Combined influences of genes, prenatal environment, cortisol, and parenting on the development of children’s internalizing vs. externalizing problems

Abstract

Research suggests that genetic, prenatal, endocrine, and parenting influences across development individually contribute to internalizing and externalizing problems in children. The present study tests the combined contributions of genetic risk for psychopathology, prenatal environments (maternal drug use and internalizing symptoms), child cortisol at age 4.5 years, and overreactive parenting influences across childhood on 6-year-old children’s internalizing and externalizing problems. We used data from an adoption design that included 361 domestically adopted children and their biological and adopted parents prospectively followed from birth. Only parenting influences contributed (independently) to externalizing problems. However, genetic influences were indirectly associated with internalizing problems (through increased prenatal risk and subsequent morning cortisol), and parenting factors were both directly and indirectly associated with internalizing problems (through morning cortisol). Results suggest that prenatal maternal drug use/symptoms and children’s morning cortisol levels are mechanisms of genetic and environmental influences on internalizing problems, but not externalizing problems, in childhood.

Biosocial approaches suggest that hypothalamic-pituitary-adrenal (HPA) axis development and function may be a mechanism of genetic, prenatal, and parenting influences on behavioral development (e.g., Belsky and Pluess, 2009; Del Giudice et al., 2011). Theory and evidence (e.g., Archer, 2006; Burghy et al., 2012; Fries et al., 2005; Kagan et al., 1988; Raine, 1996; Zuckerman, 1979) generally suggest that high HPA activity (most often assessed by levels of cortisol) may lead to more internalizing-type problems (e.g., Scerbo and Kolko, 1994; Smider et al., 2002), whereas children with low HPA activity may be more likely to engage in externalizing behaviors (e.g., Alink et al., 2008). This literature is mixed; in some samples significant effects are found in the opposite directions (e.g., Granger et al., 1998; Fairchild et al., 2008). In order to elucidate the mixed findings for cortisol-behavior associations in the literature, biosocial theorists have posited that combinations of developmental contexts and HPA activity patterns may lead to specific behavioral outcomes (e.g., Del Giudice et al., 2011). For example, a mediating role of neurobiological deficits, including HPA dysregulation, linking early developmental influences (e.g. genetic, prenatal, and early rearing environment) with behavioral development has been hypothesized (e.g., Allen et al., 1998; Gunnar & Fisher, 2006; Huizink et al., 2003; Raine, 2002; van Goozen et al., 2007). Conversely, associations of cortisol and behavior may arise because of shared upstream developmental influences contributing to both. It is critical to test an inclusive, multivariate model in order to examine whether hypothesized mediation effects are present or whether cortisol-behavior associations disappear when controlling for multiple common developmental influences. Here, we test such a model (the conceptual framework for which is presented in Figure 1) using an adoption design ideally suited for separating genetic, prenatal, and parenting influences on cortisol and behavior.

Figure 1.

Conceptual Model. This Figure represents an inclusive, multivariate model capable of testing whether prenatal exposures and morning cortisol mediate genetic influences on internalizing and externalizing problems, and whether morning cortisol mediates prenatal and postnatal environmental influences on internalizing and externalizing problems. There is empirical evidence supporting each labeled pathway (see text). Pathways generally represent positive associations, except path i which represents an inverse association of lower morning cortisol with more externalizing problems.

Genetic Influences on Cortisol and Behavior

Morning cortisol is particularly heritable (Figure 1, path a; e.g., Bartels et al., 2003; Van Hulle et al., 2012). There is also ample evidence from twin, adoption, and intergenerational studies that genetic influences contribute to the development of internalizing and externalizing problems assessed during early and middle childhood (Figure 1, path b; e.g., Bailey et al., 2009; D’Onofrio et al., 2007; Pettit et al., 2008; Rhee and Waldman, 2002; Rice and Thapar, 2009). Importantly, some of the same genes have been associated with cortisol and behavior (e.g., Jabbi et al., 2007; Kreek et al., 2005; Wust et al., 2004) which may indicate that part of cortisol-behavior associations may be attributable to common genetic influences and not causal mechanisms. Now, studies examining if the same genetic factors influence both cortisol and behavior in childhood are needed in order to understand whether or to what extent the association is attributable to shared genetic influences.

Prenatal Influences on Cortisol and Behavior

The same types of prenatal exposures (e.g. maternal substance use and internalizing symptoms during pregnancy) have been linked to both child cortisol (Figure 1, path c) and behavior (Figure 1, path d). The prenatal period is a sensitive period of offspring’s HPA development (Phillips and Jones, 2006). A mother’s biological responses to her environment can shape the fetus’s stress response system so the child is best prepared for the types of environmental challenges he or she is likely to experience after birth (Del Giudice, 2012). For example, adverse prenatal experiences may upwardly calibrate the limbic-hypothalamic arousal threshold such that the child develops higher basal cortisol or increased stress responsivity (Phillips and Jones, 2006; Pluess and Belsky, 2011). Indeed, elevated prenatal risk (i.e., drug exposure, maternal anxiety/depression) has been associated with higher morning cortisol in children even after accounting for obstetric complications and postnatal parental anxiety and depression (e.g., Glover et al., 2010; O’Connor et al., 2005). At the same time, there is evidence for cortisol-lowering effects of prenatal maternal symptoms (Laurent et al., 2012). Associations of prenatal risks and cortisol may be explained through cumulative effects of multiple exposures. Exposure to multiple distinct prenatal risks each result in maternal stress and the release of maternal glucocorticoids (e.g., Giesbrecht et al., 2012) which can influence the child’s HPA development.

These same types of prenatal risks are important and frequently studied predictors of children’s behavior problems (Figure 1, path d; e.g., Allen et al., 1998; McNeil, 1995; Williams and Ross, 2007). For example, substance use (i.e. nicotine, cocaine, alcohol), maternal anxiety and depression symptoms during pregnancy, and neonatal complications have been associated with children’s internalizing (e.g., Gutteling et al., 2005; Mattson and Riley, 2000; Knopik, 2009) and externalizing problems (e.g., Ben Amor et al., 2005; Gutteling et al., 2005; Goldschmidt et al., 2000; Knopik, 2009; Mill and Petronis, 2008). Together, similar types of prenatal exposures may influence both cortisol and behavior problems during childhood. Whether the effect of prenatal exposures operates via endocrine development, or if there are unique effects of prenatal exposures on behavior after accounting for fetal programming of the HPA axis and subsequent associations between cortisol and behavior in childhood remains unknown.

Postnatal Influences on Cortisol and Behavior

Finally, parenting influences have been linked both to children’s cortisol (Figure 1, path e) and behavior (Figure 1, path f). Early postnatal experiences, including negative rearing environments, have a long-lasting impact on the development of the HPA axis (Del Giudice et al., 2011; Essex et al., 2002; Gunnar et al., 2001; Halligan et al., 2004; Heim et al., 2002; Hunter et al., 2011; Tarullo and Gunnar, 2006; see Repetti et al., 2002). This may be because parents can help regulate the child’s perception of stressors, which impacts the activation of the child’s stress response system during childhood (Flinn and England, 1997; Gunnar and Quevedo, 2007). In particular, overreactive and inconsistent parenting has been linked to HPA function in infancy through adolescence (e.g., Azar et al., 2007; Flinn and England, 1997; Hastings et al., 2011; Marceau et al., 2013b).

There is substantial evidence from twin and adoption studies that shared environmental influences (family-level non-genetic factors that influence family members’ similarity) play an important role for internalizing and externalizing problems, especially in middle childhood (see Burt, 2009 for meta-analysis). Parenting is a frequently studied index of shared environmental influences in childhood. Specifically, as with HPA functioning, there is evidence that both mothers’ and fathers’ overreactive parenting and parenting inconsistency are associated with internalizing and externalizing problems across childhood (e.g., Lipscomb et al., 2011; Marchand-Reilly, 2012; McKee et al., 2007). Socialization theorists have historically considered the influence of parenting on child behavior to be indicative of environmental influences (e.g., Hirschi, 1969). However, a limitation of child-based twin studies and non-genetically informed studies examining parenting influences on child behavior problems is that neither can rule out the possibility of passive gene-environment correlation (rGE). That is, associations between parenting and child behavior could arise if parents parent in accordance with their genes, which they also pass on to their child, and those genes subsequently predispose the child to develop behavior problems. We use an adoption design where children were adopted at birth. Here, adoptive parents and the adopted child do not share genes or prenatal influences, so passive rGE is controlled.

Links between Genetic and Prenatal Influences

Most studies examining genetic or prenatal risk in relation to endocrine and behavioral development only assess genetic or prenatal risk and therefore cannot disentangle these influences. Theoretically, a mother’s heritable characteristics such as sensation seeking or substance use problems may put her at risk for experiencing stress or exposing her child to substances during pregnancy, which changes the intrauterine environment. Thus, the child may inherit genes that increase risk of developing behavior problems. As a consequence of heritable maternal characteristics translating to more stressful prenatal environments, the child may also experience a fetal environment that increases risk for behavior problems via a passive gene-(prenatal) environment correlation. Some of the same genes related to the maternal heritable characteristics mentioned above (e.g. genes related to regulating physiological responses to stress and arousal, or metabolizing drugs) are likely to also play a role in the amount of maternal cortisol the fetus is exposed to in utero in response to cumulative stressors. Thus, instead of or in addition to passive gene-(prenatal)environment correlation, there may be a cascading effect whereby genetic influences (some of which the child is likely to inherit) are causally linked to prenatal exposures, which then have an ‘environmental’ effect on the development of the child’s biology and behavior.

There is some empirical evidence that genetic influences are linked to the prenatal environments fetuses experience (Figure 1, path g). Some evidence from twin studies suggests that prenatal exposures (e.g. smoking during pregnancy) do not exert additional risk for internalizing and externalizing problems beyond genetic risk (see Knopik, 2009 for review). More often, prenatal and genetic influences partially overlap when predicting behavior (e.g., Ellingson et al., 2012; Maughan et al., 2004). Evidence from the adoption sample used in the current study suggests that genetic influences are associated with experiencing prenatal risk and that prenatal risk can mediate genetic influences on toddler behavior (Marceau et al., 2013a; Pemberton et al., 2011). However, there is also evidence of unique prenatal influence on children’s behavior (Knopik, 2009). Taken together, this literature suggests that prenatal risk may partially be mechanistically linked to genetic influences on child behavior problems such that genes can in part influence behavior through prenatal environments. These studies highlight the importance of examining prenatal effects in conjunction with genetic influences to understand individual and joint contributions of genetic and prenatal risks for biobehavioral development.

Evidence for the Mediational Hypothesis

The evidence presented here shows that genetic risk is linked with prenatal exposures (Figure 1, path g) and those prenatal exposures can impact HPA development consistent with the fetal programming hypothesis (Figure 1, path c), which may then impact behavioral development (Figure 1, paths h/i). Similarly, postnatal rearing environments have been linked to HPA development (Figure 1, path e), which may then impact behavioral development (Figure 1, paths h/i). Thus, cortisol may be a marker of a physiological mechanism underlying genetic, prenatal, and/or parenting influences on behavior.

To our knowledge, no studies have tested the potential mediating contribution of cortisol in the association between genetic and/or prenatal risk and internalizing problems, and only two studies have tested the potential mediating contribution of cortisol in the association between prenatal risk and externalizing problems. One study found that child cortisol did not mediate associations between retrospectively reported obstetric complications and externalizing problems in 10–12 year old children (Marsman et al., 2009). The second study also found that cortisol stress reactivity did not mediate associations between prenatal and parenting influences on reactive aggression in 8–10 year old children (Ryan et al., 2012). The present study builds on and extends this work by (1) testing whether cortisol mediates genetic influences in addition to prenatal and parenting influences on behavior problems using an adoption design, (2) investigating whether these pathways occur earlier in development (i.e., age 6 years), and (3) testing whether the hypothesized pathways emerge for internalizing and or externalizing problems after accounting for the other type of problem behavior, as internalizing and externalizing problems are often highly comorbid in children (e.g., Zahn-Waxler et al., 2008).

Present Study

Genetic, prenatal, and postnatal environments have been examined in relation to cortisol and child behavior previously in this sample (e.g., Laurent et al., 2012; 2013). However, previous studies have focused on hypotheses of moderation and have not tested multistep developmental pathways, as is the focus of this study. Our overarching hypotheses were that there would be pathways of development including 1) genetic → prenatal risk exposure → morning cortisol predicting children’s internalizing problems (indicated by significant indirect effects comprising paths g, c, and h of the conceptual model, Figure 1); 2) genetic → prenatal risk exposure → morning cortisol predicting children’s externalizing problems (indicated by significant indirect effects comprising paths g, c, and i); 3) parenting → morning cortisol predicting children’s internalizing problems (indicated by significant indirect effects comprising paths d and h); and 4) parenting → morning cortisol predicting children’s externalizing problems (indicated by significant indirect effects comprising paths d and i). We tested these hypotheses by including each of the pathways depicted in the conceptual model (Figure 1, a-i) and examining indirect effects using structural equation modeling.

The adoption design (where children are adopted at birth) is uniquely suited to test models including genetic, prenatal, physiological, and parenting environmental influences across development. Birth mothers and adopted children share genes and the prenatal environment but not postnatal environments, so correlations between birth mother characteristics and child characteristics must be due to genetic or prenatal influences. Carefully measuring birth mother characteristics and prenatal exposures separately can help clarify how genetic and prenatal pathways influence child development (Marceau et al., 2013a), as has been done to distinguish pre- and postnatal influences on child outcomes in traditional family designs (e.g., Marsman et al., 2009; Robinson et al., 2009; Ryan et al., 2012). Adoptive parents and adopted children do not share genes or prenatal environments so correlations of adoptive parent and child characteristics most likely represent postnatal environmental effects (in the absence of selective placement). Therefore, adoption designs are well-suited to examine developmental pathways leading to child behavior problems more cleanly than is possible in families where parents and children share both genes and environments.

Method

Participants and Procedures

Participants for the present study were drawn from the first cohort (N = 361 families) of the Early Growth and Development Study, a multisite longitudinal study of adopted children and their birth and adoptive parents (Leve et al., 2013), consisting of 561 linked birth mothers (BM) and adoptive families. Participants were recruited via 45 adoption agencies in 15 states across the US (see Leve et al., 2013 for further description). Participants were eligible for participation if (1) the adoption was domestic, (2) the child was placed with a non-relative adoptive family (3) and prior to 3 months of age (M = 6.20 days postpartum, SD = 12.45), (4) the child had no known major medical conditions, and (5) the BM and adoptive parents could read or understand English at least at an eighth-grade level. Children were 57% male. Over time, the first cohort of the EGDS sample (with some attrition) participated in a series of in person interviews. BMs were interviewed (usually in their homes) at approximately 4 months (n = 360), 18 months (n = 333) and 4.5 years (n = 318) postpartum. Adoptive families were interviewed when their child was 9 months (n = 358), 18 months (n = 353), 27 months (n = 340), 4.5 years (n = 304), and 6 years (n = 308) old. Information on attrition and missing data are presented below.

Measures

Genetic Risk

We assessed adopted children’s genetic risk for behavior problems by creating composites using information on a) BM psychopathology symptoms, b) BM diagnoses, c) BM age of onset of disorders and d) the proportion of BM’s first degree relatives experiencing problems with psychopathology. Three indices of genetic risk were included: genetic risk for substance use, internalizing, and externalizing problems. Substance use included symptoms, diagnoses, and age of onset of alcohol and drug dependence and abuse and tobacco dependence assessed using the Composite International Diagnostic Interview (CIDI, Kessler and Ustun, 2004) at the 18-month assessment. The proportion of first degree relatives (of mother, father, and up to three siblings) with substance use problems was calculated on a single item, “ever had problems with drugs or alcohol (e.g., drank too much or used drugs on a regular basis, got mean while drinking”. Internalizing problems included symptoms, diagnoses, and age of onset from agoraphobia, agoraphobia without panic, adult separation anxiety, dysthymia, generalized anxiety disorder, major depression, panic disorder, recurrent brief depression, separation anxiety, and social phobia assessed via the CIDI. The proportion of first degree relatives with internalizing problems was calculated on a single item, “ever been diagnosed with depression or anxiety problems that have been treated or recommended for treatment with medication or counseling”. Externalizing problems included symptoms, diagnoses, and age of onset from conduct disorder and antisocial personality assessed via the Diagnostic Interview Schedule (DIS, Robins et al., 1981). The proportion of first degree relatives was calculated on the maximum score the BM rated each relative on two items, “ever had a hot temper, been in fights frequently, or been involved in stealing regularly”, and “ever come into contact with the legal system because of things s/he has done (e.g., been arrested, spent time in jail, had a driver’s license revoked)”.

Composite scores were created by saving factor scores of principal component analyses including each of the four indicators separately for each type of psychopathology in order to create a more robust index of genetic risk in the adoption sample. Using this approach we assume that more symptoms indicate increased severity, and that having symptoms that don’t reach the threshold for disorder also constitutes a (weaker) genetic influence (e.g., Andrews et al., 1990; Levy et al., 1997, also supported by twin studies showing genetic influences on symptom severity in normative populations). We also assume that crossing from symptoms to disorder increases the genetic loading (e.g., Eley, 1997). We assume that earlier ages of onset are associated with higher genetic risk (supported particularly for depression, Cadoret et al., 1977; Levinson, 2006; Turner et al., 1993). Finally, we assume that the number of first degree relatives confers additional genetic risk, or at least is a marker of higher genetic influence (e.g., Hettema et al., 2001; Sullivan et al., 2000).

The principal component analyses yielded a single factor explaining over half of the variance for each type of genetic risk (59.49% for substance use, Eigenvalue = 2.38; 55.12% for internalizing, Eigenvalue = 2.21; 50.64% for externalizing, Eigenvalue = 2.02). Each item loaded strongly onto the genetic risk factor score such that genetic risk was comprised of more diagnoses, symptoms, and first degree relatives with problems, and younger ages of onset (diagnosis count loadings = .84 to .87; symptom count = .83 to .87; proportion of first degree relatives = .51 to 64; age of onset = −.54 to −.72).

Prenatal risk

Prenatal risk exposure was quantified from information BMs provided 4 months after birth on a pregnancy history calendar (adapted from the life history calendar, Caspi et al., 1996) focused on the period around the pregnancy and pregnancy screener focused specifically on the prenatal period. Relevant to this study, BMs reported on use of alcohol, cigarettes, and other drugs during pregnancy, 7 items from the Beck Depression Inventory (BDI; Beck et al., 1996) and 5 items from the Beck Anxiety Inventory (BAI; Beck and Steer, 1993). Responses were scored using a coding system (see Marceau et al., 2013a) based on several risk indices (e.g., McNeil et al., 1994; Kotelchuck, 1994; Williams and Ross, 2007; Van den Bergh et al., 2005).

Anxiety and Depression were the sum of anxiety and depressive symptoms, respectively, that BMs retrospectively reported experiencing during pregnancy. Anxiety and depressive symptoms were scored by creating quartile scores identifying the rank of anxiety or depressive symptoms. The bottom 25% of the sample were given a risk score of 1, 25% to 50% = 2, 50%-75% = 3, 75% − 85% = 4, and 85%-100% = 5. Approximately 30% of the BMs reached risk levels (i.e., more than moderate risk to the fetus, according to the McNeil et al., 1994 scaling criteria). Drug use included serious use of cigarettes, alcohol, eight illegal drugs, and prescription painkillers used illegally. Different weights were given to amounts of different drugs used in accordance with the McNeil-Sjostrom Scale; approximately 37% of BMs reached risk levels. Following the assignment of risk scores for each variable, we created weighted risk sum scores. If the risk score on a variable (i.e., anxiety, depression, cigarettes, alcohol, each drug) was 3 or greater (as per the McNeil-Sjostrom Scale), the risk score = the score received, if the risk score on a variable was 2 or less, the risk score = 0 (no risk). The individual weighted risk scores for anxiety, depression, and each drug were summed to create a final prenatal risk score, since the fetal programming hypothesis posits that more prenatal risk exposure would affect HPA development and subsequent behavioral development cumulatively. 52% of the sample (n = 187) reached minimum risk levels considering internalizing symptoms and drug use together. See Table 1 for sample descriptive statistics.

Table 1.

Means, Standard Deviations, and Correlations Among Study Variables

			BM Characteristics			Prenatal	Overreactive Parenting (9–54mo.)				Children’s Cortisol (4.5yrs)	Outcomes
	Variable	Mean (SD)	Substance Use	INT	EXT	Prenatal Risk	Mothers’ Average	Fathers’ Average	Mothers’ Incons.	Fathers’ Incons.	Morning	EXT	INT
BM Characteristics	Substance Use	0.00 (1.00)	1
	Internalizing	0.00 (1.00)	0.46*	1
	Externalizing	0.00 (1.00)	0.69*	0.41*	1
Prenatal	Prenatal Risk	4.31 (5.94)	0.40*	0.32*	0.35*	1
Overreactive Parenting (9–54mo.)	Mothers’ average	2.12 (.50)	−0.08	0.02	−0.06	−0.05	1
	Fathers’ average	2.11 (.52)	0.01	−0.09†	0.01	−0.02	0.30*	1
	Mothers’ inconsistency	.71 (.39)	0.05	0.03	−0.07	0.05	0.19*	−0.02	1
	Fathers’ inconsistency	.63 (.39)	0.08	−0.03	0.02	0.09†	0.13*	0.27*	0.18*	1
Children’s Cortisol (4.5yrs)	Morning Cortisolμg/dl	0.00 (.03)	−0.06	0.03	−0.001	0.11*	0.12*	0.15*	−0.06	0.07	1
Outcomes (6 yrs)	Externalizing Problems	11.24 (5.55)	0.001	0.02	−0.03	0.08	0.27*	0.17*	0.21*	0.18*	0.01	1
	Internalizing Problems	8.18 (4.78)	0.004	−0.003	−0.05	0.08	0.10†	0.10†	0.14*	0.10†	−0.23*	0.64*

INT = Internalizing; EXT = Externalizing.

^†p < .10

^*p < .05.

N = 361. BM Characteristics are standardized residuals from principal component analyses. Morning cortisol is the intercept from linear growth models (centered at zero).

Adoptive parent overreactive parenting

Adoptive parents’ level of overreactive parenting was measured at the 9 (n =330), 18 (n = 331), and 27 (n = 310) month, and 4.5 (n = 258) year assessments using the overreactivity subscale of The Parenting Scale (Arnold et al., 1993; α’s > .75 for mothers and fathers at each assessment). Higher scores reflected more overreactivity (i.e., displays of anger, meanness, and irritability), and lower scores reflected more appropriate responses to children’s misbehavior. Missing values were imputed (see below), and then mothers’ and fathers’ overall overreactive parenting was computed as the average of each mother’s or father’s four scores over time. Inconsistency of overreactive parenting across childhood was computed as the range (max – min) of the residuals from each individuals’ overall level of overreactive parenting over time, assessing how far from their mean each individual parent fluctuated over time (see Marceau et al., 2013b). See Table 1 for descriptive statistics.

Children’s cortisol

Morning and evening saliva samples were collected from children for three days around the 4.5 year assessment. Children (through their adoptive parents) were instructed to provide samples at 30 minutes after waking in the morning and at bedtime, before brushing their teeth on each of three days. Samples were returned by 70% of families participating at the 4.5 year assessment (n = 210 of 304 families participating in that assessment). Samples were stored by participants and then mailed to the primary study site and frozen until all samples were collected. Then, samples were sent to the University of Trier Laboratory and frozen at −20° C until being used for cortisol immunoassay (DELFIA procedure; see Dressendörfer et al., 1992). Samples were assayed in duplicate, with the average of the useable values taken as the level of cortisol in that sample (see Laurent et al., 2012, for further detail). Children’s morning cortisol was, on average 0.60 μg/dl (SD = 0.19 μg/dl) at 7:38 AM (SD = 43 minutes), and evening cortisol was, on average 0.06 μg/dl (SD = 0.03 μg/dl) at 8:12 PM (SD = 51 minutes).

Values more than 2.5 standard deviations from each assessment (morning or evening) mean were winsorized (replaced with 2.5 standard deviation values, < 4% of values). Cortisol values were set to missing if children used steroid medications on that day. All missing values were imputed (see below). From the repeated measures of cortisol (assessments within days within child) we extracted a measure of children’s average morning cortisol by extracting the intercept from individual-level regressions of the morning and evening cortisol values across the three days of collection (see also Marceau et al., 2013b).

Child behavior problems

At the six-year assessment adoptive mothers and fathers reported on children’s internalizing and externalizing problems using the broadband internalizing and externalizing subscales on the Child Behavior Checklist (Achenbach, 1991; α’s > .88 for mothers and fathers, n = 253). Mother and father reports were combined to reduce systematic rater bias: the maximum score of mothers’ and fathers’ reports on the internalizing and externalizing subscales was used. Raw scores are presented in Table 1. The sample had symptom levels predominantly in the normal range; 6% of children had ‘subclinical’ levels of internalizing and 5% had ‘sub-clinical’ levels of externalizing problems (T scores between 60 and 63); 5.7% of children had ‘clinical’ levels of internalizing problems and 3% of children had ‘clinical’ levels of externalizing problems (T scores > 63) according to the CBCL guidelines. In total, 18% of the sample reached threshold for elevated problems on at least one of the broadband scales.

Control variables

We also included openness/contact in the adoption as a covariate, as openness of the adoption could facilitate associations between birth and adoptive parents (see Ge et al., 2008). Openness of the adoption was assessed at each assessment via BM and AP report on the extent to which they perceived that the adoption was open on a 7-point scale ranging from 1 (very closed) to 7 (very open). The standardized mean of BM and adoptive mother and father reports was used at each assessment (i.e., BM at 4 months and AP at 9 months for the first assessment, both at 18 months for the second, and 4.5 years) except when only adoptive parents were assessed, in which case the standardized mean of adoptive mother and father reports only was used (e.g., 27 months, 6 years).

Missing data

Data were not missing completely at random according to Little’s MCAR test (Little et al., 1988), x²(3734) = 4043.1, p < .05. Therefore, missing data were imputed using SAS PROC MI; 50 datasets were imputed and aggregated (Graham et al., 2007) using a mean function (so the average imputed value for each variable was used in the data preparation steps and hypothesis testing) to reduce bias. Percentages of missing data from the full sample of 361 at each assessment were as follows: birth mother psychopathology symptoms: 8%, adoptive parent 9-month assessment: 9%, adoptive parent 18-month assessment: 8%, adoptive parent 27-month assessment: 14%, adoptive parent 4.5 year assessment: 29%, saliva samples: 42%, 6-year child internalizing and externalizing problems (outcome): 30%. We tested whether a series of demographic variables (adoptive parents’ age when the child was born and income, birth mothers’ age at the child’s date of birth and income, openness of the adoption, child’s ethnicity and sex), as well as study variables (i.e., internalizing, externalizing) assessed at prior waves contributed to whether data were missing (yes/no) at each assessment using a series of Kruskal-Wallace one-way analysis of variance tests. Of 151 tests, only nine reached significance at the p < .05 level, far below chance levels indicating these would not reach threshold for significance controlling for multiple testing. Therefore, we concluded that demographic and study variables were unrelated to missingness. Nonetheless, these demographic variables were used in the imputation model as auxiliary variables.

Analytic Strategy

One structural equation model was conducted using Mplus (Muthen and Muthen, 2004) to simultaneously test the contributions of genetic, prenatal, parenting (mothers’ and fathers’ overall and inconsistency of overreactive parenting) influences, and morning cortisol to children’s internalizing and externalizing problems. Children’s internalizing and externalizing problems were entered into the model simultaneously to account for the other type of problem behavior and test for specificity of influences on each type of problem. A single a priori model (see Table 2 for pathways which were and were not estimated) was fit to the data based on the hypotheses drawn. Specifically, we tested pathways from indexes of genetic to prenatal risk (Figure 1, path g), HPA activity (Figure 1, path a), and child outcome variables (Figure 1, path b), pathways from prenatal risk to HPA activity (Figure 1, path c) and child outcome variables (Figure 1, path d), pathways from parenting variables to HPA activity (Figure 1, path e) and child outcome variables (Figure 1, path f), pathways from HPA activity to internalizing (Figure 1, path h) and externalizing problems (Figure 1, path i), and concurrent associations among variables assessing the same general domain (i.e., among genetic risk indexes, among parenting variables, and between child internalizing and externalizing problems). Associations between genetic risk and parenting and prenatal risk and parenting variables were not included, as these pathways were not hypothesized.

Table 2.

Model Fitting Results

			BM Characteristics			Prenatal	Overreactive Parenting (9–54mo.)				Children’s Cortisol (4.5yrs)	Outcomes
	Variable	R2	Substance Use	INT	EXT	Prenatal Risk	Mothers’ average	Fathers’ average	Mothers’ incons.	Fathers’ incons.	Morning	EXT	INT
BM Characteristics	Substance Use	NE
	Internalizing	NE	.46* (.04)
	Externalizing	NE	.69* (.03)	.41* (.04)
Prenatal	Prenatal Risk	.19*	.26* (.07)	.16* (.05)	.10 (.07)
Overreactive Parenting (9–54mo.)	Mothers’ average	NE	NE	NE	NE	NE
	Fathers’ average	NE	NE	NE	NE	NE	.29* (.05)
	Mothers’ inconsistency	NE	NE	NE	NE	NE	.19* (.05)	−.02 (.05)
	Fathers’ inconsistency	NE	NE	NE	NE	NE	.13* (.05)	.27* (.05)	.18* (.05)
Children’s Cortisol (4.5yrs)	Morning Cortisolμg/dl	.06*	−.16* (.08)	.05 (.06)	−.03 (.07)	.15* (.06)	.09* (.06)	−.08* (.05)	.12* (.05)	.04 (.05)
Outcomes (6 yrs)	Externalizing Problems	.12*	−.02 (.07)	.01 (.06)	−.04 (.07)	−.10† (.06)	.21* (.05)	.10† (.05)	.14* (.05)	.10† (.05)	−.04 (.05)
	Internalizing Problems	.11*	−.02 (.07)	.001 (.06)	−.08 (.07)	.14* (.06)	−.07 (.05)	.11* (.05)	.09† (.05)	.05 (.05)	−.27* (.05)	.64* (.03)

Note. NE = not estimated.

^*p < .05.

N = 361. Standardized beta-weights presented with standard errors in parentheses.

Results

We controlled for openness of the adoption at the current assessment and prior assessments by regressing each score for all study variables on openness of the adoption variables and using the residual scores in hypothesis testing. Zero-order correlations among study variables are presented in Table 1. Children’s internalizing and externalizing problems were highly comorbid (r = .64, p < .05). Generally, correlations supported cortisol and parenting influences on children’s internalizing and/or externalizing problems. Correlations differed slightly for internalizing and externalizing problems. There were more associations of parenting measures with externalizing than internalizing problems, and lower morning cortisol was associated with more internalizing but not externalizing problems.

Model Fitting

The chi-square indicated some misfit, χ² (16) = 27.11, p = .04, however the chi-square is overly conservative in samples over 200. According to standard indices of practical fit, this model fit the data well, CFI = .97, TLI=.93, RMSEA=.04, SRMR = .03 (Figure 2). The full model predicted a significant, but modest, proportion of the variance in internalizing (R² = .11, p < .05) and externalizing problems (R² = .13, p < .05). BM characteristics (i.e., genetic risk) explained a significant proportion of the variance in prenatal risk exposure (R² = .19, p < .05), and genetic, prenatal, and postnatal environmental influences explained a significant proportion of the variance in morning cortisol (R² = .06, p < .05).

Figure 2.

Model Fitting Results. Assessments from which the data were drawn are labeled across the top. OVR = overreactive parenting. Only significant (p < .05) paths are depicted. Solid lines indicate positive associations; hashed lines indicate negative associations. Thick lines depict significant indirect pathways.

Standardized parameter estimates are presented in Table 2; only significant results are presented in the text. Genetic risk for substance use, internalizing, and externalizing problems were positively associated, ψ > .41, SE < .05, p < .05. Genetic risk for substance use and internalizing problems were related to prenatal risk, B > .16, SE < .07, p < .05 (Figure 1, path g). Genetic risk for substance use problems predicted lower morning cortisol in children at age 4.5 years, B = −.16, SE = .08, p < .05 (Figure 1, path a). Genetic risk was not directly associated with internalizing or externalizing problems not supporting Figure 1, path b). Experiencing more prenatal risk predicted higher morning cortisol in children at age 4.5 years, B = .15, SE = .06, p < .05 (Figure 1, path c) and more internalizing problems at age 6 years, B = .14, SE = .06, p < .05 (Figure 1, path d).

Mothers’ and fathers’ overall overreactive parenting and inconsistency across childhood were on the whole all positively associated, albeit modestly, ψ > .12, SE = .05, p < .05, except for the association between fathers’ overall overreactive parenting and mothers’ inconsistency in overreactive parenting. Higher levels of fathers’ overall overreactive parenting from 9 months to 4.5 years predicted higher morning cortisol in children at age 4.5 years, B = .12, SE = .05, p < .05 (Figure 1, path e). Mothers’ and fathers’ overreactive parenting inconsistency across childhood did not predict morning cortisol in children at age 4.5 years (not supporting Figure 1, path e). Higher levels of fathers’ overall overreactive parenting predicted more child internalizing problems at age 6, B = .12, SE = .05, p < .05 (Figure 1, path f). Higher levels of mothers’ and fathers’ (at trend-level) overall overreactive parenting from 9 months to 4.5 years both predicted more externalizing problems in children at age 6 years, as did mothers’ and fathers’ (at trend level) inconsistent overreactive parenting, B > .09, SE = .05, p < .05 (Figure 1, path f). Higher morning cortisol predicted fewer child internalizing problems at age 6 years, B = −.27, SE = .05, p < .05 (opposite of Figure 1, path h), but not externalizing problems (not supporting Figure 1, path i). Child internalizing and externalizing problems were positively associated, ψ = .64, SE = .03, p < .05.

Indirect Effects

All pathways to child externalizing problems were direct, and no indirect effects predicting child externalizing problems were found. Supporting the overarching hypothesis 1), there were several small indirect effects from genetic influences to internalizing problems via prenatal risk and morning cortisol. The indirect path from genetic risk for internalizing via prenatal risk to child internalizing problems was significant, B = .02, SE = .01, p = .05. The indirect path from genetic risk for internalizing via prenatal risk and child morning cortisol to child internalizing problems approached significance, B = −.01, SE = .003, p = .06. There was also a significant indirect effect from genetic risk for substance use via prenatal risk and child morning cortisol to child internalizing problems, B = −.01, SE = .005, p < .05, from genetic risk for substance use via only prenatal risk to child internalizing problems, B = .04, SE = .02, p < .05, and from genetic risk for substance use via only child morning cortisol to child internalizing problems, B = .04, SE = .02, p < .05. Additionally, there was a significant indirect effect from prenatal risk to child internalizing via child morning cortisol, B = −.04, SE = .02, p < .05. There were no indirect paths from genetic risk to externalizing problems through prenatal risk and/or morning cortisol, contrary to hypothesis 2). Supporting hypothesis 3), there was a significant indirect effect from adoptive fathers’ overall overreactive parenting to child internalizing problems via child morning cortisol, B = −.03, SE = .02, p < .05. We found no evidence of indirect effects from parenting to externalizing via morning cortisol, contrary to hypothesis 4).

Discussion

The present study tested whether there were identifiable pathways from genetic risk through prenatal risk and cortisol functioning and from overreactive parenting though cortisol functioning that predicted children’s internalizing and externalizing problems at age 6 years. Broadly, results suggested both genetic and environmental influences on children’s internalizing and environmental influences on children’s externalizing problems. However, the ways in which these influences operated differed based on the type of child problem behavior. For children’s externalizing problems, there were direct effects of parenting influences, whereas genetic and prenatal risk and child morning cortisol were not implicated. However, for children’s internalizing problems, prenatal risk and child morning cortisol transmitted effects of genetic and parenting influences via indirect pathways. There was also some evidence of direct effects of genetic and prenatal risk and parenting on children’s internalizing problems. Thus, a central inference of this study is that developmental pathways including genetic, prenatal, parenting, and cortisol influences can help to identify which children will exhibit internalizing vs. externalizing problems.

Child HPA Functioning and Behavior Problems

Our findings show different developmental pathways leading to internalizing vs. externalizing problems that are observable even in childhood. Our null indirect effects findings for externalizing problems were very consistent with the two previous studies testing whether cortisol mediated associations between prenatal risk and child externalizing problems (Marsman et al., 2009; Ryan et al., 2012). Often, there is no association between cortisol and externalizing problems reported in the literature (Alink et al., 2008). However, there is evidence that the parenting environment moderates associations between cortisol functioning and externalizing problems (Laurent et al., 2012). For example, in another study using the same sample, children with high evening cortisol who also experienced higher levels of parental depression showed the greatest levels of externalizing problems (Laurent et al., 2013). This evidence, and present findings replicating null mediation effects for externalizing problems, points to a contextually moderated role of cortisol in the development of externalizing problems.

For internalizing problems, our findings support the hypotheses that cortisol may be a marker of a physiologically mediated mechanism of genetic, prenatal, and parenting influences on child behavior. Our findings indicate that higher morning cortisol predicted fewer internalizing problems, which is consistent with some (DeBellis et al., 1996; Granger et al., 1998; Laurent et al., 2012), but not all (e.g., Feder et al., 2004; Ruttle et al., 2011) studies of cortisol and internalizing problems. Children in the present sample demonstrated a higher average level of morning cortisol than other ‘normative’ studies (~1 SD higher, e.g., McCarthy et al., 2009), consistent with other at-risk samples, including other adoption designs and children in foster care (e.g., Dozier et al., 2006; Gunnar and Donzella, 2002). This heightened morning cortisol in part may reflect the prenatal risks and genetic influences imparted by birth mothers. Surprisingly, higher morning cortisol appears beneficial in this sample, as it was associated with fewer internalizing problems. This finding may support emerging evolutionary biosocial theories (e.g., DelGuidice et al., 2011; Ellis et al., 2011) suggesting that genetic and prenatal risks organize children’s HPA functioning to be more sensitive to the environment. That is, because the overall environment that children who are adopted receive is generally positive, it may be that these children who are sensitive to the environment are able to take advantage of the positive aspects of the environment and exhibit fewer internalizing problems. However, this speculative explanation is an empirical question necessitating future mediation and moderation analyses including measures of positive parenting.

Prenatal Risk and Child Behavior Problems

Consistent with other findings from this sample (Marceau et al., 2013a; Pemberton et al., 2011), our findings suggest that genetic and prenatal influences on child behavior are associated. This may be due to an overlap between maternal symptom measures (see the limitations section); however, there is corroborating evidence for meaningful associations from other study designs (Knopik, 2009). There is also evidence that prenatal exposures may confer risk via epigenetic mechanisms linked to maternal genetics (i.e., via drug metabolizing pathways, Knopik et al., 2012) which could also contribute to associations between genetic and prenatal risk. In addition to evidence of direct influences of prenatal risk on child internalizing problems, the findings support that prenatal risk exerts influence on children’s behavior through organizational changes in physiology (Bale, 2011; Phillips and Jones, 2006) – in this case cortisol. The prenatal environment has organizing effects on other systems implicated in child outcomes, including the immune system (e.g., Coussons-Read, 2012), and so genetic and prenatal risk may further impact child behavior through unmeasured aspects of children’s physiology.

Genetic and Parenting Influences on Child Behavior Problems

We found no evidence of the direct influence of genetic risk on behavior problems, contrary to many studies of child behavior, including findings from this sample (e.g., Kerr et al., 2013). It may be that when examining multiple biosocial influences, the genetic influences assessed here do not explain sufficient unique variance in this sample at this age. We have previously shown in this sample that prenatal exposures may carry more weight than genetic influences in toddlerhood (e.g., Marceau et al., 2013a; Pemberton et al., 2011); the present findings add to this growing literature.

Finally, the present findings suggest unique roles of overall levels of adoptive mothers’ and fathers’ overreactive parenting across childhood, as well as the added role of inconsistency in mothers’ overreactive parenting. It appears that fathers’ overall overreactive parenting is more salient than mothers’ for six-year-old in terms of both direct and indirect effects on child internalizing problems. However, mothers’ overall and inconsistent over-reacting parenting were each directly associated with children’s externalizing problems, whereas fathers’ parenting only reached trend level, and only in terms of direct effects. Thus, our results suggest that generally, both mothers’ and fathers’ parenting are associated with children’s behavior problems, and further highlight different patterns of direct and indirect effects of overreactive parenting on children’s internalizing and externalizing problems.

Limitations and Future Directions

Several limitations of the present study are important to consider when interpreting the results. First, the children in our sample were primarily Caucasian (70%), and participants were limited to US domestic infant adoptions. Therefore, the generalizability of results may be specific to this population. Specific findings will likely be hard to reproduce due to the uniqueness of the sample and specific measures, especially given the modest effects found here. Future studies are needed to determine if the broad pattern of results holds for other types of adoption samples, and with other types of genetically-informed samples, including twin and sibling studies or studies examining specific genes. Further, most of the youth included in this sample do not have clinically meaningful levels of problem behavior. However, based on birth mother characteristics (genetic risk) and prenatal risk exposure, these youth are considered at-risk for developing higher levels of problems.

There were limitations regarding the methods used in this study. First, we relied solely on parent report to assess parenting. In the future, including observational data on parenting would attenuate any self-report bias. We conducted careful data preparation models to extract theoretically meaningful parameters. However, taking this approach, some parameters likely include more measurement error (i.e., overreactive parenting inconsistency) than others (i.e., overall overreactive parenting; morning cortisol). Importantly, some of the symptoms from birth mothers’ lifetime psychopathology scores may have occurred during the prenatal period, thus inflating the association between genetic and prenatal risk. That is, genetic risk for substance use as measured by birth mother characteristics and chronic use may not be conceptually different enough from her inability to stop using during pregnancy, and thus tap the same construct. This is not an issue unique to the current study, but rather applies to any study attempting to disentangle the influences of prenatal risk exposure and related maternal characteristics (either as genetic influences here or as postnatal environmental influences in traditional family study designs). Future studies using in-vitro designs and/or measuring specific gene variants would be better able to disentangle genetic from prenatal influences. Thus, we assume that some of the effects of birth mother characteristics (genetic risk) affect child behavior via changes in the prenatal environment, and that our constructs are in fact distinct. Statistically, the overlap in genetic and prenatal risk among our measures is controlled, lending support to this assumption, and the limitation in overlapping measurement is attenuated by including multiple indexes in the genetic risk scores (i.e., first degree relatives, age of onset) rather than relying solely on maternal characteristics. Including birth fathers has the potential to further attenuate this potential bias; however, only 30% of birth fathers provided data for psychopathology risk, and between 11 and 22% of families had valid data for psychopathology risk and child outcomes. Thus, we chose not to impute so much data with such limited covariance with key measures informing the imputation model. Finally, a latent framework for genetic influences wherein the three composite genetic risk scores load onto one factor may better fit the data, and both the general and specific indexes of genetic risk may be associated with child behavior problems; these possibilities should be explored in future work.

Whereas null findings for cortisol-mediated effects on externalizing problems are consistent with two previous studies, the present findings for internalizing still need to be replicated. Further, these paths should be extended into adolescence, when changes in hormone functioning have been implicated in the development of behavior problems. We also rely on the timing of assessments to judge the association between morning cortisol and later internalizing problems. The full cross-lagged model including multiple measures of cortisol, internalizing, and externalizing was not tested here because it was judged to add to much additional complexity to an already complex model. The cross-lagged model should be included in future analyses testing a more refined model. Finally, based on evidence that the family context moderates hormone influences on externalizing problems, parenting should be considered as a moderator of the mediation pathways hypothesized for genetic influences on externalizing problems in the future.

Conclusions

Even considering these limitations, the current study takes an important initial step toward understanding the mechanisms of genetic and environmental risk for the development of behavior problems. A key conclusion of our findings is that prenatal risk and HPA functioning may in part be mechanisms of genetic and environmental influences for internalizing but not for externalizing problems during early childhood. Further, we show evidence of the hypothesis that the cumulative effects of prenatal stressors jointly contribute to HPA and subsequent behavioral development. These findings potentially have clinical implications, as they provide preliminary evidence that hormone functioning and prenatal risk profiles may eventually serve as markers that could help to identify which types of problems children are likely to develop, given the presence of specific genetic and environmental risk factors. In the future, studies should continue to combine genetic, prenatal, endocrine, and family environmental influences together to investigate the overlapping and distinct pathways of the development of internalizing and externalizing problems.