 Thanks, John. It's really a great pleasure to be here and to hang out with folks from the ENCODE meeting and for me to learn more about ENCODE and how to use the vast resources. Although I would say that I've been on the sidelines watching ENCODE for many years and have benefited greatly from some of the datasets. So what I'd like to do this morning is, as John said, to tell you about our work on the selection and function of signal dependent enhancers. This is a topic that you've already heard about to some extent from Evan Rosen and others yesterday. And to focus on our work in which we study the macrophage. And so just as an introduction, we all know that different cell types arise from differential transcription of the same genome and over the past decade, the advent of massively parallel sequencing based technologies has really given us unprecedented tools to examine the mechanisms that underlie the development of each cell type from its progenitor cell. In addition, once a cell attains its specific identity, that cell has to be able to respond to internal and external signals in order to carry out its physiologic roles. And what I'm illustrating here is the ability of two different cell types to respond to the same signal in a very dramatic fashion. And here I'm illustrating macrophages and B cells. So macrophages are an immune cell that's involved in innate immunity. Among other things, they have the ability to recognize and kill bacteria. B cells are a cell within the adaptive immune system. They're specialized for making antibodies. Both of these cell types will respond to bacterial lipopolysaccharide. That's a component of gram-negative bacteria. That activates tolite receptor 4. And in each cell type, it induces a very strong transcriptional response. But you'll see from this illustration that despite the fact that both cells will induce hundreds of genes, this program of induction is largely different. With macrophages and B cells exhibiting quite distinct transcriptional programs. And if you look at the functional annotations that are associated with the induced genes, you'll see that they're quite different. So then in macrophages, the functional annotations are associated with immune response and chemotaxis. And in the B cell, these responses are associated with the processes that are required to make and secrete antibodies. And then the third point, which I'll sort of pick up on some of the things that Nancy Cox spoke about yesterday, is that signal dependent responses of the same cell type can vary among individuals. And here what I'm showing you is gene expression data taken from endothelial cells that were provided by about 150 hearts for heart transplantation. So these are aortic endothelial cells. These are critical cells for blood vessel homeostasis. And they're important cells in the development of atherosclerosis. And what is shown here is a chart in which basal expression levels for FGD6 are organized according to basal expression in about 120 samples. And then in addition to this basal expression, these cells have been treated with a substance called ox PAPC. It's an oxidized phospholipid. It's thought that this is a very important molecule in driving the pathogenesis of atherosclerosis. And you can see that each individual has a variable response to the exposure of this treatment. This is shown in the open circles, where some individuals respond quite dramatically, and other individuals respond either not at all or actually negatively. And this response is actually linked to a genetic variation. And in fact, dependent on the genotype, this represents a gene by environment quantitative EQTL. Now, these variants are in linkage equilibrium with 20 to 30 other variants. So we don't know from these experiments what the actual variant is that accounts for these changes in gene expression. And of course, this is one of the things that we would like to use genomic methods to understand. And collectively, these observations are consistent with the observation that most GWAS hits that are associated with SNPs reside in non-coding regions of the genome. And so this implies that at least some of the natural genetic variation that we observe that's associated with these differences in traits has to do with the impact on transcription. Now, the cis-regulatory elements that regulate transcription include promoters, which are the obligatory start sites for messenger RNAs and enhancers. We have known for quite some time that, well, promoters are essential for the initiation of messenger RNAs. They are not sufficient to confer the levels of expression or the tissue different responses that are required for physiologic levels of gene expression. And in order to attain appropriate developmental and homeostatic programs of gene expression, the additional functions of enhancers are required, and these operate in a tissue-specific and signal-dependent manner. And so just to review a few of the key features of enhancers, this cartoon here is meant to illustrate a functional unit of enhancer. Enhancer, these would be nucleosomes. They are remodeled to generate a nucleosome-free region. These sequences act at a distance for promoters to enhance gene expression. Their function is determined by the sequence-specific transcription factors that are bound. They exhibit a distinct epigenetic signature. We've heard a lot about this. For the purposes of this presentation, I'm just going to focus on the modifications of histone H3 lysine 4. Monomethylation is discovered by Bing Ren. In excess of trimethylation is a signature of enhancers. And enhancers that are active tend to be acetylated. And one of the acetylation marks associated with active enhancers is H3K27. Now, the work of the ENCODE consortium has really dramatically advanced our understanding of how these enhancers are selected and through the annotation of hundreds of cells using a variety of combinations of histone marks. It is suggested that there may be on the order of a million of these types of functional elements within the genome. So for me, this was really a staggering revelation. Probably, to me, one of the most important contributions of the ENCODE consortium was to tell us that there is this vast repertoire of potential regulatory information in the genome. And clearly, the number of regulatory elements greatly exceed the number of genes. And this is likely essential for the driving the complex programs of gene expression that we see in different cell types. Now, what does this look like? This is a genome browser shot that I cobbled together from data from the ENCODE consortium. What I've done here is to actually combine tracks for H3K4 trimethylation, which is present at promoters with H3K4 monomethylation and enhancers. And to do so for three different cell types. So what you're looking at from top to bottom are these composite tracks for macrophages, T cells, and smooth muscle cells. And this browser track happens to be located in the vicinity of the C-phosgene. So FOS is a member of the AP1 family of transcription factors. It's very widely expressed, and it plays different roles in different cell types. And its responses to internal and external signals differ in a cell-type dependent manner. And while the promoter of FOS is marked by histone modifications in all cell types at the promoter, what you'll observe looking upstream and downstream of the transcriptional starch site is that the genomic landscape is very cell-type specific. So for example, here in smooth muscle cells, we see histone modifications associated with enhancers that are smooth muscle cell specific, whereas here we see a modification that's specific to T cells, and here a modification that observed both in T cells and macrophages. So we can imagine that in smooth muscle cells, this region of the genome may be important in driving the expression of FOS in smooth muscle cell specific contexts. Okay, so now what I'd like to do is transition to our laboratory's work on macrophages and what we've learned about how macrophages select their enhancers. I've already briefly introduced this cell. This is an ancient cell. It precedes its present in organisms that lack an adaptive immune system. So this is an innate immune cell that you can find, for example, in Drosophila. Among their functions, they play essential roles in the response to infection and injury. They can recognize pathogens through pattern recognition receptors and initiate a process that we refer to as innate immunity, and that in turn plays important roles in instructing the adaptive immune system to combat infection. Now interestingly, this is something that the macrophage world is learning a lot about. These cells reside in all of the tissues of your body, and within each tissue they play these roles as innate immune cells, but in addition to that, they contribute to the homeostasis of that particular tissue. So for example, in your brain, you have a type of macrophage called a microglia cell. That cell within the brain is playing important roles in secreting trophic factors that are important for neurons. The cell prunes synapses. This is important during development, and it is responsible for eating protein aggregates, such as amyloid beta. And these types of tissue-specific homeostatic functions are found for virtually all of the macrophage populations that have been studied. Now, finally, one of our motivations for studying these cells is that we also know that macrophages play key roles in numerous diseases. So for example, if you take a macrophage out of the mouse, the mouse no longer has the ability to develop atherosclerosis, and we know that in humans, macrophages contribute to virtually all phases of the development of this disease. Macrophages are also involved in the generation of insulin resistance and type 2 diabetes. I've alluded to roles of microglia in the brain as being homeostatic, but there now is substantial evidence that inappropriate activation of macrophages contributes to a broad spectrum of neurodegenerative diseases, such as Alzheimer's disease. And finally, virtually all tumors contain macrophages. The content of tumors macrophages is often a predictor of disease outcome. And within tumors, macrophages contribute to tumor biology and tumor metastasis. And so we're very interested in understanding how macrophages select their enhancers, because our thinking is that the selection of the enhancer repertoire is what gives the macrophage its phenotype, and that enables it to carry out its homeostatic roles in health and its pathogenic roles in disease. Now, our real entry point into this began several years ago with the work of Sven Heinz and Chris Benner. Sven was very interested in a transcription factor called PU.1, which is required for the development of macrophages, B cells, and granulocytes. And we knew that from gene deletion experiments. And from these types of experiments and other sorts of studies, we also knew at the time that in addition to PU.1, the development of macrophages required CBPs and AP1 transcription factors, and the development of B cells required E2A, EBF, and OCT transcription factors. Sven was very interested in the question of how a single transcription factor like PU.1 could be involved in two different differentiation programs. In essence, the same transcription factor was looking at the same DNA template and yet having different functions in these two cell types. And so what he did was to perform a chip sequencing experiment for PU.1, and he identified not only common sites of PU.1 binding in these two cell types, but he also identified binding sites of PU.1 that were either macrophage specific or B cell specific. And one of the immediate outcomes of these experiments that resulted from the motif analysis that Chris Benner made available was that even at binding sites for PU.1 that were very macrophage specific on the one hand or B cell specific on the other hand, the motif, the binding motif that was observed for PU.1 was exactly the same. So the cell specific binding of PU.1 had nothing to do with the sequence that PU.1 recognized. However, in the vicinity of the macrophage specific PU.1 binding sites, Chris found that motifs for AP1 and CBP transcription factors were highly enriched. And of course this was very exciting to us because we knew from prior genetic studies that AP1 and CBP factors were required for macrophage differentiation. And conversely on the B cell side, what was found in the vicinity of PU.1 binding sites were the motifs for other lineage determining factors for B cells. Now this was not just at a subset of enhancers within the macrophage. By combining chip sequencing for HVK4 monomethyl, the enhancer mark, and chip seek data for PU.1 and CBP, we made what I thought at the time was a very surprising observation and that's illustrated here on the right. So the PI represents the total world of enhancers within the macrophage. And what we observed was that PU.1 and CBP either alone or together were residing at about two-thirds of the enhancers. So this was really quite remarkable and not only in the vicinity of genes that we think of as being highly specific to macrophages, these enhancers were in the vicinity of many of the genes. The vast majority of genes are actually expressed in macrophages. And one interesting point that fell out of this analysis was that when Chris Benner simply analyzed the motifs that were enriched in macrophages for H3K4 monomethyl, so looking at the macrophage enhancers, three motifs came out to a much higher degree of enrichment than anything else. And these were the motifs for PU.1, AP.1, and CBP. And this suggested to us that one could actually discover the motifs for the main lineage determining factors for a cell simply by looking at what was enriched in the enhancer landscape of that cell type. And Chris went on to actually use in code data to look at what was in the enhancer-like elements of many cell types. Here I'm illustrating B cells, ES cells, and liver. And in each case, the motif analysis of the enhancer-like regions returned either known lineage determining factors or what we would then speculate would be putative lineage determining factors for these cell types. We were particularly excited when we saw this outcome for embryonic stem cells because this motif analysis revealed three of the four factors that Yamanaka had found were necessary to convert fiberblasts into IPS cells. The fourth factor here is ESRR beta. This was not in Yamanaka's original mix, but the top four factors on this list actually are sufficient to convert IPS cells to convert fiberblasts to IPS cells. And we've gone on to do this type of analysis for a number of cell types and in collaboration with David Brenner's group, we've identified lineage determining factors for hepatic stellate cells. And we've also done this for kidney cells. Sorry, I couldn't get that one out. So Chris and Spen, working with Nathan Spen, performed a number of gain and loss of function experiments looking at the mechanisms that enabled P1 and CBP to bind to enhancer-like regions in the macrophage. And what they found is summarized here, and that is when regions of the genome contain binding sites for P1 and CBP that were within about 100 base pairs of each other, these were the genomic locations that were likely to be bound by P1 in the macrophage. Whereas when P1 binding sites were nearby, binding sites for B cell lineage determining factors, such as oct, then these regions of the genome would be occupied by P1, in this case in a B cell specific manner. And what we found was that these interactions and what we really studied were the places in the genome where P1 and CBP were together. These places in the genome required what we called collaborative binding. So if we took away P1, for example, in the macrophage, then at these locations CBP wouldn't bind. And if we took away CBP, P1 wouldn't bind. So we think of these as pioneering factors, but they can't find their genomic sites by themselves. So we think of this as kind of a Lewis and Clark tag team pioneering mechanism where they both have to be together. And this presence, probably at high concentrations, allows them to compete with nucleosomes to establish a nucleosome free region. Now these types of elements can be observed at tens of thousands of locations in the macrophage genome. And we know that some of these are active, but others are not. And so we became very interested in how an open region of chromatin could be transitioned from inactive to active. And we used the Toll-like receptor force signaling pathway that Ari told you about as an important type of signal to understand this better. So TLR force signaling leads to the activation of a number of latent transcription factors, and a very important target is NF-Kappa B. And what we realized by performing chip sequencing for NF-Kappa B before and after stimulation is that upon stimulation and entry into the nucleus, the binding events for NF-Kappa B were very cell-type specific. And in fact they mostly, and when I say mostly, I'm saying about between 80 and 90 percent of the binding of NF-Kappa B occurred at regions in the genome that were previously established by P1 and or CBP. And so that was the case in the macrophage and in the B cell by P1 and a lineage determining factor for B cells, such as oct. And so what this meant is was that the lineage determining factors that set up open regions of chromatin were basically instructing the signal dependent factor where to go upon activation. And this of course then determined the downstream target genes that would become activated. So we think that this is a major explanation for what I showed you at the beginning, which is why macrophages and B cells responding to the same signal have such a dramatic difference in their transcriptional output. And to give you just an example of this, this is a browser image in the vicinity of the prostaglandin e-synthase gene. This is a gene that is expressed at very low levels in a resting macrophage but is induced by orders of magnitude in response to the stimulation of the TLR4 signaling pathway. Here we're using a very pure form of LPS called KLA. And what you observe in this illustration here is that in the absence of treatment you can observe binding of P1 and CBP upstream of the prostaglandin e-synthase gene. This is associated with histone H3 lysine 4 dimethylation. But under these conditions the gene is off. Now when the cells are activated and NFKB enters the nucleus, you can see that it is binding at very precise locations that line up with the pre-existing binding sites for P1 and CBP. So we hypothesize that the binding of NFKB to these regions of the genome that are pre-existing now arms this set of enhancers and allows them to communicate to the prostaglandin e-synthase promoter and massively up-regulate gene expression. So from these types of studies we developed what we call a collaborative hierarchical model for enhancer selection and activation in macrophages and B cells. The collaborative part of the model is on the left and this is where self-fate determining factors which are expressed in cell restricted combinations collaborate with each other and additional factors to bind DNA and initiate a nucleosome remodeling. And these are factors that establish cellular identity and they have reprogramming potential. And since that number of factors at least in the macrophages is really quite small it helps us I think understand how one can actually reprogram a cell from one type to another with a relatively simple set of transcription factors. Now these self-fate determining factors we think set the stage for what we refer to as cell state determining factors so these would be factors like NFKB. These factors are broadly expressed they as I've shown you for NFKB primarily localized to pre-existing or primed enhancers and we observe that type of pattern with many different signal dependent transcription factors such as nuclear hormone receptors and these are the factors that confer responsiveness to internal and external signals. Now at this stage we wanted to test the model genetically and one thought that Casey Romanoski in the laboratory had was that we could do that by actually exploiting natural genetic variation and the idea was to use the millions of SNPs that are provided by differences between inbred strains of mice as a genome-wide mutagenesis strategy. And so we did that by performing chip sequencing for P1 and other factors in macrophages that were derived from two strains of mice C57 black 6 and valve C. Valve C has about four million SNPs compared to black 6 and the idea was to test what the impact would be of mutations that occurred within the binding site for P1 or CBP or NFKB and one of these signal dependent enhancers. And what we found was that mutations in P1 motifs not only abolished the binding of P1 but with that occurred it co-ordinately abolished the binding of CBP and conversely mutations in CBP motifs reduced the nearby binding of P1. So this was very consistent with our collaborative model. In order for one to be there, the other had to be there. Mutations in Kappa B motifs rarely reduced the binding of P1 or CBP whereas mutations in P1 or CBP frequently reduced the binding of NFKB. So that was the hierarchical part of the model. And if we looked at all of the places in these two strains of macrophages where NFKB and here we followed NFKB with a chip for P65, if you looked at all the places where it differed in one strain versus the other, only 9% of those differences could be explained by mutations in the NFKB binding site itself. And we actually observed many more frequent events that caused a loss of NFKB binding in one strain or the other as being due to mutations in one of the three lineage determining factors. And collectively if we add all this up we can explain about 35% of the NFKB binding by mutations in the lineage determining factors. Now this plus this is about 43% or 44% so we still have a lot to learn about what determines strain specific NFKB binding but this I think takes a great distance from where we were previously. And so this supported this collaborative and hierarchical model but I would also say that this model is vastly oversimplified and it has poor predictive value. What we ultimately like to get to is to be able to look at the genomic sequence and with the combination of the sequence in front of us and the knowledge of the transcription factors that are expressed in the cell to actually be able to predict what the enhancer landscape is going to look like and we are far from being able to do that. This model also fails to explain how new enhancers are selected so what I've told you about is really only relevant for a cell that has achieved its identity and presumably is living in a tissue and is looking around and responding to signals and this really does not account for how a cell gets from cell type A to cell type B during development. It also fails to account for the functions of the majority of transcription factors that are expressed in macrophages and its relevance to in vivo populations is unclear. And so very recently we've begun to try and look at how enhancers are selected and function within tissues and our first shot at this was a comparison of tissue resident macrophage subsets. One of the subsets we looked at were the macrophages that live in the brain and we also looked at two separate populations of macrophages that live within the perineal cavity. These are the large and small perineal macrophages and these have important roles in controlling the immune system in the gut. And we started simply by looking at the transcriptomes of these macrophage populations and quite remarkably maybe not so remarkably. I don't know, I was surprised by this. The transcriptomes of the macrophage that live in the perineal cavity and the macrophages that live in the brain are quite different. So there are nearly a thousand genes in each direction that are more than 16 fold differentially expressed in these two populations of macrophages. Whereas the macrophages that live together, the small and the large perineal macrophages that are in the perineal cavity, their gene expression patterns are much more similar, the small perineal cavities overexpress about 200 genes. And these differences in gene expression are actually mirrored by their enhancer landscapes. So what you're looking at here are scatter plots for H3K4 dimethyl which is marking both enhancers and promoters. The promoters are color coded in blue. And what you can see is that the perineal macrophages in the microglia have a much greater distribution of enhancer strength. These would be perineal macrophage specific enhancers or enhancer-like elements. These would be microglia enhancer-like elements. And about 15 to 20 percent of the enhancers are specific to each cell type. Whereas in the perineal cavity, the small perineal cavity, small perineal macrophages have a set of enhancers corresponding to those differentially expressed genes that I just showed you. So one of our big questions is how do you get these different enhancers? And to go back to our playbook for macrophage and B cell specific enhancers, we use PUE.1 as a way of telling us where the cell specific enhancers were and what the collaborating transcription factors are. And we know from doing a chip sequencing experiment that PUE1 is residing at many of the microglia specific enhancers and many of the large perineal macrophage specific enhancers. So we're going to basically take the same approach and we're going to ask what are the factors that PUE1 has to collaborate with in order to set up a microglia specific enhancer or a resident perineal specific enhancer. And now we're going to build on this concept that I told you about of using natural genetic variation. In the first instance of what I told you about, we used natural genetic variation basically as a way of testing the model. But now what we're going to do is we're going to actually try to use it as a way of discovering what the collaborative factors are. And we're going to extend the amount of genetic variation that we used from a single mouse to many mice. And we're going to take advantage of a lot more genetic variation here using up to 40 million SNPs that are provided by SPRET. And the basic idea of what we're going to try and do is shown here. As I've told you, PUE1 cannot effectively find its genomic binding sites by itself. It has to collaborate with other factors. And so we will assume that every place in the genome that PUE1 binds is a place where there are other factors, let's call them transcription factor X. And at most genomic locations, this collaborative interaction and this DNA target sequence will be the same. So we'll see equivalent binding of PUE1 in Strain 1 and Strain 2. But at places where variation affects the template and affects the binding of PUE1, then we will see strain specific binding of PUE1. Now, most of the genetic variation that results in very dramatic differences in the binding is due to mutations in the binding site for PUE1 itself. So that's not surprising, but it's also not informative. What we're looking for are places in the genome where there is strain specific binding of PUE1, but there is no mutation in the PUE1 binding site. So at those locations, what we are hypothesizing is that a variant has interrupted or mutated the motif for a collaborative factor. And so now the exercise, and this was developed by Casey Romanoski, the exercise is to really find evidence for this. And the approach that she took was to look at the expression of all of the transcription factors that are expressed in the macrophage. She selected the top 100 most expressed factors. She identified their motif, and then she basically looked for the presence of that motif in the vicinity of PUE1 and asked whether or not that motif had a variant comparing one strain to the next. And so an example of this is shown here. What has been done is to rank order the PUE1 binding sites with respect to their differential expression, in this case between spread and black 6, from the most black 6 binding on the left to the most spread specific binding on the right. The actual relative binding profile is shown here. And then she is color coding PUE1 binding sites as to whether or not there is an ISRE nearby that has a mutation. And what you will appreciate from this picture is that red is concentrated to the right and blue is concentrated to the left. And so what this means is that variants that are in the ISRE tend to be associated with binding strength of nearby PUE1. And this relationship is highly statistically significant. And so from this type of analysis, we would conclude that factors that bind to the ISRE are important collaborative factors for PUE1 in the macrophage. So if one then goes through that exercise for all of the 100 top express transcription factors, then we ultimately identify dozens of motifs where mutations affect the binding, the nearby binding of PUE1. The most significant of these are mutations in the S sites. So these are the sites that I told you are not very interesting. These are just sites that are probably directly affecting the binding of PUE1. But everything from here on over represents a mutation in a motif for a factor that's not PUE1. And we identify, as I said, dozens of motifs that bind factors. The names of the factors are shown below. And we observe motifs that are present in both microglia and large perineal macrophages that seem to be generally important. We observe motifs that are specific for the large perineal macrophages. And we observe motifs that are specific for microglia. And importantly, we observe motifs in more than one strain. So this is like a confirmation GWAS study. So if we find a motif that's important in the spread mouse and then find the same motif again, but for different mutations in Nod, then we're pretty confident that we found a motif that is important. And for most of the motifs that we identified, we've confirmed mutations in both strains. Now, we're of course interested in how these transcription factors actually work and to what extent do they relate to the different environments that the macrophages live in. David Gosling did a very interesting experiment in which he took the cells out of the animal and he put them in a tissue culture dish. And we know the gene expression changes when you take cells out of the body and put them into culture. But one of the interesting points that came out of this analysis was the ability to compare the macrophages from the perineum and the microglia from the brain. We saw hundreds of genes that were differentially regulated in culture. For the most part, we saw loss of gene expression. And the interesting point was that the genes that were lost in the perineal macrophages were the genes that made them most different from microglia and vice versa. The genes that made microglia, most microglia-like, were the genes that were lost compared to the perineal macrophages in culture. So basically the cells lost their distinctive personalities when they were taken out of the environment. And what this is telling us is that the environment is constantly instructing these cells as to what their identities are. And interestingly, I'm not going to show you the data. This change in gene expression was associated with the collapse of a very large fraction of the enhancer landscape. So virtually all of the enhancers that were specific, for example, to microglia were greatly reduced with respect to their epigenetic features. And consistent with this, if we now look at the factors that bind motifs that were identified as being important for P1 binding, we find that about half of the transcription factors that bind these motifs in perineal macrophages were environment-dependent. And so this now begins to give us a lot of insight into how the environment instructs the cell to set up enhancers that are important for that cell type. And that's illustrated in this cartoon here. And basically our idea is that there's a core set of enhancers that are primed that we think are common to all of the tissue macrophage populations. And these enhancers have the ability to respond to various signals, but those signals are dependent on the anatomic context. So for example, an environmental signal that's only present in the perineal cavity will act on the set of enhancers selectively in the perineal macrophage. This set of enhancers will then turn on direct target genes and included in this set of direct target genes are transcription factors that can collaborate with P1 to set up a new set of enhancers. And it's this combined output of direct targets and indirect targets that help build new enhancers that establish the perineal macrophage phenotype. And so we think that this is a process that's going on in all of the tissue resident macrophage populations. Now what we've been studying here are transcription factor combinations that bind to specific regulatory elements in mouse macrophages. We're obviously very interested in how all of this plays out in humans. A very important paper for us came out of Stam's lab last year that I think is to us very important in going further and extending at least conceptually some of the observations that we're making. And that is that if you look at individual DNA bases where transcription factor bind, there's actually very little conservation between mouse and human. As you build that up to footprints, you gain a little bit of conservation. Transcription factor to transcription factor connections, you get a little bit better. Strikingly regulatory networks are very highly conserved. And we think that what we're looking at is really a component of these transcription factor collaborative interactions fall within this general realm of regulatory network architecture. So we think that what we learn in the mouse, we will probably be able to confirm in humans, although I would make the point that there's enough genetic variation among human individuals to do a lot of what I showed you that we did in the strains in mice. So for some take home points, first I would say that knowledge of the enhancer landscape of a cell reveals much about that cell's identity and its regulatory potential. Enhancer landscapes enable prediction of key lineage determining transcription factors and sites of action of signal dependent factors. Transcription factor binding maps will inform analysis of genetic variation. And this last point for us is very important. We think that we can use natural genetic variation to discover regulatory networks that drive cell specific gene expression. Now going forward, we're very interested in trying to take what we've learned to understand and hopefully modify roles of macrophages in human disease. And so going forward, this is taking us into increasingly challenging areas where we can't isolate a million cells, we can isolate maybe 5,000 cells or 1,000 cells. And in that cell population, we want to know what's going on. So we're definitely going to need to improve the methodology to define regulatory networks and specific cell types within complex tissues. Very little of that has been done as of yet. And what we'd like to be able to do is to use these methods to determine the effects of cell-autonomous and non-autonomous disease mechanisms on enhancer selection and function. So to give you the example of the brain, we think that a primary pathological process such as an abnormality in the processing of A-beta, which is probably a primary process of neurons, is leading to the development of an aggregate which is being responded to in an inappropriate way by the macrophage. So that's a non-cell-autonomous microglia response. We'd like to understand what's going on there. And then finally, consideration of regulatory networks as complex phenotypes for therapeutic modulation. If we think we have a treatment for a condition that is really acting on these regulatory networks, we need to actually have the tools to be able to tell us that our treatment is working. And so with that, I'll thank you for your attention. I would just like to acknowledge, as I have already, Casey Romanoski, Sven Heinz and Chris Benner. The in vivo work that I showed you was led by David Goslin, working with Verena Link, a computational biologist. And we had help from Frederick Geisman and Hannah Gardner at King's College. Thank you very much. That's why you make the big bucks. I was wondering, with promoter regions, you find transcription factors, typically 500 to 3,000 base pairs upstream. The binding sites, with enhancers, how far do you typically look around an enhancer to find binding sites? Right. So the enhancers could be hundreds of kilobases away. For the purposes of what I've told you about today, we're defining enhancers as basically being anything that's actually more than 500 base pairs away. And when we look at promoters and we look at the nucleosome free region and the characteristics surrounding the transcriptional starch site, our thinking is that the promoters are a relatively discrete unit. It's about 500 base pairs or so. So anything after that is an enhancer. I haven't told you anything at all about connections between enhancers and promoters, and that is a huge area in and of itself and one that we're very interested in. I would say the rules in the macrophage are very much like they are in other cell types. So you have regions that are close to enhancers that are close by, and most of the information seems to be within a couple of 10, 20, 30 kilobases, but we're open to regulatory elements existing much further away. So I'm not sure that I've made my question clear that not the distance of an enhancer to a promoter, but if you're looking for the transcription factor binding sites that are regulating the enhancer itself, how far around the enhancer do you typically find? What's the distribution of the spacing of the transcription factors within that element? Relating to that enhancer. Related to that enhancer. So enhancers look remarkably like promoters with respect to those sorts of dimensions. So if we do an analysis, a spatial distribution analysis of how far apart things are from P1, for example, if we centered on P1, it's 150, 200 base pairs, or the plus and minus 150 to 200 base pairs is the average. Hi. My question is about what you know about the protein-protein interactions amongst these different players assembling, and specifically as anyone looked at whether mutations in the sites involved in protein-protein interactions would affect the function of these enhancers. Great question. So this actually relates to the question I just answered, which is how far apart are the motifs for P1 and CBP, for example, and their variable. So for most of the enhancers that we look at, there isn't a fixed distance between motif 1 and motif 2 that would make you think that there's a ternary complex where proteins interact and then bind. And so we don't think that ternary complex recognizing motifs is the explanation for most of what we observe. We can't exclude the possibility that proteins interact with each other in some way, and that helps concentrate them in the vicinity of an enhancer. But right now we don't see strong evidence that a particular protein-protein interaction is necessary for this initial selection of the enhancer.