 Well, thanks a lot for inviting me to give this talk, as Mary said, I am a neurobiologist in essence, and I stumble into neuroinformatics and genomics, but I'm still a neurobiologist. So I teamed up with Joseph Ecker, who is a head of the genome analysis lab at the Salk Institute, and his then postdoc, Brian Lester, who is now a professor at the University of Western Australia, and Irann Mukamel, who is a computer scientist that was a postdoc with Teri Sinovsky, and he's now an assistant professor at UCSD in cognitive sciences. So this is kind of weird. As you all know, the sequencing of the human genome led, or tried to lead to the identification of disease-causing genes, this has been pretty good, and I should say it personally, a personalized medicine advancing sequence technologies, and the foundation for the understanding of the blueprint of the human beings. However, we have learned also that DNA sequence on itself may only be half of the story. So you can notice that variations in DNA alone may not entirely account for the variations in phenotypic strains, and you have organisms like plant, insects, and mammals that show differences, which you cannot find as mutations at the level of DNA. And for instance, here is Joe Ecker's preferred plant, Arabidopsis thaliana, where you have a phenotype that appears to delay the flowering of the plant. And this is due to an epialyl that gets methylated and that delays the appearance of the flowering. The same happens in bees, which to me was very interesting, because all bees have the same genome, and the workers produce the royal jelly, which is fed to the larvae for three days, and then they stop. However, the queen gets this royal jelly for more days and very concentrated, and that is what leads to the formation of a queen. So they all start exactly the same. And recently it was found that it's a peptide that inhibits DNA methylation, what causes the possibility for that royal jelly to induce a phenotype of a queen. And also we have in mammals, the most common one is the epialyl that gives the agoutic color, that is a transposon in a specific site in the genome of the mouse, where it has an ectopic promoter, and depending on how that promoter gets methylated, you get more or less of this gene, and then you get the what is called agoutic color, which is sort of yellowish. As you can see here, it not only causes the coloring of the skin, but it also leads to obesity. So we have that one genome can lead, as you know, to all the different cell types you have in an organism. And so this patterning from this single genome to all the different cell types is controlled by transcriptional programs that are coded in the genome. And these transcriptional programs can be regulated, and we have, oops, sorry, we have what we know, the transcription factors that will bind and induce or repress a gene expression, non-coding RNAs that will do similar things. And we have epigenetic modifications, which means that modifications to the expression of the genome without altering the sequence of the genome itself. So they are layers on top of this genetic code. And those are usually covalent modifications of DNA or histone that can modulate the redox of the genome. So how can we study this? We can do it by sequencing. And these are all the different possibilities you have to study the different modifications of the DNA. So the DNA sequences, you have it there, and you have transcriptional that you can do by RNA-seq. You can do, analyze how the chromatin is exposed, which is usually related to an active state where it can transcribe. Also the nucleosomes can be analyzed and the histones modifications, and all this can be analyzed by different type of sequencing. And we are going to be talking mainly about DNA methylation and a little bit about histone modifications and chromatin accessibility, which was not done by us, but by the ENCODE group. And we were able to use their data for the analysis. So DNA methylation is modification of the cytosines in the DNA sequence by putting a methyl group. This modification has been related to control of gene expression and differentiation of cells is essential for development. So if you knock out the enzymes responsible for putting these methyl groups in the DNA, you kill the organism. It's fundamental for silencing of transposones in the DNA. So one of their theories is that in fact it came upon by doing this, and then it was co-opted to do gene regulation. But the original function of DNA methylation was silencing transposones so they cannot jump around. It's involved in X chromosome inactivation and also it's involved in cancer. So we have three players in this group, which are the ones that write the cytosine methylation. And these are two types of enzymes. The main one is DNMT1, which is the one that when DNA replicates will go and put in the new cytosine a methyl group if the parent strand had the methyl group. And then we have the NOVA DNA methyl transferases, which are the DNMT3A, 3B and DNMT3L. And these ones will just put a methyl group on a cytosine without any pre-pattern. Then we have an editing system, which is a modification of the cytosine by adding a hydroxy group and that is 5-hydroxy methyl cytosine or 5-HMC. This is an important way of taking out that methyl group. So if you want to regulate that methyl group, you can take it out. Up until a few years ago there was no way to demethylate in the mammalian brain, but it was recently shown that in fact these enzymes, the TET, which are oxidases, can add 5-hydroxy methyl cytosine and that way oxidize back to the cytosine. And then you have the readers and one of the most important readers, which are the methyl binding domain proteins, but one of the most important is MECP2, which is involved in red syndrome. How do you analyze epigenome by sequencing? You use a chemical trick. And so what you do is you shear the DNA just as if you were going to do high throughput sequencing to pieces of about 150 base pairs. You put adapters to it, but in this case your adapters are fully methylated. So you make sure you have your adapters with methyl groups, because then what you do is you treat with bisulfite. And that what it does is convert the cytosine to an uracil. And then this is read as a timing, not as a cytosine anymore. So whenever you have a cytosine without a modification, you treat with bisulfite, you will convert it to a T. And so then you do your sequencing and you will get cytosines only on those ones that were methylated. You amplify, you do high throughput sequencing and then you align your reads. And as you see, those cytosines that had a methyl group will appear as cytosines. The rest will be appearing as thymines. So this requires that you convert your genome of your preferred genome. All cytosines need to be converted to thymines to be able to map. And then, so it's a bioinformatics approach. What whole genome, a single base resolution sequence allows you is to see the context on which this modifications of DNA happen. So what we have known is that cytosines get methylated in two broad contexts. The second one is CG methylation, which is the most common and you're probably going to, whenever you read about methylation, you're going to be reading about this type of methylation, a cytosine that is followed by a guanine gets methylated. This dimers, the methyl group is about 80% of the cytosines in the genome are methylated. There is another context, which is a cytosine followed by anything but a guanine, which is non-CG methylate. And that one, up until now, it was considered rare in the mammalian genome, but we are going to discuss a little bit more about it. And so this is the way you see it if you go to our browser where you can align your genome with all the reads. Here you have the genes. I think this is chromosome two, a bit of chromosome two. You have your genes aligned there. Then you have the RNA coding for those gene. And then you have the methylation of this. Each tech here corresponds to a cytosine that has been methylated. So chromosome two, it's very compressed here. That's why you have so many genes. And in blue, you're going to see non-CG methylation. So you can see that CG methylation is much more prevalent than the non-CG methylation. But our surprise was that the brain does weird things. So you can zoom in at this point, which I did, is just clicking here to zoom in. And there you can see your cytosine that was methylated, one of these techs. And it's simply the color code is here. So when you find a cytosine in the read, it means that cytosine was methylated. So non-CG methylation was considered to be a weird thing that happened in plants only. But a Joe Ecker's group showed that, in fact, it's pretty prevalent in stem cells, where it corresponds to 24% of the cytosine in the genome. You have to think that the proportion is about one billion cytosines in a genome. CG, followed by G, is only a proportion of it. The majority is C non-CG sites. So 24% of the genome is quite a bit. And what was interesting is that this happened only in stem cells, but not in fibroblasts. So when you, and they did several differentiated cells, and it looked like that the stem cells would contain these two forms of DNA methylation. But when you differentiate, it disappears. And it was mainly in gene bodies. And the interesting thing also was that if you make pluripotent cells starting from fibroblasts, so those are the IPS cells, the non-CG methylation that you find in the stem cell would reappear. So there is, it seems that methylation is functional. The methyl, as I said, the methyl group can be further modified. And one of the important ones is the 5-hydroxy methyl cytosine. So this is done by the tetproteins, and it's a series of oxidations, where you get the 5-hydroxy, the 4-mil, and the carboxyl form, that finally oxidizes to release the methyl group that can be, again, methylated by the DNA methyl transferase. There are other pathways here that I'm not going to talk about. But this form of cytosine was known to be present in the brain. And it was an interesting. So with these tools, we went to approach the brain. And so as you know, electrical signaling in NERA networks requires a sort of stable network. But this needs to be formed by cells that have very clear phenotypes, but also are flexible. That you need cells in the brain not to change their phenotype. We know it because of if you have an alteration, for example, in an inhibitory neuron system, as it's the case of schizophrenia, you have a problem. And epilepsy is another example of changing the function of one of the neuronal systems. But again, all neurons in the brain and glia share exactly the same genome. So we thought one way of differentiating the cell types would be DNA methylation. And in fact, DNA methylation had been related to the normal development of the brain, as well as neuropsychiatric disorders like autism and red syndrome, Alzheimer's disease, schizophrenia, and normal aging. So our question was, what are the normal patterns of DNA methylation in the brain? Do they change during development? And how do they change? And are they cell type specific? So for this, we needed to separate the cell types. And to answer this, we did a whole genome sequencing. And we chose, and that was my decision, the frontal cortex, because that region is the slowest to develop in mammals, and its alteration is involved in most of the neuropsychiatric disorders we know. So its decision-making executive function seems to reside in this region. And as you know, also, we have a high complexity of cells. But mainly in the cortex, we have pyramidal neurons, inhibitory neurons. This corresponds to 80% of the population. These are 20% of the population. But we cannot forget that they sum up only 50% of the population of cells in that region. The other 50% are glia. And so it's a mixture of cell types. And so what we did was to analyze the genome, the methylone patterns throughout the lifespan of both mice and men, although there are a few women in between. But in general, it's mice and men. We have repeats here that are women. And we did the whole methylone. When I say methylone, it's whole genome DNA methylation. To analyze the translation of that, methylone changes, we did RNA-seq for the same samples. We also separated cell types as neurons and glia. And for this, we isolate nuclei. So we produce nuclei and fact sort, nu n plus and nu n minus, because we are talking about DNAs. And we also did a full single-base resolution of 5-hydroxy-methylcytosine to see where these changes was happening. This allowed us to do a series of studies that, believe it or not, the sequencing took about seven or eight months, two and a half years of analysis. It was tough. A lot of dead ends. So it required a lot of imagination and creativity, because this was 65 genomes. You have to think 3.3 billion base pairs, each of them, which you have to start comparing. So it took a while, but we got there. We are still analyzing. What is the relationship of... We can ask now, we have the whole genome. We can ask, is there any relationship between methylation of DNA and transcription? And so what we found, which was a very surprise, is that both the mouse brain and the human brain accumulate a non-CG methylation as they age. So in the fetal brain, you almost see none of the non-CG sites methylated. However, in the adult brain of both species, this accumulated pretty strongly. As we had seen in stem cells, surprisingly in stem cells, sorry, in stem cells, this correlated with increased gene expression. And that was a previous study that Joe and Ryan had done. But in the brain, it correlates very nicely with gene repression. So the accumulation of non-CG methylation in the brain seemed to be involved in gene repression. We also were able to analyze, since we had the whole structure, and we had data on DNA's hypersensitive sites, which allows you to know whether those certain parts of the genome are inaccessible to modifications. And we found that non-CG methylation correlated very nicely with this inaccessible regions of the brain, both the 5-hydroxy and the non-CG methylation, which seem to be deposited later on during development, cannot reach this inaccessible region. Just for you to know, this region that I'm showing here is the variable region of immunoglobulins, which is only active in lymphocytes, which are the ones that express that immunoglobulins. In the nervous system, it has to be shut down. And in fact, it is shut down, and it's shut down. The chromatin is like a knot. However, CG methylation in these regions is unaffected. So that is put when the cells are dividing very early on. Non-CG and 5-hydroxy-methyl cytosine could not be added to those regions. So those are inaccessible regions to this modification. What was interesting to us was that when we went and looked, OK, when does the deposition of non-CG methylation happen during development? And we found that for both mice and men, the accumulation of non-CG methylation occurs postnatal. What is interesting is that, as you know, in mice, the first postnatal week corresponds to the third trimester in human. So eye-opening would be births for human. And as you see, it's only then that you start seeing the accumulation of non-CG. And in the human, our first time point here was a 20-week gestation. And the next one is 35 months old. And this accumulates throughout the childhood and adolescence. And it seems to correlate with the period of synaptogenesis when this is increasing. DNA methylation had been implicated in brain development and plasticity when analyzing very specific sites, BDNF or FOS. And recently, another group did the study on dental gyrocells in the mouse and also found that non-CG accumulates throughout the development of the mouse at the time of synaptogenesis. So who is doing this? So we said at the beginning, there are four DNA methyltransferases, DNMT1, DNMT3A, 3B, and 3L. We found that in neurons, as it had been shown before, DNMT1 and DNMT3A are the ones that are expressed in the adult brain. The other two are more expressed in early before the time point of our fetal sample. However, what we found was DNMT3A is induced right at the time when we see this increase in non-CG methylation. And it stays high for quite a while. Indeed, data had shown that when you knock out in the nervous system DNMT3A, the animals are born healthy but start deteriorating very fast and they die very soon. So the lack of DNMT3A early on allows the animal to be born. But unfortunately, they only looked at motor neurons. So I don't know if they did any other studies on that one. But the animals die when they reach adulthood. And DNMT3A was found before to increase in the very early weeks of development and then go back down. And the double knockout, DNMT31 and DNMT, I went too fast. The double knockout for these two enzymes has a pretty serious learning and memory deficit. So we are studying now the role of DNMT3A in synaptogenesis and how this could lead to alterations in behavior. So, all is well, who is expressing this non-CG methylation that seems to be involved in the maturation of the brain? It's neuronal. When we separated the two cell types, the two main cell types in the frontal cortex, we found that in humans, it's astonishing. It's almost more non-CG methylation than CG in the adult brain. And these are one or two biological replicas. And in the mouse, this is not as great as the human, but it's also in neurons and very little in glia. And in this case, what we found was very interesting. If you look in fibroblast or in the fetal brain, there is almost no non-CG methylation. In the glia, there is a little bit. And so when we went to look in which part, which genes were showing this non-CG methylation, it was a subset of genes that had high non-CG methylation. And those ones corresponded to genes that are highly expressed in neurons and are involved in synaptic plasticity. So genes that get induced during synaptogenesis in the glial cells get highly methylated and repressed, suggesting that both neurons and glia are coming from the same precursor cells. So there is some fate determination in which the lineage of both in one needs to be told, you shut down, you're not going to become a neuron. So the CG methylation patterns were early on when the cells were defined as neuronal precursors. But then when these two go towards glia or neurons, you need another level of gene repression such that this cell does not forget it was supposed to be an astrocyte. And so absolutely speculative, must recognize. But that is my way of thinking. And so one of the questions in the field was, well, is this 5-hydroxy accumulation that is known occurs in the brain? Is it the known CG that you guys are seeing? And so we went to look. And for this, again, the enzyme responsible to put that 5-hydroxy in the methyl cytosine is the tetprotein. So we went and did full sequencing for 5-hydroxy, which was not possible before. And this is using, again, a technique taking advantage of UDP to be able to add to the methyl group using an enzyme URI glucose. And then you have modified the 5-hydroxy. And so the tetprotein cannot recognize it and oxidize. And so what you do is you pick up your genome, add the, modify all the 5-hydroxies where this glucose, and then just add tet to the system. And that one will oxidize all 5-methyl cytosines to cytosines that, upon bisulfate treatment, are going to be converted to a thymine. But the 5-hydroxy got a glucose. And that cannot be oxidized. And so that way you can sequence it. And so we compare and found out that, surprisingly, the one that gets the hydroxy is only CG. We could not find across the genome non-CG sites that would acquire a hydroxy. So it looks as if the tetproteins recognize the CG methylation. So there must be a sequence context for the recognition. Then we look also to the interindividual variability. It was so surprising how well conserved this accumulation of non-CG in mice and men that we said the position must be pretty stable. And in fact, it is. So you can see it here as the text of where the non-CG methylation occurs. Here is for the whole population of humans and mice. And it's very high. And when you compare this between the female and the male, the correlation is even better than for CG methylation, both in mice and in humans. And how does the gene body methylation change with development? And so this is a basic slide, but all I want you to look at is this is take all genes, do K-means, arrange them by gene site, and 100 kilobase upstream and downstream. So we are putting everybody under even big size, small size is relative. And you can see that way you can see what is happening during development and at the cell type level for CG methylation and non-CG methylation. So you can see that there are genes that are constitutively high or highly expressed in neurons, lose DNA methylation in neurons in development, and lose non-CG methylation in neurons. And I'm not going to detain too much. Then you have astrocytic genes are highly repressed in neurons by adding CG methylation and non-CG methylation. So the same that we had seen for the astrocytes that turn off neuronal genes, the same happens in the neurons towards astrocytic genes. They shut them down with this type of methylation. And the glial cells resemble both for non-CG and for CG methylation closely to the neuronal precursor cells. So what happens if we zoom in in this demethylated regions and we go towards enhancers? And for this, you need to, and this is something that is an ongoing problem in genomics, is how do you find, which is the way to look for differentially methylated regions? And there are tons of different algorithms to get to this. But one of the ones that we use is this moving windows to find, since you have the genomes are fortunately the same, you can go and scan to see methylation levels. And so using this, we could find that there was very striking changes in specific regions that coincided with enhancers during development. And so you have CG islands, which is regions of the genome where you have a lot of CG's higher than any other place in the genome. Those are poorly methylated, and that was well known. The promoter regions and transcription ending sites is usually pretty low. But the intragenic and the active zones, transcriptionally active zones of the genome are the ones that show highest differential methylation as you go through development. And then when we looked and we compare enhancer regions and for this, it was different methods, but what is shown here is DNA's hypersensitive side, which increases. So this is an enhancer that becomes active. Histone methylation and acetylation is the same that is none in the fetal and becomes active. And if you go down here, you have that there was a 5-hydroxy in the middle of that enhancer region. And when you look in the whole genome, you find again that in a regions where there is a demethylation between fetal and adult, you can find that 5-hydroxy in the center of the enhancer. You can look at it again here where you can see the sequence doing 5-hydroxy, where you have a few that in the adult brain disappear. And here you have the normal MCG that cannot distinguish 5-hydroxy. Now, if you compile this and you look throughout the development and now many more sites during the embryonic development, you see it's extremely dynamic. And one interesting part was that these two clusters there seem to be demethylated exclusively in neurons. And so that to us represented a very interesting because it's an identifier. And so what we are doing now is to isolate neuronal types, not just the whole family of neurons, but single neuronal types. And for this, we are using a mouse line that was created by Jeremy Nathans and Alisa Moore and John Hopkins that is a nuclear envelope protein that contains a GFP under a LOXP system. And you can cross it with a Cree line. And then you have these beautiful rings around the nucleus of the cell type expressing Cree. So in this case, I just did a immunohistochemistry for an inhibitory neuron. And you can see that there is no coincidence of the nuclear envelope with the cell type that should not express it. And using this, we have sorted this nuclear using the GFP fluorescence. And we have found that for several neuronal types, we can find very specific regions of the genome that change their methylation. So we are using these ones as identifiers now to be able to map the different cell types. So for example, in excitatory neurons, you remember you have your cortical layers. What differentiates those neurons? Why a neuron, besides the morphology and the positioning, what differentiates that neuron from excitatory neuron layer four, for example? Well, we expect with this specific methylation identifiers to be able to separate those different type of neurons in cortex. And who is demethylating? That was the question. And so our RNA sequence was showing that the three tets are expressed, but TET3 is the highest expressor in this region. So we deleted TET2. Well, we collaborated with someone that had made the knockouts for these ones, and we analyzed the brains. And what we found was that both TET2 and TET3, and mostly TET3, are involved in taking out those methyl groups from the specific sites in the brain, in the genome. And so the regions that increase their methylation or do not change during development did not show any effect on the knockouts. However, the TET2 or TET3 were hypermethylated in those regions that normally should have decreased their methylation through development. And so the idea is that the tets will mark this developmentally regulated, differentially methylated regions that will be active in the adult brain. But this requires a further oxidation by the tets. In the TET knockouts, you will not be able to produce this further demethylation. So in summary, we have found widespread and dynamic differentially methylation throughout the lifespan in two species. We found the accumulation of non-Cg methylation specifically in neurons that seems to be involved in gene repression. I didn't show this, but we also found very specific marks for a non-Cg methylation involved in chromosome X in activation. We produced the first whole genome, 5-hydroxy methyl cytosine analysis. And using this, we found that in the fetal brain, you have a poised state of specific cytosines that are going to be demethylated as you develop. And in many of them are in enhancers that become active in the adult brain. And you can find all this data. It's all public. And we have a friendly browser that you can read here, Neumorph at SOC. Just go there. This is only for the mouse. And we have everything in it. Sometimes it gets really slow because it's too much data. But you can browse there. You have the DMRs, the differentially methylated regions, the hypersensitive sites, the RNA-seq data throughout development, CG and non-Cg methylation, and also the 5-hydroxy methyl genome. And also the data is public. And you can access it, the raw data. You can access it at the geosite that is there. And with that, I want to thank all my colleagues in this adventure and people from Manuel Esteler in Barcelona, Vic Hagege in Colombia, and Dr. Ho in Stanford that provided us some samples for the human brain. So this was the 65, well, it was an 80-year-old woman that turned out to be a 65-year-old man. But that was great. Then she provided us with the male, a 53-year-old male. And Jana Rao provided us with the TET knockouts. And Dr. He was the one that performed the TAP-seq to generate the 5-hydroxy methyl cytosine. And of course, the ENCODE group for producing an enormous amount of data that we could use. And thank you very much. Millie, I cannot hear you. If you could just clarify, does the methylation pattern keep changing throughout the lifespan of a human? There is changes, but global like this, it seems to be pretty stable. So it has a dynamic regulation. As it develops, once you reach adulthood, it more or less remains pretty stable overall. Specific sites, though, that is different. So David Swett has clearly shown under fear conditioning, for example, you have very clear changes in methylation in promoter regions, which are going to lead to increased expression of genes or decreased expression of genes. Those are transient, then they come back. So that is one of the reasons why people believe the DNA methyl transphases and the TET proteins are so high in neurons. Normally, in a differentiated cell, they are nonexistent. However, in neurons, you still have both the methylation and the demethylation systems present. So one of the possibilities is because there is a plasticity ongoing that needs to, this type of change is slow. So it's not that you turn on a transcription factor and then rapidly turn it off or a phosphorylation mechanism, which will be minutes of change. This should last for a little longer. But would accumulation of these individual, I guess, cell-level changes, could they, in principle, lead to senescence or long-term change? That's one of the things that they have been looking at. And so there is one isoform of DMT3A, which they found related to cognitive decline in aging. And we want to know when we make the knockout now at the time where non-CG methylation is accumulating, what does that produce at the level of synaptogenesis, at the level of behavior of the animal? So the knockouts that have been made before are either too early, and so they end up killing the animal. You have too much alteration, or it's too late. When the recombination happened, the non-CG accumulation already reached its plateau, and so we still don't know. Is there any possibility of looking at single cell sequencing? Perhaps detect the variation among cells of the same type? Yeah, I have. You've got there. It hasn't... But it's involved with more dynamic processes in the brain, long-term memory or some such. Yeah. Single-cell genome sequencing has not yet been possible. There are attempts. We are doing some, but you cannot say much. It's... Remember, you have one copy of the genome, and it's very little. However, to this, I would add that the variability between the same type of cell in a nocule may drive you nuts if you go single one, and may not tell you much about what that nocule in the brain is doing, because it's working as a group. It's a team. And so it will show you the variability that you can have. So that is... And so that is a little bit the fear that I have. My colleague, Joe Ecker, he is single, single. I say, oh, no. We are going to drown in data and in variants. And so it's a little bit... I will... I want to go first to a group of the same type of neurons in the specific region of the brain and get the means there. What are they doing and what is the difference from the same type of neuron in the same layer in the visual cortex and compare, prefrontal with visual? And there I think we are going to get a little bit more an idea of what is the difference, because if you go single, you may see too much. As I understand it, some of the epigenetic modulations that occur, changes that occur during life, can be passed on to influence the phenotype of the offspring. And I wondered if that's true and what you think the implications might be for inheritance of characteristics in the brain. Yeah, there are studies, and some of them are pretty good, but the best ones are when you see that the change occurs in the germline. So the environment produces changes that change the methylation in the germline and those ones can be passed on. There is the maternal one, which is they have shown that the behavior of the mom in the first week of age of the animal can imprint a behavior in the female offspring that passes two generations. I don't know. It's there, but it might be we need to go and look at the germlines to see if something has been changed there. Lateral transfer of brain function, we would love it.