 So developing mouse brain atlas looks at about 2,000 genes that are pre-handpicked as hopefully being involved in development. So we've got about 700 transcription factors, axon guidance genes, some disease genes, and we look at seven developmental stages of mice in the earliest stages. The brain is still encased in the body because it is not formed enough to be taken out. When we get up to E18.5, we can then remove the brain so these first three stages are still in the body. And it goes from embryonic day 11.5 all the way through to the P56, which is our adult reference atlas mouse. This reference atlas was annotated and drawn by Luis Pueles, who's one of the foremost people in the world right now as far as looking at development of the mammalian nervous system. And he's very much a proponent of that prosumeric model that we talked about. So this atlas is done in that way. You can bring up atlas plates comparing different ages of the prenatal mouse so you can look at structures and how they evolve and how they develop. You can bring up ISH images and compare between different ages. You can back those to the reference atlas as much the same way that you can. The mouse brain atlas, all of these, I think I have this next up, we'll see. All of them are available in brain explorer as well, which means you have three dimensional control over what you're looking at. You can actually spin and zoom in to various areas across the different age groups. You can look specifically at a given age or just a couple of ages if you want. So it's fully active. In addition, in this atlas, a variety of structures have been annotated informatically based on that reference atlas, and so you get a table for each gene across a variety of areas and across age to see what the expression pattern is. You can see that this expression is limited to portion of the brain, but that it's also more highly expressed early in development as compared to adulthood. There are differentials searches that allow you to go in and select age groups versus each other age groups or areas or a combination of those. In addition, one of Lewis's colleagues has done manual annotation in the case of those earliest prenatal time points of the entire body. So he's gone into every structure and said whether or not expression exists in it, be it the body or the brain. And so there is some additional information outside of what neuroscience would generally use for that. The non-human primate atlas is currently an atlas that is looking at postnatal development of the macaque. It has both an ISH component where there were 44 key genes selected per structure to look at. And then there is a more finely tuned microarray dataset that looks at about 300 samples per case, I think, or no, I'm sorry, looks at the regions of interest that have been further annotated. So you'll be looking at different nuclei within the amygdala, for instance, or different layers of the cerebral cortex. What this allows you to do is either go in and look at different patterns for different genes across the given structure, or you could look at a given gene across different ages of development, or look at genes in a single age across different structures. The transcriptional profiling, again, as I said, you're looking at much finer detail as far as the annotations go. So instead of just these five regions that we look at with the ISH, we've broken them down into, for instance, the amygdala subnuclei, different gyri and the layers within those. And what you see, for instance, this is looking at the hippocampus, is that you get specific gene expression patterns for one component versus the other up here is the dentate gyrus. It's got a very distinct expression pattern. The pyramidal cells have a distinct expression pattern. When you look at the GABA-ergic neurons, they have a different expression pattern. The sebiculum itself, again, a different expression pattern. So you can start parsing out fine-level information. How is gene expression quantified? Most of these that you'll see, and we'll walk through it, are Z-score presentations of the raw data or of the normalized data. You do have the option of looking at raw data values, as opposed to Z-scores, depending on what you're looking at. When you download the data, it depends on what exactly you've chosen and how to look at it. Instead of downloading everything at once, it focuses on how you're using the system. Did that answer your question? In the developmental atlas, like the other ones, you have the ability to search based on age. For instance, you can go in and say, show me genes that are highly expressed at birth that are expressed much lower in adulthood, and you'll return a set of genes, and it will tell you what the fold change is and what the p-value of that fold change is across all the different samples. The brain span atlas of the developing human brain is the work of multiple groups. We have USC and Yale, a group that does transcriptional profiling. We've got Harvard that's working on some MRI stuff. We contribute ISH imagery, LCM microarray, data points, an anatomy reference framework, and the data visualization. This is an annotated prenatal human atlas. Right now, available online is 21 Weeks Post Conception. Coming up, there will be an atlas for 15 Weeks Post Conception, as well as the adult. All of the ISH is in, so you can look at 15 and 20 Weeks Post Conception. You can look at childhood or adulthood across a variety of structures. The fine-level microarray analysis, so LCM is Laser Capture Microdisection, which means we can go in on a microscope and actually dissect out at fairly fine resolution what we want to see. That allows us to look at individual layers or finer if we want. Like the other microarray studies, you can go in and look for genes that are differentially expressed between different structures. Right now, the 21 Weeks specimen, the same one used for ISH and for the reference atlas, is available with microarray. It's the opposite hemisphere. You have one case with all these data modalities on, and you have 300 samples per specimen. There will be four specimens released, two at 15 Weeks, two at 21 Weeks for the microarray here coming up in just a couple of months. That will bring up to a total of about 1200 samples that we've got microarray on for that. This project through Yale has transcriptome data as well for development. They have many more cases that they're able to look at. We've got 8 to 16 structures depending on the level of development across 12 different time periods. You can go in and look. These are, for instance, the three earliest time points, and again, ask for differential expression and find genes that are involved in a specific time point for development. The human brain atlas has a microarray component, which is the main component. It was designed as an all genes, all structures survey. We sample throughout the cortex and the subcortex, and those data points are spatially mapped back to the MRI of that case so that you can go into every individual specimen and see exactly where that sample came from. Currently online, there are three specimens. There will be six by the end of 2013. It also includes histology, MRI, DTI, the annotations, as well as visualization capabilities through BrainExplore. The way that it is set up is that the histology allows you to go in and look at, for instance, cytoarchitecture of exactly where that sample was taken from. You are able, if you're coming from an anatomy background, to say, I specifically want the cytoarchitectonic area, and I want to compare it to this other one. You can go in and look and handpick the samples that you want to use. In addition, there is an in-situ hybridization component with high-resolution cell-level gene expression for specific regions. These started off as individual studies that have now come under the umbrella of the human brain atlas. There's a subcortex study that has 55 genes in the GABA and glutamatergic pathways across subcortical regions. We have one male and one female donor for that. There's a thousand gene survey that looks specifically at visual and temporal cortex in multiple control brains. The goal is for an N of three for each of the genes. I think we have N of two to seven, depending on the gene and the area. Schizophrenia study that looks specifically at the dorsolateral prefrontal cortex of over 50 control and schizophrenia cases. About 20 schizophrenia cases, 30 control cases, and 60 genes. Those genes were picked to either show specific cell markers for either specific layers or cell types, or about half of them are included as schizophrenia-associated genes. These are genes that were identified from GWAS studies or other studies as being implicated in schizophrenia or being linked in pathways directly to other genes that are with schizophrenia. So you can go in and compare controls to schizophrenia in the same area for specific genes. Autism study works in a similar way, 25 genes from the frontal and temporal and occipital cortex, 11 controls and 11 autism cases. Now upcoming is the second half of the huge human brain atlas. It's a neurotransmitter study across the different neurotransmitter systems. 88 genes throughout the subcortex of the cerebrum, all areas of the subcortex, and 176 genes, so these 88 plus another 88 in the cortex. I want to say there's about eight different regions of cortex that we'll be looking at, and this is across four specimens. Three of these specimens have data for the microarray as well. So it's an attempt to balance and be able to look both at the ISH expression picture combined with the microarray data that we get for the same cases. Oh, that's a good question. It doesn't always relate to proteins, and we have not necessarily looked to see what that ratio is, right? In some cases it will be reflective of what the proteins are going to be, but we have seen cases where it doesn't. So for instance in the schizophrenia study, one of the genes that showed the most robust changes was CalB2, which is the gene for calretin. In protein studies, calretin is used as a control in schizophrenia studies because it doesn't show change, right? So we've got this discrepancy. There are other RNA studies, though, that show that the gene changes at the RNA level as well. So presumably the genes of interest kind of keep you walking, right? And new ones get added, so if you have like an N of 3 at this point for these 1,000 genes, what's the thinking like with every couple of years a new set of Ns with new... Probably not. Probably not. Mostly because it's a large amount of work and you're also running through specimens, right? So these are studies that are set and they have an end date in sight. So instead of thinking about it as we're going to find the answer to schizophrenia, you've got, you know, 35 genes there that have been highly linked to schizophrenia in one way or another. Yeah, well, and some of it changes, right? So in that study, they chose the genes associated with schizophrenia. By the end of the study, some of the genes that hadn't been chosen as associated with schizophrenia were now associated with schizophrenia for different reasons. So yeah, things evolve. So the problem is that when people working on autism think you might need 1,000 individuals per group, how do you suggest to compare that? Yeah, right. So yeah, and in human data, there is always that question of N because of the amount of variability. So, you know, you're kind of limited because disease states generally, they're a lot harder to get. So again, not the end, I'll be all in answering the question, just giving people the data behind some of the more research-focused genes that they can look at to help interpret their own data or launch their own studies. You could say, well, that's from the GWAS studies, is it the first to be wrong? Yes. And then you're releasing an underproud data stat, is it just the data stat? Mm-hmm. Well, and again. So what we aren't presenting here is an analysis of the data. We're presenting the data, and the data is what it is, okay? And so if it is underpowered in the manner of N to find the answer, you still have cases that you can look back to and refer to to see whether or not the patterns are consistent with your hypothesis, that type of thing. So here we go. With the human, you have this huge problem, and that's the problem of scale. So you go from a mouse brain that is this tiny up to a human brain that is much, much larger. And that's part of the reason for moving to a microarray format as opposed to an ISH format. The ISH format gives you very fine-level detail as far as what types of cells are expressing those genes. But it would be nearly impossible to do an entire brain study of ISH across the entire genome the way we did for the mouse. So the confines and that we have for the specimens also help dictate how we go about doing those studies, okay? So for the human brain atlas, again, we have that issue of size. And so it became a quite complicated pipeline for processing this because we wanted it to be multimodal, but we had limitations on how we could produce things based on the size of the tissue and the amount of the tissue and the cost. So what we settled on is that we acquire a specimen, and the first thing that happens is that we ensure that that specimen meets our criteria for control. And if it does and is consented, it goes off and we get an MRI so that we have a 3D visualization for which we could put all the data back into, depending on the source of the specimen that MRI either happens in skull or after the brain has been removed from the skull. It then goes back and is processed and slabbed into about 17 to 20 slabs of about a centimeter thick and those get frozen down. All of that has to happen within 24 hours. Those frozen slabs get shipped to us and we start off with some large format histology and microneuropathology to ensure that it really is a control case. However, our ISH and our ability to sample is limited because the largest slides that we can process are 2 by 3 inches, which is a lot smaller than a human brain. So after the large format processing, we go in and decide for each and every slab what's the best way to break it down into smaller pieces so that we lose as little data as possible. And each of those pieces goes off for additional processing for sectioning, ISH and histology and it leaves us at the very back of each of those blocks and tissue to sample from. So for the cortex, we go in and we actually dissect out by hand with a scalpel, enough tissue to send to the microarrays. For those smaller or more intricate structures, we attack it by laser capture micro dissections. So we go in and annotate the structures, send that off to molecular biology. They sit down underneath the scopes and they cut out with a laser all of the small samples. The original goal was 1,000 samples per brain due to acquisition problems for tissue that's been cut back to a single hemisphere, so we now have five to six hundred samples on one hemisphere for the last four specimens. First two specimens have just about a thousand samples per the entire brain. Those samples go off for microarray processing, come back to us for QC input into the database, mapped back to the MRI, and then visualized in 3D, in 2D, in heat maps in a variety of ways. So again it's an all genes, all structure study. So what you will see, much like the other microarrays and I promise to talk more, bore you more with the human data than the other ones, but these for instance are the structures color coded by the ontology, the different specimens up here, the genes that you've searched for here. It's a search for those genes that are preferentially expressed in the cortex, and you can see structures that these genes are highly expressed in versus those that are low expressors. It is designed to be able to be mapped back with all the histology, so what you'll see is the ability to interact with the large format sections as well as the small format sections that are on the actual nistles that are either interspersed with the samples or immediately adjacent to the samples, so that you can go in at high resolution, you can turn off these colors, you can turn off the outlines, and go in and if you come from an anatomy perspective go in and do your own cytoarchitectonic analysis of the samples or further sub-define which samples you want to use in your study. The data can be visualized in the brain explorer much the same way it is in the mouse, except that unlike the mouse where we had all of our sections 200 microns apart, so we had these beautiful voxels, we can't do that in human, and so what you have instead is each of these blobs is a data point that has been first mapped back to the MRI and then plastered on what's called an inflated white matter map, we'll talk about this right at the end a little bit more, to give you a visual localization of how the gene expression patterns are across the different brains. Like the other brain explorers, you can superimpose the actual structural information back onto the three-dimensional MRI. In the case of the human we use free-surfer to do a a persilation of the cortex because it's about 17 centimeters from front to back and if we were to do ISH at every 200 microns it would be way too much data, we'd have to really limit the number of genes probably down to under 10, so we opted for the microarray instead and it just means a different visualization process for that. Oh, good question, how many people, I've been saying microarray and people don't know what microarray is, go ahead that's fine. Okay, so with ISH I said we take a probe, we put it on the tissue and it binds to the RNA in a complimentary sense, similar to the way antibodies do when you do immunohistochemistry, right, and then you can use a color metric method to view where that probe is located. Okay, that's ISH. Micro, maybe, there we go. Okay, so and I actually took out all the slides that describe the ISH procedure and so you get get visualization of the actual tissue. For microarray you start off with a very small amount of tissue and the RNA, the tissue's lice, the RNA is extracted and purified and it's put onto slides where all those probes instead of going on the tissue are sitting on the slide and so you're taking your sample and putting it on the slide and it's binding to those probes which are then linked to fluorescence and then you excite the slide and these little tiny dots, if we go back, it'll let me go back. Whoops, oh no, you're gonna be stuck on that one. This is a, it's sort of a caricature of what it looks like, that's not quite, things are packed in more tightly but you know each dot is different probes or set of probes and so you can get an amount of fluorescence that reflects the amount of RNA in that sample for that probe. So, you basically wind up a chunk of brain and you lose all spatial information but you get a lot of gene resolution. Exactly, so previously with the human you know microarray would be like cortex and they would grab a bunch of cortex and do it and so yeah you lose a lot of spatial information which is part of the reason why we tried to be so comprehensive spatially in what we were sampling and map it back so that you kind of get some of that information back with it. But it is a comprehensive, the chip that we use is an 8 by 64,000 chip that means there's 64,000 probes on that chip for each sample, some of which are for non-coding regions, some of which have been failed out. We end up with about 55,000 probes left over at the end of the day for each sample. Is there been any analysis of the relative sensitivity? The data for that first case with the more comprehensive ISH is just starting to come into our hands so we're just starting to look at that. With the mouse they've done mouse microarray versus ISH and consistent with other studies in the field shows about a 50% correspondence which isn't great. Part of that has to do with whether or not the probe for ISH is overlapping or in the same area as the probe for the microarray and a lot of times it's not so you're gonna get exon differences and that type of thing. For our ISH component of the human brain atlas, most of those probes for the microarray were based off of probes that we already had for ISH so we should be able to get a better idea when the probes line up of how much correspondence there is. Hopefully it'll go higher than 50% and so. Is there any plan to use things like deep sequencing with the mouse code? Yeah, so the we've looked at some RNA-seq stuff. It's not clear whether we're going to go in that direction simply because the field is moving so fast by the time that we get the data out it's probably gonna be somewhat obsolete. The brain span developing human atlas does have transcriptome and exon data so it is going into RNA-seq as well. So it's a little bit deeper dive and much more complicated for analysis. So it really doesn't want to let me go past this line. Okay so with the mouse ISH studies you have the ability to bring up that reference to atlas so that you can always see where you are. There's a lot of reasons why that doesn't work with the human. One is that as we said all the human brains are distinct in their individual variations and so you can't necessarily map it back to a single reference atlas. What you can do is map it to essentially the nearest NISL image that has been annotated and for these what we have opted for for expediency-seq is what we call hot spotting and that is that on the structures there's there's a spot that you can hover over with your mouse that will identify what that structure is and it is set so that you can sync these two images up. Doesn't quite work the same way as the mouse atlas because with the mouse those sections are pretty static as far as it's a small section it gets cut goes on the slide the same way but with the human there's much larger sections and so there's a lot of pooling and twisting and stretching of stuff and so the ability to sync is there but it's a little bit more involved and a little bit more manual but it does allow you to go in and look at the ISH and say oh here's the nucleus accumbens or the clousestrom based on what you're hovering over so it works a little bit different way but you do have the high resolution viewer. As we said the neurotransmitter study that other half of the HBA brains and there are three of those specimens will have both microarray and ISH okay so there's a lot of data and when I say a lot of data like we said those are 64k chips for the microarray so for the human atlas it will eventually be looking at about 4,000 samples so 192 million data points just for the microarray going to end up having almost a hundred thousand ISH images that you can pull down and use for analysis cross just these atlases and not counting the transcriptome data and the exon data from brain span we'll be looking at almost a quarter of a billion microarray data points that you can use for analysis okay so what can you do with these it's a lot of data and besides break your computer there are a couple of things and I only just partially about that because in the early days I did try pulling down all the data for the first specimen just completely crashed my computer alright so usually you approach this from one of two different ways either you know go ahead and click the gene this slide was provided to me so there's a little bit more animation than I usually have and what you want to know is where is it expressed in the brain or what genes correlate with this expression level or you know your structure of interest or your disease state the area that you want to look at what you want to know is what genes are preferentially expressed in that area or are not expressed in that area at all so here's a couple use case scenarios for what we've done inside the Institute for some of the analyses this is a paper by Morris at all that was in PNAS and it is looking specifically at ISH data from the mouse brain and it's a semi-quantitative study looking at species differences in gene expression so what they did is they took all of these areas did a semi-quantitative analysis of expression intensity and density across seven different species of lab rats and then they looked for changes and what they found was that first off evolutionary wise the farther you get away from the C57 black general lab rat the more differences you have in gene expression across the structures and that the largest number of changes that you see tend to be in the thalamus the cortex in the hippocampal formation areas that we tend to think of as more evolutionarily evolved right so that's that's one thing you can do with ISH data second one is going at a straight quantitative analysis so this was the analysis for the schizophrenia study hopefully to be submitted as soon as I get back and what we did is we took our gene expression in all those images we outline regions of interest and using laminar marker genes we could essentially look at gene expression across the cortex so those regions of interest were divided into a hundred sub layers informatically processed to look at cell density staining intensity cell size amount of area occupied by stain we had like 40 factors that we looked at came down to two in the long run for the paper and then you can compare using these expression patterns much the same way that you do if you do Western blacks and you look at changes in in density there you could compare controls and schizophrenics across various parameters and so what we found for instance this is a chart of the p value differences that we found between controls and schizophrenics where the warm colors schizophrenics are lower and in the cool colors schizophrenia cases show higher expression and we have cell density staining intensity across the variety of interneuron markers and what we show is that one interneuron markers are are affected fairly severely in schizophrenia these were out of 60 genes that we looked at so we did this across all the genes and when we looked at specific areas which you see is that those differences are localized to very specific area of prefrontal cortex just broadman area 9 the changes in area 46 fail to reach significance with the exception of these guys right there I don't know if they're all right-handed schizophrenics really yeah I'm I would have to go back and look to see the handedness what I do know is that all the brain regions were from the same side of the brain so all those yeah yeah very true very true and so you know you can look at this in a variety of ways here is a plot of controls versus the schizophrenia cases on cell density across the variety of markers so you can see that some genes show more of a decrease than others and even in those genes that don't show a significant decrease schizophrenia tends to be just slightly lower so there's a trend for cell density of interneuron markers to decrease now this flies in the fact of the when we look at the nistles and we do cell density counts for the nistles in these cases cell density is actually increased in schizophrenia those counts are not purely neuronal so it's questionable what's going on but every single gene that we looked at that showed a decrease was in a decreased direction so the increase in actual cell count is not due to a specific cell population for any of the genes that we looked at that makes any sense so that's another thing that you can do again this is all off of ish data this is not pulling any microarray data and this was all automated image analysis written through our informatics department let's go ahead and move to the next one this is data from the developing human project and now we're looking at a combination of microarray data and ish data so again going back to the comparison between is what we see in the microarray data similar to what we see in the ish data and here we have this is cortex here a variety of genes to show preferential expression for different layers of cortex and what you can see is in the accompanying nistle here zik 1 which is almost exclusively found in the subpeel granular zone that's exactly where you see the ish move down to the cortical plate for lmo 4 and here you have expression in the cortical plate and so this really you can use the ish to help verify what you're seeing with the microarray or the other way around and because we have the ability to go in and find genes that have the same pattern you may have a gene that you know shows expression in a specific area now you can go in and find a bunch of genes that you did not know had any sort of pattern like that you can find similar patterns this is just a second panel of that figure moving into the ventricular zone so you can see again as the expression values move to the right our ish is moving to the left okay and then we move into strict microarray analysis so this paper is in publication currently at nature this is our platform paper for the human brain atlas papers headed up by Mike Carl what's an edleene and I'm going to show you a couple different ways that you can look at the data so in this figure what they looked at is they pulled out differential genes between every pair of structures so they took superior frontal gyros compared to middle frontal gyros superior frontal gyros compared to infra crossed all 176 moderate level structures we had and they counted the number of genes that showed differential expression that was in the first brain hold up for a second go ahead and go back by the time that they got done with this we had a second case in so then they did the same analysis in the second case and they took all the genes that were overlapping and they counted them up and what you see here is a diagram where the smaller the number of differential changes are in your cool colors so less than 20 and high level of differential gene expression in your very warm colors so what you're doing is comparing how similar different structures are right what we see is that cerebral cortex for the most part is very similar with the exception of some very specific regions like primary visual cortex primary motor cortex the temporal pole these tend to stand out but for the most part cortex does look very similar down in the other corner cerebellar cortex tends to look very very similar what we see down here is an actual graph kind of depicting things at a higher level most of the differential gene expression that we find is within the brainstem and this makes perfect sense because we have a lot of nuclei doing very different things in the brainstem and so our top top winners are myelin cephalon the pons the midbrain and then here comes hippocampus with a lot of differential gene expression within the different segments so that's one thing to look at when you find these differential genes you can pull them out to look at the gene expression patterns and find out what other structures hold together with the same patterns or with the same genes and because the genes cluster out as well you can go in and see what classes those genes belong to so you can find that in a hippocampus for instance you have preferential expression of a gene that may be involved with exocytosis and then a lot of the genes that show that expression are involved with exocytosis for instance and so you can see that in the hippocampus you can see that in the midbrain the pons the brainstem and again you can back it to the ISH and this here is a look at a gene that shows differential expression not just within the hippocampus but between species so here in the mouse expression is in CA1 CA2 and the dentate gyrus in the macaque have a similar pattern although your CA1 is a little bit more diffuse in the human we retain expression in the dentate gyrus CA1 is almost non-existent for expression all of a sudden so you start looking at species differences as well with this go to the next one we also were doing WCGNA analysis weighted gene co-expression analysis and this describes correlation patterns among genes so you kind of have to take a 90 degree turn in how you think about it and what this does is it assesses the pattern of expression for a given gene across all of the samples in the brain and then it looks at the next gene and it sees whether or not it has a correlative pattern or a different pattern and it starts correlating all of those genes based on the expression pattern across the entire brain across all of the samples and the output are what are called these modules okay and each of these modules then you can look at the genes involved in them see what sort of pattern it has you get what's called an eigen gene can look to see where that gene is expressed no keep it there for a second because this one gets complicated so for instance our first module which also happens to be the primary component when we do multi-dimensional analysis is this primary sensory motor cortex that pops out so you have genes that are primarily expressed in primary sensory motor areas in the cortex and you can get a list of all the genes that fall within that module many of which have not been annotated yet we have no idea what they do but they have the same expression pattern of genes that we do know so you can start to figure out what activities certain genes may be involved in this can work at a brain level or you can break it down by structure so down here we have it broken down by hippocampus and you see the same thing you see modules popping out and you start getting additional information in this case because we've limited it to the hippocampus within each region now the data is presented anterior to posterior so all of a sudden you can start seeing gradients in the hippocampus from the anterior to posterior axis you start getting additional information out of that okay I think we're still looking into that you can do the same thing with other species the question is which of the genes and you know I kind of envision it as eventually developing a cloud network of genes where you've got cores and genes in the periphery that may or may not and yeah I don't know how much species conservation there is I would assume that there's more than a little so okay and then as I mentioned you can do some multi-dimensional scaling and what we found with this is when you pull out the first three primary dimensions and you plot them down what we found is that those samples recapitulate how the brain is actually made up so in multi-dimensional scaling you're looking at how far away each point is from each other and how similar it is so this point here is going to be very dissimilar to this point up there but these two points here are going to be very similar what you see these are color-coded to match the map on the left we have the frontal cortex we have followed by the parietal cortex temporal cortex down here visual cortex in the back I have to say in as an atomist when then they show this to me I said clearly something is wrong can't possibly be telling me what this tells us is that the gene expression in structures is more like the structure next door to it in the cortex one of the major dimensions as we said was a primary sensory motor dimension we the second one is an anterior to posterior gradient recapitulates the pattern of development so some residual developmental information coming in on that and all right so those are my use case scenarios and how we've used the data in different ways just kind of give you some ideas of what you can do so again the question starts off as I know a structure I know a gene I want to know where it is or what's expressed there you can go a lot deeper in asking questions if you ask the right questions so back to a little bit of anatomy that will help you get a round the atlases if you are not coming from a biology background just real quick you have three main planes of dissection you have a coronal plane a horizontal and a sagittal plane you'll see these on the MRIs you'll also see it coded in the data when you pull up gene returns it'll tell you especially in the mouse whether the data is presented in the sagittal plane or in the coronal plane there are terms that go with a lot of the nomenclature and anatomy superior is up inferior is down anterior to the front posterior to the back we have a little bit of a problem though when we go from mice and other animals that walk about on four feet to humans that walk upright because our brain is positioned slightly differently so we have some additional terminology rostral is towards the nose caudal is towards the tail and what you see in the mouse is that caudally we have a brain stem and if you were to cut that in a coronal plane you get a nice cross section of your brain stem you also have dorsal which is towards the back and ventral which is towards the stomach which translates to up and down because in the human when we talk ventral and dorsal we're talking up and down as if we were a four legged animal and not front to back so go ahead and hit that plane so what happens is that you have a coronal cut in the human our dorsal ventral is no longer towards our back and towards our stomach unless you tilt your head back our rostral and caudal we don't really have a tail but we do use that mostly for brain stem because there are long axis is on the nose to tail dimension and go ahead and you can start clicking through these you have a true coronal plane you have a coronal plane that is more often seen in specimens we call it a nose down or a nose up depending on how the specimen is tilted you have a transverse axis of the brain stem which is neither coronal or horizontal but is is equivalent to that coronal plane in the mouse so when we do the mouse or when we do the human dissections they're done very differently and specifically for the structures so that we can retain a little bit more information now the MRIs are presented in what's called the radiologic view which can be a little confusing because left is presented on the right and right is presented on the left as if I was looking at you because your right is actually on my left and your left is on my right so you can imagine looking straight at someone's nose and then to keep it consistent when we move to the horizontal plane that also is flipped and presented right on the left and left on the right so I was told by our it team I had to bring that up because it confuses them all the time so hopefully there will be no confusion now as we talked about we've got these convolutions in the brain that makes it really hard to see where data points are especially when the points are taken within the soul side and so what you'll see are a lot of maps like these which are where we strip away the cortex apply the data points to the underlying white matter and then almost like inflate it like a balloon a little bit so that it's flat so that all those spaces come out mostly what you'll see are these maps where each data point has been plotted back to the actual position but because each brain has a different pattern of soul side and gyri going from one brain to the next go ahead and go back there for a second these data points are going to move around based on where that slab was cut in the human brain and how the gyral structure is in that brain so when you're comparing two brain sometimes this is not the best view so what we've been working on and what will hopefully come out very soon is this generic diagram of an inflated white matter map and now instead of looking at individual samples all those samples are rolled up to a given structure and we can place it in this generic map so that when you're looking for brain one to brain two to brain three it corresponds and you can look in the exact same spot on every map okay those make sense because that's going to be really important when we get to brain explorer yeah happily our IT team has made it so that in almost all cases when you click on a sample or hover over it there's a little indication of what structure you're actually in it's not as straightforward as we had originally hoped and this is kind of an intermediate improvement initially we just had spots that had no anatomic information available whatsoever so hopefully we're getting better and better at the way that things are diagramed okay all right and then just a final bit of information on naming conventions in the atlases primarily in the human atlases and that's because for one thing you want to keep all of your specimens separate there's enough individual variation unlike the mouse if you use 200 mice to do a project you don't need to keep track of which mouse is which so much but you do when you look at samples in the human okay so those samples are coded in a way to give you that information and I think they've changed the output a lot when you download it to make that more understandable as opposed to the full name of a sample that that we generally look at but when we sample we take multiple samples from a given gyrus and so to further segregate it you often see this dash i dash s type of thing or a dash m dash l and those refer to what bank of the gyrus it was sampled from now normally in a lot of cases this an i and an s is irrelevant it's in the same area but there are areas in the brain where it becomes very important and so depending on the level of resolution that you're looking at in the exact structures that you're trying to examine you may want to pay attention to that okay the brain span atlas has a really unique acronym system so we tried with the human brain atlas to at least keep the acronyms that were common enough that if you you work in anatomy and you see an acronym you can pretty well guess what it is we don't have cortical structures in the developing atlas and more important we have layers that are developing and they develop in a pattern from front to back in the different areas and so the samples are coded to tell you which layer you're in which region you're in and more specifically the sub region that you're in so these always throw me off because I wasn't involved in the naming and I always have to go back and look the developing mouse brain atlas ontology is completely different from all of that right and that it's embedded and there's a time component embedded into it in that pros and mesmer thing the way that we talked about and so what happens is that as you go deeper into the ontology you're going into more development and more segmentation of a region and so you can back out to level three and see what pros and mere mesmer you're working with but to get to the actual structures and development you need to move way down usually to like a live level 10 or 11 to get an idea of what the structure will eventually be doesn't exist quite yet on some of those and then how that's fine that was again coming off of that there are differences in the ontology we try to deal with that when you get a data return by saying hey you may be interested we have the same gene in the mouse study we've got it in human or we've got it in macaque and telling you where to go look in an easy click to get to similar information so all right questions on those apps and people should not be afraid to pull down data I mean I had a slide in here I took it out this morning just in the month of July there was almost 400 gigabytes downloaded just for the human brain atlas it's like on the order of terabytes a month that are downloaded from all over the world so if you are looking to download don't sit there and be like oh no can I really take this much information yes by all means take it we want you to have it so just be aware if you take too much your computer may not be able to keep up with it so see I took that slide out to a long time ago there are multiple publications that have come out that reference the atlas in a lot of different ways everything from you know showing that data as support for their own studies to things that look at the ontology and from more of a larger perspective look at the ontology and compare back like PubMed to get an idea of structures that are associated with each other I believe on the web page there is a link to referenced publications don't quote me on that until we get into it but we have a host of publications that have come specifically from the Institute using data from these studies from other studies that we've done but it is starting to be referenced a lot more so it is getting used testimonials of course everyone says it's great right yeah at this point not much information on dendritic stuff there are a few genes that the RNA will be located in the dendrite so you can get a bit of dendritic pattern most of that is going to come with the additional initiatives that we're taking on that will be looking more at cell structure and location of dendrites and number of synapses and stuff like that so upcoming but in these atlases you're not going to get too much of that boy just take a dive right into it yeah with the human stuff that's always a huge question right how much of what we see is technical variation versus biologic variation and some of that you have to come in with a preconceived notion of what you expect but at the same time try to attack it from a we don't know what to expect and therefore anything is possible and a lot of times what that means is you know we may get an inkling that you know there's this technical variation and we just want to get more specimens in there to see how that goes we do try to the data that you pull down is normalized to reduce some of that that technical variation so in hopes that most of what you see is biological now how much of that is structural versus specimen is I think we probably need many more specimens and we're able to do to get to that question yeah again the initial look at the mouse brain atlas showed that looking at microarray studies and comparing it to ISH studies there was about a 50 percent correspondence or so I believe again it's it's somewhat confounded by the fact that those probes weren't necessarily to the same part of the gene and so you get a host of either exon differential exon expression types of things and again when you get to the protein level we have not specifically looked at anything but we do have examples where they're disparate as far as the findings go so yeah yeah I think what they did and this was before I got there is that they looked at the expression levels and then they went back and looked at the ISH and looked for both density and intensity so when you look at the ISH you have the option to turn on a heat map of the expression which takes into account density and intensity and subtracts out background right and so I think they were able to use those calculations to look back and it may have been like a comparison of two genes which direction you know that the fold change was type of thing and it probably went as basic as you know microarray says this should be expressed and our ISH is not showing anything so that 50% you're right could mean a lot of things and I think you want to just think very globally and that about 50% of the time the ISH is going to look like you expect from the microarray or the other way around