 All right, well, thank you very much to the organizers for the opportunity to come and talk to you today. What I'm going to tell you about is some work that we've done looking at trying to understand the epigenomic and transcriptional basis, because I think those two things go really hand-in-hand of a very important medical condition that we've been, and my group have been studying for quite some time, which is to say insulin resistance, specifically a complication of obesity and associated with type 2 diabetes. I don't think I need to belabor this point, but I'm sure that most of you are really quite aware of the burden, the medical, psychological, economic burden of type 2 diabetes and obesity in our country. These are just CDC heat maps showing the increase in the prevalence of these conditions in the last several years, and you can see that my maps only go up to 2009. They've been, I didn't put it in the more recent ones, they've had to invoke all new color schemes because they're running out of red. There's only so many deep shades of red you can get. This is not just a major medical and physical burden on people, but it's actually been predicted or been calculated, I should say, to be one of the top three problems economically that the world faces following smoking and armed violence and terrorism is obesity and metabolic diseases. So what my lab studies actually are fat cells. There are many different cells that one could study in obesity, the brain, the liver, the pancreas, et cetera. We're interested in adipocytes. And my lab has already been asking what are the critical transcriptional pathways that underlie these key transitions or differentiations or distinctions in adipose biology? And we focused on, for example, what are the pathways that allow a non-fat cell to become a fat cell during a process called adipogenesis? We're also very interested in how does a fat, a mature fat cell, which is an insulin-sensitive cell, stop listening to insulin? What happens to it to make it stop listening to insulin and a corollary of that is what can we do to make it listen better? And also, how do we turn white fat, which is the energy storing cell type that we're all familiar with, to maybe a brown fat cell type, which many of you may have heard of is a more energy burning and can promote a cell type in which can promote health? So the way that we study these critical questions, actually, the way we approach these things traditionally has been to sort of, you come up with a guess as to what a candidate transcription factor that might be interesting for you to study would be, then you identify its target genes and work on its function in that way. So you might do this because there's a microarray that you did or an RNA-seq experiment, or perhaps someone made a knockout mouse for another reason and the animal is fat or it's skinny and they give us a call, or a study that's been done in lower organisms and worms or flies that affects the larval fat body, let's say. And then we go and look at the orthologous proteins in mammals. There's another way to approach this problem, which is in a more systematic and comprehensive way, which is if you can identify the cis motifs that are important, which identify regulatory elements that are contributing to the regulation of the process that you're studying, whether it's adipogenesis or insulin sensitivity. And if you can then work backward to computationally predict which transcription factors might be involved based on what the regulatory elements that you see changing in the process that you're studying, then you can go and work on the function in that way. And the way that we do that is essentially by enhancer mapping, using clues from epigenomics. So in this case, you could use DNA methylation status as your predictor. You can use non-coding RNA status in many cases. But what we use in my laboratory primarily are covalent histone modifications through a chip-seq and then identify active enhancers or I should say enhancers that change their activity during a process that we're interested in studying. And then we work backward, as I said, or as I just showed to show to identify and predict factors that might be important in that. And I'll show you some examples of that. We published a study a few years ago in which we looked at adipogenesis. We looked at both mouse cells and human cells that were undergoing differentiation in a dish toward the fat cell pathway. We accumulated various histone-marked chip-seq profiles at different time points of differentiation. And those are shown, for example, here. We have the mouse on the top and the human down below. Sorry, is there a pointer or how do I just use the mouse? Here, right here, I'm sorry, right in front of me. Yep, so the mouse up here, the human down below, and you can see here that there's promoter marks and various enhancer marks, for example, that light up in this region. And we can see that as we move through time, this is in a very early undifferentiated state. And as we move through differentiation, we can see the appearance of new alternate promoters. We can see enhancers that disappear and enhancers that appear. And we can make various predictions. And we can identify literally tens of thousands of regulatory elements that appear to be dynamic during this cell differentiation process. And then what we do is we take these enhancer regions that are interesting to us, that are changing. And we compare the DNA that's under those databases of transcription factor motifs. And we can then predict transcription factors that might be important in this process. And when we did this for adipogenesis, we came up with two lists. One of them was based on motifs, excuse me, one of them was based on enhancer peaks that disappeared during differentiation. So they were high enhancer activity during the preadipocyte stage, and then they disappeared during differentiation. And then the opposite pattern, those that were not present in preadipocytes, but appeared during differentiation. And those all predicted different transcription factors that might be involved. And everything I put a check mark next to here is something that was known from the literature at that time to be involved in differentiation. So you can see this provides an important sanity check then that you shouldn't, with this method, be finding factors that you know are important. But we, of course, wanted to identify new biology. And in this particular study, we focused on these two factors here, serum response factor and PLZF, promylocytic leukemia zinc finger protein, two very well studied transcription factors with a huge amount of literature, but no data whatsoever suggesting that they might be important in adipose tissue biology or adipogenesis specifically. And so we were able to show that when we overexpressed them, we enhanced adipogenesis. And what I'm showing here is that when we knocked down these factors, we could enhance adipogen, sorry, when we overexpressed them, we would press adipogenesis. When we knocked them down, we enhanced adipogenesis. So you can see here, the more red the dish is, the more fat has been accumulated, and various marker genes of adipogenesis are increased in these knocked down cells. And so in this way, we have now identified six novel transcription factor pathways, all of which are published, that we have identified that through this approach that we can use enhancer mapping to sort of tell us what might be important for adipogenesis. I want to shift attention now to insulin resistance. And really, our study on insulin resistance was really focused on trying to answer two different questions that have been bedeviling us for some time clinically. I am a clinical diabetologist, and I do take care of these patients. And something that always struck me was the fact that you can see insulin resistance in a lot of different kinds of patients. So we know that as companies, obesity, for example, but also once these insulin resistance in patients that don't have enough fat. This is a gentleman with a form of congenital lipodystrophy, and you can see his characteristic wasted facies. And people like this gentleman are very, very insulin resistant. People with various hormonal conditions, like the little girl up there on the top who has Cushing syndrome, and she has a tumor that causes the overexpression of glucocorticoid in her body. And the woman down below has acromegaly, a pituitary tumor that causes the overexpression of growth hormone. And both those conditions are associated with quite severe and significant insulin resistance. Also, neurodegeneration, often associated with insulin resistance, as well as even some physiological conditions like pregnancy, also highly associated with insulin resistance. And many other different disease states have insulin resistance as sort of a common phenotype as well. You can model this in a dish, by the way. You can put fat cells, grow them up in a dish, as I've shown, and you can pretty much point to anything in the Sigma catalog and dump it on top of it, and it will cause insulin resistance. And that raised the question, to what extent are the molecular pathways shared in these different conditions? And so you can imagine, then, two different scenarios. One scenario is when you have sort of all these different agents are causing insulin resistance in their own unique way, and then another possibility would be that, in fact, they're all working through some common mediator, some nodal pathway that could be integrating the response to all these different agents. So the approach that we're using, then, is we take culture to dipocytes. These are myriad of dipocytes. It's a well-established cell culture model. They have very high fidelity, and they do almost everything that a real adipocyte does. And we treat them with either TNF-alpha or with dexamethasone, and I'll explain why in just a moment. But those are well known to cause insulin resistance, and so you get this rightward and downward shift, then, in the insulin response curve with respect to glucose uptake. And what we're looking at, then, are changes. They can be transcriptomic changes or epigenomic changes or anything you're interested in that are associated with TNF treatment or dexamethasone treatment, and we're going to focus on what's in this middle set, then. What's in this center piece there, and look, then, to see if that can help direct us toward common pathways. OK, so why dex and TNF? Really, the bottom line is that these are well-established pathways. We know that if you give people or animals or cells dexamethasone or TNF, they cause insulin resistance. If you knock those things out of animals, like the enzymes or the receptors for glucocorticoids, if you knock out TNF or its receptors, you improve insulin sensitivity, et cetera. But really, the reason that we did this is because they're so very different. Remember, we're looking for something in the center of that Venn diagram, and so you really want those two circles to be as pulled apart as far as possible so that you have a very small set in the middle. That makes it a much more tractable experiment. So the really reason is that dex is the prototypical anti-inflammatory agent, TNF is the prototypical pro-inflammatory agent, and so we would expect, on first principles, that they would have a very small number of biological responses that would be overlapping. So that's the first question we wanted to get at. How do so many different things cause insulin resistance? The second question was, are there nuclear mechanisms of insulin resistance? There are literally tens of thousands of papers on the mechanisms of insulin resistance in my lab and many, many, many other labs over decades have contributed to those. And they all focus on cytoplasmic pathways, pathways dealing with signal transduction proteins, mitochondrial pathways, oxidative stress, ER stress, et cetera, et cetera. And yet there is this sort of evidence that, in fact, we should be considering other nuclear aspects to insulin sensitivity. For example, we treat insulin sensitivity in the clinic by giving drugs that activate transcription factors, for example. We also know that these cellular models of insulin resistance, such as I've just described, don't work in hours, it takes days to weeks for them to really fully develop, and that suggests this transcriptional or epigenetic component. And we also know that there is this wealth of data that links chromatin state and epigenetics to obesity and complications, and I'm referring there to things like fetal programming and the developmental origin of adult disease hypotheses where we know that, depending upon conditions that you might experience in utero or nutritional conditions, that your grandparents or great grandparents might have experience, that that has a profound influence on your risk of developing obesity and diabetes, for example. So that suggests, again, to us that there really must be some sort of nuclear component here. So the model is very simple. We take cells, these fat cells in a dish, and we treat them either with dexamethasone or with TNF. We're never treating them together. These are just two separate treatments, and we see that we're causing resistance to insulin. So this is just a measure of glucose uptake and response to added insulin, and these cut that down quite significantly. On the right is just a time course of the effect, and that just illustrates what I showed you, that you don't get full insulin resistance for about a week or so. And then at the bottom here, what I'm showing now is that we are not, with our regimen, affecting traditional signaling pathways. We're also not affecting differentiation of the cells. We're not de-differentiating the cells because this is a very important control in our field because obviously if you cause the cells to revert back to a fibroblastic state, of course they're going to be insulin resistant, and that's a trivial response, and we don't really want to study that. So we are keeping our cells healthy fat cells with our treatment regimen. We looked at the transcriptomics under these two different conditions, and you can see that there are about 1,000 genes that are up-regulated when you add dex or when you add TNF, but there's only about a little less than 300 or so that are coordinated and regulated by these two agents. And that, again, illustrates the point I made earlier that you really want to have a relatively small intersecting set to make that much more tractable to study. If we did TNF-alpha and IL-1, both of those will cause insulin resistance, but we'd imagine that they'd have an almost completely overlapping set, and that would be very difficult to parse apart. And then this just shows that if we compare our genes that are changing here, up-regulation genes here, what we're looking at, and we compare them to genes that are changing in garden variety obesity in an animal that you feed a high-fat diet. In fact, what we see is actually relatively little concordance with the dexamethasone treated genes, a little bit better with the TNF-treated genes, but that intersecting set, those 271 genes, actually are quite similar to the genes that we see when we just look at garden variety obesity. So then what we're focusing on next is epigenomic marks, and we did six different marks, I believe. It may be even a little bit more, but the one I'm going to focus on here is histone-3 lysine-27 acetylation, H3K27AC, which has been shown to be a mark of enhancer activity. And that's really what we're deriving most of our data from here, so we'll focus on that. And specifically, I'm going to focus on the H3K27AC marks that change, that are up-regulated, that are within 400 kilobases or so of genes that are also changing in the same direction. So we're looking at up-regulated genes, and now we're looking in the space around that, and looking for up-regulated K27AC peaks. And we can imagine that there are several different patterns here, you can imagine that there are sometimes you'll find a K27A peak that only changes with dexamethasone, or it might only change with TNF, or it might change co-ordinately with both. And of course those are the ones that we're most interested in. If we look then at those peaks that are changing with dexamethasone only, what we see is the top motifs that we recover under those peaks suggest actions of the glucocorticoid receptor, perhaps a very unsurprising result. And in fact, the AR, the angioreceptor and the PR receptor that you see also in that list, they actually use almost identical motifs. That's the angiogen progesterone receptor, and they're very similar steroid receptor motifs. And so what we can see here is that if we see places where dexamethasone is changing the epigenome, it seems to be doing that perhaps through the glucocorticoid receptor. Similarly, if we look at places where TNF by itself is affecting the epigenome, the top hit there is NFKB, and I think that will surprise absolutely nobody. But what's interesting is if we look at this intersecting set, the ones that were the dex and the TNF are causing similar changes. What we see again is that the top motif is the glucocorticoid receptor, and we know by permuting the labels that this is not just because there's a dex set as part of this, we know that this is truly statistically significant for this intersecting set. We also see some other interesting motifs there, and I'm going to focus on two, the GR and the VDR, which is the vitamin D receptor for the rest of this talk. So what we first of all, we're focused here on the glucocorticoid receptor motif, and we know then that dex works through the glucocorticoid receptor. It is a synthetic glucocorticoid, and that causes insulin resistance, but our work here suggests that TNF also activates enhancer marks that seem to have a GR motif, and it raises the possibility then that TNF could be working through the GR. This is a fairly heretical notion, by the way. Certainly it was very heretical to the inflammatory reviewers. I meant that they study inflammation, not that they were inflammatory, although some of them were, but what we can see is that this pro-inflammatory cytokine could be activating then an anti-inflammatory nuclear receptor. So this is just an example of what those tracks might look like here. I'll just to clue you in, dex's treatment is always going to be in blue, the untreated in black, and TNF always in red, just to make things very simple here, and what we're seeing here is also these tracks show the time course, two hours, 24 hours, and six day treatment, and you can see here that there's this cluster of genes, TMM176A and B. If anyone knows what these do, please let me know. There's really almost no literature on these two genes, but what I can tell you is that they are induced by, or I should say there's an enhancer induced by dex and methanol upstream, and that same enhancer seems to be activated by TNF, and there is a GR motif there. We can show by chip PCR that in these cells when we treat with dex and TNF that we can chip the GR preferentially off that spot. We actually went to an entirely different model, cellular model. This is primary fat cells from animals that we differentiate in a dish, not a cell line, and we treat them with dex and TNF. We also cause insulin resistance, and we also see those same ability to chip GR off that exact spot. Then I think most importantly, if we just go right into an obese animal, take the mouse, isolate its adipocytes, and check out where the GR is binding. In fact, the GR is binding right at that spot as well. Not to belabor the cell biology, but I think this is very interesting. I think it points to the sorts of biological insights one can get from these epigenomic data. We know that the GR normally resides in the cytoplasm. What happens after dex binds to is it moves to the nucleus, so it translocates in response to ligand, and you can see that here that in response to dex, you see an increase in the nuclear amount of GR and a depletion in the cytosol of the GR. This has been shown by many others as well, and what we see here with TNF is that we also can cause nuclear translocation of the GR. This is without adding any exogenous glucocorticoid. We see, in fact, though, that we're not depleting it in the cytosol. We seem to be increasing it in the cytosol as well, suggesting that one of the effects of the GR levels in the cell, and we're looking at this much more closely now in a more rigorous cell biological way. We did chip seek of GR in response to dex and in response to TNF. We can see that TNF causes about tenfold less translocation or tenfold less, I should say, binding of tenfold less spots in the genome than does dex activation of the GR. And so what we can see, though, is that those sites largely are a subset of the GR sites, and we've been looking now and trying to identify what makes those sites special. We think we've identified a co-occurring motif that directs the GR once it's been activated by TNF to those spots. And then this, for us, is really where the rubber meets the road. NR3C1 is the GR gene. We've knocked it down here with four different hairpins. And what we can show is that we can rescue completely dex-induced insulin resistance. So if you get rid of the GR, dex can't work anymore. That's not surprising. But what is surprising here, or what is very hopeful for us is that when we knock down the GR, we get rid of the full ability of TNF to induce insulin resistance. So we still have about, TNF can still do about half of its thing, but it loses about half its potency. And so that does suggest to us, then, that TNF is working, in fact, directly through the GR. And we have a variety of cell biological and biochemical studies going on, phosphor proteomic analysis of the GR and response to TNF. We've identified novel phosphorylation sites, and I won't go into all that, but just to say that all of that is an epigenomic analysis that we did at the beginning to suggest what has been a really heretical hypothesis about activation of an anti-inflammatory factor by a pro-inflammatory cytokine. Another factor that popped up on our list was the vitamin D receptor. This is very interesting to us. There are connotations in the literature that vitamin D might be involved in insulin sensitivity, but in the opposite way. It's something we can discuss later if you want. But the bottom line is that we were wondering that could the VDR be involved here? And in fact, what we can show is that there's a large gene of interest, and here's a, you can see a peak here, K2070C peak that's induced by both dex and by TNF, and it has a VDR motif, and we can chip VDR off that location. I'm not showing it here, but in this other cell model, also we see VDR in the same location, and in obese animals, VDR is also chippable from that location as well. And again, if we over-expest the vitamin D receptor, we cause insulin resistance, and if we knock down the vitamin D receptor, we can prevent the full onset of insulin resistance, and we restore a significant proportion of insulin sensitivity in response to dex and TNF. Interestingly, vitamin D receptor is elevated in the adipose tissue of obese animals, different genetically obese models, high-fat-fed animals. It's also elevated in these cells when we add dex and TNF. VDR is itself one of those 271 genes, and we can also show that it's regulated by the GR, so TNF and dexamethasone both turn on enhancers that regulate the vitamin D receptor. You can see they're different enhancers. The dex is activating through here, TNF through here, but they both have GR motifs. We can chip GR off both those sites in response to dex and TNF, and we can see binding at this site in the animals that are obese. And so what we see then are a network we're beginning to develop about how the epigenomics really can lead us to novel hypotheses about transcription factors and how they may be interrelated. In the last few minutes, what I want to talk about is new data that we've been developing. Everything I just shown you has just recently been published in Nature Cell Biology, but what I want to talk about in the final moments are new data about humans and what we're doing here. So we've tried to take the same paradigm and apply it to humans, if you will. This involves collecting adipose tissue. We are fortunate in that adipose tissue is the one tissue that people are willing to give you an awful lot of. So if we were doing neurobiology, it might be a little more difficult to convince our patients, but we do get between one and 10 kilos of fat at a time from patients. That involves a fair amount of processing. We've had to invest in industrial sausage grinders. I'm not kidding to get through this stuff very quickly. And we can isolate adipocytes. As you know, any tissue in adipose is no exception. It's complex. It's made up of very different cell types. So another advantage that we have built in to the adipose tissue is that mature adipocytes float when dissociated from the rest of their cellular matrix. And so we can use this sort of low-speed centrifugation method after collagen is treated to isolate the mature adipocytes away from the other cells in this heterogeneous tissue. And then we can use the pure adipocyte. Then we isolate the nuclei, cross-link that, and we can then use that for chip-seq. And what we get are these very beautiful profiles. I'm showing one of our profiles up here on the top. And you can see ENCODE data for adipose tissue on the purple line right below it. And one thing you can see is you can see these sort of macrophage markers in the ENCODE data, because, of course, the ENCODE data is a whole tissue and it includes a huge amount. 50% of the cells in a fat pad are not adipocytes and they're made up of immune cells, endothelial cells, and other things. And you can see here that we see the ubiquitous markers. We get a much more robust representation of the adipose tissue marker, the ones that truly belong in the adipocyte, and we basically lose the macrophage markers as well. So we're very pleased that we also lose the pre-adipocyte markers. So we're very pleased with sort of the quality of our data. What we can see is that we can begin to and for a large number of novel biology, amount of novel biology here, we're seeing novel transcripts, alternative start sites, things that are popping up in gene deserts that seem to be important in adipocytes. So what we're going to have very shortly now from literally we've now done dozens and dozens of patients is a much more detailed epigenomic map of what those adipocytes look like. But specifically what we have is now we've collected patients across a spectrum of insulin sensitivity. So we know the patients in the top and the bottom quartiles of insulin action based on a technique that we use called HOMA IR. And what you can see then is that we can already in this sort of small sample that was done a little while ago, you can see that we're already identifying enhancers, for example right here, that are relatively seem to be highly active in insulin sensitive patients, but much less active in insulin resistant patients. There are some opposite ones as well where there seem to be more activity in insulin resistant versus insulin sensitive patients. And we're seeing literally hundreds of such differentially regulated peaks now and many more patients than we have here. The ones I've shown here on the left, these represent biology that is pretty well known, adiponectin, glute 4, other adipocyte genes that we know are involved in insulin action. What we also find are as differentially regulated peaks in genes that some of which have absolutely no known biology in any system, some of them have no known biology in adipocytes, and so we're really trying to put this together now. And we have, I should say in the last week, hot off the presses, we think have identified novel factors that might be involved in insulin resistance in human disease. So to summarize what I've shown you then, DEX and TNF can cause discrete changes in the epigenome of 3T3L1 cells, these are cultured adipocytes and those say with insulin resistance, we can use motif finding in these differentially regulated regions to identify novel pathways that allows us to do a lot of cell biology. Some of what we found is that TNF causes insulin resistance in part through an independent, like an independent activation of the GR, also the vitamin D receptor seems to be important. In data that it didn't show you, we've actually identified some of the downstream target genes of these factors. So we've identified now at least four genes that went overexpressed in fat, seem to cause insulin resistance. DEX and TNF and we have our human studies ongoing which we are actively recruiting for and working on and which appear fingers crossed to be yielding some very interesting data. So with that I want to thank the people who did the work. Most of the work I showed was done by really two talented postdocs, Sona Kang and Linus Sai and Yiming Zhu was our computational biologist. Other people in the lab contributed. We had help from our collaborators and friends at other institutions including the Broad Institute where I'm also a faculty member and I think working from the roadmap was highly elemental in pursuing this and thank you very much for your attention. Great work. A couple of questions. When you looked at your K27AC chip under two conditions, you only had about 50 sites of overlap. When you did the GR chip under the two conditions, you had over 200 sites of overlap. You think there's an issue with a K27 acetylation chip, it's not like sufficient, we need to look at other marks or you think there's a bunch of GR sites that are silent even though they're recruited. Great question. There were about 200 in some odd sites that were together there. Some sites will get recruitment and not change their K27AC very much. Maybe because they're changing trading factors, maybe GR is going on and something else is going off and the overall activity and enhancer as measured by K27AC is not changing. That's one possibility. These are blunt tools. I think that's also a part of it as well. We don't tend to worry about the stuff that is trying to make use of the stuff that we do see and that's a little bit of a cop-out answer but that is how we deal with that. I agree. The second question. How are you taking the genetic heterogeneity into account in the second part of your study, the human focused samples? We are doing that right now. We are integrating the human genetic variants now. From the samples, from the individuals you're getting, are you assessing what's their identity? Do you actually be able to ascribe the one or the other if indeed your patient or your individual is? In the early samples, we were not actually consented to get genotype data from those so we've been doing a little bit with imputation but with that. All the later samples, we have changed our consent form and so we're getting to get that information and we're going to just snip genotype everybody and just figure out what to get that data. Naive question. Are there adipocyte stem cells? Yes. Do you separate them and analyze them? It's a very contentious area of research. What exactly is an adipocyte stem cell? They do exist and in our human studies they're not in the cultured cell studies that the pre-dipocytes that are using the cultured studies are from a later stage of differentiation. They're not stemmy. They sort of have lost all multi-potency and can only become a part of the human studies that gets fractionated out. Remember we collaginase them when we float off the matured dipocytes and all the stuff that goes down to the bottom which includes immune cells, stem cells, pre-dipocytes, endothelial cells, neurons, anything else that's in there is in that. So that is a separate area of investigation and flowing out the specific stem cells is something we save those fractions and we'd like to do that but the field I would say is in disarray about adipose stem cells so we're not sure what markers to use to flow out the actual stem cells.