The workshop concluded with two moderated discussions. First, Esa Davis of the University of Pittsburgh Medical Center moderated a discussion among all workshop speakers on possible future research directions, with an emphasis on expanding what scientists know about epigenetic-mediated associations between early developmental exposures and subsequent obesity-related health outcomes. Then Judith Hall of the University of British Columbia led a discussion on opportunities and challenges in epigenetics research that was open to all workshop participants.
Davis started the first discussion by listing what she observed as some consistent themes of the workshop discussion: the idea of permanency versus reversibility (i.e., with respect to epigenetic markers and their persistence over time), the need for predictive biomarkers, and efforts to understand causal mechanisms. She also observed that several speakers had called for efforts to collect prospective and longitudinal data. But, she asked, what kind of data? What are some existing cohorts or longitudinal data that could or should be used? What are some key data elements or features that any new cohorts should include?
How to Strengthen Human Observational Studies
Much of the discussion revolved around how to strengthen human observational studies in ways that will allow researchers to extract more types of useful information about the mediating role of epigenetics, especially given that, as Karen Lillycrop pointed out, many existing study cohorts were designed before epigenetics was on most researchers’ minds.
For example, according to Lillycrop, it used to be that researchers thought that one biological sample was sufficient, that is, they would collect and look at DNA one time and that would be it. Even in cases when they collected repeated samples over time, they would pool the DNA because of the assumption that DNA remains stable over time. Obviously that is not the case with epigenetic markers, Lillycrop said. She suggested the collection of repeated biological samples over time so that epigenetic changes early in life, particularly during the first 1,000 days, can be monitored. Multiple sampling over time requires a source material. She suggested buccal cells (i.e., from the inside of the cheek) as a relatively noninvasive source, although collecting sufficient DNA from buccal cells from infants is difficult.
Matthew Gillman was of the opinion that, in fact, there are many cohorts, especially in the developed world, with repeated biological sampling as well as good phenotyping, and that many of those cohorts are being used to examine epigenetic and other biological mediators between exposures and outcomes. However, he agreed that the next generation of cohort studies could be improved. He encouraged more cohorts in countries in transition, especially given the increasing prevalence of obesity in the developing world. Additionally, while it is difficult to study preconception given that many pregnancies are unplanned and that there is no ready source of recruitment, designing new cohorts in ways that would allow for the study of preconception would be helpful in his opinion. Finally, he said, because most studies to date have limited information on fathers, new cohorts that provide repeated biological and behavioral data on fathers would also be very helpful.
Repeated sampling or not, Kevin Grove observed that, often, when he asks an investigator why he or she collected a particular sample, the investigator will reply that it was available or cost-effective. “That is not a way to do a study,” Grove said, “especially when you get down to epigenetic analysis.” He emphasized the need for rational tissue selection. While it may not be as easy to get a sample of the hypothalamus, for example, as it is the muscle, Grove stressed keeping in mind that, right now, most researchers are sampling what is easy to get, not what is relevant.
Rather than generating more epigenetic data as part of either existing or future cohort studies, Caroline Relton of Newcastle University and
the MRC Integrative Epidemiology Unit at the University of Bristol suggested that it would probably be better to collect genetic data on samples for which DNA methylation and other epigenetic data have already been collected. She emphasized the need to generate genetic and epigenetic data in tandem, especially given the extreme challenge of interpreting DNA methylation and other epigenetic signatures without an understanding of the underlying genetic architecture. Doing so would be especially helpful, she said, for cohorts of different ethnic groups. Without knowledge of the underlying genetic architecture, comparing and contrasting epigenetic data across different ethnic groups will be very challenging.
Relton also suggested that intensive sampling in early life, during critical windows of epigenomic plasticity, would be beneficial, as would synthesizing the information from differently aged cohorts as a way to examine epigenetic variation over time. She agreed with Gillman that there is huge potential for those carrying out family-based study designs to improve their collection of data on fathers and siblings in already existing cohorts.
Instead of measuring DNA methylation or RNA expression in samples, Andrea Baccarelli at the Harvard School of Public Health suggested that another option, especially if cost is a limiting factor, would be to obtain information from the many online databases that include those data. In some cases, even tissue-specific data are available. He also mentioned the recently evolving concept of poised genes, that is, genes that are poised to be more easily activated. In his opinion, information collected on poised genes would be very interesting in the context of longitudinal studies.
With respect to exposure data and echoing Gillman’s call for more studies in countries in transition, Aryeh Stein of Emory University said that as an epidemiologist his interest is in the variance in prenatal, preconception, and childhood exposures. There is far more variance in the world than is concentrated in the Northern Hemisphere, yet virtually all existing cohort studies are in the north. He said, “What we need is a lot more investigation in the global south, where . . . all these problems are much larger.”
During the later discussion moderated by Judith Hall, but of relevance here, Jacob Friedman of the University of Colorado, Denver, remarked that, as a geneticist, he works with colleagues who conduct genetic screens in infants. In Colorado, the goal is to screen every infant for rare disorder. What is missing, or what needs to be done in his opinion, is a screening of infants for biomolecules that are actually within the range of normal but can be predicted by changes in fetal growth, changes in maternal weight gain, or maternal exposures. Instead of core tissue blood samples, which contain a mix of fetal and maternal blood, heel-stick samples could be used to conduct the screenings within the first 24 to 48 hours of life. In response, Hall agreed that such data would be invaluable but noted that newborn
screening regulations vary by state with respect to which tissues can be used for what type of screening.
In addition to better epigenetic and exposure measures, Linda Adair of the University of North Carolina at Chapel Hill stressed the need for better outcome measures. She observed that several workshop speakers had critiqued body mass index (BMI) as an outcome measure. She called for more precise, but field-friendly (i.e., obtainable outside a laboratory or clinical setting), measures of body composition.
Interpreting Epigenetic Data
As scientists learn more about epigenetic mechanisms, they also learn more about their complexity. For example, as Robert Waterland of the Baylor College of Medicine had pointed out earlier during the workshop, not too many years ago it was believed that a methylated gene was a silent gene. Now, however, it appears that DNA methylation can either silence or activate a gene. In fact, as Baccarelli pointed out, DNA methylation is not even always associated with gene expression. It is only one of the layers that control gene expression. The same is true of histone modifications. The growing “morass of [epigenetic] data,” as one participant described it, raised the question for Kevin Grove, does the knowledge exist and is there enough computing power to accurately predict likely biological outcomes based on those data?
Several participants agreed that computing power is not the limitation. Rather, the challenge will be to make sense of the data. Waterland pointed to the recent Nature paper summarizing the findings of the Roadmap Epigenomics Project (Kundaje et al., 2015), which, in Waterland’s opinion, is an indication of the state of the art in the ability to link epigenome-wide annotation of various markers (i.e., not just DNA methylation markers, but also histone modifications and other markers) with gene expression. For Waterland, the question is whether the political will exists to make the investment necessary for understanding inter-individual variation in mechanism, cell type and tissue specificity, and other issues. As sequencing costs continue to decline, the greater expense will be data management and analysis. Relton agreed with Waterland that the computational power exists and that the greater challenge now is to knit the accruing data together and place it in context.
Referring to Baccarelli’s statement that DNA methylation is only one of multiple layers of phenomena controlling gene expression, Marie-France Hivert of the Harvard Medical School remarked that, in her opinion, the next challenge is to interpret epigenetic data across those multiple layers. Additionally, she noted, not only are epigenetic markers tissue-specific, but they are also context-specific, with she and her research team having
observed signals present during pregnancy that do not exist outside of pregnancy. She suggested assembling teams of geneticists, physiologists, and other scientists to interpret the data.
During the second discussion, moderated by Hall, an audience member remarked that the complexity of the issues being discussed is going to increase as more types of data are collected in the future. The challenge of sorting through that complexity will be compounded by the fact that when a complex system is perturbed, the change is not necessarily linear. The system itself can change. He suggested that this is particularly true of developmentally plastic periods, when molecules develop new relationships with each other. He suggested that some of the new system science approaches may be helpful for identifying key mechanisms.
As a prelude to the second discussion, Judith Hall remarked that a principle of clinical genetics is that the “really unusual case” is important. It is those cases, she said, teach researchers about pathways. For her, the ob/ob agouti mouse (agouti mice homozygous for the ob/ob mutation, which makes them completely leptin deficient and leads to extreme obesity) is an example of the really unusual case.
Hall presented a mammalian evolutionary perspective on the issues being discussed at the workshop. For her, thinking way back “in the sands of time,” millions of years ago, the ability to be flexible and to have plasticity in response to environmental changes was how mammals evolved. While biologists have known this for some time, only recently have they begun to appreciate that flexibility and to study the multiple metabolic, psychological, and immunological pathways at play. When mammals first evolved, according to Hall, it was the placenta that allowed for mother–fetus communication. In Hall’s opinion, that’s what the Developmental Origins of Health and Disease (DOHaD), and epigenetics as part of that, is all about: the mother communicating with the fetus, via the placenta, to prepare the fetus to survive in the environment that it is about to enter.
Epigenetics is not a cool or mysterious thing, Hall said. Rather, it is simply that now that scientists have mapped the whole genome, they are “finally getting around to gene control,” she said. They are finding that genes that are turned off have different markes than genes that are turned on and that there are several different types of such markers. “It’s not magic,” she said. “It is actually just kind of what one would expect.”
Concerning the challenges in epigenetics research, Hall emphasized that there are many factors to consider when thinking about obesity, beginning with maternal birth weight. Most animal studies, she said, do not consider
maternal birth weight. The assumption is that maternal birth weights are identical. But in a litter of newborn rats, for example, not all eight rats are identical. Where they were located in the uterus may have had major effects. Hall encouraged researchers to be more careful about birth weight of the mother in both animal and human studies.
In addition to maternal birth weight, the field would benefit from greater collection of data on maternal health, stressors, diet, and related factors. The egg from which each of us was formed, Hall explained, was itself formed when our mothers were 6-week-old embryos. She disagreed with skeptics who think that maternal data are unobtainable. An estimate of birth weight, for example, can be derived simply from the mother knowing whether she was premature or not. Even better, researchers can find birth weights in hospital records where mothers were born. She herself has done that, Hall said; likewise with information about the father. Again, in her opinion, information on his birth weight, health, stressors, and diet is important. After all, sperm are formed 2 months before conception. She identified the diets of teenage boys and paternal exposures to endocrine disruptors (at any time before conception) as two areas of particular concern. For both fathers and mothers, she viewed “anything [in the environment] that is not natural,” including plastics, insecticides, artificial hormones, and antibiotics and other drugs, as factors that could potentially affect epigenetic patterning. Socioeconomic stress also likely plays a role in her opinion.
Hall also called for greater consideration of age-specific expression. Humans go through many stages, including embryo, early fetus, later fetus, newborn, and infant, and the physiology of those different stages is not well understood. In Hall’s opinion, a better understanding will come from examining which genes are turned on and off in every tissue at different stages of development. She suggested that in unfortunate situations where children in cohorts die, researchers could use tissues saved from autopsies to examine epigenetic markers. In order to do that, she added, pathologists need to be engaged in the effort.
In sum, without considering these many other factors, she said, “You don’t get the whole picture.” The challenge is, how? She sought answers from the audience.
Measuring Nutrition in Humans
An audience member asked how researchers can measure nutrition in humans, especially during pregnancy, and how they can link that information to epigenetic changes and phenotypic outcomes. Grove added that the challenge of measuring nutrition is made even more complicated by the reality that what is nutritious in one individual is not necessarily nutritious in another, depending on the body’s biochemical and molecular
capabilities—for example, how someone’s body handles different kinds of lipids or carbohydrates.
Another participant remarked that, as a nutritional epidemiologist, he has spent a fair amount of time thinking about how to measure what people eat. He has observed that researchers do not usually know what people eat; rather they know only what people say they eat, and the two are not the same. The question for him is, what is it that researchers want to know? Do they want to know what people are eating, what their microbiomes are eating, or their nutritional status? Those are different questions. If all that is sought is a measure of what people are eating, he speculated that in a few years researchers will be issuing Google glasses to study participants. That should provide a pretty good sense of what people are actually eating, he said. He agreed with Grove, however, that what that food does to the body is a different question.
The original thinking behind DOHaD, another participant said, emerged from studies relating lower birth weight to outcomes later in life, especially diabetes and cardiovascular disease. It was only later, over time, that the focus on lower birth weight shifted to a focus on maternal diet, with the term “under-nutrition” being, in the commenter’s words, “thrown around.” In the commenter’s opinion, the real object of discussion should be nutrition of the fetus, with maternal diet being only one small piece of that. Other pieces, he said, include placental function, fetal physiology, and maternal nutritional and other exposures before pregnancy. While it is very important to consider maternal diet, both during and before pregnancy, he encouraged a broader view and greater consideration of growth and health of the fetus.
Hall appreciated mention of the original thinking being the developmental origins of health and disease paradigm and the way David Barker’s focus on low birth weight infants triggered an “aha moment” for many researchers—that there was something very important about those infants. But what? In her opinion, good data are still needed.
Use of the Word “Epigenetics”
Waterland observed that much of the workshop discussion had focused on the word “epigenetics.” He reminded the workshop audience that epigenetics is just one potential mechanism of developmental programming. He agreed with Sarah Richardson that scientists need to do a better job conveying the complexity of epigenetics when communicating about epigenetic science in the public sphere. “In order to convey the complexity,” he said, “one thing we can do is not use ‘epigenetics’ and ‘developmental programming’ synonymously.” He explained that epigenetics refers very specifically to a suite of cell autonomous molecular mechanisms that stably regulate
gene expression potential. Even though researchers know very well that developmental programming of body weight regulation occurs in humans, it is unclear whether it occurs via epigenetic mechanisms.
Sleep and Metabolic Dysfunction
Referring to Antonio Convit’s discussion of sleep (see Chapter 4 for a summary of Convit’s presentation), a workshop participant remarked that sleep is known to play a role in maintaining metabolic function. For example, poor sleep patterns have been associated with insulin resistance. Given that pregnant women experience varying abilities to sleep at different times during pregnancy and even postpartum, and that baby sleep patterns are also interrupted postnatally, the participant urged a better understanding of the role of sleep both during and after pregnancy and its effect on the fetus and infant. She expressed curiosity about what researchers have learned from animal studies about the effects of sleep on metabolic dysfunction.
The Persistence of Epigenetic Markers
While there was no extensive discussion during this final session on the temporary versus permanent nature of epigenetic patterning, as there had been at earlier times during the workshop, Friedman mentioned that based on human tissue cultures sampled from obese individuals, some epigenetic patterns are reproducible. The reproducibility of those patterns suggested to him that there is some persistence, even as exercise, diet, and other factors can impact the epigenome. He suspected that researchers will be learning a lot about that persistence as they continue to study human tissue cultures.
Pediatricians in the Clinic
An audience member who described himself as a pediatric weight management clinician observed that when he first started working with children with weight problems 12 years ago, he felt like he did not have the tools to help his patients. It seemed far too simplistic to tell a child and his or her parent that the child needed to eat less and exercise more. Now, 12 years later, this greater understanding of underlying mechanisms will help to empower not only parents of children with weight problems, but also primary care providers. By thinking about obesity as a disease of biochemistry and physiology, he said, “You will look at the child and the mother totally differently.”
“In our beginning is our end.”—T. S. Eliot
In her opening remarks on the second day of the workshop, Shari Barkin of the Vanderbilt University School of Medicine began her remarks by referring to Andrea Baccarelli’s sheet music analogy, in which the sheet music is the genome, the composer’s notes are epigenetic markers, and the music produced is the phenotype. The same sheet music is played differently by the cello than by the flute, for example, just as the same genome yields different phenotypes in different tissues because of tissue-specific epigenetic markers.
Barkin identified several lessons learned from the Day 1 presentations and discussion. First, DNA is a code that has context, and its context—that is, the environment in which an individual lives and where one’s parents and grandparents lived—can affect modifiable epigenetic markers and, therefore, the way that DNA is expressed.
Second, exposure during pregnancy has phenotypic consequences for offspring. Maternal diet during pregnancy affects the patterning of the infant microbiome, which, in turn, affects infant metabolism; placental functioning affects infant inflammation; and leptin dysregulation affects appetite regulation and, later, adiposity. As Hivert discussed, just as genes are affected by context, so too are hormones, with either a negative or positive leptin regulation feedback loop operating, depending on maternal obesity.
A third point is that the paternal diet and fathers’ contributions matter. Barkin urged researchers to measure those contributions.
Fourth is the mismatch hypothesis, as first mentioned by Linda Adair of the University of North Carolina at Chapel Hill and then discussed throughout the workshop. A difference in exposures in utero versus exposures ex utero creates a mismatch, where a fetus is primed while in utero for something different than what an individual experiences ex utero. Most of the discussion around the mismatch hypothesis focused on the mismatch created when an individual is exposed to a low-fat diet in utero but a normal or high-fat diet ex utero. Barkin raised the question of what happens when a phenotype is mismatched in the other direction, that is, when an individual is exposed to a high-fat diet in utero but either a normal or high-fat diet ex utero. It appears that sometimes when individuals are exposed to a high-fat diet in utero but a normal diet ex utero, the phenotype normalizes. It is unknown whether exposure to a high-fat diet in utero and a high-fat diet ex utero actually induces better health. For Barkin, the mismatch hypothesis again highlights the importance of context. It is
important to think not just about the present moment of time, she said, but the past as well. “We are the accumulation of everything that has come before,” she said.
Another lesson learned is that the timing and duration of exposures matter. It is important to always ask the question, when did the exposure occur, and for how long?
A sixth lesson learned is what Barkin referred to as “the rule of Cs.” As you are reading the science, she said, always ask, are the results showing a correlation, causation, or confounding? She referred to Caroline Relton’s discussion of the multiple ways that this question can be examined and Relton’s call for a triangulation of evidence to address and confirm causation.
Yet another lesson learned is that when considering the clinical application of obesity prevention strategies, the sex of the child matters. Boys and girls respond differently to the same exposures. Additionally, sometimes there may be a short-term payoff but with long-term consequences. A compelling way to think about the microbiome, Barkin observed, is not just that it weighs 3 pounds, but that the majority of genes in a human body are not human. Barkin referred to Meredith Hullar’s presentation on the impact of the microbiome on human biological functioning, especially metabolism. Understanding the extraordinary symbiosis between the microbiome and human metabolism is important in Barkin’s opinion.
Lastly, she referred to Antonio Convit’s and Kevin Grove’s focus on metabolic dysfunction more generally, not just BMI. BMI itself is a crude marker of obesity in her opinion. It is the underlying metabolic dysfunction reflected in obesity that creates a pro-inflammatory state and vascular dysregulation, whether in the placenta (as Grove discussed) or the brain (as Convit discussed), with significant downstream consequences (e.g., reduced hippocampal volume and placental insufficiency).
In her closing remarks to the workshop, Barkin reiterated that epigenetics is just one approach to understanding the dynamic interaction between genetics, environment, and childhood development that affects childhood obesity and one that potentially offers insights into periods of potential reversibility and prevention.
She also stressed the need, first, to assess epigenetic variation in the context of genetic variation, as Waterland had stated during his presentation and as Relton had echoed in her comments in one of the final workshop discussions (see above), and, second, to focus on regions in the genome with inter-individual variation.
Barkin urged clarity about how the science discussed at the workshop
can benefit society. Is the goal to shed light on causal mechanistic pathways? Or is the goal to develop predictive biomarkers for either diseases or intervention responses? Or is the goal a bit of both? Which tissues should be examined depends on the question. She said, “Our approach, depending on our goal, should be different.” She also called attention to transgenerational influences and the need to consider what is meaningful and practical to include and assess.
Another issue to consider, she said, is technology. Barkin reiterated Waterland’s remarks about DNA methylation being merely one way to understand epigenetics, and epigenetics being merely one way to understand developmental programming. She stressed the need to understand what the right technology is with regard to interpreting the data. For example, the current DNA methylation technology is predominantly focused on promoter regions, but enhancer and intragenic regions are just as important to consider. Furthermore, she suggested considering the possibility that when one is working with animal models, examining DNA methylation at any genomic region might not be the best approach. A better approach might be to examine microRNA. She urged the creation of animal models that will have the greatest opportunity for translation into people and, as Matthew Gillman had suggested during his presentation, translation from people back into an animal model.
Barkin underscored the need for developing more tools to understand the complexity of interactions discussed at the workshop. What is being seen is likely the result of a cascade of factors set in motion generations ago, she said. But which epigenetic markers are “in ink” (permanent) and which are “in pencil” (transient)? And when are those markers being made?
The opportunity is vast, Barkin said, especially given today’s remarkable momentum behind big data. How then, she asked, can that momentum provide opportunities to better answer the questions posed at this workshop? The possibilities include developing systematic data elements, perhaps standard protocols, and ways to incorporate those elements or protocols across cohorts and longitudinal trials. She suggested starting not with what she called “the impossible things,” but rather with data that are simple and compelling, such as maternal and paternal birth weight. Finally, she emphasized the importance of tissue specificity, particularly the placenta, given its role as “great communicator” between the mother and child. In closing, Barkin stated, “The power of epigenetic science is understanding what we can do today to affect the health of future generations, as well as what we can do today to mitigate or modify the effects on this generation.”
This page intentionally left blank.