 But let me start you off with what is kind of my passion these days. What you're looking at here is reminiscent to me of a personal experience. I live, for those of you who aren't from Montreal, I live just in a French-Canadian village on the outskirts of Montreal. And there's daycares. If you know Quebec, you know we're big on daycare, right? We send kids to daycare, affordable, etc. And so you walk by a daycare during playtime and you see three-year-olds out, four-year-olds out in the middle of a playground, merrily enjoying themselves as you do as you see in these particular pictures. And in general, what you see across this is a group of three-year-olds just bliss, enjoying themselves across the board. And then you go dark, right? Because you know that looking at these 30-some-odd three-year-olds, that by the time they're waking their way past puberty, etc., that at least 20 to 25% of them will be suffering from a mental disorder. And that's the reality of it, right? And the question is who and what can you do about it? And the answer is not much. First problem, age of onset of psychiatric disorders peaks early. That's true for all common psychiatric disorders. They're peripubital, generally, in their onset. Girls might experience psychosis a little bit later than boys do, but for the most part, you're looking at childhood early adolescence as a peak period in the onset of mental disorders. And again, you can go through just about any particular disorder. Addictions, ADHD, autism, on and on and on. They're all peaking at about that particular point in time, which is telling you that you don't have a lot of time if you're going to bring any kind of prevention to bear. What are the other kind of challenges we face? First of all, the type of risk factors that predict an increased risk of mental disorders are far more prevalent than we assume. This is American data. It's the ASIS, the Adverse Childhood Experience, begun by Vince Filetti and colleagues. And these are obviously a range of experiences in childhood that predict a greater risk. For the most part, if you go through the documentation from the ACE consortium, which is wonderful to look at, is that people with two or more of these ASIS are at increased risk for almost anything you can imagine, ranging from breast cancer to depression and onward. And that that to and onward represents 42% of the population. That's in the Kaiser Health Network in San Francisco, right? So it is not a population enriched by any means for these particular types of experiences. It's about as close to a representative sample in Northern California as you could find. So A, onset is early. B, risk factors are far more prevalent than we'd ever assume. Okay, so what are the primary risk factors that we've identified epidemiologically? Well, maternal mental health, probably perinatal, prenatal, more so than postnatal, but that's another story. Multiple forms of childhood adversity, birth outcomes, socioeconomic status, all of these known virtually universally to predict an increased risk for mental disorders. An interesting cross-cultural comparison from working in Asia. If you work primarily, we work with primarily Chinese population, if you take risk factors for heart disease from Caucasians populations, they map poorly from the most part on to Asian populations, particularly those related to obesity, BMI, et cetera. If you do the same thing for diabetes, they work even more poorly. For example, Han Chinese women have about a 25% rate of gestational diabetes, which is about two and a half times the rates that we would find here in Montreal, and at least twice the rates that you'd find in the United States. So the risk factors for these particular metabolic cardiovascular disorders do not map easily from Caucasian populations on to Asia, but for mental health they do. We have yet to find a single risk factor that predicts or modeled predicting mental health that doesn't seem to work in both across these particulates. It's an interesting portion. The problem with that is those risk factors may be predictive at a more universal level than for other disorders, but they're not very good. So that if you take, for example, the common risk factors for mental disorders at very most, and I'm being generous when I say 10%, most risk factors account for about 5% of the variation in mental health outcomes. So A, they're early in onset, and B, we can't predict them. We are, rightfully, critical of GWAS studies for having expended billions of dollars at this point to explain 1% to 2% of the variance in common disorders. Rare diseases, different matter, but common disorders, forget it, and that's as true for diabetes as it is for mental health. But we should be equally skeptical of E. Environmental measures for mental health don't work, and it's no use kidding ourselves, and the problem with that is that we continue to invest considerable amounts of money identifying individuals on these bases, bringing intervention studies to bear, failing to recognize that most of the kids exposed to these adversities are on no trajectory that predicts a mental health outcome. So that's a huge problem. We're wasting our money, our time, and probably the thing that bothers me the most in all of this, you'll note this is a bit of a rant, is that when you come to do evaluative research, if you do evaluative research on socioeconomic status, is poverty enriched for kids who have problems with school readiness? Of course it is, but 60 to 70% of the kids who grew up in poverty are fine in school readiness. That may still be less than it is obviously in a more economically advantaged population, but in Canada 10% of the kids are below the poverty line, something like that, 90% of them are not. So if 8%, 9% of kids growing up in poverty have problems of school readiness, that's 0.9% of the population. If 3% of the kids in more affluent areas have problems of school readiness, well, you can do the math. That's over 20% of the population. Many, many more times. So we're missing target populations and we're over-including when we do intervention studies, which means our evaluative outcomes are poor, which is why we universally report very weak outcomes when we do well-thought-out, well-intentioned intervention studies. So the problem I find is that we can't really accurately identify high-risk children. So what I'm going to do is to walk you through the rationale for the idea that epigenetic measures may help us overcome this. This is a tenuous hypothesis at best, okay? How much of my earned income would I put on this working? I'm not quite sure, but I'm going to walk you through this because I think there is at least an idea that suggests we should be testing this particular idea. The other component of this, of course, that is terribly frustrating, and this is just an inherent weakness of epidemiology, it's not a critique, is that if we go from a broad-spectrum predictor of outcome to a particular outcome, right, socioeconomic status, is that the pathways that mediate those effects can be very different. And so for example, Martha Farrow and I wrote a paper a number of years ago, and the upshot of that paper was that if you're looking, for example, at language development as an outcome, then the primary mediator of socioeconomic status is the absence of cognitive stimulation within the household. If you're looking at mental health, it's more related to parental care. So the issue then is if you're going to do an intervention to compensate for what you think is a disparity in SES, you need to know what's the pathway in order to bring an effective intervention to bear. That's just another question. And this slide has nothing on it. Okay, so let me tell you where we began this and quite some time ago. Of course we start with epidemiology, right? However, we may critique the knowledge of environmental factors as good accurate predictors of outcomes. There's still our starting point. So essentially we start with this particular formula looking at how these conditions produce such a broad range. And again, the ACE studies are highly informative here in suggesting that the factors that are predicting mental health outcomes are also relevant for metabolic, cardiovascular, oncological, et cetera, which is why there's probably so much comorbidity. So what we're really asking is, can we find a biomarker that would not simply reflect adversity, but reflect the impact of the adversity at the level of the individual child, right? That's really what we want to try to do. So first of all, does it predict and reflect the adversity at the level of the individual? Does it associate with functional status of the child with respect to any particular outcome? Does it predict who's most likely to benefit from an intervention, which is key, of course, and does it reflect the effects of the intervention? And if it does so, you're off to the races. Okay, so we began, as I say with epidemiology, and I'm taking you back to some much older work that we did, but essentially the map that has merged, and you probably heard some of this earlier this morning because Sonia's work was obviously key in some of this, is that these forms of adversity operate by producing individual differences in the way people respond to stressful circumstances and that the biological response to the stressors is then active in promoting the risk for these particular disorders. At least that was the general model, and I think it still holds in many, although direct tests of the model are lacking. So our thought was that one of the ways this could operate is that these forms of adversity would act on brain regions that regulate our stress responses and that the fundamental effect here would be to alter the activity of genes within these brain regions and thus, stably modify, if you will, stress reactivity. At least that was the idea. So the question is, how to test that? So we worked with an animal model. These are Long Evans Norway rats. So they're not mice. They're far more complex than a mouse. Most of you will not. The subtleties of rodentology are lost for good reasons, right? Would you care? But I'll just, if you'll pardon a thing, is that I once spoke to a group of individuals who work with non-human primates, macaques. And I said there's a bigger difference between the rat and the mouse than there is between the non-human primate and the rat. And no one thoroughly disagreed with that, right? So that's how different those species are. But one of the reasons why the rat is fun to use is that, first of all, it's highly social, incredibly social animal. It forms social colonies that are almost indistinguishable of that of non-human primates. It has very subtle forms of aggression, unlike us. And it shows wide variation in behavior and in complex behaviors. And one of them is in maternal behavior. And so maternal behavior in the rat isn't all that complicated, but essentially what mom does, she nestles herself on top of her pups, she nurses them, and in the course of the nursing, she licks her pups, which stimulates all kinds of physiological responses, including the release of growth hormone. It's not a bad idea if you're giving somebody a whole load of nutrients to stimulate the release of growth hormone and make sure that anabolic processes are facilitated throughout this process. So the whole thing makes a lot of sense. The interesting thing is that moms vary considerably in that particular behavior. So you can easily spend, as we do, hours watching mothers lick their pups and find mothers who lick far less often than do others. And the range here is very interesting because these pups grow up just physically, just as well as through the survival rates, physical growth, you know, at a time of puberty or beyond, their body weights are indistinguishable. These guys do tend to develop diabetes, the offspring of these mothers, but that's only later on. But the point is these moms are not neglectful. They're licking their pups. They're just not licking them to this level here. Now, then what we did is exactly what you would do in a developmental epidemiological model is we allow the pups to mature to adulthood, and then we look at a whole range of outcomes, stress responses, neural development across synaptics, connectivity, et cetera. Learning and memory. And by Rose Baggett, for example, who's now a professor in our psychology department here, metabolic outcomes. As I mentioned, the offspring, particularly the female offspring of low-licking mothers are likely to develop diabetes by midlife and reproductive development. Again, in females, female offspring of low-licking mothers go through puberty earlier than do those of high-licking mothers. Anybody who knows the human puberty literature will know that maps on rather interestingly. And then, of course, you do what you are hard-pressed to do in humans. You do the cross-fostering studies, because really critically what you want to do is to link the variation of maternal care to the phenotypic outcomes. And the cross-fostering either in part or completely reverses every one of these particular outcomes. The in part part is actually kind of intriguing. We can get back to that if you'd like. But one of the phenotypes that we've really looked intensively at is the one you heard about this morning, the HPA talk. You'd almost think Lawrence put some thought into the theories and lectures. And so I don't know if you got into this, but this is time for another rant. Is that we often talk about glucocorticoids as being stress hormones, right? They're not, okay? They're not stress hormones. They're hormones that associate with stress, but they have little to do with the immediate consequences of the stressor. That, in fact, is the job of corticotropin releasing factor as well as many others, including some certain cytokines, etc. Vasopressin, again. CRF activates the autonomic nervous system response. It does so by working on midbrain structures. It activates emotional responses to stressors in the amygdala through the ventral hippocampus, etc., the BNST. All of these are coordinated beautifully by CRF, every one. Gastrointestinal responses. Again, CRF is released within the gastrointestinal tract. It produces the disruptions of which we're so fond during periods of anxiety, etc. All of that is produced by CRF. The adrenal glucocorticoids, if anything, counter-regulate the effects of CRF, right? They actually work in the opposite direction. They're actually more, as Robert Sapolsky and Alan Monk pointed out in their brilliant article, they actually have a primary restorative effect on the system, right? Now, they do have effects, as you probably heard this morning, on things like the aftermath of the stressor, learning and memory, etc. No question about that. But in terms of activating the immediate stress response, these guys are nolocontendotes. Point of fact, a good buddy of mine, Stafford Leitman, runs the largest out-of-sonians clinic in Europe. Ask Stafford, how do assins patients, right, who have adrenal insufficiency, how do they manage to deal with stress? Just fine. As long as it's not a massive viral load which can drive them into sepsis, they do fine. They do just fine. So you don't need to massively activate glucocorticoids. It's this guy here that is the primary activator of the immediate consequences that we feel during a period of stress. And evidence of that is that glucocorticoids not only work dynamically to counter-regulate CRF, they work through a longer loop as well. So glucocorticoids act on the pituitary to turn off ACTH and the hypothalamus to turn off CRF, but they also act outside of the hypothalamus to tonically inhibit CRF. And so CRF can be potentiated by stress, right? Repeated exposure to stress can cause constant increases in CRF. Glucocorticoids counteract that. They keep CRF in balance, and so they exert what we refer to as a negative feedback effect. And the brain region, there are probably many regions, but the brain region most closely associated with that particular effect is the hippocampus. So glucocorticoids act the hippocampus to inhibit CRF and the hypothalamus. Now what we know from studies with adult offspring of high and low licking and grooming mothers is that by comparison to the offspring of high licking mothers, animals reared by low licking mothers produce more CRF. And that's a basal effect. So right in the absence of any stressor, these guys are producing more CRF in the hypothalamus. When you stress them, they produce more ACTH and a greater glucocorticoid response. Makes sense. These guys produce less CRF and have a more modest stress response. And this effect is due to differences in the density of glucocorticoid receptors within the hippocampus. So glucocorticoids act on glucocorticoid receptors in the hippocampus to inhibit CRF. The more receptors you have, the greater the response to the steroid, the greater the inhibition on CRF. It's pretty simple, right? These guys produce more receptors and Rose Baggett, Chris Calgi, and others in the lab have shown that if you disrupt this, if you disrupt the effect of the glucocorticoid receptor, you completely lose the difference between these animals. So we know that it is the difference in glucocorticoid receptors in the hippocampus that drives that long-standing effect of maternal care on pituitary adrenal responses to stress. So the question we have is the ultimate developmental question, we think, and that is how is it that an experience, in this case variation in maternal care over the first week of life, can stably alter a phenotypic trait, glucocorticoid receptors in the hippocampus and thus differences in pituitary adrenal activity. And now we get to epigenetics. Okay, just for the sake of this, it has been, for anyone who grew up in biology, it was a long-standing piece of advice that when anything mysterious happened, particularly if it was developmental, it was epigenetic, okay? Because part of the problem we have in epigenetics is that every human being, if there's 40 human beings in this room, they are 40 definitions of epigenetics, right? And none of them are wrong and none of them are right. And so that made this rather convenient. But our suggestion was maternal care alters the epigenetic status of the gene that produces the glucocorticoid receptor, thus resulting in sustained differences in glucocorticoid receptor expression and changes in phenotype, phenotype here being pituitary adrenal activity. So what does this mean? So pardon me if this is dirt simple to you, but the following is just kind of a walkthrough of what we mean by epigenetic mechanisms. First point, epigenetics is not that hard. Epigenetics is simply the biochemical modifications that control gene transcription. That's it. In fact, it took me a long time to realize that I was working in molecular biology because I'd been trained in biochemistry and I didn't really know what molecular biology was and when somebody pointed out, I said, isn't that just biochemistry? They said, yeah, I know, but we call it molecular biology. Then I realized, okay, I'm doing molecular biology and somebody said, are you doing epigenetics? I said, gee, I don't know. This is what I study, the biology of gene transcription. They said, well, you do epigenetics. Cool, I do epigenetics. It's just gene transcription. There's nothing mythical about it, right? It happens very simply. It happens through a whole series of signals, right? They could be nutrients. They could be drugs. They could be hormones. I mean, on the peripheral side, they could be social interactions mediated through peptides, through various neurotransmitter systems, et cetera. And all of these extracellular signals that we associate with environmental perturbations act upon the cell to change gene expression. And they do so pretty simply because what happens is you have a class of proteins known as transcription factors. Transcription factors are lurking around in cells, inert, right? Oh, I'm just here. I'm hanging out. And then what happens is some environmental signal interacts with the receptor and activates that transcription factor. Time to get to work. Get to work for a transcription factor is bind to the DNA, right? Bind to the DNA. And then what you do, you've been activated, phosphorylation. You bind to the DNA and that turns on gene transcription. Certain transcription factors are repressors. Most are activators, right? But work for transcription factor is binding to the DNA and driving gene expression. And you'll note all of that is a consequence of some kind of environmental perturbation. That is extracellular. It could be internal to the organism or external. It doesn't matter, right? The cell doesn't know. It just knows it's time to change the expression of a gene. Now, epigenetics refers to the biochemical processes that control this interaction between the transcription factor and the DNA. And the classic, right? One's been known since the 1950s is DNA methylation, okay? So DNA methylation, which is also, by the way, if you're doing any work in humans, the easiest one to study, which is why it gets studied a lot. So how does that work? Okay, so here's the most complex part of the whole talk. And it's really not that complex. This is what DNA looks like when you do crystallography. So for anybody who did biology back in whenever, if you looked at DNA in an electron microscope, you would have seen what looks like beads on a string, right? Pearl necklace type of thing. Well, the beads are these guys, the nucleosomes. They're 146 or so base pairs wrapped tightly around these histone proteins, right? And they're electrically charged. Positive histones, negative DNA. And the reason why all that's important is that in this default mode, the nucleosome cannot be activated because it's closed. Nothing can bind to the DNA because it's tightly bound to these histone proteins. And under these particular conditions, right, transcription factors cannot bind. If you want a transcription factor to bind, you need to open up the DNA and allow the transcription factors to access their binding sites. And what DNA methylation does for the most part is it blocks that. It blocks the opening of chromatin and therefore inhibits the ability of transcription factors to bind to their receptor site. Okay, so that's part one. Part two, DNA methylation works in this way almost like a dimmer switch, okay? It's not as binary as we might think. DNA methylation basically is working to diminish the probability that transcription will occur at a particular site. And so what happens essentially is that DNA methylation, as you add methylation to these sites, it reduces the firing rate or the transcriptional rate, right? It's a progressive signal. And when you look, for the most part, at genes that are active, they tend to carry little, if any, methylation. Genes that are inactive, silenced within the tissue, tend to be highly methylated. And so it has served to allow us to understand what is one of the great mysteries of biology. How is it... Sorry, wrong sequence. How is it that we are able to produce so many different cell types from a single DNA template, right? Every cell in your body bearing somatic mutations has exactly the same DNA, and yet no one's going to mistake a muscle cell, a blood cell, a brain cell, a liver cell, et cetera. There's tremendous diversity in these cell types. Over 220 different cell types in your body, all from the same DNA. How do you produce that variation? You produce it epigenetically. And so what happens then is that if you look at brain cells and blood cells, they will both carry methylation, but the methylation that's on genes, specific genes in the brain, will differ from the methylation part of those in the blood. So there's an overlap, but it's, as I say, it's the point where you can create specificity in patterns of gene expression by having a gene that's methylated, if you will, in brain, but not in blood or vice versa. Make sense? Simple. You're just producing variation in gene expression, and you're doing it by altering the methylation that occurs at those genes across the tissue. So when you look at variation in DNA methylation, differences between tissue are incredibly important. That's what's determining which genes are expressed in this tissue versus that tissue. Okay, the other thing to keep in mind, since we're going to ultimately get into, and you've probably heard quite a bit, about the degree to which environmental circumstances can produce variation in DNA methylation. It's important to know the limitations in that particular field of study, right? Because about 70% of all the epigenetic marks are invariant across individuals. Your liver, yours and yours are going to have about 70% the identical methylation, because each of you has a liver, right? The thing is, there's another portion that does vary, and that's why your liver, yours and yours, may vary in their function. But do keep in mind about 70% of the methylome, as we call it, the methylation across the genome is invariant across individuals. We're working with a smaller portion of that signal, and the reason why that's so important is that if you start to lose methylation in these, then your cells undergo what's called de-differentiation, and that's tumorogenic, right? So that's the type of change in methylation that we don't want. But the 30% can still be massive. It can help us understand, for example, what has been one of the great issues in twin studies, and that is how there is so much discordance occurring between individuals who are virtually identical in their genome. So monosagotic twins, essentially the same DNA, not surprising mom doesn't have a lot of trouble telling them apart, right? And so you can produce this discordancy, and essentially people like Art Patronus have talked about this, is that you start off with, you know, the same DNA, and likely very much the same epigenetic, and then as development progresses there's variation that's produced as a function of all kinds of differences in experience, in utero, postnatal, etc., etc., and the variation then produces what we see as discordancy. And there is, there's not as much data on this as you might surmise, how in the two studies with twins do suggest that variation in DNA methylation gets greater and greater as they age, which is consistent with this idea. So, okay, so we go to the basic transcription factor model. This is activated by environmental signals mediated by hormones, transmitters, lipids, etc. These environmental signals activate transcription factors through phosphorylation. That process is inhibited, right, when chromatin is in this closed configuration and it needs to be oops, I just broke your microphone. There we are, I think. Let me know if that falls off, okay, is in the closed configuration. There is a very common signalling pathway by which you overcome the closed configuration of chromatin. So, when a transcription factor binds to its DNA site, it often brings along a partner. That partner is an enzyme known as histone acetyltransferase. What histone acetyltransferase does is exactly what its name implies. It acetylates various tails on these histone proteins. When it acetylates the tails, they lose some part of their charge, their positive charge, and that allows chromatin to open up. So that you're basically the signalling mechanism is open up chromatin through the histone modification, and then allow the transcription factor to bind. DNA methylation works by blocking the ability of histone acetylases to operate on the histones. So when a region of DNA is highly methylated, it brings along a group of enzymes known as histone deacetylases, and what these deacetylases do is to prevent histone acetylation. So essentially what happens is when the region is methylated, you cannot produce acetylation on the histones, you can't open chromatin, it remains closed, and you can't get transcription factor binding. And that's essentially the basic mechanism by which DNA methylation commonly works. It can work through others, but that's what we speak of when we talk about the relationship between DNA methylation and transcription. So we come back to our hypothesis then, which was that maternal care may produce sustained changes in glucocorticoid receptor activity through effects that are largely epigenetic, including those on DNA methylation. So we're focused, as I mentioned before, particularly on the glucocorticoid receptor, and the differences in receptor expression between the offspring, the adult offspring of high and low-lucky and grooming mothers. So a number of years ago, we cloned the glucocorticoid receptor gene and the RAC. We identified the promoter, that region that binds to a transcription factor and activates the glucocorticoid receptor in the hippocampus. It contains a binding site for the critical transcription factor known as NGFIA. And then when we take brain tissue from the hippocampus, from the adult offspring of high and low-lucky and grooming mothers, what you find is that that region, that critical region that binds the transcription factor is hypermethylated in animals reared by low compared to high-lucky mothers. The transcription factor is thus able to bind to that region in the offspring of lows than it is to high, and thus it produces differences in glucocorticoid receptors. And I won't chase you through the whole thing, but essentially to say that if you reverse the effect of methylation, particularly on this histone deacetylase, you can completely change the patterns of glucocorticoid receptor expression and pituitary adrenal response to stress. So mom's licking activates an intracellular signaling pathway that produces a difference in methylation. Because there's a difference in methylation, there's a greater capacity for transcription factor to access the gene in animals raised by high compared to low-licky mothers. They produce more glucocorticoid receptors and thus they have a more modest response to stress. So that's the story in the rat. And then a number of years ago we started to partner with a colleague, Gustavo Torecki, with whom we still work closely. And we started to ask the question about whether something comparable might be occurring in the human. Gustavo operates at the Douglas in our department, what is one of the largest human brain banks in the world. And so we accessed postmortem hippocampus, the same tissue that we had studied in the rat. And we obtained samples from individuals who were victims of suicide versus individuals who had died apparently accidentally. Gustavo's team does a remarkable job of maintaining the brain bank and of documenting the phenotypes of the individuals whose samples we are accessing. And that includes forensic phenotyping. So there are structured validated interviews with family members and friends to determine developmental history of the individual, as well as past and current level of psychopathology. So we had more detailed forensic information than one would normally access. And of course we can study exactly the same region we can look at the glucocorticoid receptor gene. So Patrick McGowan who's now in Department of Biological Sciences at University of Toronto was supposed to lead this activity. We are looking here at a schema of the glucocorticoid receptor gene. These are all the various individual promoters. This promoter right here is the region that is comparable to that which we studied in the rat. So we focused on that particular and we simply asked the question are there differences in glucocorticoid receptor expression between individuals whose brains are obtained from control conditions versus victims of suicide? As you can see, the suicide victims show decreased glucocorticoid receptor expression in post-mortem hippocampus. And then we took advantage of the fact that Gustavo's group and collaboration provides wonderful documentation including the developmental history of the individuals and in particular knowledge of whether those individuals were victims of abuse. And when we segmented the 24 suicide samples that we had into those who were positive or negative for a history of abuse we found that it was only amongst those who had been exposed to abuse that showed the decrease in glucocorticoid receptor expression. The individuals who had committed suicide, completed suicide without a history of abuse did not differ from controls on that measure. And the rest is exactly what you would expect. We then looked for example at methylation across this promoter to the glucocorticoid receptor and found that at individuals who were victims of abuse there was hyper-methylation of the promoter by comparison to suicide victims without abuse or to controls suggesting that at least in a correlational sense some comparable story may be occurring within the context of humans. So the idea here is that essentially the early experience is driving activity within neurons that is producing stable differences in methylation which then sets the stage for differences in gene transcription. And because DNA methylation is a chemically stable mark, it's not irreversible but it certainly is chemically stable that it can partially at least explain perhaps the persistence of these early experience effects over the course of the lifespan. Is that clear so far? That's a pretty simple story, isn't it really? So others have started to pick up on this story and there's, in fact, Gustavo and I wrote a review on biological psychiatry a couple of years ago in which the human studies there were actually only one failure to replicate any study that had looked at developmental adversity and methylation of the corticord receptor gene. There were like 23 studies in total that we had found and there are more by now. And in all cases, people were using different forms of adversity, some prenatal, some postnatal, some combination of the two. People were often using different tissue. They certainly weren't using postmortem brain tissue. They were using bucal cells, salivary cells, etc. And they were nevertheless finding these differences or these associations between exposure to adversity and methylation of the corticord receptor gene. And so that was interesting and then of course we got to the point of asking, what's the clinical relevance of this? So we started and continued to collaborate with Rachel Yehuda and Mount Sinai. And first the thing we did was to simply say, okay well Rachel you send us some samples that you've clinically phenotyped and let's see if they map on to these variations. So these of course are combat veterans, American soldiers some with, some without PTSD symptoms. We characterized the methylation of this critical promoter for the glucocorticoid receptor and found that methylation was significantly associated first of all with glucocorticoid negative feedback which was very reassuring because that was the model that emerged from the right. PTSD symptom severity avoidance depressive symptoms and peritraumatic dissociation and in all of these cases those differences were highly significant. So there was a fantastic paper I think done by Dominique de Carvin. This is really a great paper published in Journal of Neuroscience a couple of years ago working with individuals who are the survivors of the Rwanda genocide and what Dominique and his colleagues showed was that the greater the methylation along this glucocorticoid receptor in blood cells lymphocytes notably the greater, the less the risk for PTSD so in blood cells in lymphocytes the methylation difference predicted the risk for PTSD and Dominique then went on back in Switzerland to do studies and showed that the level of methylation of this site that we had studied with Gustavo and Rachel predicted the emotional memory a predicted activity in the prefrontal cortex during the learning of emotional memory and as you probably heard this morning glucocorticoids are highly implicated in the formation of emotional memories so this is not I think an unexpected finding and then we went back to working with Rachel and we published results which is very preliminary and we asked and to cover that there is a Dutch group by the way that has shown that the methylation of the glucocorticoid receptor and indeed glucocorticoid receptor levels in lymphocytes are predictive in a prospective manner of the PTSD suffered by Dutch combat veterans in the Afghan theater so these samples were all determined prior to the individuals leaving for Afghanistan, PTSD was assessed upon their return and glucocorticoid receptor dynamics predicted PTSD so the idea that these could be predictive is seemingly not without some evidence and then the question is do they predict intervention effects so this was data we've done this Rachel's gone on to replicate this this was data from again the US combat veterans exposed to psychotherapy it's a psychotherapeutic intervention for PTSD and the level of pre-treatment glucocorticoid receptor methylation predicted treatment response so that was interesting but treatment response was not associated with changes in methylation so methylation predicted the response to treatment but the efficacy of the treatment did not produce necessarily a change in glucocorticoid receptor actually it's very interesting there's another gene I'm sorry I forgot to put the graph in here but there's another gene that if you're interested in the glucocorticoid system itself is a must read Elizabeth Binder has identified a gene FKBP5 it's a chaperone it moderates the impact of glucocorticoid receptors negatively so FKBP5 binds to the glucocorticoid receptor and when it's bound to the receptor the receptor cannot be activated by the steroid so FKBP5 is kind of like it's a ying and yang thing between it and the glucocorticoid receptor FKBP5 does not predict treatment outcome but it does reflect it so there are dynamic changes in response to treatment that occur at the FKBP5 in methylation so the whole story is one in which the single epigenetic mark is probably not going to tell you the whole story but understanding the physiology including the glucocorticoid receptor itself seemingly at least in the context of PTSD is able to predict outcomes and to reflect treatment outcomes so I thought that's kind of promising so I want to take you now to a slightly different approach and this is where I'm starting to become a little more optimistic about epigenetics as a predictor of adversity so going into this the rationale is that the outcome of exposure to adversity is moderated by the genotype of the individual in other words any phenotypic outcome a function of gene by environment interaction such that some people will be genetically inherently more likely to show a particular outcome under a set of certain environmental conditions it's pretty much what we've seen in medicine for years and years there are for example those thoroughly detestable human beings who can eat almost anything without gaining weight exactly and the rest of us not so much and you know that's biology so we have to come into this with the assumption that if we're going to predict outcomes in individuals we are going to have to integrate information about environmental circumstances clinically relevant environments as well as clinically relevant genotypic information so start with that as a premise so where are we now well this slide is dated what we do when we do larger studies with human populations is we do not screen across the entire genome I mean we do in certain studies do whole genome screening for methylation but for the most part what we're doing is we're using Illumina technology to look at the methylation of hundreds of thousands of CPGs across the genome so we're doing it's a survey if you will which is the best way to put it in every CPG but we're surveying about a half a million and in the case of the newer technology the new EPIC array that came out this year we're almost up to a million CPGs which may sound like a lot but it actually is about 3% of all the CPGs in the genome so it's still pretty modest but it does allow you to survey what it really allows you to do I think best more so than anything else it allows you to understand what produces variation in epigenetic marks across the genome so that's what we use it for and the question this is what the readout looks like so I say you're getting really good I mean this technology is brilliant in terms of giving you good reliable beta values for methylation and individual CPGs its weakness is that it's not covering every CPG across the gene here's a single gene it's a map kinase gene you're looking at 34 individual CPGs across the gene and this is just the average beta value that you see in those regulatory regions where transcription factors bind to turn the gene on you find hypomethylation right and that's exactly what you would expect from a gene that's actively transcribed so I'm going to tell you about a study no I don't want to tell you about this study yes yeah I'm going to tell you about this study no I am going to tell you about this study so one of the key issues is whether or not methylation is sufficiently dynamic that it's going to inform you about environmental effects adversity as well as intervention effect so I mean I think at this particular point the evidence is quite strong that epigenetic marks at different places in the genome can reflect the impact of adversity this is a study done by Karen O'Donnell in our department showing that that seems to be also true for intervention effects so one of our partners in our studies is David Olds and the Nurse Family Partnership Program which I'm sure many of you know Nurse Family Partnership Program was developed in the United States David's at the University of Colorado it has now been rolled out many places in Great Britain and British Columbia in our country and in many other areas and what the Nurse Family Partnership does is it targets pregnant women who are at risk for abusing their child and whose families are at risk for many different types of outcomes and so the women are identified on the basis generally of the fact that they themselves were victims of abuse they are impoverished they have a poor relationship with their mother and the Nurse Family Partnership Program involves nurse visitation occurring from the time of pregnancy through to two years postnatal and it has been published in a number of vehicles and shows that it can significantly reduce the risk for abuse and primarily a lot of conduct disorder criminality outcomes are very sensitive to this as well so what we were able to do was to work with David's first cohort this was the one down in Elmira in New York we had samples about 200 subjects and the subjects in this study are not the mothers, they are the children of the mothers and they are now 27 years of age so it's now 25 years since their mothers were participating in this Nurse Family Partnership Program and we've got blood samples from them and what we're asking is very simple we're going to examine variation across the methylation across the genome in these individuals is there a component of that variation that can significantly be associated to the fact that their mothers did or did not participate in the treatment and the answer is yes there is so it's not a huge effect there's actually almost 1500 verifiably differentially methylated sites across the genome that are associated with the participation of the mother in the Nurse Family Partnership Program now is that an overwhelming effect does it explain the outcomes we have no idea it answers one simple question can you detect variation in the methylation that can at least in part be explained by whether or not the mothers of these children participated in the treatment outcome so that's where we've gotten to but at least it's a point of telling us that there is potentially a sustained relationship between the treatment the intervention early in life and later on all kinds of complications I'm happy to go through these because they bear on how we would design subsequent experiments now I want to take you to the latter part of all of this which is where I get you to the idea of thinking what measures may be interesting as far as biomarkers are concerned so first study Singapore, birth cohort well it's actually got a thousand but we got antsy and decided to look at data from the first 237 kids we've got new natal tissue we have cord placenta, cord blood we took the cord we extracted genomic DNA and we ran we did the methylation we did the methylation we did the methylation array and we have lots of environmental measures these mothers have been phenotyped as extensively more extensively than any other cohort in the world we know about their metabolic cardiovascular, psychological social, etc we know a lot about these women over the course of pregnancy and we're asking a really simple question first of all do we verify a global influence of an environmental conditions during fetal development that is detectable across the genome second are the effects of the environment moderated by the genotype of the child and do they predict functional outcomes and we're going to talk mostly about the first two so what we did was the following we've got these genome arrays on 237 kids these are the older arrays they sample about a half a million cpgs so for every individual you've got a measure of the degree to which 500,000 cpgs are methylated and now what you want to do is you want to find that 30% that vary what are the regions across the genome that show variation because the ones that don't why bother studying them so the first task is to identify variably methylated region a variably methylated region is just a region that shows individual differences in methylation that's it so we identify 1,423 of those variably methylated regions then we've genotaped these individuals we looked oh my god even in those days which was this was done some time ago before we had full imputation we had about 4 million snips that we could impute with accuracy and then we do a simple linear regression analysis we take each one of these 1,423 variably methylated regions and we regress them against every one of the four million snips so it takes a little bit of time a little computing capacity but what you end up with is a knowledge of whether the variably methylated region is associated statistically with a genetic variation and you also have of course information on the environment so now you can test the following hypothesis across these 1,423 variably methylated region is the variation between individuals best explained by an environmental exposure mums BMI mums depression level etc by the genotype of the infant or by an interaction between the two simple model for every one of the 1,423 and here's what you find so genotype the SNP best explains variation in 25% of the cases which study is that which is the publication oh TEH at all genome research 2013 TEH oh it's a bugger because every time you go right drives you mad I think it's one of the readings on your list it is, there you go 75% or so are best explained by a gene environment interaction and here you see the various environments birth weight of the child mom smoking parity maternal anxiety, maternal depression maternal BMI, lots of nutrients etc in other words 75% were best explained by a gene environment interaction now this 25% by the way is an inflated estimate because it's a multi-ethnic sample it includes Malays, Indians and Han Chinese and so any time you have a multi-ethnic sample genetics is going to be inflated in terms of its ability to predict if you simply take the 160 or so who are ethnic Chinese then now genes account for 16% so and that's actually yeah it depends on the tissue but that's probably not bad 15% purely genetic 85% G by E but you'll notice what's missing here 0% E along as a best model and that kind of created a bit of a stir and then so we collaborate with Elizabeth Binder who is fantastic scientist she's also clinical director of the Max Planck Institute for Psychiatry in Munich and Elizabeth gained access to different tissues from different cohorts six or seven different cohorts and don't tell her I showed you this this is preliminary data that's just going up publication but here you see from all the various cohorts in this I don't know what that color I guess orange of some variation that's the estimate of the degree to which environment is the best predictor right it's G by E that's primary G alone is this orangish color which is about the same as what we found interesting enough despite the fact it's a different ethnic group different tissue different cohorts we said about 16% in the Han Chinese and you'll see that estimates damn close to what Elizabeth is reporting and the rest is G by E so variation in methylation like any phenotype is reflecting the interaction between gene and environment and what we assume is occurring is that environmental adversity operates as we suggested in the rat through intracellular pathways that modify the epigenome and that these intracellular pathways in the degree to which they respond to the adversity determined in part moderated by genetic variation yes yeah you want to so Elizabeth ran two models she ran an additive and an interactive model and you know something I'm still trying because I always think of these things I'm a better biologist than an anesthetician and I'm not that good a biologist so I'm trying to wrap my head around G plus E as an additive model but I think I get it we obviously know there can be pure effects of the SNP so this would suggest perhaps that here you're seeing some pure environmental effect it's not directly moderated but somehow it's occurring in an additive I quite frankly don't know exactly how to think about it yeah it's this gene and biology I'm not quite sure to be honest and I know anyway we can take that up but I've been trying to think because I'm an author on the paper I've been trying to think about how to think about the additivity of this and it's not clear one way is that when you have genetic variation that genetic variation most commonly occurs okay? so if I took say I go back to this model right here and I look at these G effects the G effect the genetic effect is an instance in which a SNP is most is accounting for most of the variation here an individual SNP in about 80% of the cases in our data that SNP is in proximity it's about 2 to 400 base pairs away but there's a lot of cases where it's way downstream and that could reflect an instance in which there's very much indirect effects that somehow are being best accounted for by additive models that's about as far as I've gotten in terms of being able to interpret that particular effect but the G by E component works I'm just going to give you one example of a specific G by E this is again the same data set in this particular case we used a candidate gene approach so we looked at what is one of the greatest hits in biological psychiatry is a genetic variant in the brain derived neurotrophic factor BDNF gene it's a single nucleotide polymorphism it produces a a valve to MET transmutation and it's highly functional and very much related to synaptic plasticity so we asked the question would BDNF genotype moderate the impact of a specific environmental event on the methylation and in this particular case we worked with a maternal phenotype that's closely associated with BDNF this is by the way this is the infant's genotype not the mother's the question is if we look at the impact of maternal anxiety on methylation is it moderated by the infant's BDNF genotype now here is just evidence that that SNP that single nucleotide polymorphism in the BDNF gene is associated with differences in methylation so this is just within the BDNF gene itself and here you see three different genotypes and you can see the variation in methylation that occurs at least at that particular site that's a classic association between a SNP and a CPG but overall what you find is that amongst children who carry the MET MET genotype there is a substantially greater impact of maternal anxiety on the number of differentially methylated CPGs than on kids who are MET valve or valve valve and that's true whether you look at maternal state anxiety and maternal trait anxiety that's in relation to the infant's genotype if you look at the mother's genotype there's no difference so this isn't the mom's genotype that's transmitting this this is really an impact of maternal anxiety moderated by infant genotype on the variation in methylation that's just a proof of principle if you will so where do we get to general differences in epigenetic marks appear to reflect gene by environmental interactions well just like all other phenotypes right and so what if we can identify clinically relevant environmental conditions the genotypes that moderate the impact of these environmental conditions on our outcome of interest the question then would be are there a parallel epigenetic mark that reflects that gene environment interaction and if so the epigenetic marks might be the perfect biomarker because they are reflecting both the clinically relevant environment as well as the genotype of the child and that's ultimately what we want to get to okay now how many people have I lost with that one nobody or I lost you a long time ago and it's no longer relevant or you never cared to begin with you want to be elsewhere etc anyway but that's the idea if epigenetic marks reflect gene environment interactions relevant for clinical outcomes then they may be the ideal biomarker so the complication in all of this this by the way I mean okay so there's a group of individuals known as the dohad crowd developmental origins of health and adult disease etc began with David Barker and it's very big on the idea that environmental conditions operate in utero to create increased risk for disease later on and it's heavily invested in the notion that these environments are operating to produce this risk and it's generally a field that's somewhat resistant to the idea that the genotype of the individual is of any importance whatsoever get over yourself to the hell I'm too old to be diplomatic it's stupid okay I mean it's stupid I mean the notion that an environment could operate independent of the genome or even perhaps more ludicrously that the genome could operate independent of the environment makes no bloody sense at all right so move on but it's intriguing because the disappointment here the idea was that you would do what we call EWAS epigenetic wide association studies you would take a particular environmental exposure you would take methylation and associate the methylation mark to the environmental exposure and then use that to predict outcomes the tape paper by the way one of the reviewers wrote well I guess this puts an end to EWAS studies yep it sure does that's the reason why it's why GWAS doesn't work doesn't include E right so EWAS ain't going to work because it don't include G hello so anyhow EWAS studies are kindly drifting away and GWAS is becoming far more popular as G by EWAS or including environmental measures so we're getting there but the role idea is phenotype is going to be complex in terms of its development you're going to best understand developmental outcomes individual and differences in individual outcomes by knowing the relevant G by E interactions that produce them that's where you should start I mean epigenetics is downstream from all of that a good conceptual model of what are the mechanisms by which an environment operates to produce pathology is really the starting point right and epigenetics only makes sense if you have these other things coming into play I mean the reason why epigenetics works for Rachel Yehuda is that she knew so much about the clinical phenotypes the biology before she came into it right the epigenetics was just one more logical extension of all the wonderful science that she'd already done it didn't salvage it didn't make her science it just was one extension that allowed her to move to a different level of questions and I think that's pretty much what it is then I guess Lawrence the last thing I think since we got a few minutes do we have a few minutes left I have 107 yeah sorry maybe this is a better for the discussion but for some of us field workers we work with smaller samples and I totally agree with you I would love to do more SNF you know by environment kind of study but is there a rule of thumb roughly it probably depends a lot on effect sizes and things like that and distribution in the population but what kind of sample sizes are we talking about? it's a fabulous question it's exactly you put your finger right on it right that's the whole challenge now is to try to translate it I'm happy to get into the questions right now because it sort of bears on where we're trying to go we have a partnership with Harvard Center for the Developing Child Office operation and it involves a number of sites organized through the American Academy of Pediatrics and the goal is exactly that can you take these markers into the clinic and use them at the level of the individual child I mean it's pretty ambitious right and by the way it's not just epigenetics we're doing a whole series of different candidates as well we don't know almost invariably and this is an interesting question to get into here the impact of adversity will vary as a function of the age the gender the ethnicity and perhaps the cultural context of the individual all of these things perhaps you know it well right all of these things have to be basically worked and it's a bit frustrating because people are asking and say oh wow epigenetics can use these marks to predict this we have no idea and you asked about effect size it's a critical question isn't it to what extent can I with any confidence predict these particular outcomes I think it's you know the best analogy the one I use all the time although it's much it's not as complicated is the algorithm in the American Heart Association is put together right so if you want to predict the probability of stroke in an individual it's family history it's lipids do you smoke do you exercise regularly a bunch of things plug all that in AHA has an algorithm and says oh here's the probability of a stroke and in some ways better yet it can also do a little bit of precision medicine say okay if your probability here and you have a family history boom you're going on statins etc. an aspirin if your probability is high but you don't have a family history we might just go with diet and exercise alone rather than including the meds so that's kind of where that's our gold standard that's where we'd like to get to but they know what their effect size is and the value is at the level of the individual we have no idea at this particular point in time as to what that is I don't know and the part of the reason that that becomes even a bit troublesome is some of the metrics very individual variation in methylation can readily be detected but it's not that great right so if I find individual variation in 10% at a particular CPG that's a pretty good signal yes and no wide study you're talking a thousand hundreds of thousands you know people and then old genomes or big stretch of the genome etc and it's just when you're talking about a small field study I think it's very easy for some of these big labs to just dismiss some of these smaller studies and maybe the point is at the level of the meta-analysis maybe these smaller studies are something that start to add up at some point I think in a sense I almost take the opposite from GWAS and I do some GWAS is that the smallest studies to me are likely to be more informative it means almost nothing to me if I take 100,000 people and identify statistically significant signal big deal I mean it's how much variance are you accounting for and GWAS has failed in that regard you know there's a, I don't know how many familiar with 23 and me I don't know if you know this this is working intimately so there's a nature genetics paper in which there's over 450,000 subjects in a 23 and me drawing on the 23 and me database and they have the most the greatest statistical success yet in identifying low side that are significantly associated with depression they have 44 and the p-value here is 10 to the minus 8 I think it is Lawrence 10 to the minus 9 it counts for less than 2% of the variance so like show me a smaller study you know decently powered maybe a couple of hundred people or so where you can account for more variance it needs replication it needs extension but I think I find that more to grow from but my god we're such in early days in trying to address that question of putting things into a practice a pediatric practice to predict we're getting there at least we're trying right we're we're participating we have six pediatric sites who are collecting data from us biological data in the course of a routine pediatric visit right so they are going to get the samples hair salivary um bucle swab etc and they're shipping it to us now you know there's costs related issues but at least in this particular case we're working with a type of data about a sample that can meaningfully be collected in a clinical site then of course throughput becomes what's the predictive outcome what's your ability to assess the outcome etc all challenges so it's a very much work in progress we have our center here is a hub for all of this so all those samples are sent to us we work with partners like Elizabeth Binder my co bar etc we've got access to cohorts around the world and we're just trying to what extent can I derive information in this cohort and replicated in that cohort and those cohorts vary in size uh from a couple of hundred to about fifteen hundred or so um and that's what we want to see in particular is exactly as you say is effect size and replication so it's it's a challenge to take it to that particular level general question about this conceptual question what extent about research exploratory in the sense that then we interpret it as a fact and to what extent is it theory driven in the sense that we're looking in a specific place because we have a reasonable thing at this level and what are the what are some of the choices involved with you know those broadly speaking those strategies um the strategy has been uh largely exploratory descriptive take a population you hope a large population create a metric based on genetic or epigenetic data um then from that metric you can obviously just replicate it directly but you can inform yourself so give you an example this is actually not a bad example this is uh work done by a Chantella DAS so Chantella accessed um a publicly accessible data set from NIDA National Institute of Drug Abuse it's a GWAS study 4,000 subjects unlike most GWAS studies it has really good environmental information whoa wonderful so what you have are 300 3 sorry 3,292 of these subjects who were exposed to a form of adversity that statistically predicts an increased risk for psychopathology including drug abuse okay so you take those 3,292 subjects they're all exposed they're at risk all have at least one exposure that predicts a risk of addiction from those 3,292 you have some who became addicted you call them susceptible some who did not become addicted you call them resilient and then you do a GWAS and what you find is that you get a whole series of genes that distinguish the resilient from the susceptible and you create a polygenic risk score and we've replicated that in now a number of different cohorts so that's replication from a discovery but you can go further than that so what Chantella then did she took the genes that distinguish resilient from susceptible individuals and informatically she said well those genes they produce proteins and those proteins interact to produce effects within a cell can I generate that so she generated pathways that were enriched in those genes and then she said those genes also respond to transcription factors let's see if that pathway is enriched for a particular transcription factor and she did it was the estrogen receptor right so the genes that distinguish resilient from susceptible individuals both males and females by the way are enriched for sensitivity to estrogen so then she said fine I'm just going to take those estrogen sensitive genes can I then predict and she can and then we have colleagues Mount Sinai Eric Nestler's lab and Zach Lorsch who is a postdoc in Eric's lab said we're going to look in a mouse model in which we describe animals as being resilient or susceptible to stress and see if estrogen can directly change susceptibility so you're moving this way and so informatics allows you to generate hypotheses from exploratory data and so now we're working with that's just a proof of principle to be honest with you it's not getting back to the point where hey we're going in the clinic but that's a way so it starts off purely as exploratory you then validate it in replication and then explore it further for the sake of generating hypotheses that can then be tested further down line that's the way we kind of work in the center it's a good point we think and Elizabeth is a big partner in this that we can use these data GWAS data can be used to explore because the GWAS is informing you those are just pure statistical relationships they're not that strong but it might be telling you if there's smoke is there some fire it seems to be in the case of small data sets what you need is the outcome from that kind of research to that so we can actually get a small amount of data you know where to look yeah I think the big exploratory stuff you need lots of data for all of the county some of the stuff that we were talking about yesterday with the coast of the man you were applying measures for more context sensitive there are going to be better measures you were kind of free listening another kind of cognitive domain analysis they can refine more accurately kind of assess a certain kind of experience I completely agree our biggest problem is I spoke to the American Heart Association they had Framingham so they got kicked off they got a jump start from Framingham telling them look here look here look here and then all kinds of research generated on the back of Framingham at a particular point we don't have a Framingham in mental health so and the thing that beleaguers when I talk for example and others talk about the degree to which environmental factors are not great predictors it's often contextual variables that are the complication so for example the association between maternal depressive symptoms and parenting is non-significant in a higher SES it's only in the low SES that that occurs and the effects on the child are of course much greater in low SES well that's going to surprise none of you guys the problem is if you're doing a study with even a thousand subjects and you're parsing it you can't contextualize you don't have the statistical power to do so so we kind of got to the point we said okay balls on this we got to get serious so we're mounting two studies one in Montreal one in Singapore in which we hope to recruit a 10,000 mothers in each site the outcome here is going to be maternal mental health during the antenatal period and we're trying to be able to get to the point where we can start to contextualize it we can start to look I mean the other thing about context is it determines what you're looking at the thing that drives the nuts about GWAS for example I mean Eric Turkenheimer many years ago published a paper in science showing that IQ is quite heritable in more affluent individuals in low SCS settings it shows 10% heritability and why because the environments are so diverse in those particular settings and so impactful so the notion that you can just look for universal genotype-phenotype relationships ignoring context just makes fools around but to the only sample size so we're bulking up and the bulking up though is not to dismiss smaller studies it's to actually pay respect to them it's to actually try to get to the point where you can have enough statistical power to parse the contextual variables that are so so influential oh that's good enough isn't it I've said enough haven't I you do yes thank you for listening you can analyze pathways from the paternal point of view and then the genetic modifications and the genes and express response systems and then on the other hand you have correlations correlational studies so it's impossible to think about the mediating process in terms of the model understanding the social component of course don't use the term reduced biology that makes the effect on the glucocorticoid receptor in the hippocampus of the rat is that a social effect or a molecular effect it's both so it's just two different levels of the same thing and your question is spot on is about can you understand the pathways at the front end of all of this what's driving it at that particular level I think the only way to get at that is through intervention studies which is what you're suggesting and we don't have a lot to be honest so one of the things one of the objectives we're negotiating we work intensely with David Old so we're going to become that those the nurse family partnership studies you know Myra and Denver and Memphis they've been on way for a long period of time what we need is to start with baseline we need to do exactly what your suggestion is to be able to capture dynamic changes in social interactions over time and in relation to relevant outcomes I completely agree there's very very little of that that's been done today but I think you're absolutely right those are the studies that will inform us with respect to the dynamics of this particular portion I think we have I mean intervention studies are the best way to do causality research in humans so that's the path so the hope is through the newer there's a new nurse family partnership program beginning in South Carolina and our hope David's hope is that we can start to introduce right at the get-go and really start to design the study to get at those particular issues I agree completely sorry there was another question down there yeah okay so tell you what I will what was your question sorry maybe you can address that for us no no go ahead tell me because I'm going to talk a lot about that so maybe a silly kind of trivial question here about the term biomarker I'm not sure I understand so in a way I also heard other psychologists who call the biomarker for phobia would be first thing of of neuronal firing in the mech so in the sense that that's a biomarker I think in that sense what they're getting is just the phenotype or the phenomenology is rather you want to characterize it in a more objectively measurable sense so you can't just ask people how do you feel and that has a lot of bad history in psychology so in a sense if you think about it that way then ultimately you're talking about a causal chain of events that ultimately lead to the phenotype which is what you're interested in so in a sense you call it the epigenetic modifications as biomarker of course it's closer to the final outcome than the environmental and genetic factors but ultimately you can't be as close as the neurons themselves no in fact you can you know I have a colleague Megan Gunner for example who heard biomarker is eye tracking in an intentional in an executive function it can be that absolutely and you know what you might argue look if I wanted to predict ADHD Megan's biomarker is more likely to be closer to the outcome and have stronger predictive value I'd buy that I agree no no if it's predictive it's purely mathematical transgenerational inheritance oh my god who's been reading about that it's like so obviously working in the field of epigenetics I had interactions with science journalists for years and years and they've invariably good AM radio not so much but that's another game for the most part journalists are intently interested in accurately representing your story particularly in terms of epigenetics and that's gone way off the wall in terms of transgenerational so the one problem number one okay so the inheritance of epigenetic marks the inheritance of epigenetic marks so the troublesome word the word we trip over is heritability inheritance so the word heritability was developed long before we knew anything about DNA right it's people in agriculture developing you know doing animal breeding right breeding domestic animals right so if I want to produce a cow that produces more milk then I phenotype the parents right I have a mother who produces lots of milk I breed her with a bull and my hope is that I'm enhancing the trait I don't give a shit if it's DNA or not I just want cows to produce more milk so you know when Fisher developed his models they were meant to be predictive they were meant to understand to what extent should you waste your time trying to produce a cow that produces more milk is it a heritable trait right that's what it was meant to do and it's very good at that so people who work in agriculture ask about the heritability of a train they don't assume the mechanism of heritability at all they simply want to know what's the statistical reliability between the presence of the phenotype and the parent and that and the offspring tell me that off to the races for selective breeding right you figure out the mechanism I care not okay so heritability is a statistical and then you know we discover DNA and DNA becomes the mechanism of inheritance and now heritability becomes equated with DNA right questionable right but okay so you know look we know that it is a the predominant mechanism by which traits are passed from parent to offspring passively is through genetic inheritance right and then the people in behavioral ecology in ecology in general Carolyn Rossiter and many others have been talking about the maternal effects field right it's called maternal effects parental effects etc in which the phenotype of the parent influences the phenotype of the offspring independent of genomic transmission we got tons of examples of that right we used to think for example that diabetic mothers had diabetic kids because they passed like genes onto their kids who then developed diabetes and then we found that lo and behold if you treat effectively the mother's diabetes during gestation during pregnancy you can greatly diminish the chance of having a hyper somatic a large baby and a diabetic offspring right so it's the mother's phenotype it's a hyperinsulinia largely that produces the hyper somatic baby and the child adipocyte development etc etc so it's the mother's phenotype that becomes the mechanism of heritability it's heritable it's just occurring through a non-genomic mechanism and then we started to produce genomic and non-genomic effects and I thought that was where we get in some place and so for example our example of maternal care is used as an example of a non-genomic effect it's the phenotype that drives this and then people started to understand that non-genomic effects of the parent the way in which the parental phenotype might influence the phenotype of the offspring could be epigenetic the mother's phenotype father's phenotype may influence the epigenome of the offspring and thus produces and then there have been a few proof of principle studies showing that if you expose the offspring sorry, if you expose the parent to an adverse event, this are done in the mouse, Elizabeth, Mansui, Tracy Bill, these people have done they're great researchers, I mean really good scientists and they've exposed the parents to a stressor and observed a change an epigenetic change in the case of Isabel's on the CRF gene, the gene for corticotropin releasing factor when she then looks at the offspring she finds a similar epigenetic mark occurs in the offspring independent of ever being exposed to adversity so she refers to this as their heritability and it is because heritability doesn't assume the mechanism, it just means that happens in the parent, happens in the kid that's all it refers to, right well that's not quite the way it's getting played up so instead what's happening is that it's oh my god, you know you're exposed to an adversity the epigenetic marks cause you to develop PTSD and you then transmit those epigenetic marks onto your child destined to collect PTSD bullshit not a fragment of truth even close to that, nor would Tracy or Isabel even suggest that's what's happening if you're really interested the paper I think no I didn't Ann Ferguson Smith who's a co-author on one of the papers is the chair of Human Genetics at Cambridge University Ann and a colleague wrote a brilliant paper in Science 2016 I think it is on epigenetic inheritance et cetera et cetera now take either Tracy or Isabel's experiments, okay in both experiments what they've done they've exposed the paternal organism, the male to a stressor and the offspring show epigenetic resemblance they then take the sperm of the father and find that the micro RNAs, the small RNAs that are associated with the stressors are indirectly recapitulating and reproducing the epigenetic mark in the offspring this is what we call indirect passage so there's two ways an epigenetic mark could be passed from a parent to an offspring one is passive, direct and it can occur it occurs in plants it occurs in insects but plants it definitely occurs differences in methylation can be directly transmitted to the offspring okay is it common no but it does occur in plants quite reliably the second mechanism is one in which the environmental event produces an epigenetic modification all epigenetic modifications involve micro RNAs to some extent and the micro RNAs are then transmitted through the germ cell the micro RNAs directly and the micro RNAs then operate on the cell line to reproduce this epigenetic mark that's an indirect effect the mark isn't being transmitted directly it's being recapitulated because of signals coming from the parent it's no different from what we talk about with licking and grooming except licking is a social event it's a molecule but in both cases they're indirect in both cases they refer to the phenotype of the parent they are what the behavioral oncologist referred to as a parental effect in this case it is passed through the germ line but indirectly does that make sense that's critical because that really describes what we're talking about what Isabel and Tracy have identified with micro RNAs is no different from what we discovered with licking and grooming it's just that they operate through different processes then they're indirect oh we do we can yeah admittedly it's another pathway but it's a molecular pathway it's having in germ cells it's happening on a scale that requires a totally different it does but it's that's a technical limitation on the degree to which you can interfere with one versus the other I think what is really important is that conceptually they're the same they're indirect effects in which the phenotype of the parent in this case the micro RNA and the sperm of the father is influencing the phenotype of the offspring which is the epigenome it's no different from what we've seen before but it's being given this publicity and the big reason is when people see the term heritability they assume that we know the mechanism and we don't heritability simply means a statistical similarity between parent and offspring and phenotype all it means and the funny thing is that well it's not so funny but the people doing the work there's no disagreement amongst the scientists if you read Anne Ferguson Smith's paper on this there isn't a single person I know working in the field who would even slightly disagree with what Anne is writing in that paper that in complex species including mammals the most probable pathway is going to be the indirect pathway the other connotation there's the thinking is the narrative of it oh well that's huge and that's the dangerous one you're absolutely right Lawrence there's this notion that there is a determinism associated with this and so that if the parent I don't know what people are thinking if the parent was exposed to a particular traumatic event will it influence the offspring's probability of health but the probable pathways as Rachel have suggested and as you were talking about earlier are parent-child interactions and the Holocaust literature is rich in describing at least the possibilities of that particular mechanism of passage so all of a sudden we've locked in to this one thing that has this apparent error of determinism it's too bad because it's distracting but it takes you away from good science that's being done and as I said the people working in the field this is not a disagreement amongst the scientists I mean I work very closely with Tracy and I know Isabelle very well and they're wonderful scientists we don't disagree in the least on this issue not in the least so it's all in the way unfortunately that it's being presented in this particular context but you're right Lawrence that's the danger there's no forensic implications to this that's why I'm leaving town thank you very much