Human neurodevelopment requires the organization of neural elements into complex structural and functional networks called the connectome (1,2,3). While development of the connectome is contingent on many factors, emerging data suggest that prenatal exposure to maternal stress may also play a role (4,5,6,7). Stress is a signal in response to challenging and uncontrollable adverse events and perceived threat (8,9), and exposure to early life stress is a risk factor for neurobehavioral disorders ranging from autism spectrum disorder (ASD), attention deficit hyperactivity disorder and addiction, to depression and schizophrenia (10,11,12,13,14,15,16,17,18,19,20,21). Both high stress and stress-related conditions, including depression and anxiety, potently stimulate biological stress pathways (7), alter synaptogenesis (22,23), and change brain development (5,24,25,26,27,28). The prenatal period is critical for brain development, and prenatal stressors exhibit long-lasting influence on adult disorders, making stressor type and timing important factors to explore (27,29,30). Fetal sex and genetic variants may also mediate stress responsiveness (6,7,31,32,33,34).

Recent reports suggest that prenatal stress exposure (PNSE) is a global public health problem (13,29,35,36). PNSE has been reported in 10–35% of children worldwide (37). Nearly 8–23% of infants in the United States, or almost 800,000 neonates/year, experience prenatal exposure to depression (38,39), and reports from developing countries support similar numbers (40,41,42). Likewise, 1 in 7 to 1 in 13 pregnant women in the United States affirm symptoms of anxiety, while 5.6–14.8% in developing countries suffer a similar diagnosis (40,41,42,43,44). Since a nationally representative study found that more than half of the pregnant women (65.9%) experiencing depression in the United States went undiagnosed (45), these data may under-represent the problem.

PNSE is believed to both activate the hypothalamic-pituitary-adrenal (HPA) axis and result in epigenetic changes in the developing brain. This review will focus on converging preclinical and clinical imaging data to assess the impact of these changes in women’s stress-based physiology on the functional development of the human fetal brain. Prior to reviewing published data, we review common causes of PNSE and methods for measuring the structural and functional connectome. We also provide preliminary human data demonstrating increasing connectivity in limbic system structures across the third trimester of gestation.

Stress Models in Clinical and Translational Studies

While the relationship between maternal psychosocial stress and adverse pregnancy outcomes has been shown in many studies, it is important to define the nature of the stressor and the subject population (46). Stressors range from depression and anxiety to natural disasters, bereavement and steroid administration. As stressors vary, so may the outcomes. For a listing of outcomes, putative prenatal stressors and representative publications, please see Table 1 .

Table 1 Disorders and putative prenatal stressors

Depression and Anxiety

Although some older studies relied on retrospective recall measures and few evaluated the effects of increasing duration and strength of psychosocial stressors, more recent investigations have employed depression and anxiety as markers of maternal stress. Estimates suggest that 8–23% of women have symptoms of depression during their pregnancy (47). Likewise, 7.7–14% report anxiety, and there are numerous reports of coexisting depression and anxiety in the same pregnant woman at any given time. While depression and anxiety are common proxies for stress, stress in pregnant women does not always coincide with elevated depression or anxiety scales. As such, cases of PNSE may be missed in such analysis. Finally, depression and anxiety may have independent or additive effects in regards to PNSE, making it difficult to fully disentangle these effects with this model. For a more complete review of this topic, please see Suri et al. (48).

Natural Disasters

Another approach to test the hypothesis that PNSE results in neurobehavioral disorders is the use of natural disasters as “experiments of nature.” Unlike depression and anxiety, natural disasters are independent of the subject’s genetic background, personality or other confounding characteristics. Disasters strike in a random manner, similar to a randomized controlled experiment, and thus can provide data on prenatal stressors to which a given cohort of pregnant women were exposed (13). Using this strategy, the impact of disasters ranging from hurricanes to terrorist attacks on neurobehavioral outcomes of the offspring have been assessed (13,49,50,51,52).

Preconception Stress

In contrast, the influence of preconception adversity and the impact of high cumulative stress on maternal perception of prenatal stress on the developing connectome are just beginning to be explored (53,54,55). Consistent with preclinical studies showing effects of repeated stress on neural atrophy and neurobehavioral effects (56), human studies link altered structure and function of limbic, subcortical, and frontal regions to higher levels of cumulative stress (28,57,58). These data suggest that preconception adversity may shape perception and control of prenatal stress levels and should be considered in investigations of PNSE on neurobehavioral and MRI outcomes.

Prenatal Maternal Stress

PNSE has been widely associated with preterm birth, intrauterine growth restriction, and reduced fetal head growth (50,51,59,60,61). In addition, several studies have reported that increased acute maternal stress is associated with changes in fetal heart rate, activity level, sleep patterns, and higher pulsatility indices in the middle cerebral artery (21,60). PNSE can also be directly measured using prospective data collections in samples of pregnant women with questionnaires, clinical interviews, and biological samples such as cortisol from maternal saliva, blood, or amniotic fluid.

Methods to Assess Connectivity Using MRI

Advances in neuroimaging provide important information about microstructural and functional connectivity (62), and offer opportunities to understand the impact of PNSE on the developing connectome (1,2,3). In the following section, we define measures commonly used in connectomics with examples shown in Figure 1 .

Figure 1
figure 1

Examples of functional and structural connectivity. (a) Functional connectivity. Functional connectivity measures the synchrony or correlation of brain activity between two or more regions of the brain. Common methods include seed connectivity, independent components analysis, and voxel-wise connectivity. Seed-based connectivity measures functional connectivity from a predefined region of interest (ROI, or seed, shown in green) and the rest of the gray matter. Regions of positive or negative functional connectivity are shown as red and blue regions. Independent components analysis is mathematical modeling technique that parcellates the brain into independent spatial components or networks. Example components shown are the motor network and the defualt mode network. Voxel-wise connectivity methods involve correlating the time course of every voxel in the gray matter with the time course of every other voxel in the gray matter. Connectivity for each voxel are often summarized to a single number using network theory to highlight so-called hub regions in the brain. (b) Structural connectivity. Structural connectivity measures anatomical white matter connections linking different cortical and sub cortical regions. Common methods include ROI quantification, tract-based and tractography, and voxel-based morphometry. For ROI quantification, average FA across all voxels in a priori ROIs (shown in white outline and overlay) is compared across study groups. Tractography is modeling technique used to identifying white matter tracts used in further analyses. In VBM analysis, FA data from all subjects is transformed into a common space and comparison across each voxel of the white matter is performed. (Figure modified with permission, John Wiley & Sons, Hoboken, NJ)

PowerPoint slide

Functional connectivity provides information about neural regions that are physiologically functionally coupled, independent of structural connectivity (63). Based on the blood oxygen level dependent signal and derived from time series observations, it assesses “temporal correlations between spatially remote neurophysiological events.” (63) High correlation between time courses of two regions or voxels implies high functional connectivity. For the references included in this review, functional MRI (fMRI) data are largely collected in the resting state, or resting state-fMRI (rs-fMRI).

Methods to assess rs-fMRI data (64,65) include seed, independent components analysis, and voxel-wise connectivity. Seed-based connectivity is most frequently used in human studies and involves (i) selecting a predefined region of interest (ROI), (ii) extracting the average time course from this ROI, and (iii) correlating this average time course with the time courses of every other voxel in the gray matter. Independent components analysis is mathematical modeling technique that parcellates the brain into independent spatial components or networks. These networks can be compared across stubject groups or used for later analysis. Voxel-wise connectivity methods are generalizations of seed-based connectivity where many seed connectivity analyses are performed treating each voxel in the gray matter as a unique ROI. As these methods produce a large amount of data (approximately 20,000 seed-connectivity results), seed-connectivity results for each voxel are often summarized to a single number using network theory.

Anatomical connections in the developing brain represent microstructural connectivity (66). Diffusion-weighted imaging (dMRI) assesses the diffusion of water along axons and permits visualization of axonal pathways. By modeling the directional diffusion of water as an ellipsoidal shape, or “tensor”, at each voxel in the brain, dMRI permits assessment of white matter tracts. The first eigenvector, λ1, describes the direction of maximal diffusion, while the second and third define diffusivity perpendicular to this principle axis. Radial diffusivity represents the average of λ2 and λ3 and is affected by changes in axon caliber and myelination. Fractional anisotropy (FA) measures the degree to which water diffuses in one direction (along the axon) by computing the ratio of λ1 to λ2 and λ3 and is the most common measure used to assess axonal integrity. High values of FA suggest more highly organized, strongly myelinated tracts.

The three main approaches to analyzing dMRI data include region ROI quantification, tract-based analysis and tractography, and voxel-based morphometry (VBM). ROI quantification is frequently used in human studies investigating the impact of PNSE on the developing connectome. In this method, one or more ROIs are selected a priori and the average FA across all voxels in the ROI calculated. Typically, ROIs are major white matter tracts. Tractography is modeling technique used to identify these tracts. Once identified, they can be analyzed using graph theory or ROI analyses. In VBM analysis, FA data from all subjects are transformed into a common space and compared across each voxel of the white matter.

Finally, although not direct measures of the connectome, we include studies assessing brain morphometry, including cortical volumes and thickness. Morphological features of different brain regions are not independent of those of other areas, and the brain shows a high level of coordination between different structures (67). This coordination of morphological features is often referred to as anatomical covariance (67,68,69) and resembles functional and structural connectivity.

Preclinical Data Support the Impact of PNSE on Developing Connectome

Across multiple species and numerous time points, converging data suggest that gestational stress influences brain development. Similar to the human subjects, the offspring of numerous species exposed to PNSE demonstrate increases in anxiety and depression, impaired spatial memory and alterations in cognition (70,71). Systematic experimental investigations using standardized animal models and outcome measures (6,72,73) address not only the impact of PNSE on maternal endocrine functions and the “re-programming” of the fetal HPA axis (74,75,76,77,78,79), but also suggest that changes in corticogenesis contribute to the long-lasting effects on brain and behavior ( Table 2 ) (74,80,81).

Table 2 Prenatal stress and the connectome: preclinical data

MRI Studies of PNSE and the Developing Brain

While the neural correlates of acute and cumulative postnatal stress in human subjects are active fields of study, MRI research investigating PNSE in human subjects is just starting to be explored. As described below and shown in Table 3 , many investigators have interrogated the impact of PNSE on the limbic system and connected regions in the developing brain.

Table 3 Prenatal stress and the connectome: clinical data

Studies during infancy

Recent studies suggest a significant relationship between antenatal maternal depression and/or anxiety and structure and function in the developing brain. Rifkin-Graboi performed structural MRI and dMRI on 157 nonsedated 6–14-day-old newborns whose mothers participated in the GUSTO study (Growing Up in Singapore Towards Healthy Outcomes), a cohort of Asian women enrolled during the first trimester of pregnancy. Socioeconomic status, prenatal exposures, pregnancy measures, and birth outcomes were recorded, and imaging data were analyzed only for those infants who met the following criteria: (i) gestational age (GA) ≥37 wk, (ii) birth weight (BW) >2,500 g, and (iii) Apgar5 min > 7. The Edinburgh Postnatal Depression Scale and the State Trait Anxiety Inventory (STAI) were administered to all women at 26 wk of pregnancy. Adjusting for household income, maternal age and smoking exposure, postmenstrual age (PMA) at MRI, and BW, Rifkin-Graboi (24) found significantly lower FA but not volume in the right amygdala in infants of mothers with high EDPS scores. This suggests a significant relationship between PNSE and microstructure of the right amygdala, a region associated with stress reactivity and vulnerability for mood disorders.

Similarly, Qiu interrogated the GUSTO cohort to examine the consequences of PNSE to maternal anxiety on neonatal development of the hippocampus, a structure critical for stress regulation (27). Entry criteria for this analysis differed from those of Rifkin-Graboi’s 2013 study, and included both term and late preterm infants who met the following criteria: (i) GA ≥ 35 wk; (ii) BW > 2,000 g; and (iii) Apgar5 min > 9. There were 175 GUSTO infants available for this analysis; 42 underwent repeat scans at age 6 mo, and 35 (83%) had usable data. In Qiu’s analysis, antenatal maternal anxiety did not influence bilateral hippocampal volume at birth, but children of women with increased anxiety during pregnancy showed slower growth of both the left and right hippocampus between birth and age 6 mo. Subsequently, evaluating 21 GUSTO infants with high PNSE (i.e., maternal STAI > 90) and 34 with low PNSE (i.e., maternal STAI < 70), Rifkin-Graboi showed that antenatal anxiety predicted decreases in FA of regions important for cognitive-emotional responses to stress (i.e., right insula and dorsolateral prefrontal cortices (PFC)), sensory processing (right middle occipital cortex), and socio-emotional function (i.e., right angular gyrus, uncinate fasciculus, posterior cingulate, and parahippocampus) at age 5–17 d (25). Of note, infants were eligible for this analysis if met the following criteria: (i) GA ≥ 36 wk; (ii) BW > 2,000 g; and (iii) Apgar5 min > 7.

Finally, Scheinost (82) and Qiu (26) investigated prenatal depression/anxiety exposure and amygdala connectivity using rs-fMRI in preterm neonates at term equivalent age and infants at age 6 mo, respectively. These data showed that, in the neonatal period, the amygdala is functionally connected to subcortical and posterior cortical regions, and, by age 6 mo, is connected to widespread networks subserving emotional regulation, memory, and social cognition. In preterm neonates, Scheinost showed that PNSE reduces amygdalar-thalamic connectivity and is additive to effects of preterm birth. Using 24 GUSTO infants, Qui showed that infants born to mothers with higher prenatal depressive symptoms had greater rs-fMRI of the amygdala with the left temporal cortex, insula, anterior cingulate (ACC), medial orbitofrontal, and ventromedial PFC. These networks are reported in children and adults with depression, suggesting that rs-fMRI data may foreshadow future neuropsychiatric disease.

Studies during childhood

Studies of older children also suggest that maternal anxiety is associated with specific changes in brain morphology. Buss evaluated children ages 6–10 y whose mothers had been enrolled in a prospective study of pregnancy at the University of California, Irvine or Cedars Sinai Hospital in Los Angeles, CA, between 1998 and 2002 (83). Families were contacted again in 2007 and invited to participate in a follow-up study of their children to assess the influence of PNSE on brain development. At the time of this report, 35 mother–child dyads had both usable MRI data and complete maternal data. VBM on these children demonstrated that exposure to high maternal stress at 19 wk of gestation correlated with gray matter reductions in the PFC, premotor cortex, medial temporal lobe, lateral temporal cortex, post-central gyrus, and cerebellum extending to the middle occipital and fusiform gyri. Although the numbers are small and assessments of postnatal stress exposure were not included in the authors’ analyses, high pregnancy stress at 25 and 31 wk of gestation was not associated with local reductions in gray matter volume, suggesting the importance of earlier exposure to gestational psychological stress. Similarly, Sarkar performed dMRI studies to assess both FA and perpendicular diffusivity (Dperp) on 22 children ages 6–9 y whose mothers were retrospectively assessed for PNSE when the children were age 17 mo (84). For these children, PNSE was positively correlated with right uncinate FA and negatively with right uncinate Dperp, while PNSE was not associated with control tract properties.

In addition, since reduced cortical volume and thickness have both been associated with a history of depression in adult populations (85,86), Sandman measured cortical thickness in 81 school aged children whose mothers had participated in the longitudinal study described above; (83) all were prospectively evaluated for depression at 19, 25, and 31 wk of gestation (47). Prenatal maternal depression exposure was associated with thinning in the right frontal lobe, and the strongest association was with exposure at 25 wk gestation. Morphological changes were primarily found in the superior, medial orbital, and frontal pole regions of the right PFC, consistent with data in adults with depressive symptomatology (85,86). Further, the significant association between prenatal depression exposure and child externalizing behavior in this cohort of children was mediated by these changes.

Studies during adulthood

Finally, although MR studies of young adults with early life stress exposures are just beginning to emerge (87,88,89), Favaro explored the relationship between PNSE, cortical volumes and rs-fMRI in a sample of 35 healthy women aged 14–40 y (90). The sample was composed of volunteers to whose mothers a semi-structured interview assessing stress related events during pregnancy was administered. Subject scores were assigned based on interview data and used for MRI analyses. For these women, greater PNSE was associated with decreased gray matter volume in the left medial temporal lobe and both amygdalae. Strength of PNSE was positively correlated with rs-fMRI between the left medial temporal lobe and pre-genual cortex, and connectivity between the left medial temporal lobe and left medial-orbitofrontal cortex partially explained variance in depressive symptoms in this cohort.

Emerging Factors

As studies begin to investigate the impact of PNSE on the connectome, several factors from both preclinical and clinical data have emerged as key considerations for future studies. These factors include defining the normal developmental trajectories of the fetal connectome, the type, timing and duration of PNSE, and the fetal sex. Other factors not addressed within this review include assessing the role of paternal preconception stress and identifying the molecular signatures of PNSE.

Fetal networks

“Trajectory analysis is central to the assessment of the impact of PNSE on the developing brain.” (2) Fetal rs-fMRI is an emerging technology obtaining information about neural network development in utero by directly measuring the fetal brain (91). These methods are needed to investigate the prenatal connectome as it develops and pinpoint how and when PNSE alters its development. Using cross-sectional functional connectivity data between 21–38 wk of gestation, fetuses show evidence of both long-range functional connectivity and the emergence of neural networks across the third trimester, mimicking those in older children and adults (92,93). However, both longitudinal and cross-sectional data are needed to more fully characterize the developmental trajectories of PNSE. To begin to address this problem, we performed longitudinal rs-fMRI on 10 typically developing fetuses at 30–32, 34–36 wk PMA and following term delivery. This study was approved by the Yale University Human Investigation Committee, and pregnant women signed consent for the protocol. Because of its documented role in neurobehavioral disorders and alterations in studies of PNSE described above, we interrogated the emergence of amygdala networks during the prenatal period. During the 3rd trimester, left amygdala connectivity is first characterized by local circuitry, then begins to connect to ipsilateral regions in the frontal and temporal lobes, and finally develops connections to the contralateral amygdala ( Figure 2 ). The development of these important cross-hemispheric connections between the right and left amygdala develop during the end of the third trimester and likely increases the vulnerability of this circuitry to PNSE (55).

Figure 2
figure 2

Development of left amygdala functional connectivity during the perinatal period. During the perinatal period, (a) left amygdala connectivity is first characterized by a largely local circuitry, (b) then begins to connect to ipsilateral regions in the frontal and temporal lobes, and (c) finally develops connections to the contralateral amygdala.

PowerPoint slide

Timing of stress exposure

There is increasing recognition that fetal stress exposure has a particularly pronounced impact during early periods of corticogenesis, commonly known as critical periods in the developing brain. Critical periods refer to epochs characterized by both increasing plasticity and greater vulnerability; thus, these are times when the developing brain may be most easily modified in either favorable or unfavorable directions. Critical periods are thought to be environmentally sensitive, and many authors believe they underlie the developmental origins of neurobehavioral disorders such as ASD.

Typically developing fetuses with PNSE during the middle second and third trimesters of gestation are reported to be at the greatest risk for neurobehavioral disorders (13,52). Reviewing Swedish birth registries, Class examined associations between PNSE in 738,144 offspring born in 1992–2000 for childhood outcomes and 2,155,221 offspring born in 1973–1997 for adult outcomes. Although data for GA are not available, third trimester bereavement stress significantly increased risk of both ASD and attention deficit hyperactivity disorder (55). Similarly, children who had been exposed to tropical storms during gestation months 5–6 or 9–10 had 3.8 times greater risk of developing ASD than children who had been exposed to the same storms, in the same place, but during other months of gestation (52). Duration of maternal stress may also play a role. Analyzing data from 4,682 live births, Latendresse reported that children of mothers with the longest periods of prenatal depression exposure experienced more than seven times increased risk for pervasive developmental disorder when compared to children with no PNSE (53).

In contrast, in the GUSTO study, mothers were assessed for gestational depression and/or anxiety at 26 wk, and MRI measures were correlated with these data (24,25,26,27). In addition, Sandman performed depression screening on 82 pregnant mothers at 19, 25, and 31 wk gestation and found that antenatal exposure to maternal depression at 25 wk gestation was significantly correlated with cortical thinning in 24% of the frontal lobes in the offspring (47). Finally, although cortisol levels are not available for subjects in the prior MRI studies, high levels of maternal cortisol at 15 wk (but not 19, 25, 31, or 37 wk) of gestation were associated with amygdala volumetric changes in girls but not in boys (33). Since high levels of cortisol are believed to reprogram the fetal HPA axis and maternal stress has been reported to downregulate 11 β-hydroxysteroid dehydrogenase (75), the placental enzyme which metabolizes cortisol (75), future studies of maternal psychological stress during gestation should consider longitudinal assessments of maternal cortisol in tandem with fetal neuroimaging.

Sex differences in prenatal stress outcomes

The link between PNSE and outcomes may be moderated by fetal sex. The source of sex differences upon early development is unclear but may include placental functioning, exposure to adrenal hormones and testosterone and an assortment of epigenetic mechanisms (94,95,96,97). Recent fetal pathways also proposed include sex-dependent responses of the transcriptome (6,98,99,100), naturally occurring sexually-dimorphic processes mediating neuron-glial interactions (101), and differential responses of target regions in the developing brain (102). Thus, while PNSE may have consequences for both males and females, the specificity of effects may differ. To the best of our knowledge, however, only a single study has reported sex differences in MRI outcome measures. These data suggest that higher cortisol levels at 15 wk of gestation were associated with larger right amygdala volumes and more affective problems in female but not male offspring (33).

Mechanisms of Prenatal Stress and the Connectome

Taken together, published studies of PNSE suggest both proximate and long-lasting influence on the connectome. However, mechanisms of how PNSE alters the developing connectome must be explored. Mechanistic studies have focused on the HPA axis, candidate genes, and epigenetic pathways (see Table 4 ).

Table 4 Prenatal stress and the connectome: Endocrine and genetic mechanisms

Moving Forward: Investigation of the Clinical Problem, Changes in Care

Converging data suggest that PNSE alters the developing connectome. As noted by Sporns, “The placement of brain connectivity as an intermediate phenotype between environmental exposures and behavior makes it an important target for studies that link networks across levels from behavior to molecules, neurons and emerging networks in the developing brain” (62). To better address the impact of PNSE on the connectome, longitudinal studies of maternal/fetal dyads with and without stress exposure are needed. Such investigations would benefit from repeated assessments of maternal stress in order to identify type, time of onset, and duration of PNSE and correlate these data with sequential imaging. In addition, preconceptional stress may influence offspring outcome, and pregnant women should be surveyed for cumulative stress at the time of study enrollment. Likewise, both genetic variants and epigenetic changes may contribute to outcome in the offspring, and consideration should be made to include these data in PNSE-offspring outcome analyses. Finally, longitudinal fetal imaging will provide important information about target regions, and developmental trajectory analyses are well suited for interrogation of the developing connectome.

These strategies can be used to detect developmental disturbances of the connectome that may underlie the development of neurobehavioral disorders.

Statement of Financial Support

This work was supported by Gates Foundation OPP1119263, Seattle, WA, and National Institutes of Health T32 HD07094, Bethesda, MD.


None of the authors have disclosures.