In the social and clinical sciences, the goal is to understand the causes of normal and abnormal development. Features of the environment, ranging from exposure to toxins to relationships with family, romantic partners, and peers, are hypothesized to be among the causes of multiple disorders. However, a number of recent reviews have identified the challenges in identifying whether these candidate environments play a causal role in complex behaviors (D'Onofrio & Lahey, Reference D'Onofrio and Lahey2010; Foster, Reference Foster2010; Rutter, Reference Rutter2007). Of the many challenges to causal inference, we focus here on genotype–environment correlations. Our goal is to define genotype–environment correlations and then describe efforts to (a) detect genotype–environment correlations; (b) identify whether genotype–environment correlations provide alternative explanations for causal hypotheses; and (c) under certain restrictive conditions, use genotype–environment correlations to make causal inferences about environmental exposures. We illustrate this review with examples from research on harsh, coercive parenting and children's antisocial behavior where there is evidence from observational studies of a robust association (Gershoff, Reference Gershoff2002; Reid, Patterson, & Snyder, Reference Reid, Patterson and Snyder2002; Rothbaum & Weisz, Reference Rothbaum and Weisz1994).
Genotype–Environment Correlation: Definitions
Genotype–environment correlation refers to genetic differences in the exposure to a particular environment (Kendler & Eaves, Reference Kendler and Eaves1986; Plomin, DeFries, & Loehlin, Reference Plomin, DeFries and Loehlin1977). Genotype can be defined as those features of an organism that are encoded in its DNA sequence. Environment, in this context, is typically defined loosely as any nongenetic influence on outcome. Genotype–environment correlation can arise from the causal effects of genotypes and from other processes. Natural selection and genetic drift give rise to ancestral differences in genotype both within and between populations. Genetic markers for population ancestry will therefore correlate with any environmental and phenotypic features that vary by ethnic group. These spurious associations are particularly common in the presence of recent population admixture and are a source of confounding in genetic association studies.
In terms of causal processes that give rise to genotype–environment correlations, psychologists and psychiatrists commonly refer to three different categories. Passive genotype–environment correlation refers to the association between the genotype a child inherits from his or her parents and the environment in which the child is raised. In other words, the existence of passive genotype–environment correlation creates the possibility for confounding of an association between child environment and outcome by genes common to parents and children. For example, because parents who have histories of antisocial behavior (which is moderately heritable) are at elevated risk of abusing their children, maltreatment may be a marker for genetic risk that parents transmit to children rather than a causal risk factor for children's conduct problems (DiLalla & Gottesman, Reference DiLalla and Gottesman1991; Eaves, Prom, & Silberg, Reference Eaves, Prom and Silberg2010). Evocative (or reactive) genotype–environment correlation refers to the association between an individual's genetically influenced behavior and others' reactions to that behavior. For example, although arguing with a spouse may cause someone to become depressed, it is equally plausible that individuals who are prone to depression tend to provoke arguments with significant others (or other stressful life events), calling into question the direction of the effect (Kendler & Gardner, Reference Kendler and Gardner2010). In other words, evocative genotype–environment correlations are evidence of heritable person effects on the behavior of others in their milieu. Active (or selective) genotype–environment correlation refers to the association between an individual's genetic propensities and the environmental niches that individual selects. In other words, active genotype–environment correlations are evidence of heritable person effects on the environment that is due to the person's own actions. For example, individuals who have a genetically based proclivity for antisocial behavior may be especially likely to seek out a delinquent peer group (Kendler, Jacobson, Myers, & Eaves, Reference Kendler, Jacobson, Myers and Eaves2008).
These forms of genotype–environment correlation differ from Gene × Environment (G × E) interaction, which refers to genetic differences in sensitivity to particular environmental effects (Kendler & Eaves, Reference Kendler and Eaves1986). Genotype–environment correlations explain why individuals who have a genetic propensity to engage in sensation-seeking behaviors affiliate with drug-abusing peers (Yanovitzky, Reference Yanovitzky2006). G × E explains why heavy drug use is most likely to lead to psychosis only among individuals with a particular genotype (Caspi et al., Reference Caspi, Moffitt, Cannon, McClay, Murray and Harrington2005).
It bears noting that passive, evocative, and active genotype–environment correlations are special cases of more general phenomena in developmental psychology and psychopathology. For example, passive genotype–environment correlations are a specific case of unobserved heterogeneity, in which genes common to parents and children potentially confound the association between exposure and child outcome. Evocative genotype–environment correlations imply that an individual's behavior actively shapes his or her environment, a process that was described by Bell (Reference Bell1968) in the context of socialization research and was a key tenet of Piagetian theories of cognitive development (Flavell, Reference Flavell1996). The behavioral genetics perspective simply emphasizes that the behaviors that may evoke specific reactions from the environment have a genetic basis. Finally, active genotype–environment correlations imply that individuals consciously or unconsciously seek specific experiences that are congruent with their genetically based proclivities. Again, this echoes processes of selection and niche picking articulated by lifespan developmental theories (Elder, Reference Elder and Lerner1998), but it highlights the genetic basis of the traits driving the selection process.
Genotype–Environment Correlations: Evidence From Quantitative Behavioral Genetics Research
Twin and adoption studies have provided much of the evidence for genotype–environment correlations by demonstrating that putative environmental measures are heritable (Kendler & Baker, Reference Kendler and Baker2007; Plomin & Bergeman, Reference Plomin and Bergeman1991). As reviewed by Kendler and Baker (Reference Kendler and Baker2007), these include many environments that are associated with psychiatric illness, including marital quality, social support, parental discipline and warmth, family environment, and peer relationships. These also include desirable and undesirable life events, including divorce and exposure to trauma. The weighted heritability of these environments ranges from 6% to 39%, with most ranging from 15% to 35% (Kendler & Baker, Reference Kendler and Baker2007).
How one interprets genotype–environment correlations depends on the type of sample included in quantitative genetic studies. When studies involve child twins reporting on their own experiences or a parent reporting on each of her twin's experiences (e.g., of parental discipline), then heritable environments either reflect genetic influences of the child's perceptions of that experience or the extent to which the child's genetic propensities elicit or evoke particular experiences (Ge et al., Reference Ge, Conger, Cadoret, Neiderhiser, Yates and Troughton1996; Jaffee et al., Reference Jaffee, Caspi, Moffitt, Polo-Tomas, Price and Taylor2004; O'Connor, Deater-Deckard, Fulker, Rutter, & Plomin, Reference O'Connor, Deater-Deckard, Fulker, Rutter and Plomin1998). For example, Jaffee et al. (Reference Jaffee, Caspi, Moffitt, Polo-Tomas, Price and Taylor2004) found that a common genetic factor accounted for most of the observed relationship between children's antisocial behavior and their experiences of corporal punishment. This suggests that genetic factors are correlated with spanking and smacking because they give rise to oppositional and antisocial behaviors that evoke physically punitive behaviors in parents. In contrast, when studies involve samples of adult twins reporting on their experiences (e.g., of administering discipline to children), genetic influences on the putative environment reflect the extent to which the adult's genetic propensities alter their perceptions of the experience or modify or create that experience (Wade & Kendler, Reference Wade and Kendler2000).
Environments are heritable because genotype influences the behaviors that evoke, select, and modify features of the environment. Thus, environments less amenable to behavioral modification tend to be less heritable. For example, negative life events that are beyond the control of the individual (e.g., the death of a loved one, losing one's home in a natural disaster) have lower heritability than negative life events that may be dependent on an individual's behavior (e.g., getting a divorce, getting fired from a job; Kendler & Baker, Reference Kendler and Baker2007). Similarly, personal life events (i.e., events that occur directly to an individual) are more highly heritable than network life events (i.e., events that occur to someone within an individual's social network, thus affecting the individual indirectly; Bolinskey, Neale, Jacobson, Prescott, & Kendler, Reference Bolinskey, Neale, Jacobson, Prescott and Kendler2004).
The significance of this research in quantitative behavioral genetics has been to call into question the assumption that putative environments have causal effects on behavior. This work raises the alternative possibility that such environments arise because of heritable characteristics of the person or that the association between exposures and outcomes is confounded by genotype. As we will review later, this research also has implications for efforts to better understand causal processes by exploiting the existence of genotype–environment correlations.
Genotype–Environment Correlation: Evidence From the Molecular Genetic Literature
In contrast to quantitative genetic studies, which infer genetic influences from familial correlations, molecular genetic studies measure genotype directly. When we published our paper in 2007 on genotype–environment correlations (Jaffee & Price, Reference Jaffee and Price2007), these studies were thin on the ground. There remains a relatively small literature, but emerging findings are flagging up a set of issues worthy of further consideration.
Much of the recent work to demonstrate associations between measured genotypes and environments has involved relationships between maternal and child genotype and either reported or observed parenting practices. (As noted in the description of passive genotype–environment correlation, there is more than one genome of interest.) For illustrative purposes, we will focus on this specific topic, although it is far from the only area of environmental research in which genotype–environment correlations may be observed. Researchers interested in genetic influences on parenting have focused primarily on dopamine genes because of the role dopamine has been shown to play in facilitating affiliative behaviors in humans (Barrett & Fleming, Reference Barrett and Fleming2011; Depue & Morrone-Strupinsky, Reference Depue and Morrone-Strupinsky2005) and animals (Numan, Reference Numan2007; Spielewoy et al., Reference Spielewoy, Roubert, Hamon, Nosten-Bertrand, Betancur and Giros2000) and because dopamine genotypes have been associated with externalizing child behaviors (Gizer, Ficks, & Waldman, Reference Gizer, Ficks and Waldman2009; Schmidt, Fox, Rubin, Hu, & Hamer, Reference Schmidt, Fox, Rubin, Hu and Hamer2002) that may evoke suboptimal parenting practices. Researchers have also focused on oxytocin genes because of the demonstrated role of oxytocin in parent–child bonding (Numan, Reference Numan2006; Saltzman & Maestripieri, Reference Saltzman and Maestripieri2011).
Maternal dopamine genotype and parental behavior
Lee and colleagues (2010) identified a nonadditive association between maternal dopamine transporter 1 (DAT1) genotype and observed negative parenting behavior in a sample of 207 mother–son pairs, ranging in age from 3 to 7 years. DAT1 is a 40 base pair variable-number tandem repeat located in the 3′ untranslated region in exon 15 of the dopamine transporter gene (SLC6A3). The 40 base pair sequence is typically repeated 9 or 10 times and the majority of individuals are either homozygous for the 9- or 10-repeat variants (i.e., 9/9 or 10/10) or heterozygous for the 9- and 10-repeat variants (9/10).
In analyses that controlled for parental psychopathology and child disruptive behavior during the parent–child observation, results showed that mothers with the 9/10 genotype engaged in more negative parenting (e.g., using critical and negative statements, using negative physical contact) compared with either homozygous group (9/9 or 10/10), with effect sizes that were small to moderate in magnitude (d = 0.35 for 9/9 vs. 9/10 and d = 0.19 for 10/10 vs. 9/10). Mothers with the 9/9 genotype issued fewer commands to the child compared with mothers who carried the 10-repeat allele (d = 0.33). Maternal genotype was not correlated with positive parenting (e.g., praising child, positive affect, positive physical contact). Moreover, the relationship among maternal genotype, negative parenting, and maternal commands was strongest when children engaged in high levels of disruptive behavior during the parent–child interaction task. When children engaged in lower levels of disruptive behavior (e.g., at the median or lower), levels of negative parenting and the number of commands issued were uniformly low for all maternal genotype groups.
Although main effects of maternal genotype on maternal behavior have been detected infrequently, the finding that the relationship between maternal genotype and maternal behavior is conditional on individual or contextual features is one that has been replicated by other researchers. For example, three studies have shown that the relationship between maternal dopamine receptor D4 (DRD4) genotype and observed maternal sensitivity depends on characteristics of the infant or other features of the mother's environment. One found that mothers who carried the 7-repeat DRD4 allele behaved more sensitively with fussy infants than with nonfussy infants. In contrast, mothers who did not carry any 7-repeat alleles were equally sensitive to their infants, regardless of infant temperament, with this interaction accounting for nearly 3% of the variation in maternal sensitivity (Kaitz et al., Reference Kaitz, Shalev, Sapir, Devor, Samet and Mankuta2010). The second study found that mothers who carried the 7-repeat allele were more sensitive with their 3.5-year-olds if, as newborns, their children had lower versus higher levels of birth risk (as indexed by gestational age, birth weight, and admission to the neonatal intensive care unit). Again, mothers who did not carry any 7-repeat allele were equally sensitive to their children regardless of infant birth risk, with this interaction accounting for 5% of the variation in maternal sensitivity (Fortuna et al., Reference Fortuna, van IJzendoorn, Mankuta, Kaitz, Avinun and Ebstein2011). The third study found that the combined effects of DRD4 and catechol-O-methyltransferase (COMT) genotype on how sensitively mothers engaged with their toddler-age children depended on the level of daily hassles mothers reported experiencing. Mothers who had a combination of COMT and DRD4 genotypes associated with the least efficient dopamine transmission engaged in the least sensitive parenting in the context of high levels of daily hassles but the most sensitive parenting in the context of low levels of daily hassles, with this interaction accounting for approximately 2% of the variation in maternal sensitivity (van IJzendoorn, Bakermans-Kranenburg, & Mesman, Reference van IJzendoorn, Bakermans-Kranenburg and Mesman2008). Taken together, these studies suggest that the behavior of mothers who do not carry the 7-repeat DRD4 allele is relatively consistent, regardless of child characteristics or other features of the environment. In contrast, the behavior of mothers who do carry the 7-repeat allele (perhaps in combination with specific COMT genotypes) is highly responsive to context, with two of the three studies showing that under optimal conditions (e.g., low levels of daily hassles, low levels of infant birth risk), these mothers are more sensitive with their children than 7-repeat-absent mothers, but they are less sensitive than 7-repeat-absent mothers under less optimal conditions.
Child dopamine genotype and parental behavior
At least three studies have estimated the relationship between the dopamine receptor D2 (DRD2) allele in the child and parenting behavior. DRD2 has a restriction length fragment polymorphism called Taq1A (rs1800497) located on chromosome 11q. Taq1A has two alleles, known as A1 and A2 (Eisenberg et al., Reference Eisenberg, MacKillop, Modi, Beauchemin, Dang and Lisman2007). Postmortem studies and studies using positron emission topography have shown that individuals who carry the A1 allele (A1+) have fewer D2 dopamine receptors than individuals who are homozygous for the A2 allele (A1–; Jonsson et al., Reference Jonsson, Nothen, Grunhage, Farde, Nakashima and Propping1999; Noble, Blum, Ritchie, Montgomery, & Sheridan, Reference Noble, Blum, Ritchie, Montgomery and Sheridan1991; Pohjalainen et al., Reference Pohjalainen, Rinne, Nagren, Lehikoinen, Anttila and Syvalahti1998; Thompson et al., Reference Thompson, Thomas, Singleton, Piggott, Lloyd and Perry1997). In addition, children who carry the A1 allele have been shown to be more anxious and to have higher levels of negative emotionality compared with children who are homozygous for the A2 allele (Hayden et al., Reference Hayden, Klein, Dougherty, Olino, Laptook and Dyson2010; Marino et al., Reference Marino, Vanzin, Giorda, Frigerio, Lorusso and Nobile2004; Mills-Koonce et al., Reference Mills-Koonce, Propper, Gariepy, Blair, Garrett-Peters and Cox2007). Thus, Taq1A has been the focus of research on the genetic basis for child effects on parent behavior.
Two studies have shown that A1+ children receive less sensitive and supportive parenting compared with A1– children, as observed during parent–child interaction tasks (Hayden et al., Reference Hayden, Klein, Dougherty, Olino, Laptook and Dyson2010; Mills-Koonce et al., Reference Mills-Koonce, Propper, Gariepy, Blair, Garrett-Peters and Cox2007). Effect sizes were small to moderate in magnitude (d = 0.25 for Hayden et al., Reference Hayden, Klein, Dougherty, Olino, Laptook and Dyson2010; d = 0.40 for Mills-Koonce et al., Reference Mills-Koonce, Propper, Gariepy, Blair, Garrett-Peters and Cox2007). In addition, in their sample of 473 3- to 4-year-old children, Hayden et al. (Reference Hayden, Klein, Dougherty, Olino, Laptook and Dyson2010) showed that the relationship between DRD2 genotype and supportive parenting was partially mediated by child negative emotionality. Although Mills-Koonce et al. (Reference Mills-Koonce, Propper, Gariepy, Blair, Garrett-Peters and Cox2007) found a main effect of child (but not maternal) DRD2 genotype on maternal sensitivity in their sample of 172 mother–infant pairs participating in the Durham Child Health and Development Study, this effect was not moderated by other child, maternal, or family characteristics nor was it mediated by child negativity.
Other dopamine gene variants have been implicated in the genetic basis of child effects on maternal behavior. A separate analysis of children in the Durham Child Health and Development Study showed that DRD4 genotype was also associated with maternal sensitivity; children with the long DRD4 allele (7 or more repeats) experienced lower levels of warm-responsive parenting than children with the short form of the allele (d = 0.34; Propper, Willoughby, Halpern, Carbone, & Cox, Reference Propper, Willoughby, Halpern, Carbone and Cox2007). In a follow-up of 6-year-old children who participated in the study reported by Hayden et al. (Reference Hayden, Klein, Dougherty, Olino, Laptook and Dyson2010), those who carried the DAT1 9-repeat allele experienced more hostility (d = 0.22) and received less guidance (d = 0.28) from their caregivers during a structured task compared with 10/10 homozygotes, and this association was partly explained by elevated levels of negative emotionality directed at the caregiver by children who carried the 9-repeat allele (Hanna et al., Reference Hanna, Sheikh, Laptook, Kim, Singh and Klein2012).
Oppenheimer et al. (Reference Oppenheimer, Hankin, Jenness, Young, Abela and Smolen2011) estimated the association between child genotype and parental sensitivity in a sample of 237 third, sixth, and ninth graders, creating a genetic risk index that comprised DRD2 and COMT (A1 and valine [Val] alleles were identified as the risk variants). Although no main effects of youth genotype on parental sensitivity were identified, there was a significant interaction between youth genotype and parental personality characteristics. When parents scored high on a measure of extraversion and low on a measure of neuroticism, they showed more sensitivity, the more dopamine risk alleles their child had. Conversely, when parents scored low on a measure of extraversion and high on a measure of neuroticism, they showed less sensitivity, the more dopamine risk alleles their child had. These interactions accounted for about 2% of the variation in parental sensitivity. Thus, the extent to which genetically influenced characteristics of youth evoke sensitive or insensitive parenting may depend on characteristics of the parents.
Maternal oxytocin genotype and maternal behavior
Oxytocin is a hormone that plays an important role in facilitating affiliative maternal behavior in animals (Numan, Reference Numan2006; Saltzman & Maestripieri, Reference Saltzman and Maestripieri2011) and is a focus of investigation in a growing number of studies of human maternal behavior (Galbally, Lewis, van IJzendoorn, & Permezel, Reference Galbally, Lewis, van IJzendoorn and Permezel2011; Gordon, Zagoory-Sharon, Leckman, & Feldman, Reference Gordon, Zagoory-Sharon, Leckman and Feldman2010; Naber, van IJzendoorn, Deschamps, van Engeland, & Bakermans-Kranenburg, Reference Naber, van IJzendoorn, Deschamps, van Engeland and Bakermans-Kranenburg2010). For these reasons, researchers have investigated whether common polymorphisms in oxytocin genes are related to maternal sensitivity. In a sample of 176 mothers, the oxytocin receptor gene (OXTR; rs53576) was associated with maternal sensitivity, accounting for 3% of the variation; mothers with the GG genotype behaved more sensitively in observed parent–child interactions with their toddlers than mothers with the AG or AA genotypes (Bakermans-Kranenburg, van IJzendoorn, Pijlman, Mesman, & Juffer, Reference Bakermans-Kranenburg, van IJzendoorn, Pijlman, Mesman and Juffer2008). In an effort to identify physiological mediators of genetic effects, Riem, Pieper, Out, Bakermans-Kranenburg, and van IJzendoorn (Reference Riem, Pieper, Out, Bakermans-Kranenburg and van IJzendoorn2011) exposed 40 women without children to recordings of infant crying bouts and recorded their heart rate. Women with the GG genotype had greater heart rate reactivity to infant cries, but only if they did not have high levels of depressive symptoms, with this interaction accounting for 20%–25% of the variation in autonomic arousal. The authors hypothesized that at least for nondepressed women, high levels of autonomic arousal underlie quick responding to infants' needs.
Summary of findings from molecular genetic studies and future directions for research
Although the field remains small, a growing number of studies have identified correlations between maternal or child genotype and maternal behavior. These associations are typically modest in magnitude, accounting for as little variation in outcome as any single construct typically does in studies of developmental psychopathology. Often these associations are conditional on other features of the interaction, including characteristics of the child, the mother, or the broader context. In some cases, the data are consistent with the notion that some individuals are more susceptible than others to their environments in a “for better and worse” fashion (Fortuna et al., Reference Fortuna, van IJzendoorn, Mankuta, Kaitz, Avinun and Ebstein2011; Kaitz et al., Reference Kaitz, Shalev, Sapir, Devor, Samet and Mankuta2010; Oppenheimer et al., Reference Oppenheimer, Hankin, Jenness, Young, Abela and Smolen2011).
The field of molecular genetic research into genotype–environment correlation faces three major challenges. The first is to produce a body of replicated findings. This goal is undercut by the fact that a great number of researchers who publish papers on measured genes and measured environments still fail to report genotype–environment correlations, possibly because the research is usually focused on identifying G × E interactions. As just one example, in the 10 nonexperimental studies of differential susceptibility to environment reviewed by Bakermans-Kranenburg & Van IJzedoorn (Reference Bakermans-Kranenburg and van IJzendoorn2011), only 5 reported whether child (or maternal) genotype was correlated with the environmental measure. Thus, in many cases it is not clear whether or not a finding has been replicated because the test of genotype–environment correlation has not been reported. Moreover, as has been observed in the literature on Genotype × Environment interaction, efforts to replicate findings have been challenged by the use of heterogeneous measures of the environment (Uher & McGuffin, Reference Uher and McGuffin2010), making it unclear whether a specific Genotype × Environment interaction has not been replicated because the original finding was a false positive or because the replication sample involved a measure of the environment that was substantially different from the measure used in the original sample. In the first instance, replication attempts should involve measures of the environment that are operationalized as closely as possible to the original measure. Once it has been established that the original finding is replicable under these narrow conditions, the field can proceed to test the sensitivity of the genotype–environment correlation to alternative specifications of the environment.
The second major challenge to the molecular genetic study of associations between genotype and environment is for this enterprise to articulate its raison d'être. This research demonstrates an important proof of principle: genes and environments do not just act additively and interactively to influence behavior, cognition, or psychopathology, but genes get, in the words of Richard Dawkins (Reference Dawkins1982), “outside the skin” to affect the environment itself (Kendler, Reference Kendler, Kendler, Jaffee and Romer2011). A critic might take the position that beyond such generalizations this research does not tell us anything we did not know already. That is, the idea that person–environment correlations are important in shaping development has long been recognized in lifespan developmental psychology and developmental psychopathology (Elder, Reference Elder and Lerner1998; Rutter & Pickles, Reference Rutter, Pickles, Wachs and Plomin1991). The primary focus of attention should remain on the specific behaviors that serve to influence environments, not the genotypes that cause those behaviors (Moffitt, Caspi, & Rutter, Reference Moffitt, Caspi and Rutter2005).
However, if molecular genetic studies of genotype–environment correlation do not have immediate translational implications, they may still merit study. There is often a long span of time before the translational benefits of epidemiological and basic research are realized (Rutter & Plomin, Reference Rutter and Plomin2009) and the more immediate payoff may be in a better understanding of the biological basis of behaviors that shape environments crucial for our survival and healthy development. For example, how do we optimize the chances that children will be sensitively nurtured by their caregivers? Many clinicians already screen for difficulties that caregivers might face in providing sensitive care, including a lack of social support, substance use problems, postpartum depression, and domestic violence, although beyond making caregivers aware of relevant services, there is generally little follow-through to ensure that caregivers are actually coping. A current lack of joined-up services aside, imagine that as our understanding of the biological basis of children's difficult temperament advances, researchers can identify a panel of biomarkers that exceeds some threshold of sensitivity and specificity for identifying children who fail to establish regular sleeping or feeding routines, react strongly and negatively to novelty, frequently fuss and cry, and are difficult to soothe. Combining this information about children's biological risk with information about their social circumstances may help clinicians and social service workers identify particularly vulnerable infant–caregiver dyads and provide caregivers with adequate support and training to cope with challenging infants. This combination of screening and parent training could potentially circumvent evocative genotype–environment correlations in which difficult children elicit harsh and insensitive care from adults, and it could potentially do so before clinically significant child behavior problems emerge.
A third challenge in this research is methodological. To date, research on measured genes and measured environments has generally proceeded on a gene variant-by-gene variant basis. In some cases, researchers have combined information from two genes (Oppenheimer et al., Reference Oppenheimer, Hankin, Jenness, Young, Abela and Smolen2011; van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg and Mesman2008). However, the prior probability of an association between a genetic polymorphism and an environment is exceedingly low (Davey Smith et al., Reference Davey Smith, Lawlor, Harbord, Timpson, Day and Ebrahim2007). Genome-wide association studies (GWAS), with their measures of hundreds of thousands of single nucleotide polymorphisms (SNPs) provide the means to test hypotheses about genetic influences over the environment on a scale that is appropriate to our lack of knowledge about how genes act (Butcher & Plomin, Reference Butcher and Plomin2008). These vast yet affordable studies also provide opportunities to conduct pathway analyses and to group genes in functionally similar sets (Sun et al., Reference Sun, Jia, Fanous, van den Oord, Chen and Riley2010) or to aggregate multiple genes of small effect (Purcell et al., Reference Purcell, Wray, Stone, Visscher, O'Donovan and Sullivan2009).
The problem with these approaches is that GWAS often require tens of thousands of participants to generate sufficient statistical power to detect genetic main effects and, consequently, there are practical limitations as to the kinds of environmental measures that can be collected. For example, observational assessments of parents and children would not be feasible in a sample of 20,000 participants. One possible solution is to use the information generated from GWAS of personality traits and behaviors that are hypothesized to mediate relationships between genes and environments. For example, the five factor model of personality can be evaluated with a relatively brief paper-and-pencil instrument that could be administered to large numbers of individuals. These individuals could be genotyped and statistical approaches could be used to generate polygenic scores (i.e., SNP sets) that predicted substantial amounts of variance in personality factors. If one hypothesized that individuals who were high in negative emotionality and low in agreeableness were especially likely to elicit hostile reactions from their romantic partners, then one could genotype those SNP sets in a smaller sample that allowed for intensive observations of interactions between couples and test for associations between the polygenic score and couples' behavior with each other. Several caveats are worth noting. The first is that once researchers move from candidate gene approaches to polygenic scores, the possibility of testing mechanistic hypotheses is eliminated. Because they aggregate information across thousands of SNPs, polygenic scores are no more informative about the mechanisms by which genes influence brain and behavior than “black box” genetic risk estimates derived from family data (e.g., knowledge of a relative's diagnostic status). Moreover, they are usually far less predictive of outcome, although polygenic scores are able to predict disease risk within and between families, unlike family history (Ruderfer et al., Reference Ruderfer, Kirov, Chambert, Moran, Owen and O'Donovan2011). The second caveat is that although GWAS of personality traits are under way, results have been mixed, and variants that reach genome-wide statistical significance generally account for less than 1% of the variation in personality traits (Shifman et al., Reference Shifman, Bhomra, Smiley, Wray, James and Martin2008; Terracciano et al., Reference Terracciano, Sanna, Uda, Deiana, Usala and Busonero2010; van den Oord et al., Reference van den Oord, Kuo, Hartmann, Webb, Moller and Hettema2008; Verweij et al., Reference Verweij, Zietsch, Medland, Gordon, Benyamin and Nyholt2010).
Finally, some of the findings described in this section underscore the complexity of gene–environment interplay: specifically, the coexistence of genotype–environment correlations and Genotype × Environment interactions. For example, in the study conducted by van IJzendoorn et al. (Reference van IJzendoorn, Bakermans-Kranenburg and Mesman2008), mothers with Val/methionine (Met) or Val/Val COMT genotypes reported experiencing more daily hassles than mothers who were homozygous for the Met genotype: a genotype–environment correlation. Moreover, when compared with mothers who were homozygous for the Met allele, their parenting was also more adversely affected by these daily hassles, particularly if they also carried the DRD4 7-repeat allele: a Genotype × Environment interaction. As another example, although mothers who were homozygous for the 9-repeat allele of DAT1 engaged in lower levels of negative parenting than mothers who carried the 10-repeat allele (Lee et al., Reference Lee, Chronis-Tuscano, Keenan, Pelham, Loney and Van Hulle2010), child DAT1 genotype was unassociated with negative parenting in the same sample (Lahey et al., Reference Lahey, Rathouz, Lee, Chronis-Tuscano, Pelham and Waldman2011). However, the adverse effect of negative parenting on boys' conduct problems was strongest when the sons were homozygous for the 9-repeat allele (Lahey et al., Reference Lahey, Rathouz, Lee, Chronis-Tuscano, Pelham and Waldman2011). Here, then, the same genotype that was associated with lower levels of negative parenting was associated with higher levels of child conduct problems, but only when boys were exposed to high levels of negative parenting.
Genotype–Environment Correlations: Confounding Causal Inference
As described in the previous sections, quantitative and molecular genetic studies have identified genotype–environment correlations involving psychosocial risk factors generally, and parenting behaviors specifically. The existence of genotype–environment correlations calls into question the assumption that candidate environments (a) have unidirectional effects on an outcome of interest and (b) are unconfounded by “third” variables, including genotype. Researchers who are interested in making causal inferences about the effects of candidate environments must use research designs that can eliminate genotype–environment correlations as a source of confounding.
The principal method of establishing causality in health-related research is the randomized controlled trial (RCT). Subjects enrolled in the experiment are randomly assigned to treatment groups after matching on criteria of particular importance. In this way, it is possible to exclude the effects of known confounds and make confounding by any other variable (including genotype) an unlikely occurrence. If, for example, in an RCT of treatment for depression, the treated group has significantly fewer symptoms of depression at follow-up than the control group (on the basis of intention to treat), this is interpreted as evidence that treatment caused reductions in depressive symptomatology.
Unfortunately, RCTs have limitations in environmental epidemiology. First, although well-designed RCTs achieve high levels of internal validity, they often lack the external validity achieved in epidemiological research (Chambless & Hollon, Reference Chambless and Hollon1998). Second, although RCTs are often used to make inferences about the role of risk exposures, they randomly assign participants to treatments, not to risk exposures themselves. They may be informative about factors that cause change in symptoms, but the factors that led to the emergence of disorder may have been entirely different (Hinshaw, Reference Hinshaw2002). The vast majority of risk factors that are hypothesized to be involved in the etiology of psychopathology do not lend themselves to experimental manipulation. Ethical concerns preclude researchers from randomly exposing some children to domestic violence and others to more harmonious families or assigning some mothers to smoke during pregnancy and other mothers to abstain.
As a result of the ethical and practical concerns associated with experimental research into common risk factors for psychopathology, researchers have adopted quasiexperimental designs that can assess the degree to which genotype–environment correlations (and other family-wide factors) account for observed relations between risk exposures and outcomes. These include a range of genetically informative designs, including the adoption design, the classical twin design, the Children of Twins (and Extended Children of Twins) design, the Longitudinal Twins and Parents design, and the human cross-fostering design (Rice et al., Reference Rice, Harold, Boivin, Hay, Van den Bree and Thapar2009). We review these designs briefly, as they have been reviewed in detail elsewhere (D'Onofrio et al., Reference D'Onofrio, Turkheimer, Emery, Slutske, Heath and Madden2005; Jaffee, Strait, & Odgers, Reference Jaffee, Strait and Odgers2012). Again, our focus is on the relationship between harsh, coercive parenting practices and children's antisocial behavior.
There is ample evidence from twin and adoption designs that negative parenting is genetically influenced. For example, in studies of child twins, genetic factors account for approximately a fifth to a third of the variation in physical discipline (Jaffee et al., Reference Jaffee, Caspi, Moffitt, Polo-Tomas, Price and Taylor2004; Wade & Kendler, Reference Wade and Kendler2000) and parental harsh discipline (including harsh verbal as well as physical discipline; Button, Lau, Maughan, & Eley, Reference Button, Lau, Maughan and Eley2008; Pike, McGuire, Reiss, Hetherington, & Plomin, Reference Pike, McGuire, Reiss, Hetherington and Plomin1996). In addition, the covariation between harsh discipline and children's antisocial behavior is largely accounted for by a genetic factor common to both phenotypes (Button et al., Reference Button, Lau, Maughan and Eley2008; Jaffee et al., Reference Jaffee, Caspi, Moffitt, Polo-Tomas, Price and Taylor2004; Neiderhiser, Reiss, Hetherington, & Plomin, Reference Neiderhiser, Reiss, Hetherington and Plomin1999). Studies of adoptees have also found that those who are at genetic risk for antisocial behavior (based on their biological parents' history of externalizing problems) experience more harsh and coercive discipline than adoptees who are not at genetic risk (Ge et al., Reference Ge, Conger, Cadoret, Neiderhiser, Yates and Troughton1996; O'Connor et al., Reference O'Connor, Deater-Deckard, Fulker, Rutter and Plomin1998; Riggins-Casper, Cadoret, Knutson, & Langbehn, Reference Riggins-Casper, Cadoret, Knutson and Langbehn2003). Studies of adolescent twins and adoptees have shown that parent–child conflict is also moderately heritable, with genetic factors accounting for just over a third of the variation (Burt, McGue, Krueger, & Iacono, Reference Burt, McGue, Krueger and Iacono2005).
Nevertheless, there is also evidence that despite these genotype–environment correlations, negative parenting has environmentally mediated effects on youth antisocial behavior. Comparisons of twins, adoptive siblings, and children of twins have been used to test whether familial confounding, including genetic confounding, accounts for the relationship between negative parenting and children's antisocial behavior. For example, comparing the behavior of monozygotic (MZ) twins who have experienced different degrees of negative parenting is informative because MZ twins share their family environment; hence, family-wide factors cannot explain observed differences in their behavior. In addition, MZ twins have the same genotype and, presumably, the same genetic propensity to elicit negative parenting, so genotypic differences cannot explain differences in their behavior or in their experience of negative parenting. Studies of MZ twins have shown that twins have higher levels of conduct problems than their co-twins if they also experience more harsh discipline than their co-twins (Asbury, Dunn, Pike, & Plomin, Reference Asbury, Dunn, Pike and Plomin2003; Asbury, Dunn, & Plomin, Reference Asbury, Dunn and Plomin2006; Pike et al., Reference Pike, McGuire, Reiss, Hetherington and Plomin1996); more conflict with their parents, though this effect was observed only in the most discordant pairs (Burt, McGue, Iacono, & Krueger, Reference Burt, McGue, Iacono and Krueger2006); more parental criticism, although this was only found for MZ girls (Carbonneau, Eaves, Silberg, Simonoff, & Rutter, Reference Carbonneau, Eaves, Silberg, Simonoff and Rutter2002); and if they are the twin about whom the parent expresses more negative feelings (Caspi et al., Reference Caspi, Moffitt, Morgan, Rutter, Taylor and Arseneault2004; Larsson, Viding, Rijsdijk, & Plomin, Reference Larsson, Viding, Rijsdijk and Plomin2008). The extent to which within-pair differences in the experience of parenting account for within-pair differences in behavior is modest, accounting for approximately 1% to 3% of the variation in outcome (Asbury et al., Reference Asbury, Dunn, Pike and Plomin2003; Caspi et al., Reference Caspi, Moffitt, Morgan, Rutter, Taylor and Arseneault2004; Larsson et al., Reference Larsson, Viding, Rijsdijk and Plomin2008) but anywhere from 4% to 12% of the variation if the most discordant pairs were selected (Asbury et al., Reference Asbury, Dunn and Plomin2006; Burt et al., Reference Burt, McGue, Iacono and Krueger2006).
Adoption studies can also be used to test whether genotype–environment correlation accounts for the association between negative parenting and offspring antisocial behavior. In nonadoptive families, parents provide their child's rearing environment and their child's genotype. In contrast, in adoptive families, parents only provide their child's rearing environment. Thus, observed associations between parenting behavior and offspring behavior cannot be accounted for by shared genotype. In one study of 610 adoptive and nonadoptive families, higher levels of parent–child conflict and coercive parenting behavior were associated with higher levels of offspring antisocial behavior in both the adoptive and the nonadoptive families and regardless of whether parent and child behaviors were observed or self-reported (Klahr, Rueter, McGue, Iacono, & Burt, Reference Klahr, Rueter, McGue, Iacono and Burt2011). Moreover, although parent–child conflict at one point in time was predictive of adoptee conduct problems approximately 4 years later (accounting for approximately 3% of the variance), the reverse was not true: adoptee conduct problems at Time 1 were not predictive of later parent–offspring conflict (Klahr, McGue, Iacono, & Burt, Reference Klahr, McGue, Iacono and Burt2011).
Finally, the Children of Twins design can be used to test whether shared genotype (or shared family environment) accounts for the relationship between parenting behaviors and offspring outcomes. The Children of Twins design compares the offspring of twins, some of whom will have experienced different family environments. For example, children born to one twin may experience high levels of negative parenting whereas their cousins, in contrast, experience very low levels. Such comparisons are informative because they automatically control for environmental factors shared by twins in the pair (e.g., ethnicity, family of origin) and genetic factors shared by twins (completely, in the case of MZ twins). For example, in the case of MZ twin families, differences among cousins cannot be explained by differences in their genetic risk for conduct problems because the offspring of one twin are as closely genetically related to their aunts/uncles as they are to their mothers/fathers. Thus evidence for a causal effect of negative parenting on children's conduct problems is provided by two sets of findings. First, it must be shown that children who experience relatively lower levels of negative parenting have fewer conduct problems than their cousins who have experienced relatively higher levels of negative parenting. This would suggest that the association between negative parenting and children's conduct problems is not confounded by environmental factors shared by twin siblings. Second, it must be shown that the relationship between negative parenting and children's conduct problems is not significantly stronger in dizygotic (DZ) twin families than in MZ twin families. If the relationship were significantly stronger in DZ twin families, this would suggest the possibility that genetic differences between DZ twin siblings generate a spurious association between negative parenting and children's conduct problems.
Using the Children of Twins design, Lynch, Turkheimer, D'Onofrio, Mendle, and Emery (Reference Lynch, Turkheimer, D'Onofrio, Mendle and Emery2006) showed that children who experienced more harsh physical discipline than their cousins had more externalizing problems than their cousins, accounting for about 4.5% of the variance within twin families, and the magnitude of this effect was no larger in DZ twin families (where genetic confounds are uncontrolled) than in MZ twin families (where genetic confounds are automatically controlled). Thus, findings were consistent with a causal effect of harsh physical discipline on antisocial behavior.
Summary of Findings and Directions for Future Research
Data from a variety of research designs are consistent with the hypothesis that there are reciprocal effects of negative parenting and children's antisocial behavior. That is, although child effects are clearly present and may be driven by genetic factors, negative parenting nevertheless leads to increases in children's antisocial behavior. Thus, successfully reducing harsh disciplinary practices should prevent the emergence of or result in reductions in children's antisocial behavior. It is worth emphasizing that although the methods described in this section are all characterized by certain weaknesses, the consistency of findings across different methods lends them credence. Researchers should be encouraged to test hypotheses about the relationships between risk exposures and outcomes using research designs that are able to disentangle risk factors that typically go together (Rutter, Pickles, Murray, & Eaves, Reference Rutter, Pickles, Murray and Eaves2001; Rutter, Reference Rutter2007).
Using Genotype–Environment Correlations to Understand Causal Mechanisms
As reviewed so far, genotype–environment correlations have been a focus of study by quantitative and molecular genetic researchers who want to demonstrate that putative environments are under genetic influence. In quasiexperimental studies, genotype–environment correlations are treated as potential confounds that must be rejected as alternative explanations for observed associations between exposures and outcomes. A third reason investigators are interested in genotype–environment correlations is because they can be used to determine whether specific exposures (e.g., diet, health-related behaviors, toxins) have causal effects on disease. Genetic associations can be interpreted as causal. The processes that determine one's genotype, which are meiosis and possibly mutation in the parental germ-line cells, followed by fertilization, are causally prior to outcomes. This excludes the possibilities of reverse causality and transactional relationships between genotypes and outcomes like psychopathology. For example, an episode of depression cannot cause changes in the structure of DNA (although it could lead to epigenetic changes). Because the process by which the maternal and paternal chromosomes in each cell are reshuffled during meiosis to produce gametes is random, most kinds of confounding are also excluded. The major confound is population ancestry, the effects of which can be controlled for statistically if genotype data is collected on a sufficient scale (Devlin, Bacanu, & Roeder, Reference Devlin, Bacanu and Roeder2004; Price et al., Reference Price, Patterson, Plenge, Weinblatt, Shadick and Reich2006).
Risk genotypes do not necessarily have direct effects on outcomes, however, and often their effects will be mediated by measurable environments. This phenomenon can sometimes be exploited to show causal environmental effects. Let us imagine that there exists a genetic variant whose biological function mimics that of the environment, or directly influences the degree of exposure. Under certain circumstances, the genotype can act as a proxy variable or “instrument” for the environmental exposure, but one that is untroubled by confounding and reverse causality. As a consequence, association between the genetic proxy and an outcome can be interpreted as evidence for a causal effect of the environment. (It bears noting that the term “environmental exposure” in this context can mean nutritional or toxin exposures or sometimes health behaviors that are more typically studied as outcome variables in psychology and psychiatry.)
Although the concept of instrumental variables and the related statistical methodology originates in the field of econometrics, its application to genetic epidemiology has been dubbed “Mendelian randomization” (MR; Davey Smith & Ebrahim, Reference Davey Smith and Ebrahim2003). The technique is premised on Gregor Mendel's second law, the law of random assortment, that states that traits are inherited independently of each other. A genotype that influences exposure but has an independent effect on the outcome, or is linked to another genotype that does, would confound their association rather than measure it. Mendel's second law does not operate in the presence of pleiotropy, or linkage, and neither does MR. (In technical terms, these are violations of the requirement that an instrumental variable be exogenous: that is, its value is determined by factors that are external to the causal system under study.) These conditions are difficult to guarantee because we do not usually know the full range of effects of any particular genotype, not to mention the many nearby genotypes that may have correlated effects that are due to linkage disequilibrium. In addition, the range of exposures that could be considered for an MR experiment is restricted to those with documented genetic proxies that have reasonably strong effects, and this is currently a very short list.
The aldehyde dehydrogenase 2 (ALDH2) gene has been used as a genetic proxy for alcohol exposure in many studies because it fulfills these assumptions. ALDH2 metabolizes an ethanol by-product, acetaldehyde, into acetate. Homozygotes for the mutant ALDH2*2 allele have negligible ALDH2 activity and, after drinking alcohol, they experience an unpleasant flushing reaction with associated nausea, drowsiness, and headache as a result of acetaldehyde accumulation. Heterozygotes have reduced ALDH2 activity and experience less severe flushing. The ALDH2*2 allele is common in East Asian populations, and there is clear evidence of a correlation between ALDH2 genotype and alcohol exposure, with alcohol intake increasing as a function of the number of ALDH2*1 alleles (Lewis & Smith, Reference Lewis and Smith2005).
Although ALDH2 has primarily been used as a genetic proxy for alcohol exposure in studies of cancer, Irons, McGue, Iacono, and Oetting (Reference Irons, McGue, Iacono and Oetting2007) used MR to test the gateway hypothesis that alcohol use in adolescence leads directly to other forms of drug abuse and disinhibited behavior. Critics of the gateway hypothesis argue that alcohol use (or marijuana use) is correlated with the use of harder drugs because of a general propensity to try drugs and that the reason adolescents tend to try alcohol and marijuana before trying harder drugs is because opportunities to try the former tend to present themselves before opportunities to try the latter (Morral, McCaffrey, & Paddock, Reference Morral, McCaffrey and Paddock2002). MR provides a test of the gateway hypothesis because ALDH2 will be associated with alcohol use, but the law of random assortment means that it is unlikely to be associated with other individual factors (e.g., delinquency, sensation seeking, socioeconomic status) that are also correlated with alcohol use and could explain observed associations between alcohol use and later drug use.
Using data from a sample of 180 East Asian adolescents, they found that ALDH2-deficient adolescents (i.e., those who were homozygous or heterozygous for the ALDH2*2 allele) used alcohol less often than ALDH2 nondeficient adolescents, thus establishing that ALDH2 genotype was associated with alcohol exposure. However, youth who carried the ALDH2*2 allele (i.e., the proxy for low alcohol exposure) did not engage in significantly less lifetime or past-year use of other drugs (including tobacco, marijuana, or other illicit substances) or less disinhibited behavior compared with youth who were homozygous for the ALDH2*1 allele. The findings suggest that alcohol use per se is not a gateway to the use of other drugs or to disinhibited behavior.
MR may have applications in other areas of developmental psychopathology where researchers are interested in the role of dietary factors and toxin exposures in disorders like attention-deficit/hyperactivity disorder (ADHD) and in neurodevelopmental disorders (Froehlich et al., Reference Froehlich, Anixt, Loe, Chirdkiatgumchai, Kuan and Gilman2011; Nigg, Reference Nigg2010). For example, MR could be used to test whether folate supplementation in early pregnancy causes reductions in neurodevelopmental and behavioral problems in early childhood, potentially via epigenetic changes in gene expression involving methylation (Glymour, Veling, & Susser, Reference Glymour, Veling, Susser, Kendler, Jaffee and Romer2011). Folate triggers a metabolic process by which homocysteine is converted to methionine, which plays a role in DNA methylation (Lucock, Reference Lucock2000). MR studies of the association between folate intake and neurodevelopmental outcomes would require the identification of a functional gene variant that mimics the effects of folate intake. With some caveats, the gene methylenetetrahydrofolate reductase (MTHFR) meets this condition. MTHFR has a common variant (MTHFR C677T) that influences the speed at which homocysteine is converted into methionine. The 677TT allele encodes an enzyme that is associated with a slower conversion of 5,10 MTHF to 5 MTHF. Consequently, the conversion of homocysteine into methionine is also slowed, a process that mimics the effects of folate restriction. Thus, MTHFRC677T status can potentially be used as a proxy for folate status, although the validity of this approach has been questioned based on the observation that very high folate concentrations effectively silence the effects of MTHRF677TT (Schatzkin et al., Reference Schatzkin, Abnet, Cross, Gunter, Pfeiffer and Gail2009).
As a final example, there is some evidence that iron deficiency (and a range of other micronutrient and toxin exposures) is associated with ADHD (Froehlich et al., Reference Froehlich, Anixt, Loe, Chirdkiatgumchai, Kuan and Gilman2011). The biological plausibility of this association is bolstered by the observation that iron status affects myelination, dopamine neurotransmission, and cortical fiber conduction (Froehlich et al., Reference Froehlich, Anixt, Loe, Chirdkiatgumchai, Kuan and Gilman2011). Gene variants that influence serum iron concentrations, such as the hemochromatosis allele (HFE C282Y) could serve as genetic proxies for serum iron status (Schatzkin et al., Reference Schatzkin, Abnet, Cross, Gunter, Pfeiffer and Gail2009). Associations between these gene variants and ADHD would be strongly suggestive of a causal role for iron status in ADHD.
MR makes a number of assumptions, some of which may be more defensible than others. Thus, MR, like other quasiexperimental methods, cannot be considered a gold standard, but the findings from studies using MR must be viewed in the context of findings from other research designs that make different assumptions and have different strengths and weaknesses.
As GWAS identify more and more of the genetic determinants of a range of biomarkers, the potential for using MR in the context of developmental psychopathology is likely to expand. However, MR will only be a promising methodological prospect to the extent that researchers understand the biological mechanisms by which specific exposures “get under the skin” to influence behavior. Clearly, the potential for using MR is limited by the nature of the exposure, such that those involving physical effects on the organism are likely to be more empirically tractable than those involving psychological effects. For example, researchers hoping to use MR to test hypotheses about the relationship between abusive parenting or social deprivation and psychopathology are unlikely to identify plausible genetic proxies. Efforts so far to use genome-wide approaches to identify genes associated with social environments have been unsuccessful (Butcher & Plomin, Reference Butcher and Plomin2008).
Conclusions
The significance of genotype–environment interplay is that genetic risk factors do not necessarily have direct effects on disease phenotypes, but may instead modify sensitivity to environmental risk factors or influence exposure to environmental risk factors. Advances in genetic epidemiology are therefore conditional on advances in environmental epidemiology (Moffitt et al., Reference Moffitt, Caspi and Rutter2005). We have noted an important corollary: Advances in genetic epidemiology can also prompt advances in environmental epidemiology.
The recognition that genes get “outside the skin” to shape the environment has important implications for research in developmental psychology and psychopathology: namely, genotype–environment correlations overturn conventional models of the determinants of individual differences in health, behavior, and cognitions. Genes and environments do not always act independently (additively or interactively) to influence phenotypes. Rather, our genes shape our behaviors and cognitions and these, in turn, shape our environments. The possibility that genotype confounds the relationship between exposures and outcomes must be addressed explicitly in developmental science and ruled out as an alternative hypothesis. Otherwise, researchers run the risk of misidentifying the true causal agent in the relationship between a given exposure and a given outcome. A number of research designs are also available that are capable of ruling out genotype–environment correlations as alternative explanations for associations between risk exposures and outcomes. Although many of these designs require genetically sensitive data (e.g., measured genotypes or data from individuals who vary in their genetic relatedness), not all do. For example, studies of within-individual change merely require longitudinal data in order to test the hypothesis that a given event can change a person from one point in time to the next. More analytically sophisticated versions of these models allow researchers to account for simultaneous change in other aspects of the person's environment that, alongside the focal exposure, may change the individual's behavior (Sampson, Laub, & Wimer, Reference Sampson, Laub and Wimer2006). In addition, under some conditions genotype–environment correlations can be leveraged to identify genetic proxies for environments and used to make causal inferences about environmental exposures. In summary, although genotype–environment correlations have received relatively less attention than Genotype × Environment interactions, they are likely to play a key role in development. Taking account of genotype–environment correlations in research on putative environments is likely to enhance our understanding of environmentally mediated causes of individual differences in normal and abnormal development.
