Placental gene expression and offspring temperament trajectories: predicting negative affect in early childhood

Abstract

Exposure to prenatal stress increases offspring risk for long-term neurobehavioral impairments and psychopathology, such as Attention Deficit Hyperactivity Disorder (ADHD). Epigenetic regulation of glucocorticoid pathway genes may be a potential underlying mechanism by which maternal conditions ‘program’ the fetal brain for downstream vulnerabilities. The present study aims to investigate whether mRNA expression of glucocorticoid pathway genes in the placenta predict offspring negative affect during early childhood (between 6 and 24 months). Participants include 318 mother-child dyads participating in a longitudinal birth cohort study. Placental mRNA expression of glucocorticoid pathway genes (HSD11B1, HSD11B2, NR3C1, NCOR2) were profiled and negative affect traits of the offspring were measured at 6, 12, 18, and 24 months. HSD11B1 mRNA expression significantly predicted negative affect (β = −.09, SE = .04; p=.036), and Distress to Limitations trajectories (β = −.13, SE = .06; p = .016). NCOR2 mRNA expression significantly predicted Distress to Limitations (β = .43, SE = .21; p = .047), and marginally predicted Sadness trajectories (β = .39, SE = .21; p = .068). HSD11B2 and NR3C1 did not predict trajectories of Negative Affect or subscale scores. Infant negative affect traits were assessed via maternal self-report, and deviated from linearity across follow-up. mRNA expression of glucocorticoid pathway genes in the placenta may be a potentially novel tool for early identification of infants at greater risk for elevated negative affect. Further study is needed to validate the utility of mRNA expression of glucocorticoid pathway genes in the placenta.

The in-utero period is a time of rapid change and development within the fetal brain, during which downstream mental and physical health effects can begin to take shape (Wadhwa, Sandman, & Garite, 2001). The fetus is especially vulnerable to environmental insults such as stress mothers experience, including normative stress, psychopathology, physical, and social stressors (Davis et al., 2007; Wadhwa, Sandman, & Garite, 2001). Prenatal maternal stress can affect the fetal stress response circuitry of the hypothalamic-pituitary-adrenal (HPA) axis, setting the offspring on a suboptimal neurodevelopmental path (Talge, Neal, & Glover, 2007). While animal and human models have shown that prenatal maternal stress negatively impacts fetal neurodevelopment, novel research in epigenetics and molecular biology has focused on placental gene expression as a key regulatory mechanism that may underlie fetal ‘programming’ of vulnerability for later disease (J. Chen et al., 2014; Monk et al., 2016; Ponder et al., 2011).

Placental gene expression.

The placenta, which has been designated a neuroendrocrine organ, produces a group of neuropeptide hormones equivalent to those produced by the HPA-axis, such as thyrotropin-releasing hormone (TRH), and corticotropin-releasing hormone (CRH) (Lambertini, Lee, J., & Che, 2012; Marsit, Lambertini, et al., 2012). The placenta adapts to stressors via epigenetic regulation of expression of transporters and receptors that together regulate the complex interplay of maternal nutrients, hormones, amino acids, and growth factors needed to facilitate optimal development (Monk et al., 2016; Zeltser & Leibel, 2011). Epigenetic alterations are changes to gene expression that do not require changes in DNA sequence, including histone modification, DNA methylation, non-coding RNAs, and chromatin structure alterations (Monk et al., 2016). Through these processes the placenta can adapt (i.e., recruit resources for growth) to provide the fetus with the greatest chance of survival post-delivery (Hanson & Skinner, 2016). This results in a mismatch between the prenatal and expected postnatal environment, potentially leading to long-term consequences including health and behavioral diseases later in life (Myatt, 2006).

One of the most studied examples of this disruption is excess glucocorticoid exposure in-utero (Wyrwoll & Holmes, 2012). To understand mechanisms regulating the exchange of glucocorticoids, researchers have focused on key genes along this pathway in the placenta (Monk et al., 2016). These studies have primarily focused on the role of 11-beta hydroxysteroid dehydrogenase Type 1 (HSD11B1), 11-beta hydroxysteroid dehydrogenase Type 2 (HSD11B2) and glucocorticoid receptor NR3C1 (J. Chen et al., 2014; Green et al., 2015; Jensen Peña, Monk, & Champagne, 2012; Kappil et al., 2015; Monk et al., 2016; Zhang et al., 2018). These genes modulate offspring glucocorticoid overexposure, and are responsive to maternal conditions such as gestational diabetes mellitus (GDM), preeclampsia, depression, and trauma (Banik et al., 2017). Several additional genes have recently been identified as important modulators of the fetal glucocorticoid environment, including nuclear receptor corepressor 2 (NCOR2) (Zhang et al., 2018).

HSD11B1 is responsible for the interconversion of active cortisol and inactive cortisone (Dekker et al., 2012; Lemche, Chaban, & Lemche, 2016) and is a dominant regulator of circulating glucocorticoids in-utero (Morgan et al., 2014). HSD11B2 converts active cortisol into inactive cortisone (Monk et al., 2016). Researchers have observed epigenetic alterations in HSD11B1 and HSD11B2 among mothers with varying levels of stress (Jensen Peña, Monk, & Champagne, 2012). Animal models have found that inactivation of HSD11B2 results in an altered fetal stress response which attenuates throughout adulthood (Welberg, Seckl, & Holmes, 2000). Further, expression levels of both genes have been correlated with gross developmental outcomes such as birth weight and size at delivery (McTernan et al., 2001; Muramatsu-Kato et al., 2014).

Glucocorticoid exposure is additionally regulated via NR3C1, which encodes a glucocorticoid receptor (Bustamante et al., 2016; Tyrka et al., 2016; Vrijsen et al., 2015). NR3C1 has been associated with maternal stressors and has been shown to be sensitive to maternal mood (Oberlander et al., 2008). The binding of stress hormones to NR3C1 is an important part of the glucocorticoid regulatory process in the placenta, and is often posited as a potential mechanism by which prenatal maternal stress affects central nervous system (CNS) reactivity in early childhood (Oberlander et al., 2008; Reynolds et al., 2015).

Few studies have investigated the role of NCOR2 in the placenta. NCOR2 has been identified as a key gene in the cortisol-signaling pathway which modulates glucocorticoid receptor gene expression (S. Chen et al., 2017; Vakili, Jin, & Cattini, 2014). Thus, acting indirectly, this corepressor can affect the level of fetal exposure to glucocorticoids (van der Laan et al., 2008). In vivo studies have supported this pathway, suggesting that corepressors decrease glucocorticoid receptor expression, which in turn increases levels of circulating glucocorticoids in the body (Mendoza et al., 2017).

Neurodevelopmental indices.

Temperament, which refers to variations in emotional reactivity in early childhood, has been consistently associated with psychiatric disorders in later childhood (Rettew & McKee, 2005). Studies have established temperament as a moderately stable measure which is substantially affected by genetic and environmental influences, and is consistently associated with childhood psychiatric disorder (Partridge & Lerner, 2007; Rettew & McKee, 2005). Researchers have theorized that temperament overlaps substantially with neurodevelopmental indices (e.g., emotion regulation and executive functioning), and have observed that offspring exposed to increased maternal stress have increased irritability, fear, crying, and hyperactivity (Bergman et al., 2007; Brand, 2006; Davis et al., 2007, 2004). Although such characteristics may be adaptive in preparing the child for a stressful environment post-birth, these traits are often indicators of later psychopathology and neurobehavioral impairments, especially when the expected adversity does not occur (De Pauw & Mervielde, 2010). Numerous studies have found that “difficult” temperament, characterized by negative mood, withdrawal, low adaptability, and high-intensity responses, increases between 2–3 years of age, and decreases during the pre-school years (Partridge & Lerner, 2007). A study focused on negative affect in early childhood found that traits of negative affect (i.e. anger and fearfulness) increased from 4 – 16 months (Braungart-Rieker, Hill-Soderlund, & Karrass, 2010). Thus, assessing temperament during early childhood (6 – 24 months) is ideal, as negative aspects of temperament may be most likely to be observed. The temperament domain of negative affect in particular has value in predicting symptoms of psychopathology later in life. In one sample of school-aged children, negative affect in the fourth grade predicted both anxiety and depression symptoms reported in the eleventh grade (Lonigan, Philips, & Hooe, 2003). This relationship has also been demonstrated in an even younger population. Doughterty and colleagues showed that a specific trait of negative affect, irritability, as measured at age 3 predicted general functional impairment as well as diagnoses of oppositional defiant disorder and depression at age six (Doughterty et al., 2013). Furthermore, the Swedish Twin Study of Child and Adolescent Development established that irritability at ages 8–9 years may be related to downstream symptoms of anxiety and depression at ages 19–20 years, with genetic and environmental factors contributing to the relationship across development, though further monitoring of early internalizing symptoms is needed to examine causality (Savage et al., 2015). Indeed, a number of theorists have posited that temperament, especially negative affect, is an early indicator of risk for the development of psychopathology (e.g., Clark, 2005; Muris & Ollendick, 2005).

Examining mRNA expression of key HPA axis genes in a routinely discarded organ may shed light on a novel early indicator of risk for negative affect in early childhood, which could illuminate a target for intervention. Although immediate outcomes (e.g., intrauterine growth restriction) related to epigenetic alterations are well known in offspring exposed to environmental insults in-utero, long-term neuro-developmental deficits are less well understood (Barker, 2002; Zhang et al., 2018). It is essential to untangle how in utero epigenetic changes relate to long-term complications. While additional research is needed to test whether temperament based intervention are effective in mitigating risk of negative neurodevelopmental and behavioral outcomes, studies have found evidence of effective intervention in reducing anxiety (Rettew & McKee, 2005). Understanding one potential pathway (i.e., via the placental epigenome) underlying the maternal environment and offspring temperament is an important step in expanding this field.

The glucocorticoid pathway in particular, may be a good candidate to link the maternal environment during pregnancy with offspring temperament. A growing body of research has associated alterations of glucocorticoid related placental gene expression with offspring deficits in the domains of emotion regulation, stress response, and negative affect. Rodent models with experimentally reduced/inhibited HSD11B2 enzyme during pregnancy, conditioning the fetus to be exposed to active glucocorticoids, have demonstrated a direct relation with HPA axis function, amygdala glucocorticoid receptors (Welberg, Seckl, & Holmes, 2001), and elevated anxiety-like behaviors (Holmes et al., 2006). In human models, lower placental HSD11B2 was associated with newborn low birthweight and reduction in movement (Marsit et al., 2012) and higher baseline cortisol levels at 32 days (Stroud et al., 2016). HSD11B2 expression moderated the relation between maternal depression during pregnancy and female offspring baseline cortisol, whereas SLC6A4 expression moderated the relation between maternal prenatal depression and male offspring baseline cortisol (Stroud et al., 2016). Conversely, greater placental expression of HSD11B2, HSD11B1, NR3C1 and SLC6A4 was associated with greater deficits in newborn sleeping, crying, spitting, and feeding (Räikkönen et al., 2015). These conflicting findings between placental gene expression and offspring neurobehavior warrant further investigation to better understand potential fetal programming of stress responsivity and negative affect via glucocorticoid pathways.

Based on previous studies linking maternal stress to placental glucocorticoid regulatory gene expression and offspring temperament, (Brand, Engel, Canfield, & Yehuda, 2006; van der Wal, van Eijsden, & Bonsel, 2007) we formulated the following hypotheses: mRNA expression of glucocorticoid candidate genes (i.e., HSD11B1, HSD11B2, NR3C1, NCOR2) will significantly predict temperament trajectories along Negative Affect, and its 4 subscales (Falling Reactivity, Fearfulness, Distress to Limitations, and Sadness) across the follow up period, between 6 and 24 months. As exploratory hypotheses we expect that higher levels of expression of HSD11B1, HSD11B2, NR3C1 will be associated with decreased Negative Affect, while higher levels of NCOR2 expression will be associated with increased Negative Affect over time.

Methods

The present study population was composed of 318 mother-child dyads participating in the Stress In Pregnancy (SIP) Study, an ongoing longitudinal study focused on prenatal epigenetics and child development. As part of their participation, expecting mothers were recruited from two area Ob/Gyn clinics (Mount Sinai Hospital, and New York Presbyterian-Queens) in New York City, and followed throughout their pregnancy and postpartum, during which they completed questionnaires assessing demographic traits, prenatal characteristics, and offspring temperament at 6, 12, 18, and 24 months. Mothers’ placenta tissue was collected at birth for analysis. Exclusion criteria included age <15 years, HIV positivity, life-threatening maternal medical conditions, maternal psychosis, and chromosomal or congenital abnormalities in the fetus. Expecting mothers who expressed interest in participating were invited into a private research room within the clinic where research staff reviewed the list of exclusion criteria with the potential participant. If deemed eligible after this initial screening, written informed consent was obtained and the participant was enrolled in the study, subject to a second screening via chart review. Demographic traits were recorded at baseline and are presented in Table 1. All research procedures were approved by the Institutional Review Boards (IRB) at the City University of New York, New York Presbyterian Queens, and the Icahn School of Medicine at Mount Sinai. Further details regarding this study and the sample can be found elsewhere (Finik & Nomura, 2017).

Table 1.

Sample characteristics.

Demographics		N		(%)
Child Gender	Boys	163		51.3
Maternal Education	Primary / Some High School	52		16.4
	High School/GED	67		21.1
	Some College	87		27.4
	Associate Degree	32		10.1
	Bachelor Degree	45		14.2
	Graduate Degree	35		11.0
Maternal Status	Married	117		36.8
	Common Law	17		5.3
	Single	179		56.3
	Widowed/Divorced/Separated	5		1.5
Maternal Race	White	44		13.8
	Black	78		24.5
	Hispanic	156		49.1
	Asian	34		10.7
	Other	6		1.9
Obstetric characteristics		M		(SD)
Maternal Age		27.2		6.0
Birthweight	(grams)	3234.9		561.8
Gestational Age	(weeks)	39.0		1.9

Across postpartum milestones the retention of data regarding infant behavior were as follows: 235 at 6 months (74%), 156 at 12 months (49%), 160 at 18 months (50%), and 134 at 24 months (42%). While missing data is a significant methodological concern for longitudinal studies, no significant differences were detected across demographic characteristics between responders at 6, 12, 18, or 24 months. Missing data were therefore assumed to occur at random. We leveraged a robust statistical method, linear mixed effects modeling, which provides accurate estimates under this assumption.

Placenta biopsies free of maternal decidua were collected from each quadrant of the placenta midway between cord and rim within an hour of delivery to maintain RNA integrity. Biopsies were snap frozen in liquid nitrogen for 24 hours followed by storage at −80°C until use. RNA extraction from placenta biopsies were carried out using Maxwell simplyRNA Tissue Kit (Promega, # AS1280) following protocol. RNA was quantified via Nanodrop spectrophotometer (Thermo Fisher Scientific Inc., #ND-2000) using a custom-designed code set on the nCounter platform (nanoString Technologies, Seattle, WA) as described elsewhere (Kappil et al., 2015). Briefly, following hybridization, unbound probes were removed, and the purified complexes were aligned and immobilized on imaging cartridges using an nCounter Prep Station II. Cartridges were sealed and scanned in an nCounter Digital Analyser for code count detection. nCounter data was normalized using the NanoStringNorm package in R (Prokopec et al., 2012). Code count data were first normalized against the geometric mean of spike‐ in controls to account for differences in hybridization and recovery. Differences in biopsy content were adjusted for by normalizing data against the geometric mean of housekeeping genes (GAPDH, RPL19, and RPLP0). The background threshold of detection was set at two standard deviations above the mean of the included negative controls. Expression below the background threshold was set to the value of the limit of detection divided by the square root of two to maintain sample variability.

Temperament was assessed via the Infant Behavior Questionnaire-Revised (IBQ-R) (Garstein & Rothbart, 2003; Rothbart, 1981) and Early Childhood Behavior Questionnaire (ECBQ), (Putnam, Gartstein, & Rothbart, 2006) which are age appropriate assessments of childhood temperament. Mothers record the frequency of observing specific behaviors across 96-items on a scale of 1 (never) to 7 (always). These ratings comprise 14 subscales for the IBQ-R and 18 subscales for the ECBQ. 7 of the subscales are present in both the IBQ-R and the upward extension ECBQ, and 4 are equivalent subscales in both with distinct labels (IBQ-R) [ECBQ]. The 7 subscales include: Activity Level, Cuddliness, Sadness, High-Intensity Pleasure, Low-Intensity Pleasure, Fearfulness, Perceptual Sensitivity, and the 4 equivalent subscales include (Approach) [Positive Anticipation], (Falling Reactivity) [Soothability], Duration of Orienting [Attention Focusing], and Distress to Limitations [Frustration]. To avoid confusion across these subscales the IBQ-R terminology will be retained for scales that are common to both questionnaires. 3 amalgamated domains are produced from the subscales in each instrument, specifically: Surgency, Regulatory Capacity and Negative Affect. Surgency is comprised of Activity Level, High-Intensity Pleasure, and Approach. Regulatory Capacity is comprised of Duration of Orienting, Low Intensity Pleasure, Cuddliness and Perceptual Sensitivity. Finally, the Negative Affect domain is comprised of Falling Reactivity, Fearfulness, Distress to Limitations, and Sadness (Posner & Rothbart, 2007; Putnam, Rothbart, & Gartstein, 2008).

Our primary outcome of interest was the Negative Affect domain, and our secondary outcomes included the Falling Reactivity, Fearfulness, Distress to Limitations, and Sadness subscales. The domain of Negative Affect refers to negative emotional reactivity including fear, anger, and neuroticism. Falling Reactivity measures the rate at which an infant recovers from stress or arousal. Fearfulness measures the severity of an infant’s stress response to abrupt deviations in stimulation such as a new object. Distress to Limitations measures the level of an infant’s display of distress in specific situations (e.g., when confined in a place or position). Finally, Sadness measures overall lowered mood and behavior related to infant’s suffering or incapacity to carry out a wanted action (Posner & Rothbart, 2007; Putnam et al., 2008).

Confounders.

To identify the minimally sufficient adjustment set required to capture the total effect of placental gene expression on infant negative affect traits, a Directed Acyclic Graph (DAG) was created following a review of the literature (see Supplementary Figure 1). Recent advances in epidemiological research have indicated that traditional methods of adjustment can introduce conditional associations and bias, rather than reducing it. A DAG approach helps to reduce the degree of bias introduced when estimating the effect of an exposure on an outcome (Suttorp, Siegerink, Jager, Zoccali, & Dekker, 2015). Of the measured variables available, the minimal sufficient set requiring adjustment included child sex, maternal smoking, depression and trauma.

Maternal Smoking.

Mothers reported whether or not they smoked during their pregnancy at baseline (2^nd trimester). Smoking during pregnancy has been consistently linked to alterations in placental gene expression (Palma-Gudiel, Cirera, Crispi, Eixarch, & Fañanás, 2018), and an extensive history of cohort studies have found that offspring exposed to maternal smoking have higher ratings of Negative Affect (Liu et al., 2011; Martin, Dombrowski, Mullis, Wisenbaker, & Huttunen, 2006).

Child Sex.

Child sex was recorded at delivery. Previous researchers have found differential expression in a wide array of genes in the placenta according to fetal sex (Sood, Zehnder, Druzin, & Brown, 2006). While sex differences in temperament vary substantially in the literature, a systematic review of studies focusing on child temperament and gender differences found that overall, boys score higher in negative emotionality (Cosentino-Rocha & Linhares, 2013).

Prenatal depression.

Prenatal depression was assessed at baseline via the Edinburgh Postnatal Depression Scale (EPDS) (Cox, Holden, & Sagovsky, 1987). The EPDS is a validated 10 item self-report questionnaire, which assesses mother’s depressive symptomatology. Mothers record their feelings in the past week on a 4-point Likert scale. The internal consistency for this measure was .81. A cut-off of ≥12 was retained to indicate presence of maternal depression based on previous validation tudies. In the initial validation study (n=84) sensitivity and specificity for this measure was 86% and 78% respectively (Cox et al., 1987; Murray & Carothers, 1990). Maternal depression during pregnancy can impact methylation of key HPA axis genes in the placenta including NR3C1, HSD11B2, (Conradt, Lester, Appleton, Armstrong, & Marsit, 2013) HSD11B1, (Ciesielski, Marsit, & Williams, 2015) and FKBP5 (Monk et al., 2016). Maternal depression during pregnancy has been consistently linked with temperament outcomes in affected offspring across early childhood (Davis et al., 2007; Field, 2011).

Trauma.

The Posttraumatic Diagnostic Scale (PDS) (Foa, Cashman, Jaycox, & Perry, 1997) was administered to participating mothers at baseline to assess trauma. The PDS is a 49-item questionnaire that measures PTSD symptoms associated with a traumatic event. Mothers are asked to record the frequency or applicability of experiencing PTSD related symptoms. The PDS provides a total sum score and the recommended cut-off score of 15 was retained to indicate presence of maternal trauma. The PDS has an excellent internal consistency .92, and good sensitivity and specificity of 82% and 77%, respectively (Briere & Spinazzola, 2005; Foa et al., 1997; Griffin, Uhlmansiek, Resick, & Mechanic, 2004). Numerous studies have found differential epigenetic alterations (DNA methylation) in glucocorticoid candidate genes, particularly HSD11B2, FKBP5, NR3C1 and CRH, in relation to maternal trauma and psychosocial stress (Janssen et al., 2016; Monk et al., 2016; Palma-Gudiel et al., 2018). Further, children born to mothers with a history of trauma have been found to have increased levels of behavioral problems, and difficult temperaments (Plant, Pawlby, Pariante, & Jones, 2017; Haselbeck et al., 2017).

Statistical Analyses.

Although temperament is correlated within an individual across time, these measures fluctuate throughout development, and measures of temperament vary substantially between infants (Charil, Laplante, Vaillancourt, & King, 2010; Putnam et al., 2008). Thus to assess the relationship between our predictors and the longitudinal trajectory of Negative Affect and related subscales, linear mixed effect models with random intercepts and slopes were fit. To capture the Negative Affect domain and subscale trajectories, child’s age (as a time-scale) at 6, 12, 18 and 24 months at follow-up, was included in the linear mixed models. For each of the five outcomes, including one higher-order index (Negative Affect) and four subscales associated with the main higher order index (Falling Reactivity, Fearfulness, Distress to Limitations, and Sadness), the same model building approach was applied. Plots were reviewed to evaluate trends in the outcome and the distribution of residuals. All models leveraged restricted maximum likelihood for parameter estimation, which allows for the generation of estimates for missing data at certain time points with the least amount of bias. Model specifications are described below. Y is used to denote the outcome, the subscript refers to individual “i” at follow-up time “j”.

Model A: (Temperament)_ij = β₀₀ + b_0j + ε_ij

Model B: (Temperament)_ij = β₀₀ + β₁₀ (HSD11B1)_ij + β₂₀ (HSD11B2)_ij + β₃₀ (NR3C1)_ij + β₄₀ (NCOR2)_ij + β₅₀ (Age)_ij + b_0j + ε_ij

Model C: (Temperament)_ij = β₀₀ + β₁₀ (HSD11B1)_ij + β₂₀ (HSD11B2)_ij + β₃₀(NR3C1)_ij + β₄₀ (NCOR2)_ij + β₅₀ (Age)_ij + b_1j(Age)_ij + ε_ij

Model D: (Temperament)_ij = β₀₀ + β₁₀ (HSD11B1)_ij + β₂₀ (HSD11B2)_ij + β₃₀ (NR3C1)_ij + β₄₀ (NCOR2)_ij + β₅₀ (Smoking)_ij + β₆₀ (CSex)_ij + β₇₀ (Depression)_ij + β₈₀ (Trauma)_ij + β₉₀ (Age)_ij + b_1j(Age)ij + b_0j + ε_ij

Briefly, intercept only models (Model A, where we allowed for subject-specific intercepts) were fitted for each Negative Affect outcome to assess serial correlation and to compute Interclass Correlation Coefficients (ICC) (Shrout & Fleiss, 1979) to measure the degree of statistical dependency. ICCs across all outcomes indicated that a significant amount of variation in negative affect was explained by between child differences, supporting the applicability of the intended statistical approach. In Model B we specified the glucocorticoid candidate genes as our predictors, as well as the age variable. For the third model, we allowed slopes to vary across individuals. In Model C, we assessed if an unstructured or independent correlation structure should be implemented for the random intercepts and random slopes, judging by [Akaike Information Criterion (AIC), and Bayesian Information Criteria (BIC)] (see Supplementary Table 7). At each stage random effects were retained if an improvement in model fit was observed. In the final model (D), the identified measured confounders (maternal smoking, child sex, maternal depression and trauma) were incorporated into the existing model. To adjust for multiple comparisons the Bonferroni correction was applied in Models B – D (Dunn, 1961). To test whether residuals were normally distributed and independent of random effects an assessment of model diagnostics was completed. We did not observe any significant violations of these assumptions. A two-sided hypothesis was assumed for all analyses with an alpha level of .05 to indicate statistical significance. Plots and analyses were conducted in R 3.3.2.

Results

Child Temperament.

Trends in temperament trajectories over time (6 – 24 months) are depicted in Table 2. Briefly, mean Negative Affect, Fearfulness, and Sadness increased between 6 and 12 months, followed by a decrease from 18 months onward. Mean Falling Reactivity decreased slightly from 6 to 18 months, followed by an increase at 24 months. Mean Distress to Limitations increased between 6 and 12 months, followed by decrease between 12 and 18 months, and an increase at 24 months.

Table 2.

Temperament outcomes across follow up (6 −24 months).

Temperament Outcome	6 Months		12 Months		18 Months		24 Months
	M	SD	M	SD	M	SD	M	SD
Negative Affect*	3.42	.92	3.77	.79	3.29	.70	3.27	.71
Falling Reactivity	4.98	1.08	4.88	.97	4.73	1.05	4.86	1.04
Fearfulness	3.26	1.41	3.91	1.31	2.78	1.18	2.69	1.12
Distress to Limitations	3.88	1.23	4.38	1.12	3.63	1.29	3.79	1.32
Sadness	3.48	1.22	3.62	1.16	3.03	1.11	2.84	1.08

^*NB: Negative Affect is a composite temperament domain comprised of the remaining scales listed above.

ICC.

Intercept only models indicate that 40.3% of the variation in Negative Affect scores was explained by between child differences. Among the subscales, 38.1% in Falling Reactivity, 34.1% in Fearfulness, 29.9% in Distress to Limitations, and 28% in Sadness was due to between child differences across these subscales. In the subsequent model, the conditional ICC indicated that 26.3% of between child variation in Negative Affect was explained by mRNA expression of glucocorticoid candidate genes (HSD11B1, HSD11B2, NR3C1 and NCOR2). Among the subscales, 18.5% of variation in Falling Reactivity, 24.2% of Fearfulness, 6.3% of Distress to Limitations, and 23.5% of Sadness was explained by mRNA expression of glucocorticoid candidate genes. A substantial amount of the initial variation due to between child differences was explained by the addition of mRNA expression in our model, except for Distress to Limitations (29.9% vs. 6.3%) (See Supplementary Table 1).

The best fitting model (Model D) was retained for each outcome, 1) Negative Affect, 2) Falling Reactivity, 3) Fearfulness, 4) Distress to Limitations, and 5) Sadness, based on model fit statistics (AIC/BIC) to assess whether glucocorticoid candidate genes (HSD11B1, HSD11B2, NR3C1, and NCOR2) were significant predictors of temperament across follow up (6 – 24 months). All models included the following confounders identified a priori: child sex, maternal smoking, depression and trauma, however only results for the predictors (glucocorticoid genes) are reported below. For full parameter estimates from each model see Supplementary Tables 2–6.

Glucocorticoid Genes

HSD11B1 Children with higher levels of HSD11B1 expression, had lower ratings of Negative Affect (β = −.09, SE = .04; p = .036) and Distress to Limitations (β = −.13, SE = .06; p = .016) over time, as expected, on average for the population. However, HSD11B1 expression did not predict ratings in Falling Reactivity or Sadness over time. Higher levels of expression was marginally associated with Fearfulness, such that children with higher levels of expression trended towards decreased Fearfulness ratings across follow up (β = −.12, SE = .07; p = .086). HSD11B2 HSD11B2 mRNA expression did not predict Negative Affect, nor related subscales (Falling Reactivity, Fearfulness, Distress to Limitations, or Sadness) over time. NR3C1 Similarly, NR3C1 expression failed to predict temperament across Negative Affect, or related subscales across follow up. NCOR2 Children with higher levels of NCOR2 had higher rating of Distress to Limitations (β = .43, SE = .21; p = .047) and a marginally higher rating of Sadness (β = .39, SE = .21; p = .068) on average over time in the target population. NCOR2 expression marginally predicted Negative Affect, with higher levels of expression related to increased Negative Affect on average across the population over time (β = .29, SE = .15; p = .060). However, NCOR2 expression did not predict ratings in Falling Reactivity or Fearfulness across the follow up period.

Random Effects.

Positive correlations between random slopes and intercepts were observed across the trajectories of Negative Affect (r = .61), and the related subscales of Falling Reactivity (r = .62), Fearfulness (r = .43), Sadness (r = .65) and Distress to Limitations (r = .20). Children with higher scores across these subscales at baseline tended to have positive slopes across these subscales between 6 and 24 months. However, further observation is needed to assess whether the positive correlations observed attenuate over time, or if scores tend to regress toward the mean.

Discussion

The present study explored whether mRNA expression of key HPA-axis genes involved in glucocorticoid regulation in the placenta (HSD11B1, HSD11B2, NR3C1, NCOR2) predict the trajectory of offspring negative neurobehavioral traits as indicated by temperament measures at 6, 12, 18, and 24 months. A growing consensus in human and animal research has established that maternal stress during pregnancy impacts neurobehavioral development in offspring (Charil et al., 2010), which may occur via gene expression in the placenta (Monk et al., 2016). We investigated a potential marker of suboptimal neurobehavioral development via maternal-reported negative affect characteristics (Falling Reactivity, Distress to Limitations, Fearfulness and Sadness) across early childhood (6, 12, 18 and 24 months) as indicators of downstream psychopathology and neurobehavioral disorders (Posner & Rothbart, 2007; Putnam et al., 2008). Capitalizing on existing evidence indicating placental gene expression as an underlying biological pathway between maternal stress and offspring neurodevelopment, we expected to observe a significant relationship between mRNA expression of glucocorticoid pathway genes and negative affect. Further, we expected that increases in expression of HSD11B1, HSD11B2, NR3C1 and decreases in NCOR2 would be associated with decreases in negative affect trajectories over time.

In line with our initial hypotheses we found that higher levels of HSD11B1 mRNA expression significantly predicted the Negative Affect domain and Distress to Limitations subscale, while marginally predicting Fearfulness. HSD11B1 mRNA expression did not predict Falling Reactivity or Sadness. The negative directionality observed is intuitive considering decreased expression of this gene may increase the likelihood of aberrant exposure to active stress hormones in fetal circulation. Although it is difficult to assess the direct effects of this gene’s expression due to its dual functionality, it is possible that increased expression (via hypomethylation, histone modification) facilitates greater conversion of active cortisol to inactive cortisone, protecting the developing fetal HPA axis from overexposure to active stress hormones (Monk et al., 2016; Togher et al., 2014). The significant predicted decreases in Negative Affect, Distress to Limitations, and Fearfulness, may provide evidence for this protective function of HSD11B1, as these scales are suggestive of suboptimal CNS functioning (De Pauw & Mervielde, 2010). While not explored in the present study, it is possible that decreased expression could result in insufficient exposure to active glucocorticoids, also resulting in a disruption to neurodevelopment (Monk et al., 2016; Togher et al., 2014).

Although we observed an inverse association between HSD11B1 expression and Negative Affect subscales, expression failed to predict Falling Reactivity and Sadness. This may be the result of shifts in temperament development during early childhood (Posner 2007, Ruff 2001). Temperament traits are objectively different yet interdependent and hypothesized to originate from shared underlying attributes that develop differentially across time. This phenomenon, frequently described as heterotypic continuity (De Pauw & Mervielde, 2010), may underlie the distinct shifts observed across these subscales during follow up. This non-linearity (particularly across Fearfulness and Sadness) may have limited the ability of genes including HSD11B1 to predict specific subscales of negative emotionality. Interestingly, HSD11B2 and NR3C1, the most studied of the glucocorticoid pathway genes, did not predict any Negative Affect outcomes. Although study characteristics (e.g., non-linearity) may have limited the predictive capacity of HSD11B2 and NR3C1, it is possible that the interaction of the two is more consequential for the glucocorticoid pathway and the developing fetus than the individual expression of each, however this examination is beyond the scope of the present study.

Accumulating evidence supports the notion that epigenetic alterations of both HSD11B2 and NR3C1 expression are associated with infant neurobehavior (Appleton et al., 2013). Specifically, increased methylation (resulting in greater silencing of HSD11B2 and NR3C1) has been associated with negative neurobehavioral traits in newborns exposed to prenatal depression or anxiety (Conradt, Lester, Appleton, Armstrong, & Marsit, 2013). Marsit, Lambertini et al. (2012) found differing patterns of interactions between DNA methylation of NR3C1 and HSD11B1 and aspects of newborn neurobehavior. Low methylation in NR3C1 (related to increased expression) and high methylation in HSD11B2 (related to decreased expression) was associated with decreased excitability, indicative of lowered mood and negative affectivity (Marsit, Lambertini, et al., 2012; Marsit, Maccani, Padbury, & Lester, 2012). High NR3C1 methylation and low HSD11B2 methylation was associated with high asymmetrical reflex scores, which has been related to neurological problems in early childhood (Marsit, Lambertini, et al., 2012; Marsit, Maccani, et al., 2012). Although the present study found that NR3C1 and HSD11B2 were not significant predictors of negative affect, interactions between NR3C1 and HSD11B2 were not explored. Future studies are needed to elucidate how decreased expression of NR3C1 in tandem with increased HSD11B2 expression relates to infant neurobehavior, as both may decrease glucocorticoid exposure, which is necessary for optimal neurodevelopment (Marsit, Maccani, et al., 2012). Further, if NR3C1 is dysregulated, and HSD11B2 functionality is compromised, the fetus is at risk of overexposure to active glucocorticoids, which can similarly set the fetus on a suboptimal neurodevelopmental path (Sarkar et al., 2001; Appleton et al., 2015; Appleton, Holdsworth, & Ingle, 2016).

Increases in NCOR2 mRNA expression were associated with increases in Negative Affect, Distress to Limitations, and Sadness. NCOR2 did not predict Falling Reactivity or Fearfulness. Greater expression of this corepressor may expose the developing fetus to increased levels of active stress hormones, potentially altering CNS development (van der Laan et al., 2008) and increasing the offspring’s likelihood of later maladaptive neurobehavioral symptoms (Monk et al., 2016). The inability of NCOR2 mRNA expression to predict Falling Reactivity may stem from the non-linearity in this subscale over time in particular, and the previously described concept of heterotypic continuity in temperament.

Contradictions with previous findings regarding HSD11B2 and NR3C1 may be due, at least, in part to the fluctuation of placental gene expression across gestation (Gheorghe, Goyal, Mittal, & Longo, 2010). This process allows for the placenta to adapt and accommodate to the evolving needs of the fetus (Gheorghe et al., 2010). In human models we are only able to capture expression at the time of delivery in pregnancies carried to term, and thus may miss important changes in placental gene expression during gestation. The inability of candidate genes to predict ratings of Falling Reactivity may stem from its poor correlation with the remaining subscales, and consequently the amalgamated domain of Negative Affect in our sample. Falling Reactivity was moderately correlated with the Negative Affect domain (r = −.51), yet poorly correlated with Sadness (r = −.27), Fearfulness (r =−.19), and Distress to Limitations (r = −.34) (see Supplementary Table 8). This may result from the difference in individual items which load onto the domain of Falling Reactivity in the IBQ as compared to the ECBQ, where questions specific to a child’s soothability are a part of the equivalent scale.

Replication is needed to establish HSD11B1 and NCOR2 as predictors of negative temperament traits (particularly Negative Affect, Distress and Fear). However, the current findings suggest that HSD11B1 and NCOR2 may be important actors in regulating fetal exposure to active glucocorticoids, and may in turn play a role in determining fetal HPA axis vulnerability to downstream dysregulation. The inverse relation between HSD11B1 expression and negative affect and the positive relation between NCOR2 expression and distress suggest a protective role against stress and risk of internalizing problems in youth based on previous studies (Brown & Rosellini, 2011; Compas, Connor-Smith, & Jaser, 2004; Degnan, Almas, & Fox, 2010). Temperament may be an important proxy for assessing early childhood neurobehavioral development in a cost-effective and non-invasive manner. Previous intervention studies have observed improvement in temperament ratings (e.g., decreased Negative Affect) (Rettew & McKee, 2005), and suggest intervention timing could be protective against related deleterious feedback, such as negative parenting behaviors (Kiff, Lengua, & Zalewski, 2011). Providing resources during pregnancy (e.g., counseling) to reduce risk of maternal stress during pregnancy, coupled with early intervention in infancy based on temperament ratings could facilitate a strategy to minimize risk of suboptimal neurodevelopment.

Several limitations should be noted when considering the findings from this study. First, the Negative Affect domain was nonlinear across follow-up (6 – 24 months), particularly in ratings of Fearfulness and Sadness. However, this is largely in line with trajectories charted by other studies (Braungart-Rieker et al., 2010; De Pauw & Mervielde, 2010; Zhang et al., 2018). Second, the poor correlation between Negative Affect and Falling Reactivity may have limited the ability to detect meaningful differences across this subscale. Third, substantial missing data, due to reduced retention, limited the sample size to n = 134 at 24 months, which may have affected the precision of estimates reported. However, we do not foresee that missing data affected accuracy of results considering the robust statistical approach employed. In addition, no significant differences across demographic characteristics, birth outcomes, or maternal conditions during pregnancy (i.e., depression, trauma, smoking) were noted between the initial and final sample. Missing data were therefore assumed to occur at random; however, selection bias via differential drop out along unmeasured factors may have occurred. However, it should be noted that the generalizability of these findings are constrained by the limited sample size, sociodemographic characteristics of the cohort, and the subject-specific statistical approach employed. Fourth, although we posit maternal stress as a potential agent of epigenetic alteration, it was not within the scope of this study to examine the relation between maternal stress and degree of HPA axis gene alteration. While we theorize that HPA axis dysfunction underlies the observed temperament deficits, we did not measure cortisol or the structure/function of key HPA axis regions. It is therefore possible that other CNS circuits may also be responsive to cortisol exposure in-utero (e.g., amygdala, prefrontal cortex) which could contribute to the present observations. Finally, we focused on mRNA expression and did not measure epigenetic phenomena that can alter gene expression (e.g., DNA methylation, histone modification). To address certain limitations, additional measures of temperament in later childhood and a biological measure of HPA axis functioning [e.g., salivary cortisol, adrenal gland volume via computed tomography or magnetic resonance imaging (Golden, Wand, Malhotra, Kamel, & Horton, 2011)] could be leveraged to strengthen current research in this field.

Early negative affect is an important precursor to later psychopathology and neurobehavioral conditions (Garstein & Rothbart, 2003; Yehuda et al., 2005), thus a novel predictor of negative affect such as glucocorticoid gene mRNA expression, particularly HSD11B1, in a conventionally discarded organ (the placenta) may provide a new avenue for risk identification and intervention. However, the present study’s findings are limited, replication and further discovery is required to assess whether the placental glucocorticoid pathway genes predict child temperament. While constrained in scope, these findings may increase awareness of the importance of comprehensive access to prenatal care as a significant first step in facilitating psychological well-being during pregnancy. As maternal stress during pregnancy is a well-studied agent of epigenetic alterations that can negatively impact childhood developmental trajectories (Banik et al., 2017), access to comprehensive care, including screenings and alleviation of maternal stressors, may minimize the biological consequences of stressors which arise during this tumultuous time in women’s lives.

Table 3.

Summary of Fixed effects

Summary of Fixed Effects
Negative Affect	Model 1a	Model 1b	Model 1c	Model 1d
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	3.43 (.40)***	2.80 (1.66)	3.58 (1.61)*	2.89 (1.66)
HSD11B1	-------------	−.09 (.05)*	−.09 (.04)*	−.09 (.04)*
HSD11B2	-------------	−<.01 (.05)	<.01 (.04)	<.01 (.37)
NR3C1	-------------	−.07 (.18)	−.16 (.17)	−.20 (.18)
NCOR2	-------------	.21 (.16)	.22 (.15)	.29 (.15)^
Falling Reactivity	Model 2a	Model 2b	Model 2c	Model 2d
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	4.88 (.05)***	5.96 (2.03)***	5.91 (1.98)**	5.86 (2.08)**
HSD11B1	-------------	.06 (.06)	.06 (.05)	.06 (.05)
HSD11B2	-------------	.05 (.05)	.06 (.04)	.05 (.05)
NR3C1	-------------	−.24 (.22)	−.26 (.21)	−.17 (.22)
NCOR2	-------------	.04 (.19)	.06 (.19)	.01 (.19)
Fearfulness	Model 3a	Model 3b	Model 3c	Model 3d
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	3.20 (.06)***	1.56 (2.56)	1.91 (2.53)	−.55 (2.62)
HSD11B1	-------------	−.11 (.07)	−.11 (.07)^	−.12 (.07)^
HSD11B2	-------------	.02 (.06)	.03 (.06)	.04 (.06)
NR3C1	-------------	−.02 (.27)	−.13 (.27)	−.10 (.28)
NCOR2	-------------	.20 (.25)	.27 (.24)	.38 (.24)
Distress Limitations	Model 4a	Model 4b	Model 4c	Model 4d
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	3.92 (.06)***	1.84 (2.24)***	1.58 (2.21)	.01 (2.31)
HSD11B1	-------------	−.13 (.06)^	−.14 (.06)^	−.13 (.06)*
HSD11B2	-------------	<.01 (.05)	<.01 (1.36)	<.01 (.05)
NR3C1	-------------	.06 (.24)	−.02 (.24)	.02 (.24)
NCOR2	-------------	.24 (.21)**	.36 (.21)**	.43 (.21)*
Sadness	Model 5a	Model 5b	Model 5c	Model 5d
	β (SE)	β (SE)	β (SE)	β (SE)
Intercept	3.30 (.05)***	2.89 (2.16)	2.98 (2.12)	2.26 (2.24)
HSD11B1	-------------	−.09 (.06)	−.09 (.06)	−.08 (.06)
HSD11B2	-------------	.05 (.05)	.05 (.05)	.05 (.05)
NR3C1	-------------	−.28 (.23)	−.31 (.23)	−.29 (.24)
NCOR2	-------------	.32 (.21)	.36 (.21)	.39 (.21)^

NB: Child age centered at 6 months. For all parameter estimates in each model see Supplementary Tables 1–5. Unstructured covariance structure specified for all models based on an examination of model fit. (Supplementary Table 6).

^{^}p <.10;

^*p < .05;

^**p < .01;

^***p < .001