Twin and adoption designs

What twin and adoption studies have taught us about the development of psychopathology: Virtually all complex human traits are heritable and genetic influences are largely stable over time

Developmental psychopathology and quantitative genetics are both premised on the assumption that phenotypes are multiply determined by individual and environmental factors. Findings from twin and adoption studies have highlighted the importance of two types of influence in particular. First, as noted by Turkheimer (Reference Turkheimer2000, p. 160) in a paper entitled “Three Laws of Behavior Genetics and What They Mean,” “All human behavioral traits are heritable.” Twin and adoption studies have played a key role in demonstrating the pervasiveness of genetic influences on variation in behavior and traits, thus challenging the conventional wisdom that variation in children's environments determined individual differences in their behavior.

Does a child's environment have no bearing on his or her development? Quantitative genetic studies also address this question. Because genetic factors tend to account for about 50% of phenotypic variation, it must also be true that human behavioral traits are under environmental influence. It is counterintuitive that the environments that account for substantial variation in a phenotype are not typically those that are shared by children growing up in the same family but, rather, nonshared environments that make children growing up in the same family different from one another (Plomin & Daniels, Reference Plomin and Daniels1987; Reiss, Neiderhiser, Hetherington, & Plomin, Reference Reiss, Neiderhiser, Hetherington and Plomin2000; Turkheimer & Waldron, Reference Turkheimer and Waldron2000).

Second, twin and adoption studies have also highlighted the importance of genetic and nonshared environmental factors in accounting for phenotypic change and stability (Ronald, Reference Ronald2011). For example, longitudinal studies have demonstrated that symptoms of attention-deficit/hyperactivity disorder (ADHD) are moderately stable over time (Biederman et al., Reference Biederman, Faraone, Milberger, Curtis, Chen and Marrs1996). Quantitative genetic studies of ADHD have shown that symptom stability is largely accounted for by the same genes operating over time, whereas change in symptomatology is accounted for by new genetic influences and by unique environmental influences that tend to operate at one point in time but not at the next (Kuntsi, Rijsdijk, Ronald, Asherson, & Plomin, Reference Kuntsi, Rijsdijk, Ronald, Asherson and Plomin2005; Price et al., Reference Price, Simonoff, Asherson, Curran, Kuntsi and Waldman2005; Rietveld, Hudziak, Bartels, van Beijsterveldt, & Boomsma, Reference Rietveld, Hudziak, Bartels, van Beijsterveldt and Boomsma2004).

In another effort to model genetic and environmental influences on phenotypic change and stability, behavioral geneticists have integrated general growth mixture models into a biometrical modeling framework (Larsson, Dilshad, Lichtenstein, & Barker, Reference Larsson, Dilshad, Lichtenstein and Barker2011). For example, Fontaine, Rijsdijk, McCrory, and Viding (Reference Fontaine, Rijsdijk, McCrory and Viding2010) estimated trajectories of callous–unemotional traits across three time points (ages 7, 9, and 12 years) in a sample of twins. They identified four trajectory groups (stable low, stable high, increasing, and decreasing), with the stable high group also showing the highest levels of conduct problems and hyperactivity. For boys, approximately 60% to 80% of the liability for group membership was largely accounted for by genetic factors, with the rest accounted for by nonshared environmental factors. Genetic influences on membership in the stable high group were greater than genetic influences on membership in the other trajectory groups. In contrast, genetic influences on trajectory membership were relatively weaker and shared environmental influences were relatively stronger for girls. Such approaches have the potential to identify whether genetic and environmental factors differentially account for patterns of change and stability in behavior or symptomatology.

Gene–environment interplay is pervasive

At the same time that quantitative geneticists have demonstrated conclusively the genetic underpinnings of virtually all complex human behaviors, they have also demonstrated that the nature–nurture dichotomy is false. Over the past 15 years, researchers have become increasingly interested in demonstrating how genes and environments work together to produce phenotypic variation. Gene–environment interplay refers to three processes, two of which will be described in this section (Kendler & Eaves, Reference Kendler and Eaves1986; Plomin, DeFries, & Loehlin, Reference Plomin, DeFries and Loehlin1977; Rutter & Silberg, Reference Rutter and Silberg2002). First, genotype–environmental correlations refer to genetically based individual differences in exposure to environments. Second, G × E interactions refer to genetically based individual differences in susceptibility to environments. Social regulation of epigenetic and gene expression profiles will be described in the section on molecular genetic approaches.

Molecular genetic methods

Linkage studies identify genomic regions containing risk variants by matching the patterns of IBD between related individuals with their concordance for disease. The principle is that any two relatives who both suffer a heritable disease will tend to have inherited the same susceptibility loci. This approach is particularly powerful for detecting rare and highly penetrant loci (i.e., those for which carriers have a high risk of disease). Thousands of monogenic disorders have been mapped in this way, often using extended pedigrees. A more common research design for complex traits employs sibling pairs. The advantages of linkage studies are that results can be found using samples of hundreds (rather than thousands) of subjects and that genomewide scans can be conducted using panels of hundreds (rather than hundreds of thousands) of genetic markers. The corresponding disadvantages are that linkage analysis lacks power to detect the common loci with small effects that account for the majority of heritable trait variance. Moreover, linkage analysis identifies only a broad genomic region that may span dozens of genes and include many thousands of polymorphisms. Consequently, its usefulness in the mapping of complex traits has extended mainly to the identification of “positional” candidate genes located under replicated linkage peaks, a strategy that has scored such notable successes as the identification of the nucleotide-binding oligomerization domain-containing protein 2 (NOD2) risk locus for Crohn disease (Hugot et al., Reference Hugot, Chamaillard, Zouali, Lesage, Cezard and Belaiche2001; Ogura et al., Reference Ogura, Bonen, Inohara, Nicolae, Chen and Ramos2001).

Gene association studies identify candidate genes that are hypothesized to increase risk for psychopathology. Candidate genes are those about which researchers hold some a priori hypothesis. For example, there is a long history of addiction researchers studying gene variants involved in dopamine neurotransmission because of the role dopamine plays in reward (Berridge, Reference Berridge2007; Wise, Reference Wise2004). Gene association studies test whether specific gene variants predict case versus control status or variation in psychiatric symptoms.

Gene association studies have been criticized for two reasons. First, there have been very few replications in research on psychopathology, suggesting that many novel findings were false positives. Second, relatively few significant associations have been identified in protein-coding regions of the genome, suggesting that researchers do not know enough about the neurobiological basis of the phenotypes they study to construct plausible genetic hypotheses (Duncan & Keller, Reference Duncan and Keller2011).

Genomewide association studies (GWAS) were meant to be a hypothesis-free corrective to the shortcomings of candidate gene association studies. In GWAS, hundreds of thousands of common SNPs (i.e., minor allele frequencies exceed 1%) are “tagged” across the genome (i.e., they are genotyped on DNA microarray chips) and statistical analyses that make stringent corrections for multiple testing are used to determine whether any of these tagged SNPs are associated with disorder or account for significant variation in symptomatology. SNPs that meet stringent criteria for genomewide significance (typically p < 10⁻⁸) are considered to be markers for the true causal variants with which they are correlated (i.e., in linkage disequilibrium; Donnelly, Reference Donnelly2008). In theory, researchers can follow up on genomewide “hits” by fine-mapping the region around the hit and, if potentially causal variants are identified, they can use functional assays to determine which variant causes disease (Donnelly, Reference Donnelly2008). In practice, this process has proved to be challenging, and even when GWA studies have produced strong evidence of association for specific SNPs, functional follow-ups have generally not located the causal variants (Williams et al., Reference Williams, Moskvina, Smith, Dwyer, Russo and Owen2011).

What molecular genetic studies have taught us about the development of psychopathology: Common and rare variants are likely to be implicated in disease risk

Broadly speaking, psychiatric geneticists fall into two camps. Some hypothesize that most common disorders are caused by gene variants that are common in the population (meaning that minor allele frequencies are at least 1%). This is known as the common disease–common variants hypothesis (Lander, Reference Lander1996; Pritchard & Cox, Reference Pritchard and Cox2002) and it posits that most diseases are caused by a very large number of common variants, each of which has a very small additive effect on risk for disorder, potentially via their influence on gene regulation rather than protein coding (Donnelly, Reference Donnelly2008). Others have hypothesized that common disorders are caused by gene variants that are recently derived and rare in the population but have moderately high penetrance, meaning that possession of the risk allele is strongly associated with expression of the disease (Cirulli & Goldstein, Reference Cirulli and Goldstein2010). Associations between rare variants and disorder may reflect equifinality, wherein a set of rare variants increases risk for disorder in one subset of the population and a different set of rare variants increases risk for ostensibly the same disorder in a different subset of the population (Gibson, Reference Gibson2012).

To date, neither theory nor data decisively favor one hypothesis over the other (Gibson, Reference Gibson2012). GWA studies have identified scores of variants for complex diseases (see the National Human Genome Research Institute catalogue of published GWA studies at http://www.genome.gov/gwastudies), and although there are relatively few instances where genomewide hits for psychopathology have been followed by functional work to identify causal variants, it is more likely that the true casual variants are common rather than rare alleles (Wray, Purcell, & Visscher, Reference Wray, Purcell and Visscher2011). In contrast, the rare alleles hypothesis is more consistent with evolutionary theory; variants that reduce evolutionary fitness should be winnowed from the population. Moreover, with respect to psychiatric phenotypes, rare variants have been implicated in schizophrenia (Stefansson et al., Reference Stefansson, Rujescu, Cichon, Pietilainen, Ingason and Steinberg2008; for reviews, see Mulle, Reference Mulle2012; Rucker & McGuffin, Reference Rucker and McGuffin2012), autism (for reviews, see Betancur, Reference Betancur2011; State & Levitt, Reference State and Levitt2011), and bipolar disorder (for a review, see Malhotra & Sebat, Reference Malhotra and Sebat2012) with effect sizes that greatly exceed those for common variants identified in GWAS. Even when the additive effects of multiple variants are considered, not just those that exceed genomewide significance levels (Davies et al., Reference Davies, Tenesa, Payton, Yang, Harris and Liewald2011; Lee et al., Reference Lee, Wray, Goddard and Visscher2011; Yang et al., Reference Yang, Benyamin, Mcevoy, Gordon, Henders and Nyholt2010), the variance in the phenotype that is explained by genetic variance often fails to match heritability estimates, leading to questions about “missing” heritability (Manolio et al., Reference Manolio, Collins, Cox, Goldstein, Hindorff and Hunter2009). The common versus rare variants debate is unlikely to resolve in an either/or understanding of disease risk; both common and rare variants are likely to be implicated in disease. Establishing the contributions of common and rare variants will require functional approaches that delineate the mechanisms by which these variants increase risk for disorder against a given genetic backdrop.

Genotype moderates the effect of experiences and exposures on risk for psychopathology and the course of psychopathology

The possibility that individual differences in genotype explain why some individuals succumb to risk whereas others maintain positive functioning in the face of adversity is a potential explanation for multifinality in development: the notion that exposure to a given risk factor can result in many different outcomes (Cicchetti & Rogosch, Reference Cicchetti and Rogosch1996). G × E interactions predict that specific alleles will at least partly determine whether or not a particular risk exposure will result in the manifestation of psychopathology.

Although gene–environment interplay is likely to involve biological interactions, wherein one or more genes and one or more environments increase risk for disease because they are involved in the same causal pathway in an individual (Rothman, Greenland, & Lash, Reference Rothman, Greenland and Lash2008), behavioral geneticists have mainly concerned themselves with the search for statistical interactions, albeit with biologically plausible underpinnings. Statistical interactions do not allow for any inferences about the biological mode of action. They simply establish whether the effect size of an exposure on risk for disease changes across different levels of the genotype, measured in a particular scale (e.g., an additive vs. a dominant scale).

Although quantitative genetic studies published in the 1990s established that genetic influences on a phenotype were stronger under some conditions versus others (as reviewed earlier), the first demonstration of a measured G × E interaction in the psychopathology literature appeared in 2002. This study found that men who carried the low-activity variant of a 30 base pair VNTR in the MAOA gene and who had experienced maltreatment in childhood engaged in significantly higher levels of antisocial behavior in childhood and adulthood compared with those who carried the low-activity variant but had not been maltreated. In contrast, among men who carried the high-activity variant of the MAOA gene, a childhood history of maltreatment was not predictive of their antisocial behavior in either childhood or adulthood (Caspi et al., Reference Caspi, McClay, Moffitt, Mill, Martin and Craig2002). Since the publication of this and a subsequent paper by the same group (Caspi et al., Reference Caspi, Sugden, Moffitt, Taylor, Craig and Harrington2003), the literature on G × E interactions has grown exponentially (Aschard et al., Reference Aschard, Lutz, Maus, Duell, Fingerlin and Chatterjee2012).

Studies of G × E interaction not only provide examples of multifinality in development, they also demonstrate how gene–environment interplay influences the course of psychopathology. For example, in a sample of young adults, childhood maltreatment predicted persistent depression, but not intermittent depression, among individuals who carried the short allele of the serotonin transporter promoter gene-linked polymorphism. In contrast, childhood maltreatment was unrelated to either persistent or intermittent depressive disorder among individuals who were homozygous for the long allele (Uher et al., Reference Uher, Caspi, Houts, Sugden, Williams and Poulton2011). Another study found that the genotype encoding for cholinergic receptor muscarinic 2, CHRM2, differentiated youths who were on a high and stable trajectory of externalizing problems from those who were on a low and decreasing trajectory, particularly if youth affiliated with a deviant peer group (Latendresse et al., Reference Latendresse, Bates, Goodnight, Lansford, Budde and Goate2011).

The statistical interactions identified in G × E studies are consistent with both data and theory. For example, heterogeneity in response to exposures or experiences is the rule rather than the exception in epidemiological studies. Moreover, the consistently small genetic effects identified in gene association studies could be produced if there was significant heterogeneity across a sample in the magnitude of genetic effects arising from differential exposures (Moffitt, Caspi, & Rutter, Reference Moffitt, Caspi and Rutter2005). From an evolutionary standpoint, individual differences in susceptibility to rearing environments would provide a hedge against uncertainty about the future. Whereas high responsivity to the environment would confer reproductive advantages when rearing environments were good predictors of future conditions, low responsivity would confer reproductive advantages when rearing environments had poor predictive power (Belsky & Pluess, Reference Belsky and Pluess2009).

Nevertheless, there have been cogent criticisms of the literature identifying G × E interactions (Duncan & Keller, Reference Duncan and Keller2011). Criticisms of this literature fall into four related categories. The first is that positive findings are disproportionately represented in peer-reviewed journals, although it is doubtful that publication bias is any less common in other areas of the social and natural sciences. The second criticism is that most G × E studies are underpowered to detect interaction effects that account for less than 1% of the variance in outcome (a moderately sized effect, assuming that G × E interactions and genetic main effects typically account for similar portions of variance in a phenotype; Duncan & Keller, Reference Duncan and Keller2011). Although it is generally appreciated that low statistical power can lead to Type II error, it can also increase the proportion of Type I errors reported in the literature. The third criticism of G × E studies is that there is insufficient correction for multiple testing and, consequently, inflated Type I error rates (Aschard et al., Reference Aschard, Lutz, Maus, Duell, Fingerlin and Chatterjee2012), although this issue is hardly specific to research on G × E interactions (Simmons, Nelson, & Simonsohn, Reference Simmons, Nelson and Simonsohn2011).

A fourth criticism of G × E studies is that disparate exposures sometimes get called the same thing (Caspi, Hariri, Holmes, Uher, & Moffitt, Reference Caspi, Hariri, Holmes, Uher and Moffitt2010; Uher & McGuffin, Reference Uher and McGuffin2010). For example, “stressful life events” comprise self- and parent reports of early family environments, substantiated reports of maltreatment, composite measures of poverty, residential conditions, family structure, parental psychopathology, and physical illness. As a result, it is unclear whether failures to replicate G × E interactions should be attributed to measurement variance or to Type I error.

Although the G × E literature is characterized by a number of methodological shortcomings, it is not necessarily true that the identification and replication of statistical interactions between genotypes and environments should be the gold standard for determining whether or not a given G × E interaction is “real.” Statistical interactions matter only insofar as they plausibly reflect some biological process. Convincing evidence for the biological plausibility of G × E interactions comes from research that uses a range of methods and models not only to document the presence of G × E (i.e., whether the statistical interaction replicates in independent samples and across model organisms) but also to probe potential neural substrates of the interaction (Caspi et al., Reference Caspi, Hariri, Holmes, Uher and Moffitt2010). For example, although there has only been mixed success in replicating the MAOA × Maltreatment interaction (Jaffee, Reference Jaffee and Widom2012; Kim-Cohen et al., Reference Kim-Cohen, Caspi, Taylor, Williams, Newcombe and Craig2006), there is growing evidence that (a) MAOA genotype moderates the effect of early rearing experiences on aggressive behavior in experimental primate studies (Newman et al., Reference Newman, Syagailo, Barr, Wendland, Champoux and Graessle2005), (b) methylation of the MAOA promoter is associated with low levels of the MAOA enzyme (Shumay, Logan, Volkow, & Fowler, Reference Shumay, Logan, Volkow and Fowler2012) and low MAOA enzyme activity is associated with self-reported trait aggression (Alia-Klein et al., Reference Alia-Klein, Goldstein, Kriplani, Logan, Tomasi and Williams2008), and (c) the MAOA variant is associated with the activation and connectivity of cortical and subcortical regions in response to negatively valenced stimuli and response inhibition tasks (Buckholtz et al., Reference Buckholtz, Callicott, Kolachana, Hariri, Goldberg and Genderson2008; Meyer-Lindenberg et al., Reference Meyer-Lindenberg, Buckholtz, Kolachana, Hariri, Pezawas and Blasi2006; Passamonti et al., Reference Passamonti, Fera, Magariello, Cerasa, Gioia and Muglia2006). These findings make a case for the biological plausibility of G × E interactions involving MAOA and early rearing environments. Repeated exposure to negative experiences may potentiate a genetically based predisposition to aggressive behavior via corticolimbic circuits involved in emotion regulation, memory for affectively salient events, and impulsivity.

Findings from molecular genetic studies have limited clinical value thus far

One of the goals of basic research in developmental psychopathology is to translate findings on the etiology and course of psychopathology into policy and clinical practice. Behavioral geneticists share this goal. Many scientists hoped that the sequencing of the human genome in 2003 and subsequent discoveries of genes for disorders would inform genetic screening, drug discovery, and personalized medicine. These hopes have remained largely unrealized. The small size of most genetic risk factors has hampered efforts to identify an individual's risk for disease (Goddard & Lewis, Reference Goddard and Lewis2010) and simulation studies have shown that case-control studies on the order of approximately 10,000 individuals would be required to identify a set of variants that would account for more than 50% of the genetic variance (Wray et al., Reference Wray, Goddard and Visscher2007). Although it is accepted that there can be clinical utility in using genotype to predict disease risk, this is only the case for highly penetrant variants, such as the breast cancer Type 1/2 gene (BRCA1/2) loci that predispose to breast and ovarian cancer. For most multifactorial diseases, like Type II diabetes, nongenetic risk factors for complex diseases remain better predictors of disease status than genetic risk factors identified via GWA studies (Talmud et al., Reference Talmud, Hingorani, Cooper, Marmot, Brunner and Kumari2010) and genetic risk prediction for psychiatric disorder remains poor.

In the domain of personalized medicine, there has been little evidence, despite great initial excitement (Meyer et al., Reference Meyer, Rutter, Silberg, Maes, Simonoff and Shillady2000), that either candidate gene variants or significant hits from GWAS modify patients' responses to antidepressant (Keers & Aitchison, Reference Keers and Aitchison2011; Taylor, Sen, & Bhagwagar, Reference Taylor, Sen and Bhagwagar2010; Uher et al., Reference Uher, Perroud, Ng, Hauser, Henigsberg and Maier2010) or antipsychotic treatments (Malhotra, Zhang, & Lencz, Reference Malhotra, Zhang and Lencz2012). The small effects of gene variants identified in GWAS and the observation that hundreds of variants of small effects may contribute to risk for disorder means that most GWA studies have been largely uninformative for drug discovery efforts (Cirulli & Goldstein, Reference Cirulli and Goldstein2010).

Summary of molecular genetics research

In summary, there is an ongoing debate about the relative importance of common versus rare variants in predicting risk for disorder. Although GWAS have had some success at identifying genomewide hits, these variants typically account for a very small proportion of the variation in psychiatric disorders. There has also been success at identifying rare variants associated with highly heritable disorders like autism, schizophrenia, and bipolar disorder. These successes must be followed up by efforts to determine the function of likely causal variants. Although there is substantial interest in G × E interactions, there are ongoing challenges in this research area, including insufficient power to identify G × E effects, a potentially high false discovery rate, and a lively debate about what constitutes evidence of replication. Pharmacogenetic and genetic screening approaches must establish both clinical validity and utility in order to inform intervention. Doing so will involve translational efforts at different levels, of which the identification of risk variants, even in the most promising cases, is only the first step of many (McMahon & Insel, Reference McMahon and Insel2012).

Social regulation of epigenetic processes

Epigenetic processes that influence gene transcription include histone modifications and DNA methylation, the latter of which has been the focus of much of the recent work in epigenetics. Methylation refers to the addition of a methyl group to CG (CpG) dinucleotides that are sites in the genome where the nucleotide cytosine is followed by the nucleotide guanine and separated by only one phosphate. In general, DNA methylation can disrupt the binding of transcription factors (which stimulate gene expression) and attract methyl-binding proteins that are associated with gene silencing and chromatin compaction (Jaenisch & Bird, Reference Jaenisch and Bird2003).

Research on the epigenetic underpinnings of behavioral regulation was catalyzed by a series of animal studies conducted by Meaney, Szyf, and colleagues who were interested in maternal nurturing behaviors and their effects on offspring biobehavioral outcomes. Using a rat model of maternal behavior, they found that the offspring of mothers who engaged in high levels of licking, grooming, and arched-back nursing behavior were less fearful as adults and produced relatively lower levels of stress hormones (i.e., glucocorticoids and adrencorticotropic hormone) in response to being physically restrained (Liu et al., Reference Liu, Diorio, Tannenbaum, Caldji, Francis and Freedman1997). These effects of maternal behavior on offspring stress response were caused by reversible epigenetic changes to the glucocorticoid receptor (GR) gene that increased GR gene expression (Weaver et al., Reference Weaver, Cervoni, Champagne, D'Alessio, Sharma and Seckl2004, Reference Weaver, Champagne, Brown, Dymov, Sharma and Meaney2005) and by epigenetic changes to numerous other genes (Weaver, Meaney, & Szyf, Reference Weaver, Meaney and Szyf2006).

The publication of these papers and others generated enormous interest in how early adversity may increase risk for later psychopathology via epigenetic processes. Researchers have relied heavily on animal models (e.g., Mueller & Bale, Reference Mueller and Bale2008; Murgatroyd et al., Reference Murgatroyd, Patchev, Wu, Micale, Bockmuhl and Fischer2009; Roth & Sweatt, Reference Roth and Sweatt2011), but a growing number of studies use human tissue. For example, it has been shown that suicide completers with a history of childhood abuse have different profiles of hippocampal GR methylation (Labonte et al., Reference Labonte, Yerko, Gross, Mechawar, Meaney and Szyf2012b; McGowan et al., Reference McGowan, Sasaki, D'Alessio, Dymov, Labonte and Szyf2009) as well as different methylation profiles across the genome (Labonte et al., Reference Labonte, Suderman, Maussion, Navaro, Yerko and Mahar2012a) compared with suicide completers without a history of abuse. In a sample of newborns who provided cord blood, higher levels of methylation at a specific CpG site in exon 1F of the GR gene were observed among those whose mothers had higher levels of depressed and anxious mood in the third trimester (Oberlander et al., Reference Oberlander, Weinberg, Papsdorf, Grunau, Misri and Devlin2008). These methylation changes were also associated with increased cortisol in response to a visual information processing task at 3 months of age (Oberlander et al., Reference Oberlander, Weinberg, Papsdorf, Grunau, Misri and Devlin2008). Genomewide methylation scans have identified higher methylation levels across a range of genes among adults who experienced high versus low levels of childhood socioeconomic disadvantage (Borghol et al., Reference Borghol, Suderman, McArdle, Racine, Hallett and Pembrey2012), among adolescents who did or did not experience foster care (Bick et al., Reference Bick, Naumova, Hunter, Barbot, Lee and Luthar2012), and among adolescents who experienced high versus low levels of family adversity in infancy and early childhood (Essex et al., Reference Essex, Boyce, Hertzman, Lam, Armstrong and Neumann2013).

Social regulation of gene expression

Like epigenetic profiles, gene expression profiles are also socially regulated, providing a possible biological pathway by which the environment could increase risk for disease (Cole, Reference Cole, Kendler, Jaffee and Romer2012; Miller et al., Reference Miller, Chen and Parker2011). For example, elderly adults who report feeling socially isolated are characterized by upregulation of genes involved in the inflammation response and downregulation of genes involved in responses to viral infections and in the production of antibodies (Cole et al., Reference Cole, Hawkley, Arevalo, Sung, Rose and Cacioppo2007). Thus, changes in gene expression induced by a perception of social isolation could increase an individual's risk for inflammation-related diseases such as cardiovascular disease. Other studies have shown that growing up in conditions of social disadvantage (Miller et al., Reference Miller, Chen, Fok, Walker, Lim and Nicholls2009) and, particularly, perceiving one's social environment to be threatening (Chen et al., Reference Chen, Miller, Walker, Arevalo, Sung and Cole2009) is associated with upregulation of inflammation-related genes, although a warm relationship with a parent can break this link (Chen, Miller, Kobor, & Cole, Reference Chen, Miller, Kobor and Cole2011). Primate studies in which social status (i.e., dominance rank) is experimentally controlled confirm and extend results from human studies, showing that genes involved in immune function were upregulated in primates of lower dominance rank and that changes in social rank were associated with corresponding changes in gene expression (Tung et al., Reference Tung, Barreiro, Johnson, Hansen, Michopoulos and Toufexis2012). Highlighting the interdependence of endocrine and immune systems, variation in gene expression in the majority of the rank-related genes was partly mediated by variation in tissue composition and glucocorticoid regulation (Tung et al., Reference Tung, Barreiro, Johnson, Hansen, Michopoulos and Toufexis2012).

Although this literature on the social regulation of gene expression has primarily dealt with inflammation-related physical health disorders (for which it is particularly valid to interrogate peripheral blood cells versus other tissues), dysregulation of the immune response is potentially involved in some psychiatric disorders (e.g., major depressive disorder; Raison, Lowry, & Rook, Reference Raison, Lowry and Rook2010) and in response to social stressors like maltreatment (Danese et al., Reference Danese, Moffitt, Pariante, Ambler, Poulton and Caspi2008; Danese, Pariante, Caspi, Taylor, & Poulton, Reference Danese, Pariante, Caspi, Taylor and Poulton2007). The general mechanism is likely to apply to other psychiatric disorders. Interestingly, gene expression and epigenetic profiles show allele-specific patterns (Cheung & Spielman, Reference Cheung and Spielman2009; Lee, Reference Lee2012; Meaburn, Schalkwyk, & Mill, Reference Meaburn, Schalkwyk and Mill2010). If some gene variants are more susceptible to social regulation than others, then corresponding alterations in gene expression or epigenetic modifications could provide a potential mechanism by which genes and environments interact to influence health and disease (e.g., Tung et al., Reference Tung, Akinyi, Mutura, Altmann, Wray and Alberts2011).

Challenges to research on epigenetics and gene expression

Despite the excitement generated by the linking of early (and ongoing) life experience with epigenetic and gene expression profiles, this research faces numerous challenges (Heijmans & Mill, Reference Heijmans and Mill2012). The primary challenge concerns the human tissues available for research in psychiatry. Psychiatric disorders are typically disorders of the brain, but brain tissue is obviously not readily available from live human subjects. Therefore, researchers either obtain archived brain tissue from deceased subjects or peripheral tissues (usually whole blood cells) from live human subjects. Both approaches are problematic. Information about brain donors is usually limited to what is known from medical and coroner records or from individuals who were related to the deceased and who may be unreliable informants. Thus, information about an individual's early experiences may be limited. Moreover, interpretation of gene expression or epigenetic profiles in postmortem brain tissue is complicated by clinical, gender, aging, tissue, and other factors (Atz et al., Reference Atz, Walsh, Cartagena, Li, Evans and Choudary2007). Conversely, although obtaining peripheral tissues from live human subjects is a relatively noninvasive process, it is uncertain that the gene expression or epigenetic profile of those tissues is similar to that of brain or other tissues (Heijmans & Mill, Reference Heijmans and Mill2012; Rollins, Martin, Morgan, & Vawter, Reference Rollins, Martin, Morgan and Vawter2010). For example, depending on where in the genome methylation levels are being interrogated, correlations across tissue types (including brain and blood tissue) vary substantially (Davies et al., Reference Davies, Volta, Pidsley, Lunnon, Dixit and Lovestone2012). Further, certain genes whose expression is restricted to only neurons or glia will necessarily be inaccessible from the periphery.

Apart from the challenge of identifying relevant human tissues, additional work is needed to determine whether within-individual and between-group differences in epigenetic profiles are associated with corresponding changes in gene expression and, similarly, whether changes in gene expression correspond to altered protein products. Finally, epigenetic and gene expression effects are typically small in magnitude, and it is unclear what the biological significance of such small alterations is (Heijmans & Mill, Reference Heijmans and Mill2012).

Despite these challenges, there is scope for studying the social regulation of epigenetic and gene expression profiles in humans, with one goal being to identify clinically relevant biomarkers of early and/or ongoing adversity. Although acquisition of blood samples from children is not trivial, blood may represent the tissue source with the most potential to contain a predictive biomarker. An evaluation of gene expression profiles comparing patterns in blood and brain identified over 4,100 transcripts that were expressed in both compartments (Rollins et al., Reference Rollins, Martin, Morgan and Vawter2010). Further, it has been shown that plasma levels of brain-derived neurotrophic factor (BDNF) are highly correlated with brain levels of BDNF (Klein et al., Reference Klein, Williamson, Santini, Clemmensen, Ettrup and Rios2011), supporting the potential utility of this approach. Importantly, some progress has been made toward the identification of acute biomarkers for mood disorders (Le-Niculescu et al., Reference Le-Niculescu, Kurian, Yehyawi, Dike, Patel and Edenberg2009) and schizophrenia (dopamine and cyclic AMP regulated phosphoprotein-32; Torres et al., Reference Torres, Souza, Miranda, Nicolato, Neves and Barros2009). Therefore, analysis of DNA methylation profiles and/or gene expression in blood has the potential to identify a predictive risk biomarker, and discovery efforts in this domain should be pursued.

Summary

In summary, research on social regulation of epigenetic and gene expression profiles offers exciting possibilities for developmental psychopathology researchers who are interested in both the timing and episodic course of psychopathology. Because epigenetic and gene expression profiles are highly dynamic (unlike the genome itself), they offer a plausible biological mechanism by which changing exposures over time could affect the course of disorder. Repeated measurement of DNA (preferably from whole blood), exposures, and outcomes could be used to show how epigenetic or gene expression profiles vary over time in response to changes in exposures and whether these track the course of disorder. This work would require careful consideration of the density and timing of measurement to optimize chances of observing hypothesized associations.

Using behavioral genetics to hone our understanding of causal exposures

Developmental psychopathology is concerned with understanding the etiology and course of normal and abnormal development (Sroufe & Rutter, Reference Sroufe and Rutter1984), and this goal is undercut in the absence of information about whether identified risk and protective factors are actually causal. Twin and adoption studies remain relevant because they can serve as natural experiments that provide important information about the causal status of environmental exposures as well as the biological substrates of disorder. The discordant MZ twins design is a particularly powerful tool for identifying risk factors for disease because MZ twins are matched for age, sex, genetic background, and shared family experiences (van Dongen, Slagboom, Draisma, Martin, & Boomsma, Reference van Dongen, Slagboom, Draisma, Martin and Boomsma2012; Vitaro, Brendgen, & Arsenault, Reference Vitaro, Brendgen and Arsenault2009). For example, MZ twins who are discordant for the experience of having been bullied show different profiles of cortisol reactivity to a psychosocial stressor, with the bullied twin in a pair showing a blunted cortisol response in comparison to his or her nonbullied co-twin (Ouellet-Morin et al., Reference Ouellet-Morin, Danese, Bowes, Shakoor, Ambler and Pariante2011). Discordant MZ pairs also provide information about whether changes in gene expression or epigenetic profiles are associated with disease phenotypes (Kuratomi et al., Reference Kuratomi, Iwamoto, Bundo, Kusumi, Kato and Iwata2008; Rosa et al., Reference Rosa, Picchioni, Kalidindi, Loat, Knight and Toulopoulou2008). Although exposure–outcome associations detected in the discordant MZ twins design do not necessarily imply a causal association, the combination of the discordant pairs approach with other quasiexperimental and statistical matching methods does allow for stronger causal inference (Kendler & Gardner, Reference Kendler and Gardner2010).

Like the discordant MZ pairs approach, the children of twins design provides information about the extent to which familial confounding accounts for observed associations between risk factors and outcomes (e.g., D'Onofrio et al., Reference D'Onofrio, Turkheimer, Emery, Maes, Silberg and Eaves2007; Knopik et al., Reference Knopik, Heath, Jacob, Slutske, Bucholz and Madden2006). Adoption studies not only shed light on the degree to which experiences in the family have genetically versus environmentally mediated effects on children's behavior and abilities, but they can also be informative about the unique effects of the prenatal environment on children's outcomes if the child's biological mother reports on her behavior during pregnancy (Neiderhiser et al., Reference Neiderhiser, Leve, Ge, Scaramella, Conger and Reid2007).

Using behavioral genetics to understand the dynamic nature of risk factors for the development of psychopathology

The discipline of developmental psychopathology is intrinsically concerned with transitions between normal and abnormal functioning over time and with identifying factors that either minimize or enhance the likelihood of such transitions occurring. This research effort requires that researchers study individuals and their environments longitudinally in order to link changes in risk or protective factors with changes in functioning. For example, Gerald Patterson's work with the Oregon Social Learning Center showed how escalations in harsh and coercive interactions between parents and children amplified difficult or hard to manage child behaviors into clinically significant conduct problems (Patterson, DeBaryshe, & Ramsey, Reference Patterson, DeBaryshe and Ramsey1989).

Behavioral geneticists can contribute to this research effort by delineating change over time in gene expression and epigenetic profiles. For example, blood samples taken repeatedly over key developmental transitions (e.g., the transition to puberty) and assayed for gene expression or methylation profiles could be informative about the biological correlates of normal or abnormal behaviors that typically emerge during that period (e.g., the emergence of depressive disorder among adolescent girls). Such an approach would be particularly informative in conjunction with efforts to trace changes over the same period in risk and protective factors in the environment that might be regulating the genome or the epigenome (e.g., the presence or absence of socially supportive relationships). Observed associations could be followed up with experimental work to determine whether risk and protective factors were causally related to biological changes. For example, gene expression profiles could be measured before and after randomized controlled cognitive behavioral interventions designed to prevent depression. Such research would be characterized by the limitations outlined earlier, but it could provide a starting point for thinking about how developmental changes at all levels of the individual and his or her environment result in transitions into and out of psychopathology.

Bridging from behavioral genetics to neuroscience and psychology

Understanding the development of normal and abnormal behavior ultimately involves tracing paths from genes to brain to behavior and understanding where and how social experiences intersect these pathways. Behavioral genetic studies that identify associations between genes and behavior provide a starting point in this endeavor. Ultimately, however, delineating the causal mechanisms that underlie the etiology and course of normal and abnormal phenotypes will require both top-down and bottom-up approaches that span a variety of disciplines.

Bridging animal and human models in behavioral genetics research

Although there is a long history of using animal models in behavioral genetics research, this line of work has proceeded largely in parallel with human studies. The field would benefit enormously from a more integrative approach in which teams of researchers could move flexibly between human and animal models.

First, animal studies allow for a level of experimental manipulation that is not possible in human research. For example, consider the epidemiological finding that poor nutrition in pregnancy is associated with a range of psychiatric phenotypes. In humans, this association has been detected in studies that correlate mothers' self-reports of her nutrition and psychiatric symptomatology (controlling for any variables that might confound the association) and in rare circumstances when an entire population has experienced severe nutritional deprivation (e.g., the Dutch Hunger Winter; Susser & Lin, Reference Susser and Lin1992). Animal models allow for greater leverage in identifying causal effects of maternal diet on offspring outcomes because the timing and the quality of prenatal nutrition variables can be manipulated experimentally. For example, the offspring of animals fed a high-fat diet during pregnancy and lactation not only are heavier at birth than the offspring of animals fed a control diet, but they also show differential methylation of opioid and dopaminergic genes in the mesocorticolimbic reward circuitry, which is implicated in the consumption of palatable foods (Vucetic, Kimmel, Totoki, Hollenbeck, & Reyes, Reference Vucetic, Kimmel, Totoki, Hollenbeck and Reyes2010). Another study showed that the offspring of Japanese macaques who were fed a high-fat diet during pregnancy showed differential expression of serotonergic genes in the rostral and caudal dorsal raphe nucleus and elevated rates of anxious behaviors compared with the offspring of mothers fed a control diet (Sullivan et al., Reference Sullivan, Grayson, Takahashi, Robertson, Maier and Bethea2010). Thus, the experimental study of animals affords opportunities to manipulate the timing, duration, and severity of exposures and to identify biological mechanisms that mediate effects of exposures on behavior. Moreover, given shorter gestations and accelerated development in many animal species compared to humans, it is relatively easy to follow animals longitudinally or to do cross-sectional studies of cohorts at different stages of development in order to probe the effects of risk exposures over time and across multiple levels, from the biological to the behavioral.

Second, human and animal studies can be mutually informative. For example, there is a robust association in human epidemiological studies between prenatal exposure to alcohol and adverse motor, perceptual, cognitive, and behavioral outcomes in offspring, most notably fetal alcohol syndrome (Clarren & Smith, Reference Clarren and Smith1978; Huizink & Mulder, Reference Huizink and Mulder2006). These epidemiological findings are bolstered by experimental data from multiple animal models, including mouse, rat, and primate, that identify causal effects of high doses of alcohol in utero (Schneider, Moore, & Adkins, Reference Schneider, Moore and Adkins2011). Moreover, experimental animal studies have extended findings to show effects of prenatal exposure to alcohol on neural development, including dopamine receptor density in primate striatum (Roberts et al., Reference Roberts, Moor, DeJesus, Barnhart, Larson and Mukherjee2004) and multiple effects on the dopamine system as identified in rodent studies (for a review, see Schneider et al., Reference Schneider, Moore and Adkins2011). Follow-up work in humans could test whether the association between alcohol exposure and dopamine receptor density replicates and whether it potentially mediates effects of in utero alcohol exposure on hyperactivity.

The success of efforts to move between animal and human models hinges on identifying outcomes that are translatable across species (e.g., hyperactivity) and on tissues that can be readily interrogated across species; placenta, amniotic fluid, and cord blood exist that may eventually prove to contain informative biomarkers of adverse clinical outcomes in children. These tissues are readily available, and therefore worthy of investigation. The placenta represents a critical interface between the mother and developing fetus, providing nutrients as well as participating in the removal of toxins or waste. Messenger RNA profiling has successfully been used to identify altered gene expression in placentas from babies with altered birth weights (Mannik et al., Reference Mannik, Vaas, Rull, Teesalu, Rebane and Laan2010; Mericq et al., Reference Mericq, Medina, Kakarieka, Marquez, Johnson and Iniguez2009). Amniotic fluid contains cell-free fetal nucleotides (DNA and RNA), which derive from the fetus and which represent a potentially rich source of information on the state of fetal gene expression. Therefore, specific gene expression changes within these tissues may be associated with in utero exposure to alcohol and may also be predictive of adverse neurobehavioral outcomes in the offspring.

Despite promising avenues for research, there are substantial challenges in translating findings across species. Many of the risk factors that are most robustly associated with human psychopathology are multifactorial, such as social disadvantage or stressful life events. In these cases, human researchers working together with animal researchers will need to generate specific hypotheses about the “active ingredients” of these exposures that can be replicated in animal models (e.g., uncontrollable exposure to aversive stimuli, stable and unstable dominance hierarchies).

Interdisciplinary training

Although there are many developmental psychopathology researchers who embrace a biopsychosocial model of development and who are interested in integrating genetics with social and clinical science, it has, historically, been difficult to get interdisciplinary training and to do interdisciplinary research. This issue is increasingly relevant as more and more epidemiological cohort studies with strong social science orientations have collected and banked biological samples from their participants (Relton & Smith, Reference Relton and Smith2012). On many college and university campuses, collaboration between geneticists and developmental psychopathologists is hindered by several factors. Most geneticists do not study psychiatric disorders nor are they epidemiologists. They study the mechanisms of genes that underlie physical health problems, such as cancer or diabetes or cardiovascular disease. Many of them study the genetics of zebra fish, fruit flies, or even yeast and think about the environment as nothing more than a source of error variation. Under these circumstances, developmental psychopathologists and geneticists not only lack common methods, but they also lack a common phenotype, a common understanding of (or interest in) the course of disorder, and, in some cases, a common appreciation for how distal environments like social status or proximal environments like parental care interact with genetic influences on development.

The good news is that interdisciplinary training is increasingly available. Several research centers, such as the Social, Genetic, and Developmental Psychiatry Centre at King's College London, the Institute for Behavioral Genetics at the University of Colorado, Boulder, and the Virginia Institute for Psychiatric and Behavioral Genetics, offer graduate training in quantitative and molecular genetics and include developmental psychopathology researchers among their faculty. There are numerous summer school and training courses for researchers interested in learning more about quantitative behavioral and molecular genetic techniques. Interdisciplinary training, particularly for research that crosses human and animal models of disorder, could be enhanced through the development of NIH pre- and postdoctoral training programs wherein students would have the opportunity to forge links between researchers addressing similar questions from different levels of analysis.

Consortia

A notable feature of genetic epidemiology is the degree to which it is a collaborative enterprise. For meta-analytic studies of GWAS, it is not unusual for a dozen or more research groups to pool their data. This has happened for two reasons. First, researchers themselves have recognized that the sample sizes required for well-powered genetic studies exceed the resources of any one team. Second, academic journals have set increasingly stringent standards for genetic evidence that researchers can meet more successfully by pooling resources. Funding agencies recognize the need to make datasets available in public repositories so that they can be exploited as fully as possible. Meanwhile the industrialization of genetic technologies has given rise to economies of scale, concentrating data collection and analytic expertise in specialist centers whose scientists are well placed to conduct and coordinate consortia.