 So, what I will talk today is really about basically application of our effort a little bit and I'll try to keep it short and simple. I will not go into deep of the data, so feel free to ask me questions. So basically really using genomics and bioinformatics to really create a high-triple data sets and then use informatic approaches to really mine in a hypothesis-free manner the data sets and really try to come and find some things about human brain development. I will divide my presentation in three parts. First I will tell you why we are using this approach. Secondly, I will tell you about the data set we created and in third part I will show you a vignette, really several vignettes of examples of how we mine the data set and what kind of the data and some new interesting and even hypothesis that can be generated based on the data set. So do I need to turn this on? Okay. Okay, I'll start. So there are basically four problems one faces when one study human brain development. The first couple of problems are really related to the nature of human brain as biological tissue. That's first when it's complex and many of my colleagues actually showed you examples of how that complexity is both in development. And the second one which I really would like to emphasize is a very productive process and to illustrate it I'm going to show you this slide. So if you put time on a logarithmic scale, so this is basically prenatal development. This is postnatal development. You can divide it into phases such embryonic, fetal development, infancy childhood, adolescence and adulthood. And now we're just looking at the really structural development of human brain. You can see that basically takes several, seven months and your brain is still smooth. And then when you are born you have most of your primary sanitary gyroscope, even some tertiary one. But what is really interesting actually that your brain will grow three times in size in the next first five years of your postnatal life. And even then you are still making connections, you are making money and probably into late adolescence. So that tells you how. And if you look at the some of the major cellular events such as neurogenesis migrations in optogenesis and axonal growth and manipulation you can see that they are spread across the time. So basically it takes several months, around five months, to create the entire repertoire of neurons in the human cerebral cortex. That's five months basically. It will take seven days to do that all in mice. So that tells you what is the really complexity, it's not just the number. Another interesting thing about human development, if you take a mouse everything is compacted within several weeks. So once, as Tomomi showed, when you are creating layer four, telomocortical fibers are already in the cortex. Well, they are not actually in human brain, you already create layer four. They weigh there, sit there. And so this is why human brain came up with some of the really specialization such as a huge supply zone where the axons are waiting that the entire component of the cerebral cortex is generated. And I think that this is a really unexplored aspects of human development because if you take the, you know, it's called principle of heterochrome, if you take the same events and you stretch them or you compact them, you can come up with some different variations and species specific. The other two sets of problems, actually major problems are related to basically ethical and take, you know, limitations that human brain basically you cannot do experimentation. And the last problem is really, even if you want to do it, it's actually really not hard. It's hard to collect high quality tissue due to many reasons that I explained. So several years ago, actually almost a decade, I stayed the same place and I really was a great time. And I think in my opinion, technology is the major force that drives science. That's just my opinion. I think ideas are becoming cheap and what really changes science is technology. When I started my lab, like, here comes high throughput genomics. And I said, wow, I can combine these two, you know, my love and passion for human brain are really with that. And was fortunate that actually there were really people who knew what they need to do at Yale and really helped me to build that. And in the last several years, that became grew into something. And there is a consortium now of several universities listed here. And it's called the Brain Span. So you can go there. And there is an Atlas, Atlas is maintained by Alan Institute for Brain Research. And you can, there are many other things in addition to things that I will talk today. And so the first thing I did is we collected a large collection of a human brain, post-mortem human brains, that covered the entire aspect of the development of prenatal and post-mortem lifespan. So basically from five weeks of post-conception all the way to over 80 years. So we have around 200 brains. And as you can see, there are really dramatic changes in how that looks. We divided the development into 15 periods just for an adult with just for purpose of really trying to obtain multiple specimens, males and females. And so part of our part of this entire consortium and some previous work was to really profile 16 areas or regions of the human brain. Of course, you can immediately come to conclusion that my favorite part of the brain is the neocortex. So we sampled 11 neocortical areas. We were guided by two principles, functional importance. So there are four areas of the prefrontal cortex, all major sensory areas, and some of the areas involved in association temporal parietal. And also we are guided by ability to dissect this. So these were all dissected from post-mortem human brain that were collected from clinically unremarkable donors. In addition, hippocampus, amygdala striatum, medidorsal nucleus of the talums, and the cerebrocortex. In addition to looking at the gene expression or transcriptome, dynamics, my lab and several other groups have looked at many other aspects of human brain development, including reference atlas, histology, epigenome, genome, and you can all get most of these brain-span atlas. So I will specifically tell you a little bit about what we have learned and looking at the gene expression. And I'll try to keep it simple and short. So basically, the data said that we have been really focused on creating is basically around 58 clinically unremarkable specimens. The 58 were based on covering all 15 periods. In each period, we had multiple males and females. And we also tried to, we have very rigorous QC protocol to really includes they were all neuropathologically examined by trained neuropathologists. We dissected around over 1,300 samples from these brains. Half males have females. Most of are European, some African-American, Hispanic Asians, and in many cases, it's mixed. So we applied three types of analysis. One is genome. So all of them were snipped genotype using a high-density SNP arrays. And recently, we have also done whole genome sequencing using a brain cerebellum-derived DNA. We have also looked at major histomodification marks and DNA methylation, so this way, we can look at some of the epigenetic changes. And transcriptome, we used two, basically, platform. One is exon, microarrays, which primarily looks at protein-coding genes. And also, we did RNA-seq, which was done on slightly half of the samples. And the reason it's half, because in exon arrays, we look at the left and right side. And in RNA-seq, we look only on one side. And it will be clear why at the end of the talk. So that's the data set. And then what I will show in my last part is just several vignettes of type of data we have derived. And many cases, I think, will illustrate complexity of the human brain development and some new ideas about what could be happening in human brain and some species differences. So this is a multidimensional scaling plot in which we plot 1,350 dots. Each dot represents one sample. And so what we are looking at is the relationship of the transcriptome. This is not individual gene. This is a protein-coding gene. In this case, it's 17,000, because that's how many were on exon arrays. And we slightly get more on RNA-seq. Not all genes are expressed. And this is a basically plot. And then we can paint it using periods. Remember, period one is an early embryonic period, while period 15 is late adulthood. And you can see that basically what we are looking is how related are each dots to each other, looking at the whole transcriptome as a comparison. And you can see that basically you can plot it by period. And you can see that major dimensional variance is actually related to time. So basically, and you can also paint it by regions. We compressed all 11 neocortical areas into one color. Otherwise, it's impossible to see the colors. So these are multiple neocortical areas. And you can see beautiful segregation. Neocortex is here. Cerebellum is here. And you can see that. But if you look at compare these two graphs, you can see that actually major variance in this dimension is time. And what is surprising is this. This is period seven is light green samples. These are all different areas. Now they are painted by time. And in this corner, they are painted by regions. And actually two-thirds of the variance actually can be explained as prenatal development. So first, basically 10 months of your life are the Muslim enemy. And then comes the infancy. And this is 20 years to 80. It's all compressed. So that tells you basically that gene expression changes less during prenatal development, changes during postnatal development, and the least during adulthood. The most of the differences happens during prenatal development. And that is the problem, because actually this is the least what we know about the prenatal development. Due to many reasons, one of them is really collecting high quality tissue. So in addition, actually to time region, you can do this using sex and individual variances. But the major actually global difference is time, then comes region and sex and individual. So one of the tools that we have been using to really mine this data in an unbiased manner was really using a gene co-expression network. And specifically weighted gene co-expression network, which is pioneered by Steve Horvath in 2005 at UCLA and used by many colleagues, including Dengeshwin and myself and many others and people at Allen Institute. And so basically what we are looking at, spatial temporal variance. You know, we are trying to collect and see whether there is an organization of gene expression dynamics using a variance across time and regions. And basically, depending on what parameters you use, you can come up with hundreds of these networks or modules and what they are contained of genes that have same behavior. It's basically that's how Google makes money. They know that I like guitar and red wine, and my wife likes shoes. And so basically similar principle, you probably know more than me about it. And so basically you can now organize these genes into large or small modules. Modules come in two basically. They are either very large, and here are two largest one in the human developmental brain transcriptome. M means module two is just an arbitrary number that we give it. So M two is a large over 3,000 genes, and they all behave pretty much the same. And so this is a basic PC one. You can even create their eigen gene. And so all they do is that they are expressed low in during fetal development. They go up, reach plateau around childhood, early childhood, and they stay pretty much the same in all neocortical and all regions of the brain that we have analyzed. You can actually even find another large, which is M 20, it's a very large developmentally regulated, but this network is around also 3,000 genes, slightly less, I think. And what it does is basically all of these genes are basically being down-regulated across the entire. And so the second thing we can do with this type of analysis, we can then do functional notation. So we know what is the list of the genes. We can rank them by basically how close is their dynamics to eigen gene or PC one. And basically then we can look at also enrichment of gene ontologies using David and other similar databases. And so for example, for genes that are going up as you mature, you can see that most of them are enriched for membrane, proteins, calcium, signaling, synaptic transmission, as you would expect, because your brain is maturing, you're creating synapses. On the other hand, genes that are up-regulated when you are creating neurons are mainly genes related to zinc-finger transcription factor and SOC transcription factor. My other work that I'm not going to talk today has shown that many of these are important for specifying neuronal identities. So you can clearly see that these are not just meaningless networks or modules, that there are really some meaningless. And in addition to this large one that you know, you can also create or get many that are small, handful of genes that are really going up or down, depending on the time or region. And in specific, and I'll show you an example later on. So now the other interesting thing, which was one of my most disappointing things, and this is why I like hypothesis-free. So the reason we did this whole entire study, my primary motivation was find left and right asymmetry. Human brain is incredibly asymmetric. I mean functionally, structurally, comparing to other species. And so we decided, okay, let's do left and right. And many, many years of painstaking work created nothing statistically significant, which was really absolutely shocking to me, okay? And so here are two types of analysis we have done to tell you that sometimes hypothesis-free is probably the best. So what you see here is each region that we have analyzed. So these are, in this case, I'm just showing you neocortical areas. So this is a medial prefrontal cortex, orbital, those are lateral. This is Broca. This corresponds to primary auditory, which you all know that's highly, and this is a superior temporal gyrus, which would correspond to vernicare language areas. And you can see that we have covered really areas that are highly functionally, as well as structurally asymmetric. And so this is a heat map, basically showing fall differences, and you can see that majority of the gene basically fall within, you know. We see a lot of variations, left and right in individual brains, but at the population level, nothing passes even the most relaxed statistical criteria. So one thing in mind that we are looking at the population and we are grounding the tissue, so that's also another thing that you have to keep in mind. And secondly, we do see variation, but unfortunately nothing is consistent at the population level, using even same ethnicity, same periods and stuff like that. Another, as many of you probably know, there have been a lot of literature showing, okay, maybe one other model of left and right asymmetries that left and right hemispheres mature differently. That has been shown using functional studies, so here is a plot of left and right. So what we are looking at, individual neocortical areas transcriptomes, okay? We can do this based on individual genes or whole transcriptomes, so in this case, just to show you global data. So the left trajectories of individual areas are shown either in dashed or full line, and you can see that they all basically behave the same and there is no statistical difference in maturation trajectories. So what we are showing here is that at the population level, if you look at individual areas, there is no anything statistically significant and that their maturation trajectories are similar left and right. Now, there are many, many consequences of this finding and we are happy to discuss if you are interested. So another thing that really we're interested in doing this is, and Tomomi really beautifully, I mean, I'm actually very happy that she gave talk before me because I will say in a couple of minutes explaining. She showed you also examples that we actually have around 20,000 protein coding genes. I mean, we share them all with platypus. I mean, there are very few genes that are really unique to human brain and I'm not sure that they are even meaningful. Okay, that's my honest, just opinion, just the number. And we see tremendous variety and variations in morphological development. We see the cognitive development and basically it has been long known from Wilson and King in 1975 that basically what probably creates all of this is a spatial temporal variations in gene expression. And that was another reason where we were motivated to do this. And Tomomi showed you the same gene that basically is slightly differently expressed in different species and I think this is something very important. And so human cortex as you probably know, probably has more areas than other species that are commonly used in laboratories such as non-human primates or rodents. And so we are motivated to look at what's happening in dynamic of this aerial expression pattern. And so if you cluster this area, so this is a clustering of 11 neocortical areas, hierarchical clustering unsupervised based on their whole transcriptome. And you can see that we can cluster them by lobes as well. So for example, this is a prefrontal cortex. This is a motor primary, motor primary sensor. They are next to each other anatomically. But what's really also exciting is that you can also cluster areas that belong to different, in addition to topographic nearest neighbor, you can also cluster them based on whether they are functionally more related because ITC is part of the temporal cortex, but here we see clustering of angular gyrus, primary auditory and superior temporal, which are basically all important for functional posterior language areas and speech and labor. So the first thing we look actually, the number of genes expressed actually is slightly less as you mature, but it's not really big difference. But what was really surprising to us when we look at how much of difference do you see between areas? I mean, if you look at the moment's slides, you know, the freeze forgive me, you know, like when you're asleep, probably you'll forget the name of the gene. So, yeah, exactly. You can see that it's actually highly enriched in sensory areas, but there are variations across the species, but also there are variations and that's how she fished out the gene across the development. So we look at how different are neocortical areas across the entire development. And surprising this has not been done with any mouse or any other species, but our dataset allows us to do that. And so basically what we are now plotting is how many genes are differentially expressed between two neocortical areas. Keep in mind that the number of expressed genes really, it's slightly less in mature brain, but it's basically pretty much similar, okay? And this is what happens. On average, there are basically almost 1000, little bit less than 1000 genes that are differentially expressed, okay? But when it comes to early development, postnatal infants in childhood that drops dramatically and then goes up again in adolescence and adulthood. And so basically the shape is not that it goes more fetal and progressively less, it's actually does this U-shape curve, okay? What suggests that actually, and it's consistent with previous data I showed you, that most of the difference are present postnatally, but what happens that has this big dip and then again goes up and basically has this temporal hourglass pattern. And this is a Manhattan plot and basically it's a tachy test basically looking at how many combinations are of these genes between two different areas. And you can see actually, and also what we are now looking at, are all areas created equally when it comes to this differential expression or some are a little bit more differentially expressed and driving this differential expression. And you can see that actually that some areas have more differential expressed genes and here is the example. So medial prefrontal, okay? Ventralata, this is a corresponding to frontal opericum, it should be prospective Broca language area or premotor. And look at this visual, primary visual and temporal cortex. Those are basically areas that are driving mostly differential expression in the fetal development. And then suddenly the drop does not happen at birth. That was second surprising finding. The drop does not happen actually. I don't know exactly when it happens because unfortunately we don't have tissue between 24 and 35 is just impossible to collect it for reasons of heroic medicine in the United States. But we predict that drop happens in between 24 weeks of post conception and 35. Now instead of having several thousand, you basically drop to basically a few dozen of differential expressed genes. So this drop does not happen. It happens before birth, okay? And then this drop stays actually relatively constant, okay? So there are still genes that are differentially expressed and consistent especially between primary sensory areas. And then a gong goes up and not all areas are actually going up equally. Again, it's mainly driven by media prefrontal cortex and primary visual cortex. So now what are the genes that are driving this temporal hourglass? And so basically we look at, again, use the network analysis and to find. And so basically we can again cluster genes. And basically, so these are the modules. Don't worry about the numbers. But what we observed is actually exactly what I was showing you. That many of these modules are sometimes hands, a handful of genes, sometimes a few dozen of genes. Really actually they are all co-regulated in the same regions, but also what we observed are dependent on the time. This is period three, which corresponds to early fetal development. So this is a set of genes that is highly upregulated in lateral parts of the prefrontal cortex. And they are all down in non-prefrontal areas. We can find a set of genes that are actually doing this in the temporal areas. We can see that only doing in primary visual areas or medial areas and on and on and on, okay? Another feature, which I'll show you just one example later, that most of these are temporary regulated, as you would expect. Because remember that temporal hourglass shape is actually almost completely disappeared. And here is an example of one module M91. So it's a set of genes that are all highly expressed in the prefrontal area. So this is a clustering. And then basically somewhere in between six and period seven that this pattern is completely turned on. So they are all probably somehow co-regulated, either by some transcriptional makers and gene regulatory network or activity. We don't know what it is. On the other hand, many of the genes that are temporary regulated and spatially regulated in prenatal development actually turn on in all areas. So there is like a gradient. So that's basically two basic flavors. Another important thing, I'm not going to show you any data just for the sake of time. When we did the same analysis for postnatal development, especially adolescents, we actually had no gradients. Majority of gene expression modules were either related to sensory versus non-sensory areas or were expressed across the entire cortex and related to either sex differences or some other difference. So why is this interesting? Well, it's interesting because you can also do one other thing. You can give me a piece of prenatal tissue and I can tell you exactly how old is the brain and what is the part of the brain using six, 11 areas. Because each of them have a very unique transcriptional signature and it's actually changing across the time and you can divide. And at that time, it's very hard to find that there are no anatomical borders. Brain is completely smooth, I mean cortex. And what is really important that many of these genes, and Tomomi showed you some of them, are actually have been linked to major psychiatric disorders and some of them we have actually also studied in our human genetic studies. When we look at the transcave variants, I mean all genes that if you have two axon, you will be spliced at some point in development. And also, I'm not going to show you any data, except one slide. We have also compared this data with non-human primates and many of them have differential expression differences in species differences. To show you how you can really find and Tomomi stole this slide because she showed the same thing. So I was, it was too late to remove it. But, so here is an example of two genes. So this is a cerebellion two. It's a gene that was discovered in the cerebellum, was called cerebellion two. But actually what is really interesting has a really specific prefrontal gradient and you can see beautifully labeled prefrontal areas, labels all neurons. And if you look at the mouse, you can see the same thing in the mouse. However, here is an important thing. If you look at the human, it labels all layers equally. Basically this is the bottom of the cortical plate or the top, but in the mouse, it's really highly expressed in the upper layers. And this is a work from colleagues at Allen Brain Institute. And here is another gene that really was, I was so excited when I get this in situ, but, and then, so here is a gene, MP5. You all know MP5, it labels interneurons. It's one of the most commonly used markers of interneurons and you can see interneurons. But also heavily labels a visual and some temporary areas. And these are probably layer four neurons. I think they are actually, what does MP5 do that? We don't know that. If you look at the mouse, it's actually not expressed in the primary visual cortex, pyramidal cells or layer four cells. So here is a simple species difference for a very commonly studied genes. And I, to me, made the same point and I wanted to finish with this slide. And so a couple of points that I wanted to show. So basically 90% of protein coding genes are expressed in the human brain. That is going up. You have to keep one thing in mind, that this number that is a shooting and moving target, the problem is that we are grounding the tissue. So, you know, you also have blood vessels, parasites, so you know, it's not surprising actually, just to be honest with you. So basically if gene has two axons, will be spliced for sure. So basically 90% of genes are somehow differentially regulated, either splicing or expression. We see substantial sex bias, allelic expression, interingual variations. I did not talk about it. Transcriptome is organized into either small or large networks of temporary, especially, co-expressed genes. We think that they are somehow co-regulated. Prenatal development is most dynamic. I think it was neocortical transcriptome is globally symmetric and it's globally symmetric and exhibits temporal hourglass pattern. What drives that pattern? That pattern happens before birth. What happens that all those gradients are globally turned down and disappear? I don't know what the mechanisms. I can speculate on it. And also some developmentally regulated genes showed differences across species. So, this work was done primarily by these colleagues here. I'm not going to list them individually for sake of time. I would like to my colleagues and brain span. Many colleagues at Yale that basically introduced me to genomics and helped me do this. And I mean, I'm really grateful to all of them and funding, thank you. We have time for some questions. Go. Given that you're mushing the brain, it's not surprising you don't find a symmetry. And the neuronal type differences in gene expression are washed away. So, the odds of finding slight changes are going to be covered by the majority of the cells in the region, astrocytes and excitatory neurons. Yeah, so that's simple, straightforward criticism. But there have been previous reports showing left and right differences this. We actually don't see any of it. And one of the problems is also sample size. The more you add, it washes out because you also see inter-individual variations. And it's impossible to do this. It's really, and even at the levels of neurons, I've been, this really changed my thinking because I was somebody who believed in genes, but I've been really more and more activity dependent. I mean, look at the mom's work. I mean, you know, it's a gene is expressed, but it really can change the dendrites and is expressed in multiple areas. So there are two possibilities. One is what you said, which I think is likely. Our study was not designed to fish for those and detect them. So that's one caveat. But it also shows that these differences are going to be minute. They are not going to be across the entire neocortical areas. And finally, I also do think that there will be a lot of effect of activity dependent mechanisms. And brain is incredibly plastic. I mean, one thing I always tell people, I mean, if you have ever killed a child with hemispherectomy, they remove the whole hemisphere, and their English is better than mine. And I have two hemispheres and cannot get rid of my accent. And because their brain is plastic, and you know, if you remove left and right, they will develop language. And almost, you know, I'm incredibly intelligent. They go to college, and you would not recognize them. I have seen one. And I have accent, and I have two hemispheres. But it's really interesting how environment can change your brain. And I've been thinking more and more about it in the context of these problems. One possibility would be to do nuclear RNA and isolate. Yes, we are doing that, sir. We are on the same. Thank you. Beautiful talk, Nenet. And probably it's really related to your question. And I don't know anything about these things. But for the asymmetric AC, are there any possibility that Salamis is also controlling this asymmetric functioning? So again, that's, you know, if you look at the battlefield, I mean, it's incredibly plastic. I mean, one of my favorite papers was, I forget the name. It was Japanese group 1982, remember? And they'd lesion the future position of the battle cortex in the mouse. And battle cortex move completely to auditory and develop completely. So your work is also showing, I mean, if you mess up the gradients, you can have two battlefields. So it's remarkably plastic. So it's not, I mean, the only thing this tells you is not going to be simple. So that's basically what it tells you. And I was naive thinking that it would be that easy. And also guided by some previous work. Then we close this session. I would like to thank Nenet again for great work.