 Thank you Matt and David for including me and for all of your hard work organizing this and thanks to the other organizers. So we've been hearing a lot in the past day about the complexities of relating different levels of organizations, the complexities of networks of proteins within cells and I just want to pile on and tell you about another layer of complexity and that is really the complexity of cell types and I want to try to convince you that neuronal cell types really represent a sort of critical bottleneck in the biology of trying to understand how it is that the genome can actually produce circuits that ultimately produce behavior and that go awry in disease and so I'm not this is by no means a solved problem but I just want to share with you some of the approaches that we've been taking to try to tackle this problem and you know it's a it's a problem that's at different stages as it were in different kinds of preparation so if you work like Astrid does or like my colleague Eve Marder does on the stomatic asterisk ganglion of decapod crustacea it's largely a solved problem what the component cell types are and in systems like this or in the nervous system of C. elegans investigators in laboratories across the world can go and record from the same cell type or watch the same cell type and really know really work out what the details of the circuit are but in my favorite organ in preparation the neocortex of mammals that's much less of a solved problem and it turns out that in many regions of the mammalian brain there really isn't a consensus as to what the flavors of neurons are so in the neocortex broadly there are there are two main classes that everyone agrees on there are a pyramidal neurons which make long range connections they're the output neurons of the cortex they're excitatory they use the neurotransmitter glutamate and there are a diverse set of GABAergic interneurons that are inhibitory and that are the subject of of much of the efforts to classify and how these neurons get classified really depends on what you like to do for a living so if you like to look at morphology be it somatodendritic or axonal you might classify cells one way if you'd like to record their intrinsic or synaptic properties you might classify them another way but ultimately all of these cellular phenotypes presumably arise from the genes that these cells express and so the approach that we've been taking recently is to use micro arrays which you've heard a description of what they are and I can explain further if they're questions or more recently second generation sequencing to look at all of the genes that are expressed in particular cell types and so much of what I'm going to tell you about really is aimed at this so this works because the techniques for amplifying small amounts of nucleic acids are excellent so compared to the difficulty of for example isolating and the lack of an ability to amplify the proteins that these cells express their nucleic acids because they're of complementarity can readily be amplified so from only you know 50 cells or so we can amplify enough material to probe a microwave or to do next-generation sequencing and just to reiterate for those of you that aren't familiar with this technology what a microarray is it's essentially a place code for gene so each little spot on this glass chip represents a complementary sequence to one gene in the mouse genome and then an intensity code how much fluorescent nucleic acid is bound is a readout of how much of that gene is being expressed and so you can just put this through a scanner and then you use a bunch of methods to to figure out the overall expression so when we started this work a number of years ago there were not nearly as many choices as there are now for lines of mice in which to identify based on fluorescent protein expression specific sub types of neurons and so here's one this line YFPH probably the most widely studied mouse line in all of neuroscience at this point was made by Galping Feng in Josh Sainz's lab a number of years ago and it expresses a yellow fluorescent protein under the thigh one promoter thigh one it turns out is a antigen that's expressed by almost all neurons and many immune cells but because of positional effects the expression is restricted and in the cortex it's restricted to neurons that we knew and loved these are some recordings made by Jesper Jostrom a long time ago and these are actually recordings in mouse these are neurons from the rat but what we can see is that they have a very characteristic intrinsic electrophysiology they're non-adapting in response to sustained current injection whereas if one records from unlabeled cells nearby within the slice they have a variety of different firing patterns they might show repetitive bursting or adaptation the property of action potentials being more spaced despite constant current and so based both on their morphology on their intrinsic electrophysiology also on the fact that their projections are similar we think that these really correspond to a cell type a number of laboratories have have shown over the years that there really two in in layer five two main subtypes of paramedal neurons there are these so-called thick tufted layer five neurons which are the ones I just described that have a thick apical dendrite and a tuft in layer one and have non-adapting firing and then there are thin dendrite neurons that have a thinner apical dendrite and don't have a tuft in layer one and these show this property of adaptation in addition we know from work in collaboration with summa connell and others as well that a single transcription factor phezef2 is responsible for the fate decision to adopt this fate of becoming a thick tufted neuron in knockout animals that lack phezef2 there is a lack of the corticospinal projection which is made by these neurons and in fact our contribution was to show that in addition the morphologies of these neurons now resemble the thin dendrite type and in addition their their electrophysiology which is of these sort of bimodal two types in the wild type animals either adapting or non-adapting is now all all adapting so again by sort of expression of this transcription factor we really view this as a as a major cell type okay so that's an example for the paramedal neurons what about for the inhibitory interneurons this is sort of a line that labels one of the quintessential inhibitory neurons the so-called fast spiking basket cell thank you it's it's called a it's referred to as fast spiking because its action potentials are extremely rapid middle button uh there we are okay great so it can fire up to hundreds of hertz it has very specialized potassium currents that repolarize the individual action potentials very rapidly to prevent inactivation of sodium currents that then allow this to fire very rapidly they're called basket cells because their soma's actually their their axons actually wrap around the soma's and form basket-like structures and so they're very powerful sources of of inhibition in the cortex and they're labeled in this line called g42 uh my collaborator josh wang uh at cold spring harbor made this line took a huge chunk of the gad one gene this is the synthesis enzyme that actually makes gaba or one of two of them and put that in a bacterial artificial chromosome and then made a transgenic animal that had this construct but rather than recapitulating as some back trans genes do the full expression pattern of gad one it picked out only very small subsets of gab allergic neurons throughout the brain and in particular in the cortex it it labels these parvalbumin positive fast spiking basket cells so i'm going to return to this point but i want to make the point that at least in mammals it appears that these positional effects where a trans gene inserts in the genome can have profound effects on where the trans gene is expressed so rather than as i mentioned although the thigh one promoter is active in many many cells in yfph it's only labels these uh primal neurons in the various other sub types of cells similarly for the g42 neurons we know from other examples where people have used the gad promoter to drive for example gfp they label other sub types in this case three different sub types of somatostatin positive neurons and so the way that we think about this although it's really not understood is that especially in the forebrain that there are profound interactions between regulatory elements that are present just upstream of the start of transcription in the in the classical promoter but it's not just that it's also interactions with quite distal elements referred to as enhancers or oppressors i'll just call them enhancers that uh presumably loop in and regulate this transcription and so simply plunking down this promoter somewhere else in the genome is going to interact with a different set of enhancers and label different cell types and i'm going come back to this as a as a method for sort of discovering other cell types so the basic approach then is to take a transgenic mouse in which fluorescent protein is expressed in some population of neurons that we've characterized anatomically and physiologically micro dissect out a section of a brain slice in this case primary somatosensory cortex dissociate the neurons like we might do for making neuronal cultures but these are adult neurons they wouldn't survive in culture and then manually sort out the fluorescent neurons and then lice the cells get the messenger RNA amplify it and probe expression with sequencing or microarrays now one of the reasons that this works is as i mentioned that the amplification is highly efficient and so it turns out that yes if you start with only one or two or five or ten cells then you see a smaller number of transcripts present on the chip and the reason is that presumably the rarer transcripts are not present in enough abundance that they amplify so that you can see them on the chip however if you start with 30 or 40 cells or more we've gone out to 500 you don't see ever increasing numbers of transcripts so this saturates so we think we're essentially seeing everything that's there at least that's on the chip by about 40 or 50 cells the other thing that you want to be very sure of is of course that you're not seeing things that aren't supposed to be there and that turns out to be straightforward to do because for example we know lots of genes that are restricted like oligone and olig2 to glia myelin basic protein or to red blood cells this is a hemoglobin and so if you don't sort the cells at all then you see reasonable amounts of these and other transcripts it turns out that if you sort them but you don't do a good job of washing them by transferring them from dish to dish a couple of times then you see some of this but if you carry out this procedure well then you see very little contamination with things that shouldn't be there um now i'm going to take a slight detour here and just actually talk about various methods that have been used for isolating cells and it turns out that that you can do this manual method that i've just described to you you can do something called immunopanning to deplete or enrich for a particular set of cells that you like you can use fluorescence activated cell sorting you can use a method called trap in which you express in a population of neurons a specific tagged ribosomal protein and then just pull down affinity purify the mRNAs of interest or you can use a laser capture microdyssection which for example for human tissue is essentially the only game in town because you can't you can't get transgenic humans that express fluorescent proteins in particular cell types yes well it really depends on the question is can you do this on single cells yes what you can what well well no um yes you can do it if you don't care about seeing everything that's there if you care about just seeing the more abundant things then you could do it on single neurons and so it's on the list to to do more of this on single neurons yeah right so really there the devil is in the details of how carefully we've done the anatomy and physiology beforehand to characterize those those cells that's an extremely important point yes i'm going to show you that i'm going to show you that so um if one of the nice things about microarray experiments and particularly for students that are interested in computation is that all this data is publicly deposited you can't publish a paper using these sorts of techniques or sequencing without depositing the data so you can actually get that data back and you can ask all sorts of additional questions about it so one of the things you can ask because people at talks like this asked me was does how you isolate the cells matter for example you might worry that when you isolate dissociate these cells you radically change transcription you turn on all sorts of things that aren't normally there so you know we should look at look at that so um first of all i should say that all of these methods are relatively uh reproducible that is from replicate to replicate they have correlation coefficients you know in the high point nines um they differ in how much background you see so it makes sense that when you're using laser capture to sort of draw a little circle around a cell that you're actually pulling out not just that cell but some other material associated with it and so if you measure contamination or background by expression of genes that are that are known to be restricted to GABRGIC neurons or astrocytes or oligodendrocytes in samples that are not supposed to be any of these types then basically you see that some of these methods that either uh immuno purify or uh laser capture have a significant amount of background whereas these other methods that dissociate and then sort either automatically or manually have have less background um we also looked at whether there was activation of for example genes that are known to be involved in cell death genes that are known to be involved in stress and yes there are um genes that range from very active to very low in terms of expression level but there's no systematic difference between method and so it doesn't seem like we're we're turning things on with the exception of this PAM method that actually involves a little bit of culturing and also is looking specifically some of these samples are oligodendrocyte and other glia there seems to be a slight activation of immediate early genes that's the other category that we looked at but basically these differences are very are very small okay so that's a little technical aside so the basic experiment here uh in the sort of state of where this was a few years ago when we first published on this was that we looked at a dozen different cell types from the mouse forebrain most of them are from the two and south lawn there's one diencephalic cell type here of interneurons in the thalamus and we could look at for example homologous thick tufted pyramidal neurons in two different cortical areas in primary somatosensory cortex and in cingulate cortex or similar neurons that were pyramidal neurons labeled in the same line in the amygdala or in the hippocampus as well as a subset of different interneurons and we could find for each of these long lists of genes that were up-regulated shown in white or yellow in a particular cell type say the yfph strain amygdala neurons and not in other cell types or that were up-regulated in groups of related neurons for example in all GABAergic neurons that we looked at or all of the telencephalic GABAergic neurons and this turns out to be a good group to look at more closely because a lot of these genes were known from prior studies so these are three different interneuron types and these are two pyramidal neuron types and these are unsorted cells and you can see that the vesicular inhibitory amino acid transporter the canate receptor GAD1 and GAD2 the synthesis enzymes as well as other transporters and transcription factors that were known like Aristolus and DLX1 and DLX6 that were known to be expressed in pan-interneuronally are all present and so we think that this is actually quite a accurate representation of what these cells are expressing everything that's supposed to be there we see and things that are not supposed to be there we don't see. This is a more recent example that's largely unpublished in which we used a line that expresses in locus serilius neurons these are the primary source of neurodinergic modulation of the cortex and the rest of the forebrain and they're labeled in a line in which tyrosine hydroxylase promoter drives GFP and here too we could find a long list of things some of which shown in green were known like tyrosine hydroxylase itself this is a transcription factor this is the transporters this is dopamine beta hydroxylase because these are neurodinergic neurons et cetera and then many things that were not known and one of the nice things about having access to the allen brain atlas is we could not have to do all of the in situs ourselves to go verify this but all the ones with the star we could just look up and see that in fact there were nice examples of labeling in the locus serilius which is all sort of collected into a nucleus under the near the cerebellum and confirm that in fact those things were were expressed by in situ so the original motivation for this was to have a sort of unbiased method for classifying cells and indeed it seems to work well for that so first of all things are reproducible so all of the circles of the same color wind up quite close to each other in this unsupervised clustering and similarly as one might expect all of the glutamatergic neurons wind up close to each other and separated from the gabarergic neurons there also seems to be a huge difference between the gabarergic neurons in the telencephalon and those in the diencephalon we haven't sampled quite densely enough yet to know whether transmitter phenotype or major brain division is more primary you could see that things in neocortex wound up very close to each other and closer than for example other things in the limbic forebrain like amygdala and yeah this is no this is essentially the whole genome we actually what we did is just take all the genes that are differentially expressed across all the samples but that's like 2,000 genes or something like that but you get almost exactly the same clustering if you even include all genes this is a larger data set now that where we include some for example experiments on developing interneurons and on catecholaminergic neurons which wind up again widely separated so we think that this approach would likely scale for as we do a larger number of of cell types which is something that we're doing in collaboration with genelia farm so i want to as a way of introducing some of the things that we've used this for actually raise some of the caveats to this approach for for classifying neurons and these should all be things that are kind of obvious to biologists or to anyone that thinks about them for some time but but they're important to sort of think about how to address the first is development so you know expression changes dramatically not just early in development to the mid you know young adulthood that many of us study in slice experiments but even it continues to to evolve and we've looked at this in detail for in the case of these fast biking basket cells so here are the baskets that are formed around the somas of a pyramid and you can see how distinctive their electrophysiology here is by this fi curve that plots the firing rate as a function of injected current for a bunch of different interneuron subtypes and here's the fast biking cells you'd see they get up to incredible rates of firing relative to the other interneuron subtypes but also takes a lot to get them going they have very low input resistance and of course this does not is not the case when the animal is younger it's something that emerges over development and so if you record and first identify these cells as having arrived in the cortex they actually migrate in from the presumptive striatum at about a week of age they are able to sustain firing at much lower rates and they have a much higher input resistance and this then progressively drops over time the amount of current and voltage that it takes to get them to fire the amount of current that takes into fire increases quite a bit and their maximum firing rate increases and so we we asked ourselves could we actually relate these changes in electrophysiological properties to the changes in gene expression over these over this period and the changes in gene expression are enormous so something like 2000 genes change their expression over a particular choice of significance and most of these things are either monotonically increasing or decreasing and so that's expressed here as just doing a principal component analysis of you know the relative to these ages that we've looked at and you can see that most things either have a large positive or negative principal component one that is there they're either decreasing with age thing or increasing with age and very few things have a low principal component one and a high principal component two which sort of measures how peaky the the expression is so perhaps if we'd recorded if we'd done this earlier in development we might have seen many genes that turn on very transiently and then turn off but we didn't see that from this period of phenotypic development yeah this is um somatosensory uh cortex um we have i'll get to that in a minute for the intern arms there's a very little difference actually in gene expression or or in properties so not too surprisingly perhaps given these electrophysiological changes one of the biggest categories of things that were changing both up regulating and down regulating were uh ion channel genes some of these were things that were known before like um kcnc1 and c2 uh or kv3.1 and 3.2 is what the proteins are referred to are these very rapidly activating potassium conductances that are known to be critical for the fast spiking if you knock these out they can't fast spike anymore um and they're very specialized for these cells um what was not known was uh that they also up regulate these uh leak channels uh task one and twic one are the names of the of the proteins kcnk1 and k3 are the gene names and uh we hypothesized that perhaps actually it was this that was contributing to the progressive drop in input resistance of these cells the fact that it's took more and more current to make them fire and in fact we could show at the protein level uh at least for task one for which there was a good antibody that that also increased over this period uh and then there aren't it turns out good drugs for um uh task one but there's a somewhat uh dirty uh drug bupivacane um which blocks this leak current and we could show that the proportion of the total leak that was blocked that was bupivacane sensitive also increased and then about the same time Bernardo Rudy's lab um found the same thing and they went one step further and actually knocked out uh task one and showed that that in fact that had a dramatic impact on input resistance so um the the uh second sort of caveat to classifying cells by uh their expression profile is that the same cell type in different regions of the cortex to get to your question peter um actually can can have somewhat different properties and um i'll give you one particular example of that but ultimately what we would like and and what people at the allen institute would like is to be able to recognize the same cell types across species and across regions and we've heard a lot about this you know sort of anatomical complexity i'm not going to talk at all about how conserved things are across uh species um but i do want to address a little bit um how uh how conserved things are across areas and so we saw some of that in our original data that is that that looking at the these pyramidal neurons for example from uh primary sonata sensory cortex or from uh singulate cortex there was a very little difference in them in the most expressed genes if you go way down the list you could find some very subtle differences actually for this g30 interneurons um you you couldn't really find any at the level of stringency that we were typically using difference between the um uh neurons in these two different regions of cortex but subsequently and somewhat fortuitously we found um a very dramatic difference in uh the electrophysiological properties of one of these cell types as a function of uh region and that is these thick tufted pyramidal neurons that i told you were essentially non-adapting that is that there was a constant interval between their action potentials um it that turns out is true everywhere in the cortex except in motor cortex and in motor cortex they actually accelerate their firing with time this acceleration turns out to be due to the expression of this depolarizing ramp which in turn we could show is due to a slowly inactivating potassium conductance and um i won't go through all the gory details of which specific subunits but these are basically shaker subunits that contribute to uh this slowly inactivating uh so-called decurrent which you can block with very selective toxins and uh which basically convert the firing type of these motor cortex neurons into those of the other neurons so did do they express different subunits um the answer is yes at the protein level so it turns out that um neural mabs have made beautiful antibodies to these different shaker subunits and we could show uh that the somatic and proximal dendrites express different levels of one two three and five which agreed with the toxin data that i'm not showing you um uh but if you look at the mRNA level there was no significant difference between the corresponding subunits so whatever is causing this and we don't fully understand the regulation is is happening at the um whoops is happening at the uh protein level and not at the there it is um at the mRNA level and so that's really just by way of caveat to remind you that there's a whole lot of biology that happens after uh gene expression and that that could also be cell type specific there may be cell type specific differences in translation in post translational modifications in trafficking etc and these approaches won't address that so what can you do with this what what can you uh use this set of tools for um we as i mentioned we were motivated initially for sort of understanding circuits and classifying the cells that comprise them i think i showed you one example where we could start to get at some of the uh molecular causes at least for ion channels of cellular phenotypes and we're doing a lot more of this but i want to also go through an example where we've used this as a more sensitive assay for uh changes in expression that might underlie disease states or forms of plasticity these are much more subtle differences in in expression and the the particular example that i'll uh describe is um the uh mouse model of ret syndrome so um some of you may be familiar with us the underlying protein mecp2 was discovered by adrian bird who's here and uh who does jogby then discovered that mutations in this protein mecp2 uh cause most cases of this uh genetic monogenic form of uh autism spectrum disorder although there are important differences phenotypically between ret syndrome and uh sort of most other forms of autism it's it it shares with it um a tremendous impairment of of language and of other cognitive functions it's particularly devastating for the families that are affected because the parents initially think that they have a perfectly normal girl i'll explain why i said girl in a moment and uh they what they then uh see is after an initial year year and a half of normal development these girls begin to regress they lose what language they've acquired they lose purposive movements of their hands and are often confined to a wheelchair and have a bunch of other problems as well um so this gene mecp2 is is x-linked and so although this syndrome does happen in boys um in boys who only have a one copy of mecp2 if they if that's a mutant copy they typically have a very rapid course and often this doesn't get recognized as ret syndrome which is occurs about one in ten thousand one and fifteen thousand uh female uh births um neuropathologically uh the disorder is rather subtle it's not a neurodegenerative disease it's not that there is a loss of a particular cell type um much more subtly than is indicated in this uh cartoon from a review by Houdijagbi there are some simplification of the dendritic trees a reduction in the number of spines a smaller soma size of uh ret neurons and therefore the whole brain is actually reduced in size and so we've been working on uh sort of two ends of this problem one trying to use electrophysiology to actually understand what's altered in the cortical circuit and two use the sort of gene expression approaches to see what genes are misregulated and uh so we found initially that uh if you record from neurons in somatosensory or motor cortex under conditions in which there's spontaneous activity and those are conditions where you actually reduce the uh concentration of calcium and magnesium ions and increase the concentration of potassium then uh there's much less activity in the slices from the mutant animals these are recordings from layer five neurons and um we could see that even before the animals become uh symptomatic so they're they're really sort of um three general ways in which you could imagine making neurons in a recurrent excitatory inhibitory circuit less um uh active and what are those just to see if you're awake any of the students want to venture a guess what what could you change if you if you had a model cortical circuit and now you want to make it less active change potassium okay well you could change the intrinsic properties of the neurons themselves right so we could test that we block all synaptic transmission and record from the cells and they're actually exactly the same so it's not that what else could you change about the circuit well this is a recurrent circuit right there are excitatory synapses there are inhibitory synapses so I could either make less excitation or I could make more inhibition and it turns out that both of those are true so if you record spontaneous synaptic input to these neurons then there's a modest increase in the total integrated inhibitory charge which we haven't really pursued much and then there's a much more substantial reduction in spontaneous excitatory currents and it turns out um that this change in the balance between excitation and inhibition may in fact contribute to the to the uh fact that it's much more difficult to induce long-term plasticity in slices from these animals and presumably therefore to the fact that these animals have a harder time learning but there's somewhat of a chicken and an egg problem here because is it really the fact that the ME CP2 has somehow targeted genes involved in LTP and you've blocked LTP and therefore synapses excitatory synapses are weaker or is it that excitatory synapses are weaker and inhibitory synapses are stronger therefore it's harder to induce LTP so the way to get at this is to actually try to just isolate a pair of neurons and the synaptic connection between them and then ask whether you can still induce normal LTP and so using either a spike timing type paradigm that's illustrated here uh or a pairing paradigm where you basically just depolarize the cell uh and and activate the pre-synaptic cell um uh Vardan Danny was able to show that actually you can induce perfectly normal long-term potentiation at these synapses so it seems like whatever ME CP2 is doing it's not disrupting the ability of the synapses to undergo LTP and in fact instead what Vardan showed was that as the disease becomes symptomatic the amplitude of unitary synaptic connections between pyramidal neurons or the probability of finding any two cells that are connected both go down significantly so basically you're you're losing recurrent synapses between excitatory neurons. Now how does this come about from loss of ME CP2? Well I have to tell you something more about what ME CP2 does well the way Adrienne Byrd discovered it was he was actually looking for proteins that bound to methylated DNA so I've just finished telling you that cell types express different genes one of the reasons that they do is because of DNA methylation this is presumed and not entirely known but but the thought the classic thought is that genes that are going to be shut off in a particular cell type their promoters or various other parts get methylated that then recruits ME CP2 and other methyl DNA binding proteins that then recruit repressor complexes that lead to transcriptional silencing. As I mentioned there this is somewhat a debated point and there's some suggestion that actually binding of ME CP2 to some methylated DNA could actually activate it but that's I'll leave that alone. Anyway the reason that we got into this in the first place is that when Rudy Yanich's lab and others first made the mouse knockouts and knocked out ME CP2 these animals got sick and died and they recapitulated many of the features of the human disease but yet when they ground up the cortex or the cerebellum or whatever they found really subtle changes in gene expression so if there was a you know seven percent difference in this gene or a five percent difference in this gene they could if you looked across a 10 gene classifier you could say this was mutant and this was wild type but no single gene was significant and so that of course led us to think well maybe it's because you're grinding up different cell types in which different genes are affected by loss of ME CP2 and this sort of take home message is that that's the case. So this is what we refer to as the dilution problem and that is if you imagine that you have a big change a tenfold change in one cell type as a result of some manipulation plasticity disease state whatever but that cell type only comprises one percent of the total tissue and now you assay it in the whole tissue you you're obviously going to see a much smaller change a barely significant change and so we reasoned if we could actually pull out individual cell types in the ME CP2 background we could test this idea and so we did it for four different cell types that were chosen to be as different as possible. These thick tufted pyramidal neurons fast spiking basket cells in the cortex and then these lococerellius catecholaminergic neurons and prokengene neurons in the cerebellum that also are labeled in this G42 line and the first thing we saw is that there were actually large differences so you know 10 50 fold changes in expression and this just shows for three of the of the cell types. These are not as big as the difference as the baseline differences between gene expression between cell types but they're still large relative to what was seen in tissue. Are not what? Oh well no I mean collagen it turns out most genes are not expressed uniquely in the nervous system they most have function so collagen is important in our connective tissue it's also important in fast spiking basket cells and you know the extracellular matrix that gets constrained and so on so and that's true some of these are genes that were really isolated first in the nervous system this is a presynaptic protein etc so but that's an important point so this is this is a representation of all of the genes that were affected in the knockout. First of all we saw more things that were up regulated than down regulated as you might expect if you were mostly if this were mostly a repressor and you could also see that it's really the minority of genes that are altered in multiple cell types for most things they're altered in one cell type but not in the others and so it makes sense that if you grind everything up together you're actually going to see a reduced signal in the tissue. Now I should say that although it was largely different genes that were affected in these different cell types there was one category of genes namely cell adhesion molecules that was affected in all of the four only three of which are shown here cell types that we looked at and so this is this is intriguing obviously many of these fell in families that have well described roles both in the nervous system and outside of the nervous system. Kidherans do a lot else outside of the nervous system collagen's, proto-kidherans, contactants etc and so this is a is another sort of representation of this that asks the question of how likely would this occur by chance and and essentially what the color here indicates the how many cell types we saw the change in and what this is a representation of the whole gene ontology set of categories which are hierarchically organized which were altered and so basically you can see that this is huge relative to these other categories and this is actually a statistical test by Monte Carlo simulation of how often you would expect this by chance and this is the the probability that you get for the for the results. So we think that this is real is it is it causal I mean that is really the big problem here so the hypothesis is that loss of m e c p 2 leads to a cell specific defect in cell adhesion molecule expression and that then has diverse effects in different circuits in the cortex that causes a loss of recurrent excitatory synapses that kind of shuts down activity but really to to get at this what we need to be able to do is manipulate these cell adhesion molecules in specific cell types in the way that is is evident in the disease state and this is true more generally I mean in order to figure out the mapping between expression and neural function you've got to not just measure it you've got to be able to manipulate it and so what I want to leave you with in the last few minutes is are some recent efforts to sort of build some better tools for how you would how you would do this in the nervous system this is something that that many people are interested in we've heard about some efforts from the allen institute to develop transgenic lines the basic strategy that's that's used is a combinational one that is a combination of two different alleles a driver allele that expresses a molecule like the crea recombinase that can recognize sites that you put into the dna and knock out a gene or a transcriptional activator like the tet transcriptional activator and then this the driver allows this recombinase or transcriptional activator to only be expressed in a certain subset of cells and then you have a responder allele which is your particular gene of interest that either has the locks p sites that are recognized by crea or has the tet response element that's recognized by tet and the point of doing this separately is that now you can examine many genes in a particular cell type or examine many cell types in there in the importance of a particular gene so oops the sort of standard approach for doing this is to knock in a crea recombinase or an inducible form of crea recombinase that can be activated by giving tamoxifen into a locus that you know is active in a particular cell type and so together with Josh Wang we've characterized a bunch of lines that that he's made that allow us to target interneurons and and these are widely available from jacks and other places the the problem with this approach is that it's just not specific enough so although in general we like the idea that you know one one gene identifies one cell type the reality is that actually any given cell type is really represents the conjunction of multiple genes and any given gene is active in multiple cell types and i'll just give you a brief example of the pitfalls that come from this problem so we were interested in the issue of how do these fast biking cells remember who they are what aspects of their biology tell them to keep making the same proteins and keep having the same neuronal identity and so it turns out that in addition to gene expression there's regulation of translation through microRNAs as we heard how important are these microRNAs in keeping their neuronal identity and so the way to look at this is to knock out a key bottleneck in the production of microRNAs the enzyme dicer and so conveniently there is an an allele a flox dicer and we have a animal that expresses cre-recominase just in these parvalbumin positive neurons which in the cortex are these fast-biking basket cells and axo-axonic cells but there's lots of other parvalbumin expressing neurons across the brain and so the question is is there some effect on these neurons of knocking out dicer and initially we were extremely excited to see a very powerful effect of of knocking out dicer on the behavior of mice so this is if you knock out dicer everywhere it's lethal that animal was normal it was a heterozygous this animal just lacks dicer and parvalbumin positive neurons and you can see that it has some major motor deficits and i'll spare you the actual um physiology but uh essentially it was all uh negative and uh oops we'll look at that again so essentially the fast-biking neurons in the cortex are totally normal um in these mice and uh uh presumably this is a problem in parvalbumin positive neurons in the spinal cord or somewhere else so that that sort of serves as motivation for an alternative strategy um which is uh referred to as an enhancer trap strategy the idea is that you insert a weak promoter that by itself is not really able to drive transcription and you randomly insert it throughout the genome into different locations and uh it lands next to different enhancers that can then enhance transcription from this promoter so if you're familiar with gal four lines in flies that's the basic strategy this is this is a strategy in mice um that that was done in collaboration with Carlos Lois and basically uses lentiviral transgenesis to insert uh a this weak heat shock promoter uh driving a reporter and driving as you'll see TET um in random places in the genome and we then just screen mice and see if they have interesting patterns of expression um in the brain i won't go through the sort of uh strategy so here's one example this is a strain of mice uh in which the neurons that are labeled are cahal retzi cells these are excitatory neurons that sort of pioneer the cortex and here they are they uh still around in the dentate chyrus um if you look earlier you can see them in the neocortex as well this is another line that labels a specific population of thalonic relay neurons this line labels one of two main subtypes of pyramidal neurons in piriform cortex so-called semi-lunar cells um and um uh a variety of these label um neuronal subtypes in the neocortex this labels just uh most of the uh neurons in the lateral geniculate nucleus of the thalamus so it turns out that there are many um genes that are expressed throughout the entire thalamus there is no gene that is restricted to the lgn that just lights up the lgn so this is giving you a sense that this is a more potentially selective uh labeling what you can see is the lateral geniculate nucleus and then the axons in layer four and layer six and layer one of primary visual cortex um here's a couple of other examples this driver line labels a layer four neurons in primary sensory areas this labels a subset of layer five neurons just in prefrontal cortex these seem to be cells that project to the raffia nucleus and other modulatory nuclei um this labels a subset of layer six neurons it turns out that these are corticothalamic neurons you can see some labeling in the thalamus here but it turns out that there are actually two subtypes of corticothalamic neurons there's some that project back from a sensory area to the primary nucleus so in somatosensory cortex it's the vvpm in the lateral geniculate nucleus for the visual cortex shown here you see these don't project to the lateral geniculate nucleus they project outside of it into p o in the case of the somatosensory system or the pulvenar equivalent in the case of the visual system so the point is not the sort of fine details of these particular connections but just that this is a potential strategy for picking out a very very selective subsets of cell types and so it's something that we're actively pursuing this is the sort of summary of where we were a couple of months ago screened about 150 different lines and roughly a third of them to a quarter of them have interesting expression patterns so i think these are really interesting times for vertebrate for mammalian neuroscience because some of the kinds of tricks that have long been available for my colleagues that work on inverted nervous systems are now starting to really mature and be feasible in the mouse and you know i want to just leave this group with the idea that in addition there's a tremendous sort of computational problem of how you constrain expression in these particular cell types what actually makes a cell type how is it that they know to express some genes and not others in addition to the computational issues that we've known about for a long time about how do you actually build neural circuits that do things and i'll stop there after thanking some of the people that have done the work much of this work was started in my lab by kensigino and chris hemple with help on the physiology from these individuals mouse making in my lab has been largely the efforts of yasushima in collaboration with josh wayan carlos lois and i talked about the projects in the retz mouse that was in collaboration with these individuals thanks very much that specific genes make specific defects and those with the specificity of each sort of genetic clothing is going to be different or different genes are attacking similar pathways and it's actually a stage development that comes from at least a similar thing so it is there really typically markers, fairs and more of an axis of life have you done any comparison yet or what do you think of those two different ways of looking at their relevance? yeah i mean it's it's sort of the boring answer but i think that both are true it really depends on the gene so in the case of fragile x and mecp2 like diser we're talking about a very fundamental very common piece of biology that's going to do lots of things in lots of different cells but clearly there are channelopathies where it's a single you know mutation in a in a sodium channel or a potassium channel that gives you a very well-defined disease i think in general the the classical molecular biological one gene one function approach is going to be a very rare case and the very fuzzy systems biology it's all a network is going to be unfortunately the the reality for most of it for most phenotypes so you know i think it really is a complex mapping from every level to the next level and that's you know that we're going to need to get in those weeds to to figure it out yeah i'm wondering if you or anybody is mining these data to move from kind of second order statistics and that the reason is you probably know from the from the martyrs that in this amount of gas we're getting even more finding that while individual mRNA coffee numbers for different items vary all over the place specific cell types have like fixed ratios for pairs or even components of mining channel types yeah so that you can see by looking at and you would be searching those stuff just like only by right so um it's important to note um two very important differences between this sort of experiment and the experiment that you're referring to one is that we're always looking at at least 30 or 40 neurons so we're averaging out those cell to cell differences and it's very much on the books to ask whether we see similar things and there's plenty of transcripts that are abundant enough that you could do that for even ion channels um the other thing is that we're looking at genetically identical animals that have been raised very similarly as opposed to crabs or lobsters that are very divergent from one another genetically and have had very different experience crawling across the ocean floor for years before the fishermen grab them and give sell them to eve or you so I don't I I suspect we're not going to see that level of variability there might be some and it might be quite interesting um but it's it's an it's an interesting question how much of that variability that you see that eve sees is experience dependent versus how much of it is is genetic certainly there are big strain differences we heard about an illusion to that um earlier uh there can be very significant strain differences uh in some genes um so be it'd be interesting to to mine that right so we I should say we haven't tried to do that um and it depends on what method you're so first of all I should say that the micro array method and the sequencing method are highly quantitative the method that most people use for analyzing single cells rt pcr um is not highly quantitative it's very difficult to quantify that well um so if one wants to do lots of single cell micro arrays or single cell sequencing maybe you could address this um the second important point is it totally depends on who your out group is so what I mean by this is let me give you an example um everywhere besides somatosensory cortex in the cortex if you express parvalbumin you are a fast-biking neuron however there are many other cells in the whole brain that express parvalbumin if you allow all of those cells to be an out group then having parvalbumin doesn't tell me that much but if you tell me oh I'm in the cortex and I'm not in somatosensory cortex and I say not in somatosensory cortex because strangely there layer 5 parvalbumin neurons can express some low level of parvalbumin but anyway it it totally depends on who you're comparing to how many genes you need and we haven't tried to actually calculate that but it is an interesting issue no no it's it's levels of expression that's what I'm saying we could see over you know I don't know 10 log 2 unit something like that differences in expression maybe even more well I mean that there are interest interest and and uh usefulness for classifying art actually can be found at all expression levels so for example cam k2 is present in all excitatory neurons and not all inhibitory neurons so that gives you a pretty important binary distinction but it's a very pretty abundant protein but yes I mean it's it's much harder to classify just on rare markers you know and unpoorly expressed markers okay thank you very much