Decreased amygdala-sensorimotor connectivity mediates the association between prenatal stress and broad autism phenotype in young adults: Project Ice Storm

Abstract

Studies show that prenatal maternal stress (PNMS) is related to risk for child autism, and to atypical amygdala functional connectivity in the autistic child. Yet, it remains unclear whether amygdala functional connectivity mediates the association between PNMS and autistic traits, particularly in young adult offspring. We recruited women who were pregnant during, or within 3 months of, the 1998 Quebec ice storm crisis, and assessed three aspects of PNMS: objective hardship (events experienced during the ice storm), subjective distress (post-traumatic stress symptoms experienced as a result of the ice storm) and cognitive appraisal. At age 19, 32 young adults (21 females) self-reported their autistic-like traits (i.e., aloof personality, pragmatic language impairment and rigid personality), and underwent structural MRI and resting-state functional MRI scans. Seed-to-voxel analyses were conducted to map the amygdala functional connectivity network. Mediation analyses were implemented with bootstrapping of 20,000 resamplings. We found that greater maternal objective hardship was associated with weaker functional connectivity between the left amygdala and the right postcentral gyrus, which was then associated with more pragmatic language impairment. Greater maternal subjective distress was associated with weaker functional connectivity between the right amygdala and the left precentral gyrus, which was then associated with more aloof personality. Our results demonstrate that the long-lasting effect of PNMS on offspring autistic-like traits may be mediated by decreased amygdala-sensorimotor circuits. The differences between amygdala-sensory and amygdala-motor pathways mediating different aspects of PNMS on different autism phenotypes need to be studied further.

Keywords:

• Amygdala
• functional connectivity
• prenatal maternal stress
• broad autism phenotype
• postcentral gyrus
• precentral gyrus

Introduction

Prenatal environmental exposures have long-lasting effects on offspring neurodevelopment (Lautarescu et al., 2020; van den Bergh et al., 2018). For example, childhood autism spectrum disorder (ASD) is associated with different types of prenatal maternal stress (PNMS): depression (Gao et al., 2015; Say et al., 2016; Zhang et al., 2010), bereavement (Class et al., 2014), life events (Beversdorf et al., 2005; Rai et al., 2012; Varcin et al., 2017), and hurricanes and tropical storms (Kinney et al., 2008).

The association between PNMS and child ASD may be the result of altered amygdala activity. The dysregulation of the fetal hypothalamic-pituitary-adrenal (HPA) axis may be one possible mechanism by which prenatal stress influences offspring physiological and psychological outcomes (Howland et al., 2017). The amygdala has been implicated in the regulation of the HPA axis in the following ways. During stress, corticotropin-releasing hormone (CRH) is synthesized in the paraventricular nucleus of the hypothalamus and released into the hypophyseal portal blood CRH binds to its receptors on pituitary corticotropes, stimulating the release of adrenocorticotrophic hormone that binds to its receptors in the adrenal cortex and that stimulates the release of glucocorticoids (Howland et al., 2017). CRH-producing neurons in the paraventricular nucleus of the hypothalamus are largely innervated by excitatory projections from the amygdala (Herman et al., 2005). It could also be plausible that prenatal stress operates directly on fetal neurodevelopment; stress stimulates the release of maternal glucocorticoids, and the elevated maternal glucocorticoid levels attenuate 11β-hydroxysteroid dehydrogenase type-2 expression and cause elevated fetal glucocorticoid load (Mandy & Nyirenda, 2018). The fetal amygdala is particularly sensitive to glucocorticoids given its high concentration of glucocorticoid receptors (Lautarescu et al., 2020).

It is evident that the amygdala is involved in the stress response and the regulation of stress memory (Roozendaal et al., 2009). Studies utilizing resting-state functional MRI (rs-fMRI) have shown the vulnerability of the offspring amygdala to PNMS exposure: prenatal maternal depression (Posner et al., 2016) and prenatal maternal life events (Humphreys et al., 2020) are associated with weaker resting-state functional connectivity (rs-FC) from the amygdala to the prefrontal cortex (PFC) in neonates.

Altered amygdala rs-FC has then been observed in individuals with ASD. Given social impairment as the core symptomatology of ASD, attention has been paid to the amygdala that plays a critical role in social information processing (Fishman et al., 2018). Studies using rs-fMRI have reported abnormalities in rs-FC from the amygdala to subcortical and cortical regions in individuals with ASD compared to controls, with lower rs-FC from the amygdala to the medial PFC, bilateral temporal lobes, striatum, thalamus, cingulate cortex, and cerebellum in preschool-aged children with ASD (Shen et al., 2016), with lower rs-FC from the amygdala to the thalamus and putamen in adolescents with ASD (Guo et al., 2016), and with lower amygdala-occipital rs-FC and higher amygdala-pericentral rs-FC in adolescents with ASD (Fishman et al., 2018). In addition, among the adolescents with ASD, the lower amygdala-thalamus/putamen rs-FC has been associated with more severe social deficits (Guo et al., 2016).

Unlike naturally-occurring life events (e.g., job loss or divorce), disasters tend to have sudden onsets and to be independent of the control of individuals (King & Laplante, 2015). Prenatal exposure to a natural disaster is a type of PNMS that is independent of potential confounders (e.g., maternal personality and socioeconomic status). In the case of the Project Ice Storm, the family socioeconomic status is uncorrelated with the degree of objective hardship (r = −0.03) and is only weakly correlated with the women’s subjective distress (r = 0.22) (Laplante et al., 2007).

In January 1998, three separate weather systems passed through southern Quebec producing five continuous days of freezing rain and leaving up to 100 mm of ice on the region. The weight of the ice toppled electrical poles and high-tension power lines and pylons leaving more than 1.5 million households without power for as long as 45 days in the coldest month of the year. With $1.5 billion in damages and more than 27 deaths, the 1998 ice storm is one of the worst disasters in Canadian history. In June 1998, we launched Project Ice Storm, in which we recruited women who gave birth in 1998, assessed three aspects of their stress caused by the ice storm, and prospectively followed up their children development. We found that prenatal maternal objective hardship and subjective distress were associated with autistic-like traits at age 6½ years as rated by their mothers, especially when exposure was in early pregnancy (Walder et al., 2014). At age 19, the children self-reported their symptoms of the broad autism phenotype (BAP) including aloof personality, pragmatic language impairment and rigid personality, and we found that different aspects of PNMS explain different phenotypes, with maternal objective hardship and cognitive appraisal related to aloof personality and rigid personality, and maternal subjective distress related to pragmatic language impairment (Li et al., 2023).

Given the vulnerability of the amygdala to PNMS, and given its possible involvement in ASD, atypical amygdala rs-FC may mediate the association between PNMS and ASD symptoms, although this has never been demonstrated. Here, we aimed to determine (1) amygdala rs-FC network in the young adults of mothers exposed to disaster-related PNMS; and (2) the extent to which the association between disaster-related PNMS and BAP is mediated by atypical amygdala rs-FC.

Methods

Participants

Project Ice Storm: Women with due dates between January 9, 1998 and December 31, 1998 were identified by their physicians who mailed the questionnaire to them on June 1, 1998. The initial Project Ice Storm cohort consisted of 176 women who responded to the first questionnaire and who agreed to further contact. These mother-child dyads were followed up, and the children’s development was assessed, periodically between the ages of 6 months and 19 years. At age 19, 33 youths completed the Broad Autism Phenotype Questionnaire (BAPQ); of those, 32 underwent structural MRI and rs-fMRI scans. The most common structural MRI sequence is the T1-weighted (T1w) scan, a 3D image of the entire brain that is represented in a volume of a given matrix. The most common functional MRI technique uses fluctuations of blood-oxygen-level-dependent (BOLD) contrast based on the magnetic susceptibility of hemoglobin. During rest (absence of tasks), participants were instructed to relax and not to think of anything in particular, and their level of spontaneous brain activity was measured, which was referred to as rs-fMRI (Glover, 2011).

Ethical approval

All phases of this study were approved by the Research Ethics Board of Douglas Hospital Research Center in Montreal, Canada. All participants gave written informed consent.

Prenatal maternal stress

In June 1998, three aspects of PNMS were measured by the postal questionnaire: objective hardship (i.e., events experienced during the ice storm), subjective distress (i.e., post-traumatic stress symptoms experienced as a result of the ice storm), and cognitive appraisal (i.e., judging the consequences of the ice storm). These three measures were only weakly correlated with each other (|r|s = 0.211 − 0.372). The severity of objective hardship experienced by pregnant women during January 1998 ice storm was assessed based on four dimensions of exposure: Threat (e.g., injuries), Loss (e.g., loss of personal income), Scope (e.g., duration without electricity), and Change (e.g., temporary shelter) (Bromet & Dew, 1995). There were eight points for each dimension, and a total score (Storm32) was calculated by summing scores across all four dimensions (Laplante et al., 2007). Storm32 has shown predictive validity by correlating significantly with a number of child outcomes including functional play development (Laplante et al., 2007) and cognitive and language development (King & Laplante, 2005) at 2 years of age.

Women’s subjective distress at the time of recruitment was assessed using a validated French version (Brunet et al., 2003) of the Impact of Event Scale—Revised (IES-R) (Weiss et al., 1997). This 22-item scale has been widely used for assessing the severity of post-traumatic stress disorder symptoms in the preceding week including three categories: Intrusive Thoughts, Hyperarousal, and Avoidance. The IES-R has also predicted a number of child behaviors including autistic-like traits at 6½ years (Walder et al., 2014) and externalizing behaviors via enlarged amygdala volume at 11½ years (Jones et al., 2019). In the current study, we used log-transformed values of the total score due to the skewed distribution.

Women’s cognitive appraisal of the ice storm was assessed using the following question: “Overall, what were the consequences of the ice storm on you and your family?”; the five response options were coded into three levels: “negative” (“-1”), “neutral” (“0”), and “positive” (“1”). Women’s negative cognitive appraisal has altered several child outcomes (e.g., BMI and C-peptide) via DNA methylation at 13½ years (Cao-Lei et al., 2016, 2018).

Broad autism phenotype

The BAPQ was specifically designed to assess autistic-like traits in non-clinical populations, and this self-report tool has good validity in clinical assessment of the broad autism phenotype (Sasson et al., 2013). The BAPQ includes 36 items assessing three domains: Aloof Personality (12 items), Pragmatic Language (12 items), and Rigid Personality (12 items). Each item is rated using a 6-point Likert scale ranging from “very rarely” (1) to “very often” (6) (Sasson et al., 2013). Cronbach’s alpha coefficient was 0.94 for the Aloof Personality subscale, 0.91 for the Rigid Personality subscale, 0.85 for the Pragmatic Language subscale and 0.95 across all 36 items. Aloof personality is defined as a lack of interest in, or enjoyment of, social interactions (corresponding to social deficits of autism). Pragmatic language impairment refers to deficits in the social aspects of language, resulting in difficulties communicating effectively or in holding a fluid, reciprocal conversation (corresponding to communication deficits of autism). Rigid personality is defined as little interest in change or difficulty adjusting to change (corresponding to restricted and repetitive behaviors of autism) (Hurley et al., 2007).

MRI data acquisition

MR images were acquired using a 3 T Siemens MAGNETOM Trio TIM Syngo MRI scanner, with a 12-channel head coil. Anatomical images were obtained using a T1w Magnetization Prepared Rapid Gradient Echo sequence (192 slices; TR = 2400 ms; TE = 2.43 ms; slice thickness = 1 mm; flip angle = 8°; matrix = 256 × 256). Resting-sate functional images were acquired using a T2*-weighted echo-planar imaging sequence (42 slices; TR = 2600 ms; TE = 30 ms; flip angle = 90°; slice thickness = 3.4 mm; FoV = 218 mm, matrix = 64 × 64). Throughout the 5:01 min resting state scan, participants were instructed to lie still with their eyes open.

MRI data preprocessing

fMRIPrep 1.5.7 (Esteban et al., 2019) was used for preprocessing. T1w images were corrected for intensity non-uniformity with N4BiasFieldCorrection (Tustison et al., 2010), distributed with ANTs 2.2.0 (Avants et al., 2008), and used as T1w-reference throughout the workflow. The T1w-reference was then skull-stripped with a Nipype implementation of the antsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as target template. Brain tissue segmentation of gray matter, white matter, and cerebrospinal fluid was performed on the brain-extracted T1w using fast FSL 5.0.9 (Zhang et al., 2001). Volume-based spatial normalization to the Montreal Neurological Institute (MNI) space was performed through nonlinear registration with antsRegistration (ANTs 2.2.0), using brain-extracted versions of both T1w reference and the T1w template. For each of the average BOLD runs found per subject, the following preprocessing was performed. First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. A deformation field to correct for susceptibility distortions was estimated based on fMRIPrep’s fieldmap-less approach. The deformation field is that resulting from co-registering the BOLD reference to the same-subject T1w-reference with its intensity inverted (Huntenburg, 2014; Wang et al., 2017). Registration is performed with antsRegistration (ANTs 2.2.0), and the process regularized by constraining deformation to be nonzero only along the phase-encoding direction, and modulated with an average fieldmap template (Treiber et al., 2016). Based on the estimated susceptibility distortion, a corrected echo-planar imaging reference was calculated for a more accurate co-registration with the anatomical reference. The BOLD reference was then co-registered to the T1w reference using flirt FSL 5.0.9 (Jenkinson & Smith, 2001) with the boundary-based registration (Greve & Fischl, 2009) cost-function. Co-registration was configured with nine degrees of freedom to account for distortions remaining in the BOLD reference. Head-motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) are estimated before any spatiotemporal filtering using mcflirt FSL 5.0.9 (Jenkinson et al., 2002). The BOLD time-series were resampled onto their original, native space by applying a single, composite transform to correct for head-motion and susceptibility distortions. The BOLD time-series were then resampled into the MNI space. Automatic removal of motion artifacts using independent component analysis (ICA-AROMA) (Pruim et al., 2015) was performed on the spatially-normalized, preprocessed BOLD on MNI space time-series after removal of non-steady state volumes and spatial smoothing with an isotropic, Gaussian kernel of 6 mm full width at half maximum. In addition, we removed signals of white matter and cerebrospinal fluid from the BOLD time series and temporally bandpass filtering (> 0.01 Hz).

Resting-state functional connectivity analysis

Functional connectivity analyses were conducted using CONN functional connectivity toolbox 19b (Whitfield-Gabrieli & Nieto-Castanon, 2012). For amygdala (left and right) masks, we used the Harvard-Oxford structural atlas. In the individual level, seed-to-voxel analyses were implemented to examine the amygdala rs-FC network: Pearson’s correlation coefficients were calculated between BOLD time series extracted from each amygdala mask (left and right, separately) and the time courses of all voxels across the brain, for each participant. The resultant correlation coefficients were converted to normally distributed z-values. Group-level one-sample t tests were used to map amygdala rs-FC network. Positive mean z-values indicate positive functional connectivity between BOLD signals in the amygdala and a cluster (i.e., the two regions are likely to be active at the same time), while negative mean z-values indicate negative functional connectivity between BOLD signals in the amygdala and a cluster (i.e., when one region is active, the other one is not active and vice versa). Thresholds for all significant results were set to voxel-level uncorrected p < 0.001 for voxels, cluster-corrected p < 0.05. We extracted mean z-values between the amygdala and the clusters.

Mediation analysis

We extracted mean z-values between the left amygdala and the 23 clusters, and between the right amygdala and the 13 clusters (shown in Supplemental Table S1). Mediation analyses were conducted to determine the extent to which amygdala rs-FC network (left and right, separately) mediates the association between PNMS and BAPQ scores (i.e., total and three domain scores). The SPSS 26 PROCESS macro (Hayes, 2017) was used to calculate the mediation model, and the mediation Model 4 template is presented in Supplemental Figure S1. Given the small sample size, bootstrapping with 20,000 resamplings was used to determine the significance of the mediation effect. As shown in Supplemental Figure S2, PROCESS Model 4 included BAPQ scores as the outcome, PNMS as the predictor, and amygdala rs-FC (i.e., the mean z-values) as the mediator. A mediation effect was considered significant (p < 0.05) if 0 was not included in the bootstrap confidence interval.

Results

Sample description

The characteristics of the mother-child dyads are presented in Table 1. Regarding parental socioeconomic status at 1998 recruitment, 28.10% (9/32) were middle class; 46.90% (15/32) were upper middle class; and 25.00% (8/32) were upper class. None of the families were lower or lower middle class. On January 9th, 1998 (the date at which the storm peaked), 25.00% (8/32) of the mothers were exposed within the 3-month preconception period; 25.00% (8/32) were exposed in the 1st trimester of pregnancy; 28.10% (9/32) were exposed in the 2nd trimester; and 21.90% (7/32) were exposed in the 3rd trimester.

Table 1. Demographic characteristics of 32 mother-child dyads.

	Range	Mean ± SD
Mothers
Objective hardship	4.00 − 24.00	10.88 ± 4.14
Subjective distress	0 − 40.00	10.43 ± 10.89
Subjective distress (log-transformed)	0 − 3.71	1.91 ± 1.12
Cognitive appraisal	-1–1	0.16 ± 0.85
SES, n (%)
Middle class	9 (28.10)
Upper middle class	15 (46.90)
Upper class	8 (25.00)
Number of children, n (%)
0	9 (28.13)
1	13 (40.63)
2	5 (15.62)
3	4 (12.50)
4	1 (3.12)
Age at child’s birth (year)	20.23 − 41.17	30.68 ± 4.86
Young adults
Sex (F/M)	21/11
Gestational age at birth (week)	33.43 − 41.00	39.51 ± 1.36
Birth weight (g)	1850 − 4432	3506.16 ± 491.66
Age at scan (year)	18.10 − 19.62	18.75 ± 0.36
Full-Scale IQ^a	90 − 137	113.67 ± 11.14
Handedness (R/L)	29/3
In-scanner head motion (mm)	0.04 − 0.20	0.09 ± 0.04
BAPQ
Total score
Aloof Personality	1.33 − 4.73	2.95 ± 0.99
Pragmatic Language	1.00 − 4.33	2.30 ± 0.70
Rigid Personality	1.83 − 4.83	3.15 ± 0.78

BAPQ: Broad Autism Phenotype Questionanire; F: female; IQ: intelligence quotient; L: left; M: male; R: right; SD: standard deviation; SES: socioeconomic status.

^a IQ was assessed at 19 years of age using the Wechsler Adult Intelligence Scale–Third Edition short form.

The young adult offspring (21 females; 11 males) were aged from 18.10 to 19.62 years. The Full-Scale intelligence quotient was above 90. Regarding the BAPQ, 15.63% (5/32) of the young adults met total score cutoffs; 43.75% (14/32) met aloof personality cutoffs; 21.88% (7/32) met pragmatic language impairment cutoffs; and 46.88% (15/32) met rigid personality cutoffs. Although these rates seem high, they are not dissimilar from those (25.30% met BAPQ total score cutoffs) found in a sample of 170 college students who completed the BAPQ (Dovgan & Villanti, 2021).

Amygdala functional connectivity network

The functional connectivity analysis results are presented in Figure 1 and Table 2 and Supplemental Table S1. In total, we found that left and right amygdala functionally connected with multiple brain regions (shown in Figure 1). Supplemental Table S1 presents the details that the left amygdala connected with 23 clusters and that the right amygdala connected with 13 clusters. Table 2 summarizes the results by grouping the clusters according to different functional networks including the emotional regulation network (bilateral insula and prefrontal cortices), the sensory and perceptual network (motor-sensory cortices, basal ganglia, parietal, temporal and occipital cortices, and cerebellum), and the emotional memory network (bilateral hippocampus, parahippocampus and temporal pole).

Figure 1. Amygdala functional connectivity network.

The top panel presents the left amygdala functional connectivity network; The bottom panel presents the right amygdala functional connectivity network. Clusters of positive functional connectivity are depicted in red and yellow colors. Clusters of negative functional connectivity are depicted in purple and blue colors.

Table 2. The amygdala functional connectivity network.

Left amygdala	Right amygdala
Emotional regulation network
Left/Right insula	Left/Right insula
Left frontal orbital cortex	Right frontal orbital cortex
Left frontal pole	Right frontal pole
Frontal medial cortex
	Left superior frontal gyrus
	Left/Right paracingulate gyrus
Sensory and perceptual network
Left/Right postcentral gyrus	Left/Right postcentral gyrus
Left/Right precentral gyrus	Left/Right precentral gyrus
Left/Right cuneus	Left/Right cuneus
Left/Right parietal operculum cortex	Left/Right parietal operculum cortex
Left/Right lateral occipital cortex
Left/Right accumbens	Right accumbens
Left/Right putamen	Left/Right putamen
Left/Right superior temporal gyrus	Left/Right superior temporal gyrus
Left/Right temporal fusiform cortex	Left/Right temporal fusiform cortex
Left/Right planum temporale	Left/Right planum temporale
Left inferior temporal gyrus
Left middle temporal gyrus	Left middle temporal gyrus
Posterior cingulate cortex	Posterior cingulate cortex
Precuneus
Left supramarginal gyrus	Left/Right supramarginal gyrus
Right supracalcarine cortex	Left/Right supracalcarine cortex
Brain stem	Brain stem
Left/Right cerebellum	Left/Right cerebellum
	Left/Right intracalcarine cortex
	Left/Right Heschl’s gyrus
	Right occipital pole
	Left lingual gyrus
Emotional memory network
Left/Right hippocampus	Left/Right hippocampus
Left/Right parahippocampal gyrus	Left/Right parahippocampal gyrus
Left/Right temporal poles	Left/Right temporal poles

Following up on the mean z-values from the 23 clusters for the left amygdala and from the 13 clusters for the right amygdala, 23 simple mediations were tested for the left amygdala rs-FC and 13 mediations were conducted for the right amygdala rs-FC. The significant mediations are presented below.

Decreased amygdala-somatosensory functional connectivity mediates the association between objective hardship and pragmatic language impairment

As illustrated in Figure 2, higher maternal objective hardship was marginally significantly associated with weaker rs-FC from the left amygdala to the right postcentral gyrus. The weaker rs-FC in that connection was in turn marginally significantly associated with greater pragmatic language impairment (reflecting communication deficits). The mediation effect was significant (index of mediation = 0.0182 and bootstrap 95% CI = [0.0021, 0.0515]): the positive association between maternal objective hardship and pragmatic language impairment was mediated by decreased rs-FC from the left amygdala to the right postcentral gyrus.

Figure 2. Mediation effect of left amygdala-right postcentral rs-FC.

Path a presents the association between maternal objective hardship and left amygdala-right postcentral rs-FC. Path b presents the association between left amygdala-right postcentral rs-FC and pragmatic language impairment when controlling for objective hardship. Path c’ presents the association between maternal objective hardship and pragmatic language impairment when controlling for the rs-FC. rs-FC: resting-state functional connectivity.

Decreased amygdala-somatomotor functional connectivity mediates the association between subjective distress and aloof personality

As shown in Figure 3, higher maternal subjective distress was significantly associated with weaker rs-FC from the right amygdala to the left precentral gyrus. The weaker rs-FC in that connection was in turn significantly associated with greater aloof personality (reflecting social deficits). The mediation effect was significant (index of mediation = 0.1357 and bootstrap 95% CI = [0.0008, 0.3782]): the positive association between maternal subjective distress and aloof personality was mediated by decreased rs-FC from the right amygdala to the left precentral gyrus.

Figure 3. Mediation effect of right amygdala-left precentral rs-FC.

Path a presents the association between maternal subjective distress and right amygdala-left precentral rs-FC. Path b presents the association between right amygdala-left precentral rs-FC and aloof personality when controlling for subjective distress. Path c’ presents the association between maternal subjective distress and aloof personality when controlling for the rs-FC. rs-FC: resting-state functional connectivity.

Discussion

Using the 1998 Quebec ice storm as a source of PNMS, the current study indicates the mediator role of amygdala rs-FC on the association between PNMS and autistic-like traits in young adults, such that weaker amygdala rs-FC with the sensorimotor regions, including the right postcentral gyrus and the left precentral gyrus, mediated associations between greater maternal objective hardship and subjective distress and more severe communication and social deficits in young adults.

First, we found that the amygdala was functionally connected to networks associated with emotional regulation, sensation and perception, and emotional memory in the young adults of mothers exposed to PNMS. These findings are consistent with the afferent and efferent anatomical connections of the amygdala observed in rats, cats and monkeys (Sah et al., 2003). These functional connections of the amygdala were also seen in a previous study of 6-month-old infants whose mothers had prenatal depressive symptoms (Qiu et al., 2015). Our results suggest that these positive functional connections of the amygdala, observed in the context of PNMS, may persist from infancy into adulthood.

Our results also raise a discussion on hemispheric differences of the amygdala contributing to the emotional regulation network and the sensory and perceptual network. Indeed, several hypotheses about putative differences between the left and right amygdala have been suggested (Baas et al., 2004). For example, one hypothesis suggests that the right amygdala is more active when the emotional property of the stimulus is visual in nature, whereas the left amygdala may be more active when the emotional property of a stimulus is learned through verbal communication (Phelps et al., 2001). In our study, the right amygdala, not the left amygdala, was more functionally connected to the visual cortex including the left/right intracalcarine cortex, right occipital pole and left lingual gyrus, which can be partly supported by this hypothesis. In addition, the laterality of the amygdala is supported by functional neuroimaging studies showing the left amygdala is more often activated than the right amygdala in top-down emotion generation, whereas both the left and right amygdala are activated in bottom-up emotion generation (Ochsner et al., 2009). Inconsistently, we found that the left and right amygdala were functionally connected to the frontal medial cortex and left/right paracingulate gyrus, respectively; these regions are involved in top-down emotion generation. Although our results cannot be thoroughly explained, the hypotheses extend our understanding of the neural basis of emotion processing.

The amygdala plays a critical role in both bottom-up and top-down emotion generation processes (McRae et al., 2012; Ochsner et al., 2009). The bottom-up process activates the amygdala and the parietal and occipital cortices implicated in detecting affectively arousing stimuli and in modulating their encoding into memory. The top-down process activates the amygdala, PFC and temporal cortices implicated in working memory and in the retrieval of information from semantic memory (Ochsner et al., 2009). In the current study, the amygdala functionally connected with sensory (i.e., somatosensory, visual, and auditory modalities) cortices and temporal cortices, forming the circuit for bottom-up emotion generation. In addition, we found that the amygdala was positively functionally connected with the PFC, forming the top-down emotion generation circuit.

Relevant to the bottom-up process of the amygdala, we found that, for Path a, maternal objective hardship and subjective distress were associated with decreased amygdala-somatosensory rs-FC and decreased amygdala-somato­motor rs-FC in young adult offspring. Neuroimaging research has shown that the amygdala and somatosensory cortex co-activate when processing emotional stimuli (Toschi et al., 2017; Van den Stock et al., 2011). Decreased amygdala-sensorimotor rs-FC has been observed in post-traumatic stress disorder (Belleau et al., 2020; Liu et al., 2021; Simmons et al., 2011), indicating reduced integration of the sensorimotor input to the amygdala circuits of stress regulation. For Path b, the decreased amygdala-somatosensory rs-FC and decreased amygdala-somatomotor rs-FC were associated with pragmatic language impairment and aloof personality, consistent with a previous study showing decreased amygdala-sensorimotor rs-FC associated with language and social deficits in adolescents and young adults with ASD (Fishman et al., 2018). Our finding that pragmatic language impairment and aloof personality were associated with the decreased amygdala-somatosensory rs-FC, supports the current consensus that atypical responsiveness to sensory stimuli is associated with core ASD symptoms (i.e., communication and social difficulties) (Mikkelsen et al., 2018). This is not surprising because atypical response to sensory stimuli may underlie difficulties understanding and communicating with other people who respond “typically” to the same stimuli (Hilton et al., 2010). It has been reported that approximately 90% of children with ASD experience atypical sensory responsiveness; as such, “hypo- or hyper-reactivity to sensory input” was added to the diagnostic criteria of ASD in the DSM-5 (American Psychiatric Association [APA], 2013; Marco et al., 2011).

Of note, we found that decreased amygdala-somatosensory rs-FC mediated the association between maternal objective hardship and pragmatic language impairment in young adult offspring, and that decreased amygdala-somatomotor rs-FC mediated the association between maternal subjective distress and aloof personality in the offspring. These findings suggest that different aspects of PNMS may have distinct correlates with respect to amygdala circuits and autistic-like phenotypes in young adulthood. Future studies with larger sample sizes are needed to clarify the differences between amygdala-somatosensory and amygdala-somatomotor pathways involved in the etiology of different autistic-like phenotypes.

Relevant to the top-down process of the amygdala, the current study found no associations between maternal objective hardship or subjective distress and amygdala-PFC rs-FC in the young adult offspring, which is inconsistent with a prior study showing that higher levels of prenatal maternal depressive symptoms were associated with greater amygdala-medial PFC functional connectivity in the infant offspring (Qiu et al., 2015). The discrepancy may be attributed to the different type of stress (objective hardship vs. depression) and offspring age (19 years old vs. 6 months). Different types of stress could have different and even opposing effects on offspring neurodevelopment; for example, unlike greater prenatal depressive symptoms associated with stronger amygdala-medial PFC rs-FC in infants (Qiu et al., 2015), in another study, more prenatal stressful events were associated with weaker amygdala-medial PFC rs-FC in infants (Humphreys et al., 2020).

Some limitations should be considered. First, after long-term follow-up in Project Ice Storm, attrition led to a small sample size at age 19 years, thus constraining statistical power; we used bootstrapping of 20,000 resamples to minimize this limitation. Second, to preserve statistical power, we did not correct the results for multiple comparisons, which included three aspects of PNMS, BAPQ scores (i.e., total and three domain scores), 23 clusters for the left amygdala rs-FC, and 13 clusters for the right amygdala rs-FC. Finally, the sample at age 19 was of higher socioeconomic status relative to the recruitment sample, which may limit the generalizability of the findings to lower socioeconomic status. Our study also had strengths. First, because we examined a sudden-onset natural disaster as the source of PNMS, exposure to objective hardship was independent of maternal characteristics. Second, we measured three different aspects of PNMS soon after the ice storm, and identified distinct associations with three different BAP domains. Finally, because the ice storm had a known and sudden onset, we were able to accurately measure timing of exposure to the ice storm.

Conclusions

Our study highlights amygdala-sensorimotor rs-FC as a mediator between disaster-related PNMS and autistic-like traits in young adulthood, suggesting long-lasting fetal programming of the transmission of maternal stress to the offspring. Our results also suggest that the differences between amygdala-somatosensory and amygdala-somatomotor circuits may explain associations between different aspects of PNMS and different autistic-like phenotypes.

Acknowledgements

We would like to thank all Project Ice Storm participants.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This study was supported by Canadian Institutes of Health Research [MOP-57849; MOP-111177; MOP-125892] and by the McGill University Stairs Memorial Fund. X.L. was supported by the China Scholarship Council and McGill University Grad Excellence Award.

Notes on contributors

Xinyuan Li

Xinyuan Li, M.D., Ph.D. , completed her PhD in Neuroscience at McGill University in Montreal, specializing in the effect of prenatal maternal stress on offspring neurodevelopment. She is currently a postdoctoral fellow at Alberta Children's Hospital in Calgary, Canada.

Muhammad Naveed Iqbal Qureshi

Muhammad Naveed Iqbal Qureshi, Ph.D. (Biomedical Engineering, Computational Neuroscience, and Artificial Intelligence) completed post-doctoral fellowships at the Department of Psychiatry and Neurosurgery at McGill University, and at the Department of Neuroscience at The Hospital for Sick Children, University of Toronto, Canada.

David P. Laplante

David P. Laplante, Ph.D. is a Senior Research Associate at McGill University’s Lady Davis Institute-Jewish General Hospital in Montreal. His interest lies in examining the interplay between prenatal exposure to maternal adversity (mood, disaster stress), child characteristics, early environmental influences, and youth mental health.

Guillaume Elgbeili

Guillaume Elgbeili, M.Sc. is a statistician working at the Douglas Research Center in Montreal specialized in modeling concurrent and longitudinal associations between parental and child physical and mental health, perinatal and early life environment, and genetics and epigenetics.

Vincent Paquin

Vincent Paquin, M.D. is a Psychiatry Resident and candidate in the Clinician Investigator Program at McGill University, Canada. His research focuses on socio-environmental determinants of mental health in youth.

Sherri Lee Jones

Sherri Lee Jones, Ph.D. (Psychology, Behavioral Neuroscience) completed post-doctoral fellowships at the Department of Psychiatry at McGill University, specializing in the roles of parental factors (e.g. stress, mental health) in offspring neuroendocrine and behavioral development. She is currently a Research Associate at the Douglas Research Center in Montreal, Canada.

Suzanne King

Suzanne King, Ph.D. is Professor Emerita (retired, active) in the Department of Psychiatry at McGill University, and Principal Investigator at the Douglas Hospital Research Centre. To increase understanding about prenatal stress in humans as risk factor for child development her research program focuses on women pregnant during natural disasters (www.mcgill.ca/spiral).

Pedro Rosa-Neto

Pedro Rosa-Neto, M.D., Ph.D. is a Professor of Neurology and neurosurgery and Psychiatry. Dr. Rosa-Neto directs the Translational Neuroimaging Laboratory (TNL) at McGill. University.