Abstract
Hostility often co-occurs in parents and associates with increased aggression and inattention problems in children. In this population-based cohort of 484 mother–father–child neuroimaging trios, we investigated the degree to which associations of prenatal and childhood parental hostility would be associated with maternal, paternal, and child brain structural differences. Also, we examined whether hippocampal volumes of the parents or child mediate the association of prenatal parental hostility with child externalizing behaviors. Maternal and paternal hostility was assessed with the hostility subscale of the Brief Symptom Inventory at three time points: prenatally at 30 weeks’ gestation and when the child was 3 and 10 years old. During adolescence assessment wave (age 14), maternal, paternal, and offspring assessment included a magnetic resonance imaging. Child externalizing problems were assessed with Youth Self-Report Child Behavior Checklist. Our findings suggest that maternal and paternal hostility were each associated with smaller gray matter, white matter, and hippocampal volumes of their own and their partner's brain. Prenatal maternal but not paternal hostility was associated with smaller total gray matter, white matter, and hippocampal volumes in the offspring. The child's hippocampal volumes partially mediated the associations of prenatal parental hostility (latent construct) with adolescent externalizing behavior, even after adjusting for prior child externalizing problems. Moreover, parental psychopathology may have long-lasting neurodevelopmental correlates in children that underlie the intergenerational transmission of behavioral problems. The behavior of family members results from a system of interdependent dyadic relationships over time that associate with specific brain structural differences.
Significance Statement
Parental hostility often co-occurs in the parents. Research suggests that what transpires in one family subsystem, e.g., hostility among parents, is related to what transpires in other subsystems, e.g., mother–child or father–child, and can negatively impact child development. To understand the neurobiological effects of parental hostility on the families, these can best be studied with trio analysis as parents and children may all be affected. Overall, the findings elucidate how hostility of a parent negatively relates to different family subsystems and associated brain characteristic, such as the hippocampal volume. Our findings suggest that the behavior of family members results from a system of interdependent dyadic relationships over time that associate with specific brain structural differences.
Introduction
Parental hostility is defined as overt behavior and communication between parents and children that is characterized by arguing, angry comments, contempt, yelling, name-calling, and physical aggression (Cummings et al., 1989; Davies et al., 2012). Hostile behavior of a parent is associated with many aspects of child and family functioning including high levels of parental conflict, depression, and poor parental–child relationship quality. Reviews of the literature within the framework of familial high-risk environments have shown that aggression and disengagement in the interparental or parent–child context are risk factors for the persistence of negative parent and child mental health outcomes (Repetti et al., 2011; Davies and Sturge-Apple, 2014). Children exposed to higher levels of parental hostility also appear to live in a state of chronic psychological stress (Gottman and Katz, 1989), which may lead to disruption of brain development during childhood and beyond (Hanson et al., 2015; Chad-Friedman et al., 2021).
Parental hostility often co-occurs in the parents (Low and Stocker, 2005). Research suggests that what transpires in one family subsystem, e.g., hostility among parents, is related to what transpires in other subsystems, e.g., mother–child or father–child, and can negatively impact child development (Schermerhorn and Cummings, 2008). A large body of evidence shows that family relationships function as unitary systems and are built on the ongoing transactions between family members (Schermerhorn and Cummings, 2008; Peltz et al., 2018). Childhood behavioral problems, and in particular aggression and rule breaking behavior, are often the result of interparental conflict and interrelated parental psychopathology (Erel and Burman, 1995). Although mothers and fathers have different kinds of relationships with children that evoke different behaviors (Achenbach, 2006), one parent's hostility is thus likely to affect all members of the family and relationships within family.
Functional imaging studies suggest that stress and psychopathology in mothers after the birth of a child correlate with pronounced long-term changes in the mother's brain (Feldman, 2015; P. Kim et al., 2016). A recent study that assessed brain-to-brain synchrony between mothers and their children at age 3–4 and its association with stress suggests that higher parenting stress experienced by mothers is associated with reduced brain-to-brain synchrony (Azhari et al., 2019). Research is beginning to address the effects of parenthood on the father's brain, suggesting similar responses as found in mothers including sensory/motivation, reflexive caring, emotion regulation, and social cognitive networks (Swain et al., 2014).
In familial high-risk environments, parents and children often experience some form of psychopathology, and they are likely to exhibit similar neuroendocrine, immunological, and cardiovascular correlates of persistent stress (Repetti et al., 2002). In particular the evidence that persistent stress can affect parents and children is consistent (Thompson, 2014). These environmental and biological changes that occur in pregnancy and early childhood could be accompanied by maternal, paternal, and offspring brain differences in specific brain areas such as limbic and frontotemporal brain regions (Feldman, 2015). To understand the neurobiological effects of parental hostility on the families, these can best be studied with trio analysis as parents and children may all be affected.
Designed within a general pediatric population setting, we examined associations of prenatal and childhood parental hostility would be associated with differences in maternal, paternal, and child brain structure if analyzed together, i.e., as trios. The design enabled us to utilize a within-family-based approach that controls for family effects such as confounding by genetic or environmental (e.g., parenting or health behavior) in population-based studies. First, we aimed to examine the associations of parental hostility measured repeatedly over time with child brain development. Second, we examined whether each parent's hostility was associated with parental brain morphology (within-parent analyses) or with differences in the partner's brain morphology (across-parent analyses). Our third aim was to quantify the extent to which the prospective relation of prenatal parental hostility with adolescent externalizing behavior is mediated by subcortical brain volumes of mothers, fathers, or children.
Materials and Methods
Participants
Data were collected in the Generation R Study, a multiethnic population-based cohort from fetal life onward. The Generation R Study has been described in detail previously (Kooijman et al., 2016). Briefly, all pregnant women living in Rotterdam, the Netherlands, with an expected delivery date between April 2002 and January 2006 were invited to participate. Of the 8,879 women included during pregnancy, we excluded 1,763 mothers with no partner data and 490 with missing parental hostility data, leaving 6,626 eligible mother–child pairs with 4,489 actively participating fathers. We randomly excluded 32 siblings and selected only one sibling for inclusion in these cases. Neuroimaging data were obtained for 3,992 children from the late childhood (mean child age 10 years) assessment wave from 2012 to 2015 (White et al., 2018b). We excluded 657 children with poor imaging data (n = 638) quality or incidental findings (n = 19; Muetzel et al., 2019). We additionally excluded 1,231 children with no hostility data, leaving 2,104 eligible mother–child pairs. Between 2017 and 2020, neuroimaging data for parents were obtained as part of the ORACLE project, a subsample cohort of Generation R (Lamballais et al., 2021). T1-weighted data were available for 2,083 parents (1,362 mothers, 721 fathers/partners). We excluded 48 parents (12 mothers, and 36 fathers) with insufficient image quality and 13 parents with incidental findings. Of the remaining 1,337 mothers and 672 fathers, those with no data on parental hostility or missing partner or missing child imaging (853 mothers, 188 fathers) were excluded. We excluded an additional 963 children with no maternal or paternal imaging data. That is, all individuals were nested within families. The final sample consisted of 484 mother–father–child trios with mother, father, and their child that all had neuroimaging data.
The study was conducted in accordance with the guidelines as set by the World Medical Association Declaration of Helsinki. The study was approved by the Medical Ethics Committee of the Erasmus Medical Center, Rotterdam (registration number MEC 02.1015). Written informed consent was obtained from all adult participants and from both the parents of minors.
Parental hostility
Parental hostility was assessed with the Brief Symptom Inventory (BSI). Mothers and fathers reported their hostile symptoms at 20 weeks’ pregnancy and again when their child was 3 and 10 years old. The BSI is a widely used instrument to measure self-reported psychological symptoms in samples of psychiatric patients and community nonpatients (Derogatis and Melisaratos, 1983). We used the hostility subscale of the BSI, a validated self-report questionnaire answered on a five-point scale, ranging from 0, “not at all,” to 4, “extremely” (Derogatis and Melisaratos, 1983; de Beurs, 2004). This instrument encompasses symptom dimensions covering clinically relevant psychiatric and psychosomatic symptoms (Derogatis and Melisaratos, 1983). The hostility subscale consists of five items: “easily becoming bored or feeling irritable,” “uncontrollable bursts of anger,” “an urge to hit, injure or cause pain to others,” “an urge to damage or break things,” and “often getting involved in arguments.” High validity and reliability were reported for the Dutch translation (De Beurs and Zitman, 2005). In the current study, internal consistencies (Cronbach's alpha) for parental hostility ranged from 0.60 to 0.71.
Child externalizing behavior
The 118-item Youth Self-Report for older children (YSR/11-18; Achenbach and Rescorla, 2001b; Achenbach et al., 2011) was used to obtain standardized child reports of problem behaviors. Similar to the Child Behavior Checklist (CBCL/6-18), the YSR/11-18 yields syndrome scales and the three broadband scales Internalizing and Externalizing, and Total Problems with ratings based on the preceding 6 months (Achenbach et al., 2011). Each item is scored on a three-point rating scale, 0, “not true”; 1, “somewhat or sometimes true”; and 2, “very true or often true,” based on the preceding 2 months. Good reliability and validity have been reported for the CBCL/6-18 and YSR (Achenbach and Rescorla, 2000, 2001a). We will use the continuous Externalizing Problems score as our outcome measures. In the current study, the internal consistency (Cronbach's alpha) for externalizing problems was 0.85.
Image acquisition
Neuroimaging data were acquired with a 3 T GE Discovery MR750w MRI System (General Electric) and an 8-channel receive-only head coil (White et al., 2018b). High-resolution, T1-weighted structural MRI data were acquired using a coronal inversion recovery fast spoiled gradient recalled sequence. Structural MRI data were processed through the FreeSurfer analysis suite, version 6.0 (Fischl et al., 2004; Athinoula A. Martinos Center for Biomedical Imaging). T1-weighted images were acquired in the children with a coronal inversion recovery fast spoiled gradient recalled sequence (TR, 8.77 ms; TE, 3.4 ms; TI, 600 ms; flip angle, 10°; field of view, 220 mm × 220 mm; acquisition matrix, 220 × 220; slice thickness, 1 mm; number of slices, 230). The parental images were collected with a similar sequence but with an axial orientation (Lamballais et al., 2021).
Image processing
The T1-weighted images were processed through the FreeSurfer analysis suite, version 6.0. The details of the processing steps for the child data have been described elsewhere (Muetzel et al., 2019), and the parental images were processed with the exact same protocol. Briefly, nonbrain tissue was removed, voxel intensities were normalized for B1 inhomogeneity, whole-brain tissue segmentation was performed, and a surface-based model of the cortex was reconstructed. In our group, we have developed a metric of image quality which automatically characterizes the amount of motion/artifact based on signal intensities outside of the brain (White et al., 2018a). In the image processing, we additionally control for a metric for that described motion artifact and quality. Global metrics of volume were extracted (e.g., total brain volume and subcortical volume), and subcortical and cortical structures (amygdala, orbitofrontal cortex, etc.) were automatically labeled. The averaged left and right hemispheres for all measures were used in primary analyses. Surface reconstructions were visually inspected for accuracy and data not suitable for statistical analysis were excluded (Muetzel et al., 2018).
Covariates
Maternal and paternal age was assessed at intake in the study. Child age at MRI (based on date of birth) and sex were obtained from birth records. Parental national origin was categorized into three groups: Dutch, other Western, and non-Western national origin (Netherlands Statistics, 2006). Mothers were asked about their country of origin at enrollment. Given the diversity of Rotterdam and its surrounding area, participants reported a wide spectrum of ethnic backgrounds, including Dutch, Moroccan, Turkish, Dutch Antillean, Surinamese, Cape Verdean, and several other non-Dutch origins. We operationalized these data following prior work in the Generation R Study as a three-category variable we refer to as “maternal or paternal national origin.” Categories include Dutch and non-Dutch [European (non-Turkish), Turkish, Moroccan, Surinamese] and Other National Origin. Further, the associations remained when we accounted for six additional national origin categories (results not shown). Parental education was classified in three levels: “low” (maximum of 3 years general secondary school); “medium” (>3 years general secondary school; intermediate vocational training); and “high” (bachelor's degree or higher academic education). Information about smoking during pregnancy (three categories: no smoking; smoked until pregnancy recognized; and continued smoking), alcohol intake during pregnancy [four categories: no alcohol consumption; alcohol consumption until pregnancy recognized; continued occasionally (<1 glass/week); and continued frequently (≥1 glass/week)], and parity was assessed prenatally using self-report questionnaires. With our data, we were not able to differentiate children who were exposed to multiple separation/divorces from those exposed to a single such event. For our study, different partner/informant was included for all subsequent time points.
Statistical analysis
Trio design consideration
To study the associations of parental hostility and structural brain morphology of both parent and children, we treated mother, father, and child as a family unit, i.e., a trio (Kenny et al., 2020). The parent–child trio design (nested family or group level design) has some important advantages for studies of family relationships such as psychopathology effects, which are both complex and heterogeneous in their etiology. First, the design is used to examine “nonindependence.” Nonindependence refers to when three scores from the three family members (mother, father, and child) of the trio are more similar to (or different from) one another than are three scores from three people who are not members of the same trio (Kenny et al., 2020). Statistically, nonindependence is captured by making the trio rather the individual the most important unit of analysis.
For example, there are three processes that may produce nonindependence between the two individuals in one family: (1) a partner effect that occurs when a behavior (e.g., aggression) of one parent affects his or her partner's and child outcomes; (2) mutual influence occurs when both parents’ aggression directly affect one another and the child; and (3) joint confounding that occur when both parents (family members) are exposed to the same causal factors. Second, all the data are analyzed in one analysis. However, we did not combine maternal, paternal, and child brain outcomes in one family score. Rather, we used distinct variables, and thus not averaged scores of dyads, to test whether the two hostility scores differ within families. Thus, each parent is paired with partner and child. This analysis takes into account the effects of subsystem dynamics, such that correlation values between mother, father, and child brain morphology were used to adjust similarity between family members (within family correlations) in each model.
In addition to correlation between parental hostility over time measuring similarity of the parents, we computed the within-family interactions between the parent's own aggression (X1) and the partner's aggression (X2) over time with maternal, paternal, and child brain structures (Kenny et al., 2020). That is, we tested whether parental hostility over time vary significantly across families. These interactions are carefully adjusted for confounders that can induce spurious similarities.
Next, we examined how similarity between parental hostility over time and maternal, paternal, and child brain structures differs across families. Mother–father–child correlation group coefficients of the repeated measures of parental hostility, total gray matter, white matter, hippocampus, and amygdala volumes were compared to detect whether the correlations from the same family differ in magnitude to other family members.
The structural model effect estimates used in the current study take multiple sources of variance into account by dividing the variance to be explained across the family members. This allows to examine the extent to which one parent's hostility is associated independently of the other parent's in relation to each maternal, paternal, and child brain structures. Analyses defined the parental hostility measures paired with each other as independent variables and maternal, paternal, and children brain measures (including total white matter and gray matter volumes, hippocampus, or amygdala volumes) as the dependent variables (i.e., all individuals were nested within families). If we observed an association with any of these brain structures, subsequent analyses of substructures were conducted to facilitate interpretation of results obtained with the primary outcome measures.
Structural models
We then used structural equation model (SEM) path analysis to test whether maternal and paternal hostility from pregnancy onward is associated with maternal, paternal, and child brain morphology. The general strategy in specifying path models with triadic data is that each parent's hostility score is used as predictor variable for the brain structures of themselves, their partner, and the child. This allows for the estimation of each of the different variances of the associations within the family (mother–child, father–child, and mother–father). The model included paths from maternal and paternal hostility (e.g., prenatal, age 3, and age 10) to both parent's and child brain measures. This model addresses whether there are differences between each family member. Then two sets of models were constrained to be equal for mothers and fathers to test the model fit. Basically, parallel SEMs are created for each family member, and correlations across members are added to model nonindependence. That is, we correlate all outcome measures across parents and children, and they can be viewed as correlations between mother, father, and children, controlling for both parents’ hostility exposures. In addition, this model addresses whether there are differences between each family member. This is useful when dealing with the interdependence or statistical nonindependence of triadic data (484 families) with 1,452 distinguishable members in the family. We performed all models adjusting for the confounders described above and intracranial volume (ICV). Next, in addition to the three main maternal ethnic groups (i.e., Dutch, non-Dutch Western, and non-Western), we further explored whether the maternal-reported hostility on child brain morphology was similar across maternal national origin subgroups as our cohort is a multiethnic sample. These analyses were performed in six ethnic subgroups: Dutch, non-Dutch Western, non-Western—Caribbean, Moroccan/Turkish, African, Indonesian, and others.
Three indices were used to assess the model fit. Root mean square error of approximation (RMSEA) ≤0.05 and the comparative fit index (CFI) ≥0.90 examine the general fit in the SEMs. When comparing the estimated SEMs, goodness of fit was also evaluated using the chi-square statistic (χ2; Kline, 2015). Models with different intercept and slope estimates for each parent and children fitted the data best (Fig. 1, χ2 = 45.27, p = 0.61, RMSEA = 0.024, CFI = 0.92; Fig. 2, χ2 = 56.4, p = 0.31, RMSEA = 0.010, CFI = 0.94; Fig. 3, χ2 = 41.5, p = 0.16, RMSEA = 0.018, CFI = 0.94).
Association between parental hostility with gray matter volumes of parents’ and child. a, Correlation between maternal and paternal hostility prenatal, when child was 3 and 10 years old. The dark blue color indicates higher values were significantly correlated with maternal and paternal hostility score. b, Each plot depicts the association between parental hostility (mother and father prenatal, child age 3 and 10) with gray matter volumes of mother, father, and child measured using magnetic resonance imaging (MRI). The greatest effect size was found for the association between prenatal mother-reported hostility and child gray matter volume. We found that maternal and paternal hostility measured during pregnancy and early childhood was associated with gray matter volumes of the respective parent (i.e., within-parent analyses), wherein the strongest effect sizes for each parent were observed at child age 10. Maternal and paternal hostility scores associated with their partner's gray matter volumes (across-parent analysis). Models are adjusted for child age at brain MRI scan, sex assigned at birth, and age, national origin, education, parity, and smoking and alcohol consumption reported by mother and father. Global brain measures are not adjusted for total ICV. Number of tests correction for multiple testing = 24 (4 outcomes and the 3 relevant exposure periods). Critical value for FDR p < 0.05. * Survived multiple testing correction (includes only within-parent and child analysis; FDR p < 0.05 corrected). c, The adjusted correlation between child, mother, and father gray matter volumes adjusting for both parents’ hostility exposures.
Association between parental hostility with white matter volumes of parents’ and child. a, Adjusted correlation between maternal and paternal hostility prenatal, when child was 3 and 10 years old. All measures were significantly correlated. The dark purple color indicates higher values were significantly correlated with maternal and paternal hostility score. b, Each plot depicts the association between parental hostility (mother and father prenatal, child ages 3 and 10) with white matter volumes of mother, father, and child measured using magnetic resonance imaging (MRI). The greatest effect size was found for the association between prenatal mother-reported hostility and child white matter volume. We found that maternal and paternal hostility measured during pregnancy and early childhood was associated with white matter volumes of the respective parent (i.e., within-parent analyses), wherein the strongest effect sizes for each parent were observed at child age 10. Maternal and paternal hostility scores associated with their partner's white matter volumes (across-parent analysis). Models are adjusted for child age at brain MRI scan, sex assigned at birth, and age, national origin, education, parity, and smoking and alcohol consumption reported by mother and father. Global brain measures are not adjusted for total ICV. Number of tests correction for multiple testing = 24 (4 outcomes and the 3 relevant exposure periods). Critical value for FDR p < 0.05. * Survived multiple testing correction (includes only within-parent and child analysis; FDR p < 0.05 corrected). c, Depicts the adjusted correlation between child, mother, and father white matter volumes adjusting for both parents’ hostility exposures.
Association between parental hostility with hippocampus volumes of parents and child. a, Adjusted correlation between maternal and paternal hostility prenatal, when child was 3 and 10 years old with hippocampal volumes. All measures were significantly correlated with the hippocampal volumes. The dark green color indicates higher values were significantly correlated with maternal and paternal hostility score. b, Each plot depicts the association between parental hostility (mother and father prenatal, child ages 3 and 10) with hippocampal volumes of mother, father, and child measured using magnetic resonance imaging (MRI). The greatest effect size was found for the association between prenatal mother-reported hostility and child hippocampal volumes. We found that maternal and paternal hostility measured during pregnancy and early childhood was associated with hippocampal volumes of the respective parent (i.e., within-parent analyses). Maternal and paternal hostility scores associated with their partner's hippocampal volumes (across-parent analysis). Models are adjusted for child age at brain MRI scan, sex assigned at birth, total ICV, and age, national origin, education, parity, and smoking and alcohol consumption reported by mother and father. Number of tests correction for multiple testing = 24 (4 outcomes and the 3 relevant exposure periods). Critical value for FDR p < 0.05. * Survived multiple testing correction (includes only within-parent and child analysis; FDR p < 0.05 corrected). c, The adjusted correlation between child, mother and father hippocampal volumes adjusting for both parents’ hostility exposures.
Latent construct
As hostility is measured in both, we used a latent construct in the mediation model combining maternal and paternal hostility. Latent factor analysis on maternal and paternal-reported parental hostility was modeled as latent variable via common confirmatory factor analytic (CFA) methods. The models were allowed to correlate and were estimated with the robust maximum likelihood estimator using standardized latent variables. The latent parental hostility factor was used in mediation model to test whether the associations between prenatal parental hostility factor and child externalizing behavior was mediated by child hippocampal volume. The association between the latent construct of parental hostility with child hippocampal volumes and externalizing behavior captures covariation across raters or the extent to which a given dimension is reflected both across parents (i.e., a “between-rater” dimension factor). The latent constructs showed good model fit as judged with the CFI (acceptable fit ≥0.90; McDonald and Ho, 2002). The goodness of fit of these models was compared with the Bayesian information criterion (BIC) and Akaike's information criterion (AIC). A lower value for AIC and BIC indicates a better fit (Hu and Bentler, 1999).
Mediation model
As psychopathology interferes with children's and parents' abilities to regulate stress physiology and may be associated with disruption in typical brain development, we finally examined whether brain morphology of mothers, fathers, or children mediated the association between prenatal maternal- and paternal-reported hostility and child externalizing behavior at age 14 years. The mediation processes in trios allow an efficient evaluation if the hostility measure in both parents can be expected to influence both parents and children. Based on the results, we only examined mediation for the hippocampus in the association between maternal and paternal hostility and adolescent externalizing behavior. We also examined if the association of parental hostility (jointly measured in a latent construct) with adolescent externalizing behavior was mediated by the hippocampus. Mediation analysis provides estimates of the natural direct effect, the natural indirect effect, and the total effect (Valeri and Vanderweele, 2013). All models were adjusted for baseline confounders, total ICV (global brain measures are not adjusted for total ICV), and child externalizing behavior when the child was 1.5 years old to help rule out a reverse association (VanderWeele et al., 2020).
Adjustment for multiple comparisons was made using the Benjamini–Hochberg method (Benjamini and Hochberg, 1995) to obtain a false discovery rate (FDR) of 0.05. We adjusted for multiple hypothesis testing of four outcomes (total gray and white matter volumes, hippocampus, and amygdala volumes) and three exposure periods (prenatal, mid childhood, and late childhood), yielding 24 comparisons. The unstandardized coefficients (B) and 95% confidence intervals (CIs) were calculated. All analyses were performed using SAS 9.4 software. All effect sizes for the brain measures were generated using the Python-based nilearn package v0.5.2 (https://nilearn.github.io/).
Trio design modeling code
Below is an example of the trio model code used to examine whether prenatal and childhood parental hostility would be associated with differences in maternal, paternal, and child brain structure in “SAS” version 9.4 of the regions of interest that included both random intercept and random slope components (with an unstructured covariance matrix between the two):
proc mixed covtest;
class triadid;
model satisfaction = maternal_hosttility paternal_hostility / solution
ddfm=satterth;
repeated / type=cs subject=triadid;
Results
Mother, father, and child characteristics are described in Table 1. The mean age of the children at scanning was 13.6 years (SD = 0.8), and 50.2% of the children were girls. In total, 28.1% of mothers and 24.9% of fathers had a non-Western national origin. The mean age at scanning of the mothers was 46.6 years (SD = 4.5) and fathers was 49.3 years (SD = 5.2). Means and standard deviations across family similarity are described in Table 1.⇓
Sociodemographic characteristics of the 1,452 Generation R participants (n = 484 trios)
Within-family interaction of parental hostility with brain structures of parents and child
Adjusted correlations (similarity) across parents and children
Maternal and paternal reports of hostility were all positively correlated both in the prenatal period and during childhood (Figs. 1a, 2a, 3a, 4a). Within-informant longitudinal stability in hostility ratings (e.g., mothers’ ratings at different time points; rs = 0.27–0.29) was higher than cross-informant longitudinal stability (e.g., mothers’ ratings at one time point and fathers’ ratings at another time point; rs = 0.19). Furthermore, white matter, gray matter, and hippocampal volumes between mother, father, and children were significantly correlated (see, e.g., maternal and paternal cortical gray matter volume; rs = 0.18; Figs. 1c, 2c, 3c). Overall, adjusting for baseline confounders typically reduces correlations with dyadic indexes (mother and father hostility), as the correlations became much smaller. No significant correlations were found between maternal, paternal, and child amygdala volumes (Fig. 4c).
Association between parental hostility with amygdala volumes of parents’ and child. a, Correlation between maternal and paternal hostility prenatal, when child was 3 and 10 years old. All measures were significantly correlated. The dark orange color indicates higher values were significantly correlated with maternal and paternal hostility score. b, Each plot depicts the association between parental hostility (mother and father prenatal, child ages 3 and 10) with amygdala volumes of child, mother, and father measured using magnetic resonance imaging (MRI). We found that maternal and paternal hostility measured during pregnancy and early childhood was associated with amygdala volumes of the respective parent (i.e., within-parent analyses), but not across-parent analysis. Models are adjusted for child age at brain MRI scan, sex assigned at birth, total ICV, and age, national origin, education, parity, and smoking and alcohol consumption reported by mother and father. Number of tests correction for multiple testing = 24 (4 outcomes and the 3 relevant exposure periods). Critical value for FDR p < 0.05. After adjusting for multiple testing correction (FDR p < 0.05), we found no association between maternal and paternal-reported hostility over time with amygdala volumes of child, mother and father. c, The adjusted correlation between child, mother, and father gray matter volumes adjusting for both parents’ amygdala exposures.
Comparison of group correlation coefficients between parental hostility (during pregnancy and when children were 3 and 10 years old) with parental and child total gray matter volumes showed significantly higher similarity in mother–child, father–child, and mother–father dyads within families compared with between families (r(32) = −0.018; p = 0.013).
Higher similarity, as measured by correlation, between parental hostility (during pregnancy and when children were 3 and 10 years old) and total white matter volumes was also observed in mother–child, father–child, and mother–father dyads within families compared with between families (r(27) = −0.015; p = 0.023). Correlation coefficients of parental hostility over time and hippocampal volumes in mother–child, father–child, and mother–father dyads were significantly different within families when compared with across families (r(25) = −0.009; p = 0.035). No similarity, which is higher within-family as between-family correlation, was observed between parental hostility over time and amygdala volumes (r(12) = 0.003; p = 0.465).
Within-family interaction
Tests for within-family interactions of the varying parental hostility effects over time at different ages showed a significant interaction between levels of hostility over time in association with child white matter (B = −0.16; p = 0.013) as well as gray matter volumes (B = −0.12; p = 0.026). The direction of this effect suggests that parental hostility vary over time and had a negative effect on child brain structures at high levels of parental hostility, and that pregnancy is a more vulnerable period for the effects of parental hostility on gray and white matter development.
We also found evidence for an interaction effect of hostility over time and maternal white matter volume (B = −0.18; p = 0.038) and gray matter volume (B = −0.21; p = 0.020), with a small effect size. Similar associations were observed for parental hostility over time and paternal white matter (B = −0.14; p = 0.041) and gray matter volumes (B = −0.17; p = 0.026). The evidence for interaction implies that parental hostility had a more negative effect on maternal and paternal brain structures when children were 10 years old (Table 2). No interaction effects were observed between parental hostility over time with any of the parent and child amygdala and hippocampal volumes.
Parental hostility and parental brain morphology
Next, we tested whether parental hostility was associated with parental brain structures of mothers and fathers. We found that both maternal and paternal hostility measured during pregnancy and early childhood were associated with smaller white matter, gray matter, and hippocampal volumes in within-parent analyses even after correction for multiple testing. Moreover, both maternal and paternal hostility scores during pregnancy and early childhood were associated with their partner's smaller white matter and gray matter and hippocampal volumes brain morphology 10 years later in across-parent analysis (adjusted betas per unit increase of maternal and paternal hostility; Figs. 1b, 2b, 3b). Again, no associations were observed between parental hostility with amygdala volumes of the partner (i.e., across parents; Fig. 4b).
Parental hostility and child brain morphology
Prenatal maternal hostility was associated with smaller child cortical gray matter volume (B per unit hostility = −1.33; 95% CI, −1.46, −1.20; p = 0.013; FDR corrected; p < 0.05; Fig. 1b), as well as with child smaller cortical white matter volume (B per unit hostility = −1.09; 95% CI, −1.21, −1.06; p = 0.010; FDR corrected; p < 0.05; Fig. 2b). We also found that prenatal maternal hostility was associated with child smaller hippocampal volume (Beta per unit hostility = −0.10; 95% CI, −0.16, −0.06; p = 0.024; FDR corrected; p < 0.05; Fig. 3b). These coefficients indicate how a unit change in parent hostility relates to the average change in the respective brain structure per cubic centimeter. We observed no associations between prenatal father hostility and any of these child brain measures. No associations were observed between prenatal parental hostility and child amygdala volume (Fig. 4b).
No associations were observed between early or late childhood mother or father hostility measure and child brain morphology. Additional adjustment for parental smoking and alcohol consumption did not meaningfully change these associations.
Taken together, these results indicate that the observed associations of maternal and paternal hostility co-occur and prenatal parental hostility was related to individual differences of parent and child brain structures in a trio model that accounted for the family structure. Parental socio-economic status remains another important confounder potentially underlying our observations. However, when we adjusted for parental educational achievement, we found that the association between parental hostility and brain structural volumes remained.
In addition, standard regression models were used to examine period-specific exposure associations of both maternal and paternal hostility with each parent and child brain structures. Results were generally similar though the effect estimates (see, e.g., prenatal maternal hostility with child cortical gray matter volume B = −1.55%; 95% CI −1.61, −1.47; p = 0.001) were slightly higher than the effect estimates of the respective path coefficient in the structural equation modeling (Table 3). However, these models do not account for the correlated nature of family data and are thus likely to be biased by unmeasured familial factors.
Linear regression analysis of the associations between maternal hostility and child brain morphology (n = 484)
Further, we observed no associations between any parental hostility score and temporal, frontal, and parietal lobar measures of mothers, fathers, and children (p > 0.05; results not shown). Lastly, associations between maternal-reported hostility and child brain outcomes did not differ by different maternal ethnic/nationality subgroups (Table 4).
Linear regression analysis of the associations between maternal hostility and child brain morphology adjusting for separate ethnic groups (n = 484)
Mediation analysis
As psychopathology interferes with children's and parents’ abilities to regulate stress physiology and may be associated with disruption in typical brain development, we finally examined whether brain morphology of mothers, fathers, or children mediated the association between prenatal maternal- and paternal-reported hostility and child externalizing behavior at age 14 years. As Figure 5 illustrates, hippocampal volumes of the mother partially mediated the association of prenatal maternal hostility and adolescent externalizing behavior (indirect effect: B = 0.08; 95% CI, 0.04, 0.12; 5.4% of the total effect; p = 0.008). In contrast, the hippocampal volumes of the father did not mediate this association. Next a latent factor of prenatal mother and father hostility was constructed, to examine whether hippocampal volumes of the child mediate the association of combined parental hostility and adolescent externalizing behavior. The latent variable model captures the covariance between mothers’ and fathers’ hostility (Materials and Methods). We found evidence that smaller hippocampal volumes of the child could partly underlie the observed associations between combined prenatal mother and father hostility and adolescent externalizing behavior during late childhood (B = 0.06; 4.2% of the total effect; 95% CI, 0.02, 0.10; p = 0.016). Additional adjustment for pre-existing child externalizing behavior at 1.5 years, resulted in no meaningful change in mediation results, suggesting that the behavioral changes did not precede the observed brain differences in the child. Lastly, timeline of data collection, correlations between maternal and paternal psychopathology, and flowchart of the study sample are presented in Figures 6, 7, and 8.
Mediation analysis of hippocampal volumes of parents and children in association with prenatal maternal and paternal hostility with adolescent externalizing behavior. Path A is the association of prenatal parental hostility with maternal, paternal, or child hippocampal volumes, and path B is for the association of maternal, paternal, or child hippocampal volumes with adolescent externalizing behavior. Path C (in black) is the total association between prenatal maternal or paternal hostility and adolescent externalizing behavior with hippocampal volumes not in the model. Path C′ (in gray) is the direct association between prenatal parental hostility and adolescent externalizing behavior factor with hippocampal volume in the model. Model is adjusted for child age at brain MRI scan, sex assigned at birth, total ICV, maternal age, national origin, education, marital status during pregnancy, parity, smoking and alcohol consumption, and prior child externalizing behavior when child was aged 1.5 years. * The latent construct of maternal- and paternal-reported hostility. Parental hostility factor captures covariation across raters or the extent to which a given dimension is reflected across parents (i.e., a between-rater dimension factor) (c).
Timeline of data collection: parental hostility with maternal, paternal, and child brain morphology.
Correlation of the maternal and paternal report hostility. All hostility measures were significantly correlated. Correlation is significant at the 0.01 level (2-tailed).
Inclusion of the study sample.
Discussion
This population-based neuroimaging study suggests that each parents’ hostility from pregnancy onward associates with the behavior of other family members and their brain morphology. Using a triadic approach to neuroimaging data of mothers, fathers, and children, we were able to obtain the following three key findings. First, our results suggest that within parents, maternal and paternal hostility was associated with differences in their total white matter, gray matter, and hippocampus volumes. Importantly, parental hostility was also associated with morphological brain differences across parents; more hostility in one was related to less total white matter and gray matter volumes in the other parent. Second, we observed that children exposed to maternal hostility occurring during pregnancy had smaller total gray matter, white matter, and hippocampal volumes in mid childhood. Third, we showed that the association of prenatal parental hostility with child externalizing behavior was partially mediated by the hippocampal volume of the child and importantly, also the hippocampal volume of the mother. These associations remained after adjusting for baseline family factors and multiple testing.
Previous studies indicate that the brain development is vulnerable to persistent stressors and these environmental factors are particularly significant early in life (Fox et al., 2010; Belsky and De Haan, 2011). Our findings provide evidence that the offspring of mothers with prenatal maternal hostility has smaller gray matter and hippocampal volumes. Several mechanisms could explain this association. First, mothers with high levels of intrusive behavior during interactions with partners likely experience stress, which in turn dysregulates the hypothalamic–pituitary–adrenal axis and in turn affects child development. Stress could also affect brain development through inflammatory responses and changes in the balance of the autonomic nervous system (Hompes et al., 2012). Changes in white and gray matter volumes may be attributable to a number of factors, such as learning about emotions by observing parents’ emotional displays and interactions. Modeling has long been demonstrated as an important mechanism through which children learn specific behaviors (Bandura, 1977). Children learn that certain situations provoke emotions, and they observe the reactions of others in order to know how they “should” react in similar situations (Denham et al., 1997). For example, parents’ overall expressivity may affect children and adolescent's modeling of emotions and behavior and continuities in detrimental environments that are not influenced by contextual stressors examined in the current study. Although we cannot demonstrate causality, our findings suggest that parental hostility has the potential to compromise child neurodevelopment in the long term.
There is some evidence that maternal depression and anxiety during pregnancy and early childhood are related to structural differences in the maternal brain (Feldman, 2015). These studies suggest that psychopathology is expressed in the brain as reduced activation to infant cues. In particular reward and empathy circuits found in maternal depression are linked with lower maternal sensitivity and reward from parenting (Laurent et al., 2011; Musser et al., 2012; Apter-Levy et al., 2013; Ho et al., 2014). Similarly, our findings underscore that differences in maternal hostile behavior were associated with maternal brain morphology. In a previous study, positive experiences during early postpartum period co-occurred with structural differences in mothers’ brain regions susceptive to stress exposure including the prefrontal cortex in a study of 19 women (P. Kim et al., 2010). Several mechanisms could explain the observed associations between parental hostility of both parent and child brain morphology. Pre-existing familial vulnerability factors may explain the association with parental brain differences. Importantly, such associations could also be explained by genetic influences. Recently, a genome-wide association study identified a few genetic loci associated with hippocampal volume (van der Meer et al., 2020), which could be (indirectly) associated with parental psychopathology. Our findings indicate that in addition to pre-existing familial factors and genetic predisposition, different influences of environment such as learning through observation, parenting practices, and emotional climate in the family (trio model) can affect both parents and children psychopathology (Morris et al., 2007; Rutherford et al., 2015).
We found fathers who were more hostile during pre- and postnatal period had lower total white matter, gray matter, and hippocampal volumes, but not amygdala volume. Indeed, the accumulating evidence suggests that fathers’ psychopathology poses many challenges to men's lives and mental health (P. Kim and Swain, 2007). The timing of paternal psychopathology before and after child's birth is just recently being studied and placed into a biological framework that could involve brain morphology (Swain et al., 2014). Researchers have consistently reported that adults with psychopathology have smaller hippocampus and amygdala volumes (Besteher et al., 2020). For example, smaller amygdala and hippocampus have been shown in mothers with postpartum depression, trauma, or substance use (Landi et al., 2011; Chase et al., 2014; S. Kim et al., 2014). Overall, our findings indicate that the family relations is formed by the individual characteristics of each member relationships across trios.
Our findings show that one parent's hostility occurs together with his or her partner's hostility, which has important implication for family health and well-being. Higher similarity within families (e.g., mother–child, father–child, and mother–father) than across families was further observed for measures of gray matter, white matter, and hippocampal volumes, but not amygdala volumes. Evidence for interaction between the maternal and paternal hostility dyads and brain morphological structures implies that pregnancy and early childhood is a vulnerable period when development in response to parental care disruptions is maximally dynamic. The findings highlight the role of each parent's ability to transpire hostile behaviors to his or her partner and the child. This could be the results of assortative mating (Maes et al., 1998), which suggests that mothers’ and fathers’ psychopathology may be similar before engaging in a relation (Frisell et al., 2012). That is, both partners may, for example, share common environments affecting the main trait with genetic and phenotypic resemblance.
Our findings are based on within-family methods nested in a large population-based neuroimaging study. Estimates from within-family based are more robust but often less precise (typically reflecting the smaller sample size) than estimates of studies sampling unrelated individuals. However, this approach controls for shared environmental (often poorly controlled SES-related or parenting traits) and genetic confounders (genetic variants related to both brain and behavior). An illustrative example was recently published by Brumpton et al., the association between taller height and lower BMI with better educational attainment was strongly attenuated in within-family reflecting dynastic effects such as the inheritance of a healthier environment or genetic confounding (Brumpton et al., 2020). The present within-family estimates of the specific associations in 484 mother–father–offspring trios can be considered relatively well powered. Some authors have suggested that certain associations such as intrauterine exposures can only be thoroughly addressed by family designs (including discordant sibling designs). In neuroimaging studies, such contemporary implementations of modern epidemiological methods have rarely been performed to date (Hwang et al., 2021; Frach et al., 2023). This likely reflects the lack of neuroimaging data collected from families at a scale sufficient to be appropriately powered to study the modestly strong brain–behavior associations.
We found that the association between prenatal parental hostility and child externalizing behavior was partially mediated by the child's hippocampal volumes. Although the common variance is shared across mother- and father-reported hostility, the association with prenatal exposure suggests that direct maternal physiological influences may underlie the findings. This is consistent with prior research in the present cohort showing that smaller offspring hippocampal volumes partially mediate the association between prenatal family disruption and child adjustment problems (Xerxa et al., 2020). However, maternal hippocampal volumes also partially mediated the associations of prenatal parental hostility with child externalizing behavior. This could in part be explained by the specific role of the mother in the postnatal period which—together with the intrauterine period—is considered a critical period for optimal child development (Lisofsky et al., 2019). As infants are highly dependent, this critical period of nurturance and care requires a tremendous maternal investment. Whether this reflects that pregnancy has a long-lasting impact on maternal brain or background risk factors cannot be disentangled with our study design. However, we show that the maternal hippocampal volume is related to maternal hostility and that this may underlie child problems.
There are several limitations of this study. First, as with many observational studies, the relative homogeneity of the population limits its generalizability. Second, given the lack of the repeated measures of both parents and child brain morphology, we cannot assess the directionality of the associations between parental hostility and the brain characteristics of the parents. Third, parental genetic variation could possibly predispose parents to higher levels of hostility while also affecting their child's brain structure. The strengths of the study lie in its large population-based sample and the SEM approach in the unique triadic data to testing the associations of prospectively measured exposure data with two parents’ and child brain outcomes measured at 14 years of age. Furthermore, we included repeated maternal and paternal reports hostility and therefore could examine how one parent's hostility affects all members of the family.
To date, interventions targeting parental psychopathology among high-risk parents has typically focused on the clinician–patient dyad rather than clinician–patient–family trio. The value of trio design is clear; many genetic and environmental changes are able to be ruled out by comparing the children's sequence with their parents’. As the associations of parental hostility to parental and child brain are small, the trio (mother–father–child) approach have the potential to transform communication within clinical encounters and the quality of patient and family care. Thus, interventions aimed at parental psychopathology that include a child component would likely be only marginally more effective.
In conclusion, our findings show that prenatal parental hostility is associated with smaller volumes of total gray matter and the hippocampus in children. This suggests that parental psychopathology may have long-lasting neurodevelopmental correlates in children. Moreover, maternal and paternal hostility were each associated with differences in his or her own brain morphology as well as his or her partner's total gray matter, white matter, and hippocampus volumes. Our findings suggest that the behavior of family members results from a system of interdependent dyadic relationships over time that associate with specific brain structural differences.
Data Availability
Data can be obtained upon request. Requests should be directed to the management team of the Generation R Study (secretariaat.genr@erasmusmc.nl), which has a protocol of approving data requests. Because of restrictions based on privacy regulations and informed consent of participants, data cannot be made freely available in a public repository.
Footnotes
The Generation R Study is conducted by the Erasmus Medical Center, Rotterdam in close collaboration with the Faculty of Social Sciences of the Erasmus University Rotterdam, the Municipal Health Service, Rotterdam Homecare Foundation and Stichting Trombosedienst & Artsenlaboratorium Rijnmond. We thank the contribution of participating parents and their children, general practitioners, hospitals, midwives, and pharmacies. H.T. was supported by a European Union Horizon-Staying-Healthy-2021 Grant (EU grant number 101057529). Y.X. was supported by the Consortium on Individual Development (CID) which is funded through the Gravitation Program of the Dutch Ministry of Education, Culture, and Science and the NWO (grant number 024.001.003), and Erasmus MC2 - Research Innovation Grant (grant number 114799). High-performance computing for image analyses was supported by the Netherlands Organization for Scientific Research (NWO, Surf, project 2021.042).
The authors declare no competing financial interests.
- Correspondence should be addressed to Yllza Xerxa at y.xerxa{at}erasmusmc.nl or Henning Tiemeier at tiemeier{at}hsph.harvard.edu.
