Sex-specific associations between placental corticotropin releasing hormone and problem behaviors in childhood

Placental corticotropin-releasing hormone (pCRH) is a neuroactive peptide produced in high concentrations in mid-late pregnancy, during key periods of fetal brain development. Some evidence suggests that higher pCRH exposure during gestation is associated with adverse neurodevelopment, particularly in female offspring. In 858 mother-child dyads from the sociodemographically diverse CANDLE cohort (Memphis, TN), we examined: (1) the slope of pCRH rise in mid-late pregnancy and (2) estimated pCRH at delivery as a measure of cumulative prenatal exposure. When children were 4 years-old, mothers reported on problem behaviors using the Child Behavior Checklist (CBCL) and cognitive performance was assessed by trained psychologists using the Stanford-Binet Intelligence Scales. We fitted linear regression models examining pCRH in relation to behavioral and cognitive performance measures, adjusting for covariates. Using interaction models, we evaluated whether associations differed by fetal sex, breastfeeding, and postnatal neighborhood opportunity. In the full cohort, log-transformed pCRH measures were not associated with outcomes; however, we observed sex differences in some models (interaction p-values ≤ 0.01). In male offspring, an interquartile (IQR) increase in pCRH slope (but not estimated pCRH at delivery), was positively associated with raw Total ( β = 3.06, 95%CI: 0.40, 5.72), Inter-nalizing ( β = 0.89, 95%CI: 0.03, 1.76), and Externalizing ( β = 1.25, 95%CI: 0.27, 2.22) Problem scores, whereas, in females, all associations were negative (Total Problems: β =(cid:0) 1.99, 95%CI: (cid:0) 3.89, (cid:0) 0.09; Internalizing: β


Introduction
A central tenet of the Developmental Origins of Health and Disease (DOHaD) framework is that during development individuals actively adapt their physiology in response to environmental cues to enhance survival and fitness in that specific setting (Barker, 2007).This adaptive capacity is made possible by tremendous plasticity during early development and is particularly evident in the human fetal brain where as many as 250,000 new neurons may be produced every minute, with 40, 000 synapses created per second (Cowan, 1979;Levitt, 2003).In line with the DOHaD perspective, during sensitive periods, the development of this immense neural network, including neurogenesis, migration, synaptogenesis, may be programmed by stimuli such as maternal psychosocial or physiological stress (Van den Bergh et al., 2020).
Stressors can also upregulate the mother's hypothalamic-pituitaryadrenal (HPA) axis activity, leading to higher circulating levels of the stress hormone, cortisol (Van den Bergh et al., 2020).Cortisol suppresses production of the hypothalamic corticotropin releasing hormone (hCRH) through negative feedback but in a unique feature of pregnancy physiology, cortisol also upregulates placental CRH (pCRH) production through placenta-specific transcriptional regulators (Ma et al., 2001;Robinson et al., 1988).As a result, pCRH rises exponentially across pregnancy and plays an important role as a "placental clock" regulating the timing of parturition (Herrera et al., 2022;McLean et al., 1995).CRH measured in maternal circulation during pregnancy is almost exclusively of placental origin and mid-late pregnancy pCRH may be an important predictor of preterm birth (Herrera et al., 2021;Ruiz et al., 2016;Wadhwa et al., 2004).
The high levels of pCRH produced in mid-late gestation are also notable given that the molecule is neuroactive and can cross the fetal blood-brain barrier during sensitive periods of fetal brain development (Kastin and Akerstrom, 2002).Because pCRH levels are also responsive to both internal and external cues (e.g., maternal depression, anxiety, trauma history, infection), pCRH has been posited as a pathway by which internal and external signals are integrated and transmitted to the fetus to shape subsequent developmental trajectories.To that end, a small number of studies have examined the extent to which neurodevelopment may be responsive to pCRH activity during gestation, with a particular focus on temperament and behavior.
An early study reported that pCRH concentrations in mid-gestation (25 weeks) were associated with higher levels of parent-reported fearful and reactive temperament in infants, however no associations were observed with pCRH measured at other points in pregnancy (19 and 31 weeks; n=248) (Davis et al., 2005).A reanalysis of those data explicitly focused on sex differences indicated that associations between pCRH and temperament were limited to female infants, with no associations observed in males (Sandman et al., 2013).In a subset of the same cohort (n=91), a stronger increase in pCRH in mid-gestation was later associated with greater internalizing behavior at age 5 as measured by an observational task; however, this time, no sex differences were observed (Howland et al., 2016).These small studies are complemented by research in a cohort of children born very preterm (<28 weeks gestation) and therefore at heightened risk for neurodevelopmental concerns.In that cohort, few associations between placental pCRH mRNA concentrations and a large battery of neurodevelopmental measures at age 10 were observed, although there was some suggestion that higher pCRH expression was associated with greater social impairment, particularly among females (Leviton et al., 2018).The mechanisms that might underlie these preliminary associations remain unclear, however, epidemiological evidence suggests that higher gestational pCRH levels may be linked to greater cortisol reactivity in infancy as well as reduced cortical thickness in certain brain regions including the frontal and temporal lobes later in childhood (Rinne et al., 2022;Sandman et al., 2018).
Taken as a whole, this small body of research offers equivocal support for the hypothesis that pCRH may impact the developing brain.
However, well-powered, prospective longitudinal research is essential to replicate prior studies and clarify inconsistent findings on associations between fetal exposure to pCRH and neurodevelopmental problems, particularly in lower medical risk children.Questions also remain regarding which factors might moderate the impact of pCRH exposure on child neurodevelopment.Given evidence of sexually dimorphic responses to prenatal stress (Barrett and Lessing, 2021), and pCRH exposure specifically (e.g., Sandman et al., 2013), as well as sex differences in CRH receptors in the brain (reviewed in Bangasser, 2013), fetal sex is a plausible moderator of associations.Further, protective factors in the postnatal environment may attenuate associations and provide insight on potentially modifiable treatment targets to support child development.
Here, we used data from a large, diverse U.S. pregnancy cohort to address these questions and extend the literature on pCRH and child development.We hypothesized that higher pCRH exposure would be associated with neurodevelopment problems, including more emotional and behavioral problems and lower cognitive performance, at age 4 years.We examined two measures of pCRH: (1) rate of pCRH change across mid-late gestation (slope); and (2) a novel measure of pCRH at delivery which can be considered an estimate of cumulative exposure to pCRH across gestation, approximating exposures over the full period of prenatal brain development, rather than shorter windows.Given prior research, our analyses additionally tested for sex differences in associations.Finally, we evaluated the extent to which postnatal factors, specifically, breastfeeding and quality of neighborhood resources, may protect against potential adverse impacts of pCRH.

Overview of study design and population
The Conditions Affecting Neurocognitive Development and Learning in Early Childhood (CANDLE) study recruited pregnant participants in Shelby County, Tennessee, USA from 2006 to 2011 (LeWinn et al., 2020;Sontag-Padilla et al., 2015).Eligibility criteria included age 16-40 years carrying a singleton fetus, between 16 and 29 weeks gestation, and plans to deliver at a participating hospital.Participants were low medical risk at the time of enrollment, which was defined as lacking a major medical condition including insulin-dependent diabetes, chronic hypertension, and major endocrine disorders.Each participant engaged in two study visits occurring roughly in mid-(16-29 weeks) and late (22-39 weeks) pregnancy.Flexibility in prenatal visit timing was deliberately planned to coordinate with clinical care and reduce the burden on participants.Mother-child dyads engaged in study activities across infancy and early childhood.Outcome data for the current analysis are based on in-person visits that occurred when the index children were 4 years old.The Institutional Review Board at the University of Tennessee Health Sciences Center (the main data collection site) approved the study and all participants provided written informed consent.The current analysis was made possible through the ECHO-PATHWAYS consortium, part of the NIH's Environmental Influences on Child Health Outcomes (ECHO) program (LeWinn et al., 2022).

pCRH
At both prenatal visits, blood samples were collected using EDTA plasma separator tubes.After collection and processing, blood was frozen at − 80⁰C until shipment to University of Newcastle (Australia) on dry ice.At Newcastle, pCRH was measured according to previously published radioimmunoassay protocols (Smith et al., 2009;Steine et al., 2020).Extraction recovery was 87% and intra-and inter-assay coefficients of variation were 7.3% and 8.7%, respectively.Due to non-normal distributions, we log transformed pCRH concentrations for subsequent analyses.
Based on the predictable, exponential rise in pCRH across gestation, E.S. Barrett et al. using the two measured pCRH concentrations per participant (in midand late pregnancy) and gestational age at delivery (abstracted from the medical record), we operationalized prenatal exposure to pCRH in two ways: (1) rate of change (slope) and (2) estimated pCRH concentration at delivery.Briefly, we used the R nlme package to fit a linear mixedeffects model treating the relationship between log(pCRH) and gestational age as a fixed effect, with random slope and intercept for each participant (Pinheiro and Bates, 2000).We then used this model to predict log(pCRH) levels on the date of delivery of each participating child.pCRH at delivery is strongly correlated with pCRH area under the curve across pregnancy (r=0.99;not shown) and is interpreted here as an estimate of cumulative exposure to pCRH across the gestational period.Considering evidence that both pCRH concentrations and rate of change in pCRH may be important predictors (Steine et al., 2020), we used the mixed models results to calculate pCRH slope (rate of change) across pregnancy.Given the well-established exponential increase in pCRH during mid-late pregnancy, both pCRH measures were log-transformed for all analyses, effectively shifting the exponential scale to a linear one, making it possible to use linear models.Although some prior studies have focused on pCRH measures at individual timepoints during gestation (Davis et al., 2005;Howland et al., 2016), here, using these two measures (pCRH rise and estimated pCRH at delivery) was preferred due to potential error stemming from variation in the timing of the visits when pCRH was measured.

Behavior and emotional problems
At the age 4 visit, child behavioral problems were assessed using the raw scores from the Internalizing and Externalizing broadband problem subscales of the widely used Child-Behavior Checklist, which has wellestablished psychometric properties (CBCL; Achenbach, 2011;Achenbach and Rescorla, 2000).This parent-reported questionnaire assesses emotional and behavioral problems occurring in the last six months.It consists of 99 items, and response options range from 0 (not true) to 2 (often or very true).The Total Problems score was obtained from summing the internalizing and externalizing scores along with items on sleep and one open item about problems not otherwise captured.Our primary models used the raw, discrete CBCL scores.Secondarily, we evaluated behavioral outcomes dichotomized based on clinical thresholds (T-score=64) (Lande et al., 2009).

Cognitive performance
At the age 4 visit, trained psychologists administered the Stanford-Binet Intelligence Scales (5th edition), from which full-scale IQ (FSIQ) scores were derived (Roid and Pomplun, 2012).These scores were used continuously in analyses as our primary cognitive performance measure.Secondarily, we evaluated verbal and non-verbal sub-scores.

Covariates and effect modifiers
Covariates were selected a priori based on the literature and our prior work on pCRH, CBCL, and IQ data in this cohort.Covariate data derived from multiple sources including pre-and postnatal questionnaires and biospecimens, medical record review and linkage to external databases.Data on maternal age, child birth order, prenatal supplement use, prepregnancy BMI, and prenatal alcohol use were obtained from prenatal questionnaires.Cotinine was measured in maternal prenatal spot urine samples as an indicator of smoking and/or environmental smoke exposure.Maternal preeclampsia, gestational hypertension, and gestational diabetes were determined by participant report and subsequently confirmed through medical record review.Child biological sex and gestational age at birth were abstracted from the medical record.During childhood, mothers reported on their child's race and ethnicity, postnatal smoking (both by the mother as well as other household members), maternal education, household income, and marital status.Mothers additionally completed the PROMIS depression scale (Pilkonis et al., 2011) and the Wechsler Abbreviated Scale Intelligence (WASI) (Weschler, 2011) as a measure of maternal cognitive performance.For the purposes of this analysis, for measures assessed at multiple timepoints (e.g., maternal depression, postnatal smoking), we used the values reported at the same visit as the CBCL administration.When those values were not available, we used the values from the prior visit.We explored two potential effect modifiers in addition to child sex.First, at postnatal visits, mothers reported how long they breastfed the index child and for the current analysis, responses were categorized as never, 0-6 months, and greater than 6 months.Second, based on geocoded address data collected at the age 4 visit, we calculated Child Opportunity Index (COI) scores (Acevedo-Garcia et al., 2014), a measure of the quality of neighborhood resources; the current analyses used the total score composite of resources across three domains: neighborhood social and economic opportunity, educational opportunity, and health and environmental status, with higher composite scores representing greater opportunity.

Statistical analysis
We used descriptive statistics (mean, standard deviation, median, min, max, n, frequency, missingness) to summarize participant characteristics as well as the distributions of the exposure and outcome measures.Consistent with our approach in prior ECHO-PATHWAYS analyses (e.g., Ni et al., 2022;Wallace et al., 2022), we fitted a series of staged multiple linear regression models with robust standard errors to examine associations between the exposure and outcome with varying levels of adjustment for covariates.We used the same sets of covariates in analyses examining each of the six continuous outcome measures (Total Problems, Internalizing, and Externalizing Behaviors, FSIQ, verbal IQ, and non-verbal IQ).In minimally adjusted models, we adjusted for child sex and age at CBCL administration (continuous).In our main, fully-adjusted models, we included an extensive set of additional covariates including maternal age at delivery (continuous), child birth order (first born vs later born), preeclampsia/gestational hypertension (yes/no), maternal education at the age 4 visit (less than high school; high school/GED/technical school, college, graduate work/graduate degree), household income regionally adjusted income by household size at the age 4 visit (continuous), insurance status at the age 4 visit (none, Medicaid/Medicare only; Medicaid/Medicare and private, private only), marital status at the age 4 visit (married/living as married vs other), total COI score (continuous), prenatal urinary cotinine (continuous), breastfeeding (categorical by duration, e.g.never, 0-6 months, 6+ months), postnatal secondhand smoke exposure in the household at the age 4 visit (any vs. none), pre-pregnancy BMI, use of alcohol during pregnancy (any/none), prenatal vitamin use (any/none), and maternal IQ as measured by the WASI (continuous, for models examining IQ only).In extended models, we additionally included child race as an imperfect measure of unmeasured factors related to systemic racism (categorized as Black, White, other races) as well as a set of covariates that could be considered confounders but could alternatively be on the pathway between exposure and outcome, including gestational age at birth (continuous), gestational diabetes (yes/no), and maternal depression scores at the age 4 visit (continuous).To complement models examining the three continuous CBCL outcomes, using multivariable Poisson regression models, we secondarily considered the T-scores dichotomized at the 64th percentile based on cut-offs indicative of clinically significant levels of behavioral problems.For ease of interpretation, log(pCRH slope) was rescaled by interquartile range (IQR) for all models considering that exposure.
After fitting the models described above, we further explored effect modification.Specifically, we examined effect modification by child sex at birth (male/female), breastfeeding duration (never, 0-6 months, 6+ months), and quality of neighborhood resources (COI score above versus below median).We did so by including interaction terms for each of E.S. Barrett et al. these potential moderators (in separate models) with the pCRH measures, which allowed us to estimate the statistical significance of the interaction.We additionally reparameterized these models in order to estimate slopes for each level of the moderator (e.g., for males and females) without sacrificing power by stratifying the sample.In post hoc analyses informed by the main results and limited to male offspring, we evaluated effect modification by breastfeeding duration and quality of neighborhood resources by including interaction terms.
Finally, we conducted two sensitivity analyses to evaluate the robustness of our results.In the first one, we refitted models excluding participants with gestation age <37 weeks given prior studies demonstrating lower neurodevelopmental performance among children born preterm (Cheong et al., 2017;Hee Chung et al., 2020).In a second sensitivity analysis involving only cognitive outcomes, we excluded individuals with IQ ≤ 70 to reduce their potential influence on the results.All analyses were conducted in R, version 4.1.3(R Foundation for Statistical Computing, Vienna, Austria).

Descriptive statistics
Of 1503 CANDLE participants initially recruited, 1324 had pCRH data and of those, 968 had data on child CBCL and/or IQ at age 4. Data on one or more covariates were missing for 110 participants, yielding a total of 858 mother-child dyads who contributed data to the current analysis (Table 1).When descriptively comparing the two groups of participants, more participants who were not included in the analysis due to missing data reported demographic characteristics reflecting lower levels of social and economic resources (e.g., public health insurance, "less than high school" as the highest level of education, having lower household income, being divorced/separated/widowed/single) than levels reported by those included in the analysis (Table A.1).On average, mothers in the analytic data set were 26.98±5.56years old with a pre-pregnancy BMI of 28.19±7.83kg/m 2 .Forty percent were nulliparas and 60% were married or living as married at the time of delivery.Roughly half (53%) had public insurance only, with the remainder having private insurance only (43%), both public and private insurance (3%), or no insurance (<1%).Most participants had a high school/GED/technical school level education (47%) or less (9%), and the mean household income (at the age 4 visit) was $22,559.Most mothers reported prenatal vitamin use (95%) and a small fraction reported prenatal alcohol use (8%).The two study visits occurred at mean 22.93±3.03and 31.85±1.72weeks gestation, and at those visits, pCRH concentrations were 56.47±78.59and 359.14±428.58pg/mL, respectively.Children were born at 39.03±1.39weeks gestation on average and 51% were assigned female at birth.

Models examining behavioral and emotional problems
In minimally adjusted models, few associations between pCRH measures and problem behaviors were observed (Table A.2). Fully adjusted models examining the whole cohort were similarly unremarkable, with pCRH measures generally weakly positively associated with problem behaviors but all confidence intervals including the null (Fig. 1,  Table A.2).For example, a one unit increase in estimated log pCRH at delivery was associated with 0.64 (95% CI − 0.70, 1.98) point higher Total Problems score, whereas an IQR increase in log(pCRH slope) was associated with a 0.45 (95%CI: − 1.21, 2.11) point higher Total Problems score.Extended model results were similar to those from fully adjusted models, with no strong associations observed (Table A

.2).
Estimates for secondary models evaluating problem behavior T-scores were also close to 0 with 95% confidence intervals that included the null (Table A.3).
When we fitted models that additionally included a pCRH*sex interaction term, we observed evidence of sex-dependent associations as indicated by statistically significant interaction terms for models examining pCRH slope in relation to all three problem behavior outcomes (Fig. 1, Table A.4).We observed significant interactions by sex in In models considering effect modification by potential moderators, interaction terms between pCRH measures and breastfeeding duration (never, 0-6 months, 6+ months) and the quality of child's neighborhood resources (COI score above versus below median) were not statistically significant (Table A.5).In post hoc analyses limited to male offspring, there were no statistically significant interactions between the pCRH measures and duration of breastfeeding (Table A.6) nor between the pCRH measures and COI score (Table A.7).

Models examining cognitive performance
In minimally adjusted models, estimated pCRH at delivery was positively associated with all measures of cognitive performance (FSIQ: β=3.08, 95%CI: 1.90, 4.27; Verbal IQ: β=1.73, 95%CI: 1.08, 2.38; Nonverbal IQ: β=1.35, 95% CI: 0.76, 1.95), whereas no associations with  pCRH slope were observed (Table A.2).These associations were attenuated in fully adjusted and extended models in which both estimated pCRH at delivery and pCRH slope showed weak, non-significant positive associations with cognitive performance scores (Fig. 1, Table A.2). Estimates for secondary models evaluating problem behavior T-scores were also close to 0 with 95% confidence intervals that included the null (Table A.3).In analyses examining pCRH*sex interactions, interaction terms were non-significant (Table A.4) as were the interactions between pCRH*breastfeeding history and pCRH*quality of child's neighborhood resources (Table A.5).
In sensitivity analyses excluding 68 children who were born preterm (<37 weeks), results for both behavioral and cognitive performance outcomes were similar to our primary analyses (Table A.8).In analyses restricted to children with FSIQ>70 (n=781), results were similar, with 95% CIs that included the null (not shown).

Discussion
In this study, contrary to our original hypotheses that were based upon evidence in smaller samples with limited adjustment for confounders, we observed few associations between maternal pCRH and age 4-6 neurodevelopmental outcomes in the whole cohort.However, there was evidence of sex-specific associations.Specifically, the rise in pCRH in mid-late pregnancy was associated with higher levels of behavioral problems in males.In females, the opposite pattern was observed, whereby pCRH rise was associated with fewer behavioral problems, particularly for internalizing symptoms.These opposing sex-specific associations were specific to pCRH rise (slope) and were not observed in models examining estimated pCRH at delivery (representing cumulative exposure) or in analyses examining children's cognitive performance.Overall, these findings complement a small body of prior work linking higher mid-gestation pCRH to behavioral problems in infants and young children, some in a sex-specific manner (Davis et al., 2005;Howland et al., 2016).Given the size (n=858) and socioeconomic and racial diversity of our cohort, these results may be more widely generalizable than prior studies on this topic, which were mostly based on small, homogeneous samples or clinical populations (Davis et al., 2005;Howland et al., 2016;Leviton et al., 2018).Other important advances in the current analysis include our novel methodology to estimate pCRH at delivery, representing cumulative pCRH exposure across gestation, and the consideration of postnatal factors that may modify the impacts of gestational pCRH exposure on neurodevelopmental outcomes.
Despite considerable interest in the potential impacts of pCRH on brain development over the last two decades, the literature on this topic remains limited, with most studies coming from what appears to be a single research sample drawn from California, US, and focused on internalizing symptoms.The earliest study on the topic suggested that higher pCRH in mid-pregnancy (N= 248) was associated with fearfulness and distress in early infancy, with follow-up analyses indicating that the associations were limited to females (Davis et al., 2005;Sandman et al., 2013).Subsequent follow-up of 91 children at age 5 again suggested positive associations between pCRH and internalizing symptoms, but the sex difference was no longer observed (Howland et al., 2016).While supporting the same general premise that pCRH may play an important role in neurodevelopment, our results have some notable differences from these prior findings.In contrast to the prior studies and despite the larger size of our cohort, we did not observe associations between pCRH measures and internalizing behaviors at age 4 when considering the whole cohort.Evidence of associations with internalizing behaviors (as well as externalizing and total problem behaviors) was only evident when we evaluated sex-specific associations.
Whereas some of those prior studies have suggested that higher pCRH exposure may have more adverse impacts on behavior in female children (Davis et al., 2005;Sandman et al., 2013), here, we instead observed adverse associations in male offspring and protective associations in females.Inconsistencies in patterns of findings for sex differences are recapitulated, to some extent, in the larger literature on prenatal stress hormone pathways and child neurodevelopment.That male fetuses may be more susceptible to gestational insults is well known and often colloquially characterized as "male vulnerability" or "male disadvantage" (Barrett and Lessing, 2021).Pregnancies with male fetuses are at greater risk of a number of complications including gestational diabetes, preterm birth, stillbirth, and postnatally, low Apgar scores, respiratory issues, and neurodevelopmental impairments (Hintz et al., 2005;Mage and Donner, 2004;Verburg et al., 2016).
What is less clear, however, is why increases in pCRH associated with lower risk of behavioral problems in girls.On the one hand, a viabilityvulnerability trade-off has been proposed, whereby female fetuses may be more resilient to gestational insults resulting in more favorable perinatal outcomes, but also more vulnerable to longer-term impacts of early adversity later in development (Sandman et al., 2013).Several studies of infant neurodevelopment support this framework, with associations between maternal affective symptoms or cortisol levels in pregnancy and temperament stronger among daughters compared to sons (Braithwaite et al., 2017a(Braithwaite et al., , 2017b;;Sandman et al., 2013).Alternatively, however, some research suggests female brain development may be sensitive to maternal prenatal cortisol exposure, with cortisol-linked changes in brain functioning mediating increased risk of female internalizing problems (Graham et al., 2019;Kim et al., 2017).There is also the possibility of gender-related differences in maternal report of children's symptomatology that may have some association with factors related to maternal pCRH.
That neurodevelopment may be differentially impacted by prenatal HPA activity (including pCRH) in males and females is plausible given the extensive sexual dimorphism of the brain evident even in infancy, including neural circuitry involved in the stress response such as the hypothalamus, amygdala, hippocampus, ventromedial and orbital prefrontal cortices, and anterior cingulate gyrus (Knickmeyer et al., 2014;Oyola and Handa, 2017).These tissues are rich in glucocorticoid as well as sex steroid receptors (McEwen et al., 2016) and in several studies, maternal cortisol in pregnancy has been associated with alterations in neural connectivity, although sex-specific associations have varied across studies, many of which have been quite small and not explicitly designed to study sex differences (Graham et al., 2019;Kim et al., 2017).Of greatest relevance to the current study, in follow-up of the California cohort, higher average maternal pCRH across gestation was associated with greater cortical thinning at age 6-9 years, particularly in the temporal and frontal lobes (Sandman et al., 2018).Some sex-specific patterns were noted whereby associations with pCRH at some exposure time points (19 and 31 weeks gestation) tended to be stronger in females compared to males.Additionally, associations with pCRH and cortical thinning tended to be local (for instance thinning in the temporal lobe) in females, but global in males, and it was hypothesized that associations between pCRH and neurocognitive development may be attributable to these anatomical changes in the brain.
Of note, in our study, there were no substantive associations between pCRH measures and child cognitive performance in the full cohort or differentially by child sex.To our knowledge, the prior studies on pCRH exposure and neurodevelopment have largely ignored measures of cognitive function, although one related study reported that higher maternal plasma cortisol levels were associated with lower verbal IQ scores at age 7 (LeWinn et al., 2009).Research in children born extremely preterm (<28 weeks) has examined pCRH gene expression in relation to a range of neurocognitive measures reporting that both low and high pCRH mRNA were associated with adverse outcomes including evidence of brain alterations, some Bayley sub-scale scores (particularly motor index), cerebral palsy, and motor dysfunction (Leviton et al., 2016).However, given the high-risk nature of that cohort and the known hyperproduction of pCRH that often accompanies preterm birth, applicability of those findings to a typically-developing population is questionable.Further, child neurodevelopment in that cohort was assessed at age two; in a follow-up study of the same children at age 10, few E.S. Barrett et al. associations were found (Leviton et al., 2018), suggesting that any risk pCRH might confer for neurodevelopment early in development may attenuate over time.Overall, to date, there is little evidence to suggest that in relatively healthy pregnancies, variation in pCRH is associated with child cognitive development.
Given a priori hypotheses that pCRH exposure would be associated with adverse neurodevelopment, we explored postnatal factors that might buffer associations, specifically breastfeeding and child-focused neighborhood opportunity, a geospatially-linked environmental measure encompassing socioeconomic and educational resources.In both cases, the interaction terms were not statistically significant.Recent research using the Childhood Opportunity Index suggests experiences of discrimination may blunt potential benefits conferred by access to neighborhood resources (Okuzono et al., 2023) and race-based discrimination is likely to be a common experience for many in this sample.
Our study has several important strengths.It is the largest study of pCRH and child neurodevelopment in a community sample to date and is based on a sociodemographically diverse cohort.With a nearrepresentative sample of Shelby County, Tennessee residents, CANDLE is one of the few pregnancy cohorts located in the underrepresented U.S. South.We used gold standard assays to quantify pCRH, a notoriously difficult analyte to measure accurately, and, based on those pCRH data, we leveraged a traditional measure of rise over gestation (slope) and also developed a novel measure of pCRH at delivery, representing cumulative pCRH exposure across gestation.We assessed multiple domains of neurodevelopment in childhood using both a widely-used validated behavioral questionnaire and a clinician-administered measure of cognitive performance.Finally, the rich characterization and long-term follow-up of this unique cohort allowed us to fit staged models adjusting for numerous covariates (including some that are often overlooked such as maternal cognitive performance and prenatal vitamin intake) and to consider potential postnatal protective factors.This rigorous adjustment for covariates may be one factor in our lack of detected main effects found in smaller, less adjusted analyses.
At the same time, we note some important limitations.We only assayed pCRH at two timepoints in mid-late pregnancy whereas some prior studies have measurements at up to 5 timepoints across pregnancy (Davis et al., 2005;Howland et al., 2016;Sandman et al., 2013).As a result, we cannot evaluate associations with potential windows of neurodevelopmental vulnerability occurring during early pregnancy.However, given that pCRH production is far higher in late pregnancy and brain development continues through (and even after) birth, we believe late gestation is a period of great biological importance.In addition, this was a medically low-risk cohort at maternal recruitment.While this can be considered a strength, it also means we are underpowered to look at higher-risk children including those born preterm.Our most notable associations were observed for problem behaviors and while the CBCL is a widely-used, validated tool, it is based on parental report which can be vulnerable to biases.Additionally, the possibility of residual confounding is an issue in any pregnancy cohort, and, although we adjusted for a wide range of covariates in a staged manner, there may be important confounders that were not measured.Finally, we fitted numerous models which could raise issues around multiple comparisons and increase the possibility of a Type I error.

Conclusions
In summary, we observed no association between pCRH and child cognitive performance after adjustment for covariates.Similarly, few associations were observed between cumulative pCRH exposure and child behavior, all of which suggests, at least among children born from full-term pregnancies, exposure to increased levels of pCRH during midlate gestation may confer little risk for children's future neurodevelopmental outcomes.However, the male-specific associations we observed between the rise in pCRH and problem behaviors warrant further study, as rise in pCRH may also be a risk factor for preterm birth, which itself is a risk factor for adverse neurodevelopment.Future studies may examine the extent to which preterm birth lies on the pathway between pCRH and child behavior.Further, the unexpected potential protective effects of rise in pCRH exposure on girls' behavioral problems calls for replication and further inquiry to understand these sex differences, ideally using large, diverse samples with pCRH measured at multiple points in gestation.We suggest that continued postnatal followup is warranted as our participants age which may allow us to better understand the discrepant sex differences observed here compared to prior studies.

Fig. 1 .
Fig. 1.Fully adjusted multiple linear regression models examining two measures of pCRH (log pCRH at delivery and pCRH slope) in relation to child behavior (CBCL scores) 1 and cognitive performance (IQ) 2 at age 4. Sex-specific estimates were derived from reparameterized models. 3Log(pCRH slope)was rescaled by IQR (0.047).CBCL: Child Behavior Checklist; FSIQ: full scale intelligence quotient; IQ: intelligence quotient.1 Adjusted for maternal age, child age at study visit, pre-eclampsia/gestational hypertension, birth order, maternal education, household income, insurance status, marital status, child opportunity index, maternal cotinine, duration of breastfeeding, childhood secondhand smoke exposure, maternal pre-pregnancy BMI, alcohol use in pregnancy, prenatal vitamin use.Models looking at all children are additionally adjusted for child sex. 2 Models with IQ as the outcome include the covariates above plus maternal IQ. 3 The n's are 844 and 858, respectively, for models examining CBCL scores and IQ scores.Reparameterized models include the full cohort (male and female) while also permitting the estimation of slopes for each level of the moderator (e.g., for males and females) without sacrificing power by stratifying the sample.For analyses of CBCL scores, the sample includes 418 males and 426 females.For analyses of IQ, the sample includes 421 males and 437 females.

Table 1
Characteristics of mother-child dyads included in the current analysis (n=858*).

Table 2
Distribution of child neurodevelopmental outcome scores among CANDLE participants included in the current analysis.