A major overarching theme of the workshop was the emerging nature of the evidence for epigenetics as a key component of the “Early Origins of Obesity” model, which was represented by the workshop infographic shown in Figure 2-1, and whether experimental findings indicate causal versus correlational or confounding associations between observed epigenetic changes and either exposures or outcomes. Session 1, moderated by Matthew Gillman of the Harvard School of Public Health, set the conceptual stage for this discussion. This chapter summarizes the Session 1 presentations and discussion.
Robert Waterland of the Baylor College of Medicine described what is arguably the clearest example of the causal role of epigenetic dsyregulation in obesity in an animal model: genetically identical agouti mice developing into either lean (brown) or obese (yellow) mice, depending on the degree of DNA methylation (one of several types of epigenetic marker) at the agouti locus. Waterland cautioned, however, that there are a multitude of obstacles to understanding how epigenetic dysregulation might similarly cause obesity in humans. He called for more prospective studies in humans to help infer causality. He also identified the need to assess epigenetic variation within the context of genetic variation and to focus on tissue-specific epigenetic patterning.
Andrea Baccarelli of the Harvard School of Public Health elaborated on the challenge of differentiating causality from correlation. He described a recent study in which the authors associated methylation of a gene in fat
FIGURE 2-1 Genetically identical agouti mice, which were phenotypically indistinguishable at birth, grew up into very different phenotypes depending on the level of DNA methylation at the Avy locus.
SOURCE: Presented by Robert Waterland on February 26, 2015.
tissue with body mass index (BMI). They concluded that methylation did not determine BMI; rather, BMI determined methylation (Dick et al., 2014). Regardless of causality, Baccarelli expressed hope that in the future epigenetic markers at birth can be used to identify newborns at increased risk of childhood obesity. But again, as Waterland cautioned, many obstacles will need to be overcome before reaching that clinical point. Baccarelli speculated that epigenetics might be experiencing the same “winner’s curse” that befell the field of genetics when geneticists started reporting genome-wide associations between genomic patterning and disease. Initially, many reported effect sizes were overestimated.
Much of what Robert Waterland discussed was built on ideas put forth in Waterland and Michels’s (2007) review of the epigenetic epidemiology of the developmental origins hypothesis and Waterland’s (2014) update of that review with a focus on obesity.
The developmental origins hypothesis proposes that during critical periods of development, transient environmental stimuli can have a persistent, even lifelong, impact on gene expression, metabolism, and risk of disease. Lucas (1991) put forth the concept of “biological programming” to describe these effects. Waterland and Garza (1999) built on Lucas’s idea of biological programming and devised a mechanistic construct that they called “metabolic imprinting.”
1 This section summarizes information and opinions presented by Robert Waterland, Ph.D., Baylor College of Medicine, Houston, Texas.
Metabolic imprinting refers to adaptive responses to early nutrition that are characterized by (1) a limited period of susceptibility (i.e., a critical window effect), (2) a persistent effect that last through adulthood, (3) a specific and measurable outcome, and (4) a quantitative relationship between exposure and outcome. Waterland pointed out that the first two characteristics are very similar to the concept of ethological imprinting proposed by Konrad Lorenz in the 1970s. The latter two characteristics were intended to guide mechanistic studies of metabolic imprinting. Waterland and Garza (1999) proposed four different potential mechanisms of metabolic imprinting, namely, alterations in organ structure, alterations in cell number or ploidy, clonal selection, and epigenetic mechanisms.
Epigenetics and Epigenetic Mechanisms
Waterland stressed that today, many people talk about developmental programming as if it is “all about epigenetics,” or all about epigenetic mechanisms, when in actuality it is very likely that epigenetics is just one of several different mechanisms interacting and that researchers should be studying all mechanisms in an integrative fashion. That said, the focus of his presentation was on epigenetics, which he defined as the study of “mitotically heritable and stable alterations in gene expression potential that are not caused by changes in DNA sequence.” The key to this definition, Waterland stated, is the stability of the alterations. While there are many ways to alter gene expression in a cell, epigenetic alterations are long-term stable alterations.
The best way to think about the stable nature of epigenetic alterations in gene expression, in Waterland’s opinion, is to think about the many different tissues and cell types in the human body, all with the exact same complement of DNA (i.e., the entire human genome), but each expressing very different subsets of that DNA. Cell type–specific expression is established early during development, and it is persistent. Even as many cells are replaced over time by progenitor cells, their progenitor cells “remember” to generate the same epigenetic markers.
Another way to think about epigenetics is to remember that the word “epigenetics” literally means “above genetics.” Epigenetic mechanisms are gene regulatory mechanisms layered on top of the DNA sequence information.
There are several epigenetic mechanisms, Waterland said. Methylation of cytosine within CpG dinucleotides is clearly an important one, and one that he would be discussing in detail.
Histone modifications are another potential epigenetic mechanism. Various modifications to the amino terminal tails of the histone proteins that package DNA in the nucleus of each cell are known to be highly correlated
with transcriptional activity and chromatin structure and therefore clearly play a role in regulating gene expression potential. However, as pointed out by Henikoff and Shilatifard (2011) and others, it remains unclear whether histone modifications have the definitive epigenetic characteristics of mitotic heritability, that is, whether specific established histone modifications can convey information over mitosis.
On the other hand, autoregulatory transcription factors have been recognized for decades as being able to function epigenetically (Riggs and Porter, 1996), yet they receive very little attention these days, Waterland observed. For example, the MyoD transcription factor, which plays an important role in muscle development in mammals, binds to and regulates its own transcription. During cell division, once MyoD is turned on, the MyoD protein, which is in the nucleus, is partitioned to both daughter nuclei, perpetuating its feed-forward auto-regulation. Many other transcription factors work in the same fashion.
Finally, noncoding RNA, another epigenetic mechanism, works in a similar way in terms of being partitioned in the nuclei during mitosis and being delivered to both daughter cells.
Waterland emphasized that all of these mechanisms and potentially others as well work in a synergistic fashion to maintain different regions of the chromatin in either a more transcriptionally silent or more transcriptionally active state.
Why Focus on DNA Methylation?
Of all the various potential epigenetic mechanisms, Waterland observed that most of the presentations at the workshop would focus on DNA methylation. He asked, why? First, DNA methylation is the most stable epigenetic mark, making it a very good candidate for conveying the type of long-term memory effects of relevance within the context of the developmental origins paradigm. Additionally, researchers understand its mechanism of mitotic heritability, knowledge of which makes it a bona fide epigenetic mark, in Waterland’s opinion. Also, DNA methylation can be measured in minute quantities of DNA. Moreover, it can be measured in a molecule-specific fashion, allowing for precise quantitation of the genetic influences on epigenetic outcomes.
To provide some background on DNA methylation, Waterland explained that, first, most cytosines within CpG dinucleotides are methylated at the number 5 position, converting cytosine to methyl-cytosine, a covalent modification that affects gene expression by regulating the affinity of methylation-sensitive DNA-binding proteins.
Another feature of DNA methylation to keep in mind is that tissue-specific patterns of CpG methylation are established during development.
Shortly after fertilization, the vast majority of methylation in both the sperm and egg genome is erased. Then, at about the time of the early embryo’s implantation, methylation patterns are reestablished in a lineage-specific manner as part of the differentiation process. The reestablishment process proceeds during fetal development and even during postnatal life. Each period of developmental establishment of DNA methylation patterns constitutes a “critical window” in which the environment, including nutrition, can affect the process.
Yet another feature of DNA methylation is that it requires dietary methyl donors and cofactors. And finally, and very importantly in Waterland’s opinion, DNA methylation is mitotically heritable and researchers understand the mechanism underlying its mitotic heritability. He explained that a CG sequence on one strand is also a CG sequence in the opposite direction on the other strand, allowing for semiconservative replication of established DNA methylation patterns during mitosis.
Is Epigenetic Dysregulation Contributing to the Obesity Epidemic?
Waterland discussed evidence—mostly from animal models but also from humans—demonstrating how epigenetic mechanisms can affect obesity. For example, when mice and other mammals are cloned, they often are born with a slightly elevated weight and develop adult-onset obesity. Waterland showed an image from Tamashiro et al. (2002) of two genetically identical mice, one produced by cloning, with the cloned mouse also being obese, its obesity clearly an epigenetic, not a genetic, effect.
In humans, the neurodevelopmental syndrome known as Prader–Willi syndrome is a good example, in Waterland’s opinion, of an epigenetic dysregulation that can cause obesity. Although the syndrome is most commonly caused by a genomic deletion of a large portion of chromosome 15, a subset of sporadic cases are caused by epigenetic silencing of the same genomic region.
Agouti mice are a third example of epigenetic dysregulation known to cause obesity. Again, Waterland showed an image of two genetically identical mice, this time two newborns who were indistinguishable at birth but who, because of an epigenetic difference at the agouti viable yellow (Avy) locus, grew up into very different phenotypes. One grew up to be yellow and obese, the other brown and lean (see Figure 2-1), with the obese mouse having a very low level of DNA methylation at the Avy locus and her lean sister being very highly methylated at the same locus.
Alleles that behave like the Avy, that is, with dramatic inter-individual variation in DNA methylation even among genetically identical individuals, are called metastable epialleles. Waterland and others have shown that nutrition and other environmental stimuli, both before and during preg-
nancy, can affect the establishment of DNA methylation at metastable epialleles with persistent and permanent phenotypic consequences.
A fascinating feature of metastable epialleles, in Waterland’s opinion, is the systemic nature of the inter-individual variation, with essentially the same level of methylation present in all of the different cells of the body. Consequently, one could take a drop of blood from an agouti mouse, measure the methylation at Avy, and predict with absolute certainty whether the mouse would become obese in adulthood.
Obstacles to Understanding the Epigenetic Contribution to Human Obesity
While many clinicians and epidemiologists would like to have an epigenetic biomarker in humans like the differentially methylated Avy locus in agouti mice that could be used to predict who will become obese, Waterland cautioned that finding such a marker will not be a simple task. He identified several obstacles to understanding how epigenetic dysregulation contributes to human obesity, not the least of which is that genetic variation is an important determinant of epigenetic variation. If one was to conduct a case control study of obese versus lean individuals, one could certainly find epigenetic differences between the two groups, he said. However, it would be difficult to rule out that the observed differences in epigenetic regulation (and obesity) were caused by genetic differences between the two groups.
Another obstacle to understanding the epigenetic contribution to human obesity is the largely cell type–specific nature of epigenetic regulation. Although clinicians and epidemiologists would like to be able to study DNA from easily obtainable samples (e.g., peripheral blood or buccal swabs), Waterland observed that in most cases those samples will not provide much information about epigenetic regulation occurring in tissues of greater relevance to obesity, such as hypothalamus or adipose tissues.
Yet another obstacle is poor characterization of epigenetic regulatory regions, though the situation is improving, Waterland observed. He referred to just-published data from the National Institutes of Health (NIH) reference epigenome mapping project (Kundaje et al., 2015). One of the biggest insights provided by those data, in Waterland’s opinion, is the importance of epigenetic regulation in enhancer regions. Most epigenetics researchers have been focused over the past couple of decades on promoter regions, that is, regions at the beginnings of genes. Enhancers are regulatory regions often located hundreds of thousands of base pairs away from genes. It appears now that epigenetic regulation at enhancers plays a critical role in tissue-specific and cell type–specific gene expression. In Waterland’s opinion, inferring tissue-specific epigenetic dysregulation is going to be very difficult in human studies.
Also with respect to the poor characterization of epigenetic regulatory regions, while the general rule is that DNA methylation is a silencing mechanism, Yu et al. (2013a) reported a large class of genes in the human genome that are actually transcriptionally activated, not silenced, by methylation at the 3′ end of a gene. In sum, Waterland said, “We clearly have a lot to learn about how epigenetic regulation works.”
Finally, the disease process itself can affect epigenetic mechanisms, raising questions about causality. The best example, in Waterland’s opinion, is in cancer epigenetics. It has been known for decades that tumors are characterized by dramatic epigenetic dysregulation. However, it was unknown until recently whether epigenetic dysregulation actually caused the cancer. With respect to obesity, when epigenetic changes are observed in lean versus obese individuals, the direction of causality is still unclear.
The Way Forward
Waterland suggested some steps forward to help move the field past these many obstacles. First, controlled studies in appropriate animal models are urgently needed to advance researchers’ understanding of epigenetic mechanisms underlying the developmental programming of obesity. For example, Waterland pointed to the significant advantages of using inbred mouse models: the removal of genetic variation as a factor, the ability to observe a single life span from embryonic development to adulthood in only 1 year, and the ability to obtain all relevant tissues.
In terms of human studies, Waterland reiterated the need to assess epigenetic variation in the context of genetic variation, to study appropriate tissues (or, if that is not possible, then at least confirmation of systemic variation), to focus on genomic regions with known functional inter-individual variation, and to conduct prospective studies that allow for causal inference (i.e., rather than just cross-sectional studies).
In closing, Waterland showed for a second time the image of the genetically identical but epigenetically different (at the Avy locus) agouti mice. He remarked that although it is customary to think that all phenotypic variation is genetically based, clearly epigenetics has a large role in determining phenotype.
Imagine a musical score with a very specific sequence of notes. Like DNA, the sequences of notes are translated into a phenotype, that is, music.
2 This section summarizes information and opinions presented by Andrea Baccarelli, M.D., M.P.H., Ph.D., Harvard School of Public Health, Boston, Massachusetts.
Andrea Baccarelli showed an image of a music score and the great composer Herbert von Karajan at the podium, in front of an orchestra, translating the score into a phenotype (see Figure 2-2). If Baccarelli himself was at the podium instead of von Karajan, he said, the phenotype would change dramatically. That is an example, Baccarelli said, of how the exact same sequence of notes, or DNA, can yield completely different phenotypes. If Baccarelli had looked at the music score beforehand, he would have noticed that several parts of the score had marks written above the sequence of notes, marks indicating, for example, “louder” or “softer.” Those marks, he said, are what epigenetics is all about: marks added to a sequence of notes, or a sequence of DNA, that do not change the sequence but do change the phenotype. The marks can be written in either pencil or pen, with notes written in pen representing, in the genome, permanent epigenetic markings established during fetal life. Notes written in pencil, in contrast, like the methylation markings on inflammatory genes, are reprogrammable and can change within a matter of minutes.
Baccarelli proposed a model to explain the epigenetic influence on obesity risk, with the fetus playing a central role and in utero exposures affecting the embryo and fetus in ways that potentially modify the epigenome
FIGURE 2-2 Comparison of the relationship between DNA, epigenetics, and a phenotype to the relationship between the sequence of notes on sheet music, marks scribbled on the score by the composer, and the music produced.
SOURCE: Presented by Andrea Baccarelli on February 26, 2015.
at birth and, in doing so, program risks associated with obesity (Fleisch et al., 2012). Baccarelli suggested that not only are in utero exposures important, but preconception exposures may be as well, particularly exposures of the gametes, given that gametes have their own epigenomes that can be influenced and modified by environmental exposures. In other words, what parents do before they have children, including what they do during their own childhood, potentially influences risks in their children. There is no clear evidence indicating that a parent’s epigenome directly impacts the fetus, other than via gamete exposure; in fact, according to Baccarelli, some experts would argue whether such a direct impact (i.e., not via the gametes) is possible.
Based on this model, and borrowing concepts from Gillman and Ludwig (2013), Baccarelli proposed that fetal life exposure programs the epigenome at birth, but that the epigenome at birth is modifiable and can change in postnatal life. Fetal exposures can potentially be correlated to risk of obesity, and its sequelae, via the epigenome, include high maternal body mass index (BMI), high gestational weight gain, gestational diabetes, and other environmental factors.
Correlating DNA Methylation and Body Mass Index: The Challenge of Reverse Causation
In an epigenome-wide association study (EWAS), Dick et al. (2014) measured BMI and DNA methylation in blood samples from individuals participating in three different cohort studies: the Discovery in the Cardiogenics Consortium (n = 479), MARTHA (n = 339), and KORA (n = 1,789). Importantly, Baccarelli noted, both BMI and DNA methylation were measured at the same visit. The researchers identified three methylation sites in the hypoxia inducible factor 3a, or HIF3A, gene that were positively correlated with BMI across all three cohorts. Additionally, one of the three sites also correlated with BMI when DNA methylation was measured in adipose tissue (in addition to blood), but not in skin. However, based on a Mendelian randomization analysis, they concluded that perturbation of HIF3A methylation might play a role in the response to obesity, but is likely not a cause of obesity. (See the summary of Caroline Relton’s presentation in the next chapter for a more detailed discussion of the use of Mendelian randomization to interpret the Dick et al.  findings.)
Baccarelli explained that instead of the epigenome modifying the phenotype, in this case BMI, it is not surprising to find, that it is the phenotype modifying the epigenome (Relton and Davey Smith, 2012). He suggested that longitudinal studies might be a preferred study design for human epigenetics with respect to identifying the temporal, or causal, sequence of events.
Learning from Past Experience: Lessons from Genetics
One of the first genome-wide association studies (GWASs), published in 1990, reported an odds ratio of 8.7 for an association between alcohol dependency and a polymorphism in the dopamine receptor, with people who had the polymorphism 8.7 times more prone to be dependent on alcohol. Not only was the odds ratio high, Baccarelli explained, but so were the confidence intervals. Subsequent studies provided a different perspective (Smith et al., 2008). Over time, studies became larger, confidence intervals shrank, and the odds ratios being reported suggested that the increased risk associated with the polymorphism was actually only 40 percent higher. That odds ratio, Baccarelli said, is not nearly as high as 70 percent. There are other examples, he said, of odds ratios initially being reported as high as 9 but, over time, being reported as zero. He wondered whether the same phenomena, known as “the winner’s curse,” might be happening now with epigenetics. Are larger effects being reported that will become more moderate over time?
The winner’s curse is similar to what happens on an Internet auction site, Baccarelli said, when an item is first auctioned and people overbid because they do not know the true value of the item. Initially, there is a high variance in the estimated dollar value of the item, and the “winner” is likely to be “cursed” because he or she has bid the most. In genetics, the first to publish positive GWAS findings often overbid, or overestimate, the effects. Again, he asked, is this happening now with EWAS findings? If so, how can it be avoided?
Additional Questions to Consider
In conclusion, Baccarelli observed that most NIH funding for epigenetic research is focused on DNA methylation, with less funding directed toward research on microRNA and histone modification (Burris and Baccarelli, 2014). He wondered whether opportunities exist to study other epigenetic mechanisms and what should be done to create those opportunities.
In terms of new research directions, he asked, which are the most promising? Which mechanisms are less explored? Is looking at the effect of the environment on the epigenome enough? Drawing another parallel between genetics and epigenetics, Baccarelli pointed out that most researchers are focusing on how the environment modifies the epigenome and how the epigenome, in turn, modifies disease risk. In genetics, researchers also examine how the environment modifies, in its case, the genome, for example, how carcinogens damage DNA. But geneticists have also been thinking a lot about how the genome makes individuals more or less susceptible to environmental influences via gene-by-environment interactions. Baccarelli
suggested that permanent epigenomic markers (i.e., markers “drawn in ink”) might similarly be modifying the effect of environmental influences on phenotype (Bollati and Baccarelli, 2010).
Most importantly, in Baccarelli’s opinion, is the need to consider the meaning of the research conducted. Why are researchers studying this? Are they looking for mechanisms or biomarkers? What are the benefits to patients and to society?
In the panel discussion following Baccarelli’s presentation, workshop participants considered a range of topics: the tissue specificity of epigenetic markers and changes, nutritional exposure to dietary methyl donors, the impact of assisted reproductive technologies on epigenetic patterning, and the temporary nature of many epigenetic markers. This summary of the discussion is organized by topic.
Kevin Grove of the Oregon National Primate Research Center asked how common it is for epigenetic phenomena to be tissue- or cell type–specific. He wondered whether heterogeneous epigenetic markers in different liver cells, for example, might contribute to the heterogeneity in observed responses. Robert Waterland responded that, yes, at least with DNA methylation, there are dramatic differences between different tissues and cell types. Regarding the liver, he said that he was unaware of the extent to which researchers have examined the heterogeneity of epigenetic regulation across different hepatocyte populations, but he suspected differences. While studying the hypothalamus in mice, he and his colleagues have observed dramatic differences between neurons and glia in epigenetic changes between birth and weaning (at day 21). Specifically, they have observed much greater increases in methylation in neurons compared to glia. He said that if they were to examine different classes of neurons in the hypothalamus, they would probably see dramatic differences there as well.
Baccarelli added that when measuring methylation in a tissue, researchers measured the proportion of cells that are methylated. When they measure it that way, there is a lot of variation, he said, between subjects.
Nutritional Exposure: Dietary Methyl Donors
An audience member asked Waterland which dietary methyl donors he and his colleagues have used in their research. Waterland replied that in the Avy agouti mice studies and in subsequent studies with related models, they
used what he referred to as a “methyl donor cocktail.” They supplemented the mouse diet with extra folic acid, vitamin B12, betaine, and choline. In his opinion, a combination of nutrients is required. They have unpublished data suggesting that folic acid supplementation alone does not produce the same pro-methylation effects in the offspring. He referred the questioner to a recent paper that he worked on in collaboration with Andrew Prentice showing that maternal nutrition status at around the time of conception affects DNA methylation at human metastable epialleles in offspring (Dominguez-Salas et al., 2014).
Assisted Reproductive Technologies and Epigenetics
When asked about assisted reproductive technologies and the risk of epigenetic dysregulation associated with it, Waterland replied, “The issue of assisted reproductive technologies is clearly very important. I find it just amazing that there seems to be more regulation about what goes into our breakfast cereal than what is the specific composition of the media that are used for these early embryos during the in vitro fertilization process.” Early studies suggested that individuals conceived via assisted reproductive technologies run a higher risk of certain developmental diseases. However, according to Waterland, it is still unclear whether those diseases result from the process itself or from epigenetic aberrations that existed in either the sperm or the egg and that contributed to the infertility in the first place.
Reversible Epigenetic Markers (or Markers “Made in Pencil”)
Grove asked if the reversibility of (reversible) epigenetic markers was due to cell turnover. He pointed out that there is a lot more cell turnover in the liver or skin, for example, than in the brain. Waterland responded, “Absolutely.” He explained that there are two potential ways to reverse DNA methylation markers. One is active demethylation, which occurs via a ten-eleven translocation (TET) methylcytosine-mediated process, and the other is failure to maintain established methylation patterns during mitosis, which leads to erasure.
Waterland reiterated that metastable epialleles, which are established during early development, persist into later life. He and his team have data from mouse studies showing that measures of methylation and inter-individual variation in methylation obtained early in life are exactly the same as those obtained later in life. The same has been observed in humans, with methylation in one region measured at age 7 being predictive of what will be observed at age 17.
Related to the issue of reversibility, Caroline Relton of Newcastle University asked whether any epigenetic markers made early in life as a result
of early life exposure disappear later in life. She wondered whether if so, those transient markers may nonetheless set in motion a physiological chain of events that influence risk of obesity or other chronic disease. Moderator Matthew Gillman closed the session by suggesting that Relton’s question be kept in mind for the remainder of the workshop. As he rephrased it, “Even if DNA methylation or other epigenetic processes are transient, could they set in motion a programmatic phenomenon?”
This page intentionally left blank.