sub-layers for example F2 is expressed in these superficial layer 2 cells while MPCD for example is excluded from the superficial layer of layer 2 but expressed in subset of 2B and 3 etc. For some of these antibodies for some of these genes with antibodies so here are three more examples Wielin as Charles Greer has pointed out is fairly specific for excitatory neurons in layer 2A it's also expressed in some internal neurons in 1 and 3 Cox 1 is not present from this layer 2A but it's present in a large population of deep piriform neurons and we have several markers that identify a layer specific subset such as this one barred cell 1 so these are nice genes to identify types of neurons and look at the organization in layers doesn't tell us anything about connectivity yet so in order to get there we re-validated some of the piriform target areas by simply injecting an AAV channel YFP in piriform neurons so channel YFP is very efficiently expressed not some termini and then cut the brain from anterior to posterior see that piriform neurons project to the olfactory bulb this is what Florian just talked about in his previous talk in the previous talk they project to many other cortical areas and so we particularly looked at the medial prefrontal cortex at the cortical amygdala and at the antiviral cortex and that is because they are far apart and so we can now use retrograde tracing experiments to see how neurons projecting from piriform to these target areas are organized within piriform so that's such an experiment where we use colatoxin B as a retrograde tracer injected in bulb prefrontal amygdala and antiviral cortex and what you see immediately and I think what you can appreciate quite readily is that these neurons in piriform projecting to these different target areas now line up in different layers and not so different from the layers of gene expression patterns I showed you before so for example bulb neurons are deep neurons they're excluded from layer to A the frontal neurons are primarily present in 2B in contrast cortical amygdala and antiviral cortex are fairly excluded from deep layers but present primarily in superficial layer too so in the following experiments I'll try to convince you that the genes and connections match we did that in three different scenarios so first we ask whether genes that specify different layers or cell types in different layers specify the same cell types that are segregated in terms of projection patterns in other words are the cox1 neurons in deep layers the same neurons that project to the bulb and are the real in 2A neurons the same neurons that project to the cortical amygdala and the short answer is yes so about 80-85% of all cox neurons do project to the olfactory bulb so they're co-labeled by retrograde phases from the olfactory bulb with the cox1 transcription factor and they never express real in in contrast all the real in 80-85% of the real in neurons project to the cortical amygdala and cortical amygdala projecting neurons and never cox1 post so neurons in different layers match up well with transcription factor profiles that are layer specific we can push this one step further in fact we can also characterize these neurons morphologically by now sparsely label them with a retrograde the transported virus so we can inject the calf to Cree with a conditional morphological marker and peer reform and we see that these cox1 positive bulb projecting neurons are constituted by a fairly diverse set of pyramidal cells while the cortical amygdala projecting neurons are exclusively these semi-lunar cells which are characterized by the lack of basal dendrites so we can push this matching of genes and connections one step further and we can ask whether neurons that sit in the same territory but project to different target areas can be discriminated using the genes we look at so that's for example the case of bulb and prefrontal cortex projecting neurons they largely overlap in this layer to be of peer-reformed cortex so we can ask can we find molecular signatures for neurons that sit or occupy overlapping territories but project to different target areas and the answer is yes so the combination of two transcription factors cox1 and CTIP gives us a good prediction for the connectivity of patterns of these neurons so this is the overall expression patterns cox1 is in deep layers in peer-reformed CTIP is primarily in these superficial cells but they do overlap if we zoom in to layer 2B where these neurons sit then we see cells that express either CTIP alone in red cox1 alone in green or cox1 CTIP double positive cells which are shown in yellow and if we combine these molecular markers with tracing studies we find the following distribution the bulb projecting neurons summarized in these pie charts are either cox1 positive alone in green or cox1 CTIP positive in yellow there's very few CTIP-only cells in red in contrast prefrontal neurons again in the same area within peer-reformed are either CTIP-only or CTIP-cox1 but they're never green alone so they're never cox1 alone and if we now inject two different color retrograde faces in prefrontal and non-factory bulb we can identify neurons that project to both areas so that they bifurcate and project to both target areas and all of these neurons are both cox1 and CTIP so these experiments identify three types of neurons cox1 neurons that project to the bulb CTIP neurons that project to prefrontal cortex and cox1 CTIP double positive neurons which co-project to both target areas so these data suggest that we have molecular markers that can identify the connectivity patterns of neurons that are either segregated in layers or overlapping within a layer and so the last experiment in this context we did is we asked whether these molecular signatures of connectivity hold in the absence of all layers if we scramble the organization of prefrontal cortex and that's an experiment that was done 50 years ago by the characterization of these real amutin mice which have massive defects in the lamination of neoportex which is similar defects in the lamination of prefrontal so you see that in the density of the neurons you cannot detect any lamination based on the density of neurons real expression is lost due to the mutation of the real and locus however all the other genes that we've looked at in terms of layer specific markers remain to be expressed at numbers that are fairly similar to control only that they are now distributed throughout the prefrontal cortex so for example cox1 which is a marker for deep cells is now found without any bias across the entire depths of pierform so we can ask our connections to maintain in the first place we focused on pierform connections to bulb and the cortical mucular that's because these structures are reasonably well maintained in the real amutin mouse and we see that pierform neurons do project to the bulb and to the cortical mucular at numbers that are fairly similar to controls only again that we lose the lamina segregation and finally we can ask whether the markers we identify as delineating the connectivity patterns of these neurons are maintained in this scenario and the answer is yes so despite the fact that cox1 neurons are now positioned without preference across the depth of pierform 85% of cox1 neurons do project to the bulb very similar to what we see in controls and the real in cells so we cannot use real in as a marker for these superficial cells but we can use FETSF2 which overlaps with real in these semilunar cells and despite the fact that they are distributed or scrambled throughout pierform depth all of these or 85% of these cells still maintain the specific connections with the cortical amucular so to summarize this set of data we've identified using this laser capture deep sequencing approach we've identified genes that delineate distinct pierform layers and cell types these genes either alone or in combination can tell us about the specificity of pierform projection patterns and connectivity and this link between the molecular identity of neurons and their connectivity is maintained in the scrambled cortex of a real mutant animal which suggests or which sort of highlights the importance of intrinsic genetic programs in specifying connectivity rather than sort of local spatial information Do these different things? Yes, so I think by and large that's what we're doing now we're not specifically looking at this set of neurons but the hypothesis is that different types of neurons do different things and so the goal is to match up identity and connectivity with the functional properties of these neurons we don't have a lot of data yet but that's where we're going Right, so in fact maybe I can ask that in a minute it will come to that Okay, what happens to other transcription? Right, so the problem there is that these transcription factors tend to be expressed along the entire pathway so starting in the sensory neuron and the bulb in different subsets of bulb neurons I think we'd have to aim at conditional knockouts in piriform which is a hard thing to do Right, so we use viruses a lot It's... Yes, so it's a large area to manipulate with the virus So we are both... So from this deep sequence in there we have some candidate genes that the promoters that appear from specific and we are trying to generate free lines to see if they'll be useful for that purpose We don't have any yet We're hoping that other people generate them But in fact I'll go to the next slide in terms of understanding the genetics of the system a little better So quickly, future ongoing experiments and future perspective and the first thing I really would like to point out is sort of striking observations we made when we looked at gene expression patterns in piriform and each time you look you obviously have the neocortex on the hippocampus side by side and the prediction could have been that piriform is a simple paleocortex only contains three layers and maybe a fewer genes or fewer genetically defined cell types in piriform than you have in the six-layered neocortex Now that, at least for us, is very obviously not the case Every single gene that people have looked at in neocortex is a gene that defines the identities and connectivity of neocortical neurons is expressed in piriform cortex However, the basic organization of these genes is very, very different and so that's illustrated here in this slide So while Cux1 is a marker for superficial neurons it's clearly a marker for deep neurons in piriform and while CTIP is a marker for deep neurons in neocortex it's clearly a marker for superficial neurons in piriform This is not always the case we get all kinds of combinations it makes no sense to us What is also striking is that the genes that are never co-expressed in neocortex are very happily co-expressed in piriform and this is important because people have postulated that sort of cross-repressive networks specify these neurons in neocortex and clearly the situation is very different in the piriform cortex So, however, one you know, this apparent inversion of gene expression patterns these layers reflect the sequential generation of neurons during development and the observation that some of these genes are somewhat upside down in piriform suggested to us the idea that piriform is developed in a very different way and so we did an experiment that is identical to what Charlie described in the first day of the meeting Yes, we don't In fact, yes, so in fact that's a good so in fact we do now so in sections this is very painful because everything you want to look at ends up not being on your section so we're doing everything in cleared brains now where we, you know, particular any sort of transition areas a bit puzzling to us how do you, you know, make things connect from one area to another in terms of gene expression patterns So we did EDU birth dating experiments and what we see is what Charlie presented is that in contrast to neocortex at least piriform layer two develops outside DNA as opposed to inside out so in neocortex the early born neurons inhabit the deep layers in piriform you see that the early born neurons here indicated by EDU labeling at 11.5 are the superficial cells quantified here whereas the later born neurons in the 14.5 are clearly excluded from the more superficial layers and are predominantly present in the deep layers so in some way despite the fact that we have the same if not the same largely overlapping sets of genes that specify these neurons in piriform and neocortex the logic of circuit assembly appears to be fairly different and so what we're doing now is we're using the feds of two gene as an entry point in studying that so feds of two is a key determinant of deep layer cells in the neocortex as primarily shown by the nazi stance lab so it's essential for the specification of corticospinal tract neurons which also has big implications for mammalian evolution et cetera in contrast we see feds of two very robustly expressed in the superficial care from neurons that are very different in terms of the morphological characteristic intrinsic properties, connectivity, et cetera so what we're doing is we're following now using transcriptomics and enhancer accessibility studies the specification of these feds of two lineages to try and sort of identify key decision points in the specification of different cortical neural cell types that may also determine the connectivity and intrinsic properties and I think it will be important to sort of understand how basic neural circuit motifs are specified genetically what we also hope to generate by doing these experiments is more specific markers for peripheral neurons and that's sort of in response to Linda's question before that will allow us to target these cells in a more refined way so the other sets of experiments we're currently doing is bringing together and that's what Pierre-Marie pointed out the genetic and imaging and behavioral tools eventually with the idea in mind that different types of neurons do different things it's somewhat trivial but it's not a very prominent concept or hasn't been a very prominent concept in the field so we can target these different types of neurons now using mass genetics we can use imaging to assess the functional properties and optogenetics and chemogenetics to manipulate their activity we are also particularly interested in how information is being transmitted throughout the nervous system and that's again very similar to what Florian discussed with the idea in mind that different features of information are not transmitted randomly throughout the entire cortex and that they may be used in differential ways depending on the behavioral context and we can target so we can do that by targeting neurons based on their connectivity patterns using retrogradely transported viruses and finally we can do that in a way behaving animals so this takes a lot of weight off my shoulders by either by using Greenland's technology both in these miniscope types of experiments but also in two photon head fixed experiments and so the idea is to understand how different types of neurons, how circuit functions as I pointed out in the introduction emerge from the concerted activity of different types of neurons and how information about the stimulus about all the identity, intensity, valence, et cetera is being routed throughout the nervous system in a behaviorally dependent or behavioral context dependent with this I'll stop and I'll acknowledge the people so the genetic work was done by a student of the ODATU who was in fact a student here at CISA before she joined my lab imaging experiments were done by Benjamin Roulin in my lab and a lot of the, this was done in close collaboration with Kevin Frank said Duke was done electro-extracellular recordings and come to somewhat similar yet complementary conclusions we got a lot of help from BISPA TV in terms of data analysis Andreas Schaefer, Gloria Toy and Sonia Garell for the molecular studies and this is our funding and thank you very much for your time. Thank you. Thank you.