Major Gender Differences in Relations between Life Stressor Frequency and Gray Matter in Adolescence and Emerging Adulthood

Abstract

Adolescence and emerging adulthood is likely a sensitive period for the neural effects of stress due to increasing life stress, onset of stress-related disorders, and continued gray matter (GM) development. In adults, stress is associated with GM differences in the medial prefrontal cortex (mPFC), hippocampus, and amygdala, but little is known about these relations, and whether they differ by gender, during adolescence and emerging adulthood. Further, it is unknown whether dependent (self-generated) and independent (fateful) stressors have distinct associations with GM, as each have distinct relations with internalizing psychopathology. We tested relations between recent dependent and independent stressor frequency (ALEQ-R) and GM structure using MRI in a priori regions of interest (mPFC, amygdala, and hippocampus) and across the cortex in youth from the Denver/Boulder metro area ages 14–22 (N=144). Across both genders, no effects passed multiple comparison correction (FDR q>.05). However, there were significant differences between male and female youth (FDR q<.05), with opposite relations between dependent stressor frequency and cortical GM thickness in the salience network and emotion regulation regions and with surface area in default mode network regions. These results motivate future investigations of gender differences in neural mechanisms of stress generation and reactivity.

Adolescents and emerging adults experience many life stressors, which increase risk for psychopathology (Kendler & Gardner, 2016). Stressor exposure is additionally associated with differences in gray matter (GM) structure in adult humans, most notably in regions associated with psychopathology (e.g., Radley et al., 2015), and can alter GM in rodents. As GM develops during adolescence and emerging adulthood, individuals may be especially vulnerable to these neural impacts (Andersen & Teicher, 2008; Romeo, 2017), but it is unknown how stress experienced during this developmental period relates to GM structure. Additionally, previous research does not distinguish between dependent stressors (i.e., self-generated, e.g., getting a bad grade) and independent stressors (i.e., fateful, e.g., illness in the family), despite evidence for a stronger link between dependent (vs. independent) stressors and psychopathology (Technow et al., 2015). The current study addresses these gaps by testing how the frequency of recent dependent and independent stressors are associated with GM morphometry in a sample of adolescents and emerging adults.

In rodents, stress alters GM in areas high in glucocorticoid receptors, increasing vulnerability to depression- and anxiety-like behavior (Sousa, 2016). Stress causes dendritic branching in the amygdala, a key region for fear learning, salience encoding and activating the stress response, and dendritic atrophy in the medial prefrontal cortex (mPFC) and hippocampus, regions crucial for regulating the stress response. Similarly, in human adults, life stress is associated with lower GM volume (GMV) in the hippocampus and mPFC (Ansell et al., 2012; Gianaros et al., 2007; Papagni et al., 2010), with mixed findings of increased or decreased GMV in the amygdala (Gerritsen et al., 2014; Ganzel et al., 2008; Lotze et al., 2020; Sublette et al., 2016). Importantly, in humans, life stress is additionally associated with GMV differences in regions associated with psychopathology and involved in detecting and reacting to stressful stimuli. These regions include the insula (Ansell et al., 2012; H. Li et al., 2014) and anterior cingulate cortex (ACC; Kuhn et al., 2015; Y. Li et al., 2017; Papagni et al., 2010), key nodes of the salience network crucial for integrating internal and external signals and involved in cognitive emotion regulation (Ochsner et al., 2008; Stevens et al., 2011), and the parahippocampal gyrus (H. Li et al., 2014; Papagni et al., 2010), a region involved in emotion regulation (Phillips et al., 2008).

However, these regions are inconsistently identified across studies, potentially due to methodological issues. Many studies use a stress measure that combines stressor frequency, appraisals, perceived coping ability, and distress, which is problematic because these dimensions have distinct relations with psychopathology (Fassett-Carman et al., 2020; Hewitt et al., 1992; Krueger & Markon, 2006) and therefore likely have distinct neural correlates.

Further, research has not distinguished between dependent and independent stressors. Dependent stressors, or those that individuals likely contribute to causing (e.g., getting bad grades, fights with friends), are more strongly associated with internalizing psychopathology than independent, or fateful stressors (e.g., illness in the family, disruptive neighbors; Connolly et al., 2010, Kendler et al., 1999). This difference in association with internalizing symptoms is possibly due to bidirectional effects. Consistent with the stress generation hypothesis, internalizing symptoms affect behavior in ways that generate more dependent stressors (for review, see Hammen, 2018). In turn, consistent with attributional style theory, attributing negative events to the self, which is more likely for dependent stressors, is depressogenic (for meta-analysis, see Huang, 2015). Despite differences in links between dependent and independent stressors and mental health, it is unknown if dependent and independent stressors are distinctly and/or similarly related to GM structure, which could help elucidate pathways between these stressor types and internalizing symptoms.

Adolescence and emerging adulthood may be a sensitive period for stress effects, marked by increased life stressors, the onset of stress-related internalizing psychopathology, heightened HPA axis reactivity to stressors, and ongoing development of brain regions involved in facilitating and regulating stress responses (e.g., mPFC, hippocampus, and amygdala; Romeo, 2017; Hankin et al., 2016). The major developmental trend during adolescence is a reduction in GMV across the brain, due to neurodevelopmental mechanisms such as synaptic pruning, that can affect cortical thickness, and can be influenced by environmental factors (Tau & Peterson, 2010). The timing of pruning varies across regions, including those involved in stress reactivity: amygdala and hippocampus GMV increases until late adolescence (~17 and 19 years respectively), then declines (Wierenga et al., 2014), whereas insula, mPFC, and ACC cortical thickness decreases across this period (Tamnes et al., 2017; Weirenga et al., 2014; Ducharme et al., 2016). In contrast, cortical surface area (SA)—largely shaped by the number of cortical columns (Rakic, 1988; 2009; Tau & Peterson, 2010)—is less affected by environmental input (Aleman-Gomez et al., 2013, Seldon, 2005). It shows inconsistent developmental trajectories during this period across studies, with some evidence for gradual decreases (Tamnes et al., 2017) and other evidence for increases (Vijayakumar et al, 2016), including in areas associated with stress reactivity such as the PFC (Wierenga et al., 2014, Vijayakumar, et al., 2016).

Surprisingly, little research exists on the association of brain morphology and stressful life events that occur during this sensitive adolescent and emerging adult period, compared to effects of early life adversity (for review, see Teicher & Samson, 2016) or childhood and early adolescent trauma, which captures more extreme stressors such as abuse and neglect (for meta-analysis, see Paquola et al., 2016). Only one study tested whether common adolescent stressors are associated with GM structure and found that being disliked by peers was associated with reduced normative decline of PFC GMV and increased typical hippocampal GMV growth between ages 14 and 17, demonstrating that this stressor can affect neurodevelopment across adolescence (Tyborowska et al., 2018). However, these results could be caused by other factors (e.g., behavioral factors that affect interactions with peers, psychopathology) and may not generalize to other stressors.

The current study addresses these gaps by investigating GM correlates of stressful life events in a community sample of adolescents and emerging adults using a robust measure of dependent and independent stressor frequency. Because there is limited previous research during this developmental period, we based preregistered exploratory hypotheses¹ on overlap of regions identified across rodent and adult human research: (1) greater stress frequency will be associated with GMV differences in the amygdala (no directionality hypothesis due to mixed evidence), lower GMV in the hippocampus, and lower cortical thickness and SA in the mPFC, and (2) these effects will be stronger for dependent than independent stressors, as dependent stressors are more predictive of psychopathology than independent stressors, and are likely particularly salient to adolescents and emerging adults, as they include social and academic stressors. We also performed exploratory whole-brain analyses. We tested for gender differences given gender differences in GM developmental trajectories (Gennatas et al., 2017) and due to the greater strength of life stress-psychopathology associations in females than males (Hankin et al., 2007; Fassett-Carman et al., 2019), potentially suggesting differences in stress-GM relations. We additionally tested age moderation due to the somewhat wide age range of our sample across adolescence. Finally, we conducted follow-up analyses testing moderation of the stress-GM association in ROIs by pubertal development, given links between pubertal and GM development (Herting & Sowell, 2017; Juraska & Willing, 2017).

Method

Participants

Participants (N=144, 50.1% Female) were unselected community adolescents and emerging adults ages 14–22 drawn from the greater Denver/Boulder metro area as part of the Colorado Cognitive Neuroimaging Family Emotion Research (CoNiFER) study. Participants were originally recruited for the GEM (Genes, Environment and Mood) study (NIH Grant R01 MH077195) and an associated follow-up study (R21MH102210) via public schools and direct mail to target zip codes to maximize demographic and socioeconomic diversity. Inclusion criteria were no neurological insult, English as a first language, and no dyslexia or difficulty reading. Participants gave informed consent (ages 18–22) or assent with parental consent (ages 14–17), and received cash compensation for their participation. All procedures were approved by the University of Colorado Institutional Review Board.

The recruited sample size was determined by the needs of other analyses that were the focus of the Colorado CoNiFER study. Power analysis demonstrated the sample size provided >80% power to detect medium effect sizes in regression analyses. All participants that completed the neuroimaging portion of the study (n=126) as well as the ALEQ stress measure (n=129) were included in the current analyses (final n=114).

Procedure

Analyses presented here use Time Point 1 data from the Colorado CoNiFER longitudinal study. Time Point 1 consisted of two 1-hour structural and functional neuroimaging sessions, and one behavioral tasks and psychopathology assessment session.

Measures

Demographics.

Participants selected their gender from the options “Male” or “Female” and report their current age.

Stressful life events.

The Adolescent Life Event Questionnaire – Revised (ALEQ-R; Fassett-Carman et al., 2019) assesses occurrence in the past 6 months of 24 dependent and 25 independent negative life events commonly experienced by youth (See Fassett-Carman et al., 2019 for details regarding classification of items as dependent or independent). For nine infrequent events (eight independent and one dependent stressor; e.g., “Parents getting divorced”), participants indicated “This event did not happen” [0] or “This event happened” [4]. For all other items, participants rated how often each event occurred from 0 (“Never”) to 4 (“Always”). Dependent and independent stressor frequency scores were calculated by summing frequency ratings of items within the stressor category for each participant. Severity and controllability ratings were collected but are the focus of a separate manuscript (Fassett-Carman et al., 2022).

Pubertal development.

The Pubertal Development Scale (PDS; Carskadon & Acebo, 1993) measures changes that typically occur throughout puberty. Boys and girls are both asked about growth in height and body hair, and skin changes. Boys are additionally asked about deepening of the voice and facial hair growth. Girls are additionally asked about breast growth and menstruation. Participants indicate (1) “Not yet started”, (2) “Barely started”, (3) “Definitely started”, and (4) “Seems complete” for each item except the menstruation item, which is rated dichotomously (0=menstruation has not begun, 4=menstruation has begun). Scores are calculated by averaging the ratings of the five items applicable to each participant. The PDS shows good reliability and validity (Carskadon & Acebo, 1993).

Structural MRI acquisition and surface-based morphometry.

Whole brain high-resolution structural MRI data were acquired at the Intermountain Neuroimaging Consortium located at the University of Colorado Boulder using a Siemens 3-Tesla PRISMA MRI scanner for all but 18 participants, who were scanned on the pre-upgrade version of the same magnet (TIM TRIO). A 32-channel headcoil was used for radiofrequency transmission and reception. GM structure data were acquired via a T1-weighted Magnetization Prepared Gradient Echo sequence in 224 sagittal slices, with a repetition time=2400 ms, echo time=2.07 ms, flip angle=8 degrees, field of view=256 mm, and voxel size=.8 mm³.

T1-weighted structural images were brain extracted using a hybrid watershed/surface deformation procedure (Segonne et al., 2004), followed by a transformation into Talaiarch space, intensity normalization (Sled et al., 1998), tessellation of the gray/white matter boundary (Fischl et al., 2001), and surface deformation along intensity gradients to optimally differentiate gray matter, white matter, and cerebral spinal fluid (CSF) boundaries (Dale et al., 1999; Fischl and Dale, 2000). The resulting segmented surfaces were registered to a standard spherical inflated brain template (Fischl et al., 1999a, b), parcellated according to gyral and sulcal structure (Desikan et al., 2006; Fischl et al., 2004), and then used to compute a range of surface-based measurements, including cortical SA and thickness. Prior to running surface-based analyses, data quality assurance was checked using FreeSurfer’s standard quality assurance tools (https://surfer.nmr.mgh.harvard.edu/fswiki/QATools), including visual inspection of the relevant parcellations and checking for outliers in morphometry measures, signal-to-noise ratio, and white matter intensity.

Analysis

To test whether frequency of dependent and independent life stressors is associated with ROI GM structure, we conducted separate linear regressions for each of these stressor types with GMV as the outcome measure for subcortical ROIs (left and right amygdala and hippocampus), and SA and thickness as outcome measures for cortical ROIs (left and right mPFC). Additional linear regressions tested for gender and age moderation of these relations. Whole-brain exploratory analyses were conducted with the Permutation Analysis of Linear Models (PALM) tool (Winkler et al., 2014). Data at each vertex were shuffled 10,000 times, using sign-flipping for main effect and age-moderation models, and regrouping for tests of gender moderation. Exchangeability blocks accounted for dependence between data points created by 14 sibling-pairs and one sibling-trio in the dataset.

All analyses controlled for age and gender. Cortical SA and subcortical GMV analyses were conducted with and without controlling for average SA and total brain volume excluding ventricles, respectively, because these GM measures are associated with total head size, whereas cortical thickness is not (Barnes et al., 2010).

For clusters in which gender moderation passed FDR correction, we conducted follow-up analyses to determine what drove the moderation. We used the FreeSurfer function “mri_label2label” to map clusters from fsaverage space to subject space and subsequently extracted SA and thickness measurements for each participant. These values were used as outcome variables in gender-specific linear regressions, with stressor frequency as the predictor and age as a covariate.

Given the wide age-range of the sample and links between pubertal and neural development (Herting & Sowell, 2017; Juraska & Willing, 2017), we conducted follow-up analyses investigating whether pubertal development moderated the associations between dependent and independent stressor frequency and ROI gray matter morphometry.

To account for the large number of analyses conducted, but balance this fact with the exploratory nature of the study (given limited prior research as described in the Introduction), we analyzed and present the data at multiple levels of correction. For ROIs analyses, we present all uncorrected analyses in the tables, but focus the results and discussion on analyses that pass FDR correction (q<.05). For whole-brain analyses, we present uncorrected clusters with a vertex-wise threshold of p<.01 and a maximum vertex of p<.001 in tables and figures, but focus the results and discussion on clusters that pass FDR or FWE correction (q<.05). We present both FDR and FWE corrections for whole-brain analyses to illustrate which vertices passed the most stringent correction, and which passed a more lenient yet still stringent threshold.

Transparency and Openness

We report how we determined our sample size, data exclusions, all manipulations, and all measures. Data and analysis code for ROI analyses performed using SPSS, version 27, are available². Whole brain data are not available on OSF. Analyses were preregistered (footnote 1).

Results

Outlier datapoints (+/− 3 SD from the mean) in ALEQ or ROI measures were excluded from analysis (n=2 for dependent stress frequency and n=1 each for left and right hippocampus and amygdala GMV). On average, participants endorsed (i.e., rated frequency above “Never” [0]) 5.98 dependent stressors (SD=3.86, range: 0–17) and 2.74 independent stressors (SD=2.30, range: 0–11; Table S1). Dependent and independent stressor frequency correlated with each other and with internalizing symptoms, consistent with past work (Fassett-Carman et al., 2019; 2020; Supplemental Method and Table S2). Table S3 gives descriptive information regarding pubertal development of the sample.

ROI Results

No relations between dependent or independent stressor frequency and ROI GM passed FDR correction (Table 1). No gender (Table S4 & S5) or age (Table S6) differences in the relations between stressor frequency with ROI GM passed FDR correction. Pubertal development moderated the association between independent stressor frequency and left and right mPFC SA in regressions without controlling for whole-brain SA, and right mPFC SA when including whole-brain SA as a covariate: greater independent stressor frequency was associated with less SA earlier in puberty and greater SA later in puberty (Table S7, Figure S1).

Table 1.

Associations between Dependent and Independent Stress Frequency and ROIs Controlling for Age and Gender, with and without Whole Brain Control

			With whole brain control					Without whole-brain control
	ROI	b	β	SE	t	p	b	β	SE	t	p
Dependent Stressors	L. Amygdala	−1.498	−0.048	1.989	−0.753	.453	−0.650	−0.021	2.526	−0.257	.797
	R. Amygdala	−4.682	−0.146	1.983	−2.361	.020	−4.331	−0.135	2.500	−1.732	.086
	L. Hippocampus	5.769	0.091	4.684	1.232	.221	7.720	0.121	5.751	1.342	.182
	R. Hippocampus	2.412	0.040	3.876	0.622	.535	4.313	0.072	5.067	0.851	.396
	L. mPFC
	Area	1.006	0.029	2.781	0.362	.718	1.795	0.052	3.122	0.575	.566
	Thickness	-	-	-	-	-	0.000	−0.010	0.003	−0.102	.919
	R. mPFC
	Area	1.991	0.065	2.230	0.893	.374	2.882	0.094	2.756	1.046	.298
	Thickness	-	-	-	-	-	−0.001	−0.043	0.003	−0.454	.651
Independent Stressors	L. Amygdala	.239	0.006	2.421	0.099	.921	−1.763	−0.046	3.052	−0.577	.565
	R. Amygdala	−2.139	−0.055	2.438	−0.877	.382	−4.268	−0.110	3.000	−1.422	.158
	L. Hippocampus	.672	0.009	5.763	0.117	.907	−3.494	−0.044	7.055	−0.495	.621
	R. Hippocampus	−.312	−0.004	4.761	−0.066	.948	−4.353	−0.059	6.196	−0.702	.484
	L. mPFC
	Area	−2.115	−0.050	3.349	−0.631	.529	−2.748	−0.064	3.759	−0.731	.466
	Thickness	-	-	-	-	-	<0.001	0.011	0.003	0.120	.905
	R. mPFC
	Area	.224	0.006	2.686	0.084	.934	−0.496	−0.013	3.325	−0.149	.882
	Thickness	-	-	-	-	-	−0.004	−0.124	0.003	−1.330	.186

Note. Bold=p<.05. Outcome measure for subcortical ROIs=GMV; Whole brain control for subcortical ROIs=total brain volume excluding ventricles; Whole brain control for cortical surface area=whole brain average area. No whole-brain control for thickness analyses as GM thickness is not associated with head size (Barnes et al., 2010). No results passed FDR correction for multiple comparisons.

Main Effects of Stressor frequency (Table S8, Figure S2)

No clusters passed multiple comparison correction.

Gender Moderation (Tables 2 & 3, Figures 1–4)

Table 2.

Gender moderation of Vertex-Wise Cortical Thickness Correlates of Dependent and Independent Stressor Frequency.

Stressor Type	Hemi	+/−	Region	Size (mm2)	MaxP (log(10)p)	X	Y	Z	Passes FDR	Passes FWE	Yeo Network
	l	-	insula	780.25	4.000	−37.2	3.4	12.6	y	y	VA
	l	-	SFG	366.33	4.000	−8.8	49.1	21.4	y	y	DMN
	l	-	mOFC	299.68	4.000	−6.6	24.4	−9.9	y	n	Limbic
	l	-	preC	239.73	4.000	−55.3	6.3	27.1	y	n	DA/SM
	l	-	rMFG	211.18	4.000	−22.9	38.4	22.9	y	n	DA/DMN
	l	-	SFG/MFG	190.20	4.000	−21.4	23.6	32.5	y	n	DMN
	l	-	MTG/temp pole	616.07	3.699	−48.1	1.5	−28.3	y	n	Limbic/DMN
	l	-	rMFG	399.48	3.699	−37.0	48.0	7.1	y	n	FPCN
	l	-	SMG	276.41	3.699	−51.2	−43.4	43.6	y	n	FPCN
	l	-	SPL	124.68	3.699	−24.3	−50.0	60.5	y	n	DA
	l	-	rACC	473.53	3.523	−5.8	38.0	5.7	y	n	DMN/FPCN
	l	-	SPL	126.14	3.523	−31.4	−38.0	41.1	y	n	DA
	l	-	EC	273.75	3.398	−22.4	−9.4	−27.2	y	n	Limbic
	l	-	IPL	127.20	3.398	−30.0	−60.2	40.0	y	n	DMN
	l	-	STG	118.25	3.398	−59.6	−45.1	19.3	y	n	DMN
Dependent	l	-	IPL	97.01	3.398	−42.1	−52.2	24.0	y	n	FPCN
	l	-	lOFC	392.40	3.301	−33.6	29.1	−12.3	y	n	DMN/limbic
	l	-	SMG	85.31	3.046	−43.9	−51.5	38.3	n	n	FPCN
	r	-	MFG/FP/lOFC	1316.52	4.000	22.5	58.7	−2.1	y	n	Limbic/FPCN/DMN
	r	-	EC/fusiform	650.18	4.000	35.9	−16.0	−26.2	y	n	Limbic
	r	-	pars operc/preC/postC	519.26	4.000	43.2	9.3	17.4	y	y	VA/DA/SM
	r	-	insula	442.16	4.000	32.3	8.1	9.7	y	y	VA
	r	-	IPL/SPL	289.10	4.000	32.8	−62.5	37.7	y	n	DA/FPCN
	r	-	cACC/PCC/SFG	182.51	4.000	10.5	10.5	33.2	y	n	VA
	r	-	postC	169.82	3.523	52.0	−16.9	47.3	y	n	SM
	r	-	STG	733.25	3.398	63.6	−13.5	2.5	y	n	SM
	r	-	lOFC	233.81	3.398	26.1	18.5	−18.7	y	n	Limbic
	r	-	MTG	168.38	3.398	59.8	−52.4	5.5	y	n	VA
	r	-	mOFC	134.43	3.398	9.6	14.2	−15.1	y	n	Limbic
	r	-	ICC	121.54	3.398	18.1	46.7	3.8	y	n	visual
	r	-	ITG	248.15	3.301	51.1	−43.9	−14.2	y	n	FPCN/DA
	r	-	lingual	114.85	3.046	6.7	−85.8	−4.6	y	n	Visual
	r	-	precun	128.15	3.000	5.1	−56.6	27.7	y	n	DMN
Independent	l	-	SPL	144.50	3.301	−17.9	−64.8	36.9	n	n	DA
	l	+	insula	207.73	3.097	−36.1	2.2	−9.7	n	n	VA
	r	-	SPL	122.08	3.222	22.3	−52.7	54.2	n	n	DA

Note. All results passed a vertex-wise threshold of .01 with a max p less than .001. Clusters with sub-clusters passing FDR and FWE corrections are indicated. y=yes; n=no; hemi=hemisphere; r=right; l=left; SFG=superior frontal gyrus; mOFC=medial orbitofrontal gyrus; preC=precentral gyrus; rMFG=rostral middle frontal gyrus; MFG=middle frontal gyrus; MTG=middle temporal gyrus; temp pole=temporal pole; SMG=supramarginal gyrus; SPPL=superior parietal lobule; rACC=rostral anterior cingulate cortex; EC=entorhinal cortex; IPL=inferior parietal lobule; lOFC=lateral orbitofrontal cortex; FP=frontal pole; fusiform=fusiform gyrus; pars operc=pars opercularis; postC=postcentral gyrus; cACC=caudal anterior cingulate cortex; PCC=posterior cingulate cortex; STG=superior temporal gyrus; ICC=isthmus cingulate cortex; ITG=inferior temporal gyrus; lingual=lingual gyrus; precun=precuneus.

Table 3.

Gender moderation of Vertex-Wise Cortical Surface Area Correlates of Dependent and Independent Stressor Frequency.

Stressor Type	Cortical Measure	Hemi	+/−	Region	Size (mm2)	MaxP (log(10)p)	X	Y	Z	Passes FDR	Passes FWE	Yeo Network
	Area	l	+	precun/ICC/SPL	1879.11	4.000	−6.9	−48.2	46.0	n	y	DMN/FPCN/DA
	Area	l	+	STG	314.45	4.000	−58.3	−12.8	−3.0	n	n	DMN/SM
	Area	l	+	IPL	297.33	3.699	−36.2	−80.7	24.1	n	n	visual/DA
	Area	l	+	postC/preC	107.90	3.301	−60.6	−7.3	10.9	n	n	SM
	Area	r	+	precun/ICC/SPL	2300.75	4.000	14.2	−48.6	34.0	y	y	DMN/FPCN/DA
Dependent	Area	r	+	SFG	1083.61	4.000	21.0	20.0	50.7	y	n	DMN/FPCN
	Area	r	+	IPL	548.67	3.301	42.2	−46.4	38.5	y	n	FPCN
	Area	r	+	IPL	382.14	3.301	42.7	−70.2	19.9	y	n	DA
	Area	r	+	SPL	347.33	3.046	34.1	−45.5	58.2	y	n	DA
	Area w/ WBC	l	-	cuneus/precun/periC	1315.15	3.699	−12.4	−74.8	24.1	n	n	Visual
	Area w/ WBC	l	+	precun	804.60	4.000	−8.4	−46.6	45.6	n	n	DMN
	Area w/ WBC	l	+	STG	150.04	3.301	−60.4	−11.9	0.1	n	n	SM
	Area w/ WBC	r	-	pars operc	536.72	3.097	49.1	6.8	8.5	n	n	FPCN/DA/VA
	Area w/ WBC	r	+	precun/ICC	775.64	4.000	14.9	−44.5	36.8	n	n	DMN
	Area w/ WBC	r	+	SFG	548.74	3.301	21.2	23.9	49.5	n	n	DMN/FPCN
	Area	l	+	SPL	288.91	3.097	−14.6	−62.1	57.8	n	n	DA
Independent	Area	r	+	SPL	906.59	4.000	34.0	−42.0	47.2	n	n	DA
	Area	r	+	precun	224.31	3.301	7.2	−64.4	34.6	n	n	DMN
	Area w/ WBC	l	-	fusiform	608.17	3.046	−31.1	−73.6	−7.0	n	n	Visual
	Area w/ WBC	r	-	banksSTS	300.47	3.699	51.3	−33.1	6.0	n	n	SM
	Area w/ WBC	r	+	SPL	256.85	3.699	33.5	−41.7	45.8	n	n	DA
	Area w/ WBC	r	+	precun	217.26	3.699	8.9	−63.7	35.0	n	n	DMN

Note. All results passed a vertex-wise threshold of .01 with a max p less than .001. Clusters with sub-clusters passing FDR and FWE corrections are indicated. WBC=whole brain control (average area); y=yes; n=no; hemi=hemisphere; r=right; l=left; precun=precuneus; ICC=isthmus cingulate cortex; SPL=superior parietal lobule; STG=superior temporal gyrus; IPL=inferior parietal lobule; preC=precentral gyrus; postC=postcentral gyrus; SFG=superior frontal gyrus; periC=pericalcarine; pars operc=pars opercularis; fusiform=fusiform gyrus; banksSTS= Banks superior temporal sulcus.

Figure 1.

Note. All results passed a vertex-wise threshold of .01 with a max p less than .001. Sub-clusters passing FDR and FWE corrections are indicated. RH=right hemisphere; LH=left hemisphere; SFG=superior frontal gyrus; mOFC=medial orbitofrontal gyrus; preC=precentral gyrus; rMFG=rostral middle frontal gyrus; MFG=middle frontal gyrus; MTG=middle temporal gyrus; temp pole=temporal pole; SMG=supramarginal gyrus; SPPL=superior parietal lobule; rACC=rostral anterior cingulate cortex; EC=entorhinal cortex; IPL=inferior parietal lobule; lOFC=lateral orbitofrontal cortex; FP=frontal pole; fusiform=fusiform gyrus; pars operc=pars opercularis; postC=postcentral gyrus; cACC=caudal anterior cingulate cortex; PCC=posterior cingulate cortex; STG=superior temporal gyrus; ICC=isthmus cingulate cortex; ITG=inferior temporal gyrus; lingual=lingual gyrus; precun=precuneus.

Figure 4.

Relation between Dependent Stressor Frequency and Cortical Surface Area for Male and Female Youth in Clusters demonstrating Significant Gender Moderation in Whole-Brain Analyses

Note. Clusters passed vertex-wise threshold of p<.01, with a max p<.001, and contained a sub-cluster that passed FWE and/or FDR correction; *p<.05; rh=left hemisphere; rh=right hemisphere; precun=precuneus; ICC=isthmus cingulate cortex; SPL=superior parietal lobule; SFG=superior frontal gyrus; IPL=inferior parietal lobule.

Only dependent, and not independent, stressors had gender moderation effects that passed multiple comparison correction. Gender moderated the relations between dependent stress frequency and cortical thickness in 32 clusters that passed FDR correction, four of which passed FWE correction (Figure 1, Table 2). These clusters were widely distributed, located bilaterally in the insula, superior frontal gyrus (SFG), medial orbitofrontal cortex (mOFC), precentral gyrus, middle frontal gyrus (MFG), middle temporal gyrus (MTG), superior parietal lobule (SPL), inferior parietal lobule (IPL), entorhinal cortex, superior temporal gyrus (STG), and lateral orbitofrontal cortex (lOFC). Further, gender moderation clusters were found in the left temporal pole, rostral anterior cingulate cortex (rACC), and supramarginal gyrus (SMG), and the right frontal pole, fusiform cortex, pars opercularis, postcentral gyrus, caudal anterior cingulate cortex (cACC), posterior cingulate cortex, postcentral gyrus, isthmus cingulate cortex (ICC), inferior temporal gyrus, lingual gyrus, and precuneus. Gender moderation was driven by greater dependent stress frequency being associated with lower GM thickness in female youth and greater GM thickness in male youth (Figure 3, Table S9). In most clusters these relations were significant for both genders.

Figure 3.

Relation between Dependent Stressor Frequency and Cortical Thickness for Male and Female Youth in Clusters demonstrating Significant Gender Moderation in Whole-Brain Analyses

Note. Clusters passed vertex-wise threshold of p<.01, with a max p<.001, and contained a sub-cluster that passed FWE and/or FDR correction. *p<.05; SFG=superior frontal gyrus; mOFC=medial orbitofrontal gyrus; preC=precentral gyrus; rMFG=rostral middle frontal gyrus; MFG=middle frontal gyrus; MTG=middle temporal gyrus; temp pole=temporal pole; SMG=supramarginal gyrus; SPPL=superior parietal lobule; rACC=rostral anterior cingulate cortex; EC=entorhinal cortex; IPL=inferior parietal lobule; lOFC=lateral orbitofrontal cortex; FP=frontal pole; fusiform=fusiform gyrus; pars operc=pars opercularis; postC=postcentral gyrus; cACC=caudal anterior cingulate cortex; PCC=posterior cingulate cortex; STG=superior temporal gyrus; ICC=isthmus cingulate cortex; ITG=inferior temporal gyrus; lingual=lingual gyrus; precun=precuneus.

Gender additionally moderated the relations between dependent stressor frequency and cortical SA in clusters largely non-overlapping with the thickness clusters (Table 3, Figure 2). Clusters passing correction (five FDR, two FWE) were located bilaterally in precuneus, ICC, and SPL, and in the right SFG, IPL, and SPL. Gender moderation was driven by greater dependent stressor frequency being associated with lower SA in male youth and greater SA in female youth (Figure 4, Table S10). These relations were significant for both genders in most clusters. Controlling for whole-brain average area, no SA gender moderation clusters passed correction for multiple comparisons.

Figure 2.

Note. All results passed a vertex-wise threshold of .01 with a max p less than .001. Sub-clusters passing FDR and FWE corrections are indicated. RH=right hemisphere; LH=left hemisphere; control=whole brain average area; precun=precuneus; ICC=isthmus cingulate cortex; SPL=superior parietal lobule; STG=superior temporal gyrus; IPL=inferior parietal lobule; preC=precentral gyrus; postC=postcentral gyrus; SFG=superior frontal gyrus; periC=pericalcarine; pars operc=pars opercularis; fusiform=fusiform gyrus; banksSTS=Banks superior temporal sulcus.

Gender Moderation Follow-Up Analyses

Given the widespread gender moderation of relations between dependent stressor frequency and cortical thickness and SA, we tested whether this moderation extended to whole-brain mean thickness and SA. Gender moderated the relation between dependent stressor frequency and mean thickness (β=−.261, p=.005), driven by a significant negative relation for girls (β=−.349, p=.007) but not boys (β=.198, p=.143). There was a non-significant trend of gender moderation in the relation between dependent stress frequency and whole-brain average SA (β=.140, p=.057): split by gender, there was a significant positive relation for girls (β=.265, p=.042) but not boys (β=−.100, p=.464).

Age Moderation

Age moderated the relation between independent stressor frequency and GM thickness in one cluster in the IPL which passed FWE correction (Table S11, Figure S3). No other age moderation effects passed correction (Table S11, Figures S3 & S4).

Discussion

In the full sample of adolescents and emerging adults, there were no associations that passed multiple comparisons corrections between dependent or independent stressors and GM morphometry in the mPFC, hippocampus, and amygdala, or across the cortex. However, there were significant gender differences in relations between dependent stressor frequency and cortical thickness and SA, as well as pubertal developmental differences in relations between independent stressor frequency and mPFC SA.

Effects of Dependent vs. Independent Stressors

Our results supported our hypothesis that dependent stressor frequency would be more closely related to GM morphometry than independent stressor frequency: most relations that passed multiple comparison correction were for dependent rather than independent stressors. There are multiple mechanisms that may account for this. First, dependent stressors may cause greater stress reactivity than independent stressors, due to the knowledge that one contributed to causing these events and subsequent negative cognitions such as self-blame and rumination that can increase and prolong biological stress reactivity (LeMoult, 2020). These cognitive reactions or the greater biological reactivity could lead to structural GM changes seen for dependent but not independent stressors. Second, dependent stressors may be particularly common and salient during adolescence and emerging adulthood, given the heightened complexity and salience of peer relationships and greater academic pressure, which can lead to dependent stressors. This heightened exposure and salience of dependent stressors could increase their impact on adolescent brain development (e.g., Fuhrman et al., 2015), consistent with their heightened relation to psychopathology (Kendler et al., 1999; Technow et al., 2014). On the other hand, pre-existing structural differences may affect neural function and behavior in ways that increase the generation of dependent stressors, but not independent stressors. This possibility is consistent with the stress generation theory and evidence that traits or experiences (e.g., internalizing symptoms, poor EF, rumination) can influence behavior in ways that increase stressful events (Hammen, 2018; Snyder et al., 2016; 2019; Stroud, 2018). This can create a cycle whereby life stress increases symptoms and increased symptoms generate more stress.

Observed Gender Differences in GM Correlates of Stress

Greater dependent stressor frequency was associated with thinner GM for female youth and thicker GM for male youth in portions of the salience network (insula, ACC) and regions involved in cognitive aspects of emotion regulation in the frontoparietal control network (FPCN; MFG, lOFC, IPL, rACC) and limbic network (SMG, MTG and STG). Further, greater dependent stress frequency was associated with a thinner mean thickness across the entire cortex in girls but not boys. Thus, the gender differences in links between dependent stressor exposure and cortical thickness are widespread and may not be specific to the regions that passed correction for multiple comparisons. This possibility is consistent with recent evidence that chronic stress in rodents results in whole-brain volumetric changes compared to unstressed controls (Magalhães et al., 2018). Given that all neurons contain glucocorticoid receptors, the morphometric effects of high levels of glucocorticoids resulting from chronic stress may occur throughout the brain, rather than being limited to regions with the highest glucocorticoid receptor density (amygdala, hippocampus, mPFC). This possible generalized effect of stress may characterize adolescence, a critical developmental period, whereas more specific targets may be associated with life stress in adulthood. Further, these glucocorticoid effects may differ for girls and boys (Shanksy & Murphy, 2021), potentially leading to the widespread gender differences identified here.

Despite the possibility that the clusters with significant gender differences are part of more widespread effects, many of the regions identified in the current study are implicated in stress reactivity and regulation. The salience network is important for responding to stressful stimuli and coordinating neural and autonomic responses (Menon & Uddin, 2010; Gogolla, 2017). The insula and ACC, salience network hubs, are thought to integrate internal and external stimuli and use this information to regulate the engagement of the FPCN and default mode network (DMN; Menon, 2015; Gogolla, 2017), which showed gender differences in relations between dependent stressor frequency and cortical thickness and SA respectively in the current study.

Critically, the FPCN and DMN are both involved in stress processing. The FPCN is involved in cognitive regulation of stress and emotion, while DMN activity underlies inwardly-directed thought processes triggered by stressors, such as rumination. For example, FPCN-sites of gender-specific relations between GM thickness and dependent stress frequency in the current study, such as the dlPFC, vlPFC, and IPL, are crucial for the executive function (EF) processes that underlie cognitive reappraisal, a cognitively demanding strategy for emotion regulation. These regions support holding of reappraisal goals in working memory, selectively attending to information helpful to the goal of regulating emotion, and inhibiting interpretations of information that may interfere with appraisal goals (Ochsner et al., 2012; Banich et al., 2009). Further, the SMG and temporal regions (e.g., MTG, STG) identified in the current analyses outside of the FPCN are involved in representing perceptual and semantic knowledge and are activated during reappraisal, potentially as reappraisal works to change perceptions of stressful stimuli (Ochsner et al., 2012). Thus, the regions showing gender-specific relations between dependent life stress and GM thickness are implicated in the stress-emotion experience.

In addition to gender differences in the relations between stress and thickness, we observed opposite relations between SA and dependent stressors for male and female youth but in largely non-overlapping clusters (precuneus, isthmus cingulate, SFG, SPL and IPL) and in the opposite direction as for thickness: greater dependent stressor frequency is associated with less SA in male youth and greater SA in female youth. Different effects for SA and thickness are understandable given that these appear to be independent properties of brain morphology (Winkler et al., 2010). Most of the SA clusters identified are in the dorsal attention network or DMN, which subserve externally directed attention and internally directed thought, respectively (Corbetta & Shulman, 2002). The largest clusters identified (bilateral precuneus, right SFG) are part of the DMN which is associated with rumination (Zhou et al., 2020), a maladaptive response to life stress characterized by internally focused, repetitive negative thinking (Nolen-Hoeksema et al., 2008). Higher exposure to stressors increases rumination (Snyder et al., 2016), which is a strong, transdiagnostic risk factor for internalizing psychopathology (Aldao et al., 2016). Importantly, female youth are more likely to ruminate than male youth (Rood et al., 2009), potentially putting them at heightened risk for internalizing psychopathology. However, it is not clear whether this aspect of morphology might compromise the ability to deal with stressors and/or lead to behaviors that would increase the effects of stressors (e.g., rumination) in female youth.

Potential Mechanisms of Gender Differences in Stress-Brain Relations

Here we consider why male and female youth show opposite relations between dependent stressor frequency and regional GM thickness, in particular. First, there may be gender differences in the neural impacts of stressors. Although the current study asked participants to indicate their gender and did not collect information regarding biological sex, we can draw from rodent model research investigating sex differences in the neural effects of stressor exposure for possible underlying mechanisms of the gender differences found. Rodent models reveal well-documented effects of chronic stress exposure on dendritic structure in male rats (atrophy in mPFC and hippocampus, hypertrophy in amygdala; Radley et al., 2016). In contrast, female rats exhibit largely opposite effects: chronic stress elicits dendritic shrinkage in the amygdala (Blume et al., 2019), no atrophy (or hypertrophy) in the hippocampus (Galea et al., 1997), and dendritic branching in the mPFC (Garrett & Wellman, 2009; Shansky et al., 2010; Farrell et al., 2016) that is dependent on the presence of estrogen (Garrett & Wellman, 2009). These results highlight that sex hormones are involved in shaping neural stress outcomes.

Studies in humans also provide evidence for sex differences in neural reactivity to acute stress, which could contribute to different relations with GM structure during development. Although different from the stressful events assessed in the current study, a recent study found that men and women showed opposite patterns of neural reactivity to aversive images in prefrontal regions (sgACC, OFC, dmPFC; increased BOLD signal for men, blunted signal for women) and limbic regions (hippocampus, insula/putamen, pallidum; increased BOLD signal for women, blunted signal for men; Goldfarb et al., 2019). Importantly, sex differences in neural reactivity to stressors could impact developmental GM thickness trajectories by driving synaptic pruning and branching, potentially leading to different associations between life stressor exposure and GM thickness in male and female youth.

Alternatively, gender differences could be due to sex differences in GM developmental trajectories. Male youth appear to lag 1–2 years behind female youth in their cortical GMV and SA trajectories and puberty onset (Lenroot et al., 2007). There is evidence for sex differences in the rate of GM thickness change across childhood and adolescence (Zhou et al., 2015) and region-specific sex differences in cortical thickness development (Gennatas et al., 2017; Ducharme et al., 2016) suggesting potential sex-specific trajectories depending on region in humans (see Kaczkurkin et al., 2019 for review).

It is thus possible that recently experienced stressors affect male and female youth differently due to neural sex differences in ongoing developmental processes. Perhaps female youth, having undergone more substantial pruning compared to the male youth (e.g., Zhou et al., 2015; Drzewiecki et al., 2016), demonstrate a stress-GM profile more similar to adults: stress, either having a neurotoxic effect or causing excessive pruning in these regions due to repeated activation during development, leads to excess thinning. Indeed, reduced GMV in prefrontal and salience networks are associated with internalizing psychopathology in adulthood, which is more prevalent in women than men (Lai, 2013; Stratmann et al., 2014; Herring et al., 2012). Male youth, being at an earlier stage in GM developmental or undergoing different region-specific developmental timing, may exhibit opposite effects of stress. This possibility lends further support to calls for investigations of sex-specific effects of developmental timing on the neural impacts of life stressors (Helpman et al., 2017).

A third possibility is that pre-existing differences in GM thickness could influence the frequency of dependent stressors via stress-generation. Individual differences in the morphology of brain regions involved in cognitive regulation might affect behavior in ways that bring about stressors. FPCN regions, and their regulation by the salience network, are crucial not only for emotional regulation but also for non-emotional aspects of cognitive control. Atypical GM thickness in these networks might impair cognitive control, both directly via GM structural abnormalities in the FPCN, or due to problems engaging the FPCN resulting from a dysfunctional salience network. This possibility is consistent with evidence that lower EF longitudinally predicts greater dependent stressor frequency, leading to psychopathology (Snyder et al., 2016). Further, EF improves across adolescence as FPCN areas undergo cortical thinning (Kharitonova et al., 2013), highlighting the possibility that individual differences in GM developmental trajectories could affect stress generation. It is unclear why the structural effects promoting stress generation would be opposite in male and female youth, but could be an instance of equifinality in which different neural mechanisms bring about a similar behavioral phenotype (Shansky & Murphy, 2021). Investigating how the GM clusters identified in the current study relate to behavior will be important for understanding their potential effects on exacerbating or protecting against dependent stress generation.

Pubertal Development Moderation of Associations between Independent Stressor Frequency and mPFC Surface Area

Pubertal development moderated links between independent stressor frequency and right and left mPFC SA such that greater independent stressor frequency is associated less mPFC SA earlier in puberty and greater mPFC SA later in puberty (Figures S1). This result for independent stressor frequency contrasts with the widespread gender differences seen in the neural correlates of dependent stressor frequency in the current study, and suggests that fateful life stressors are differently associated with SA at different points of pubertal development. This result builds on previous work suggesting that developmental timing of stressful experiences may affect their neural impacts (Teicher et al., 2016; Helpman et al., 2017), and highlights that pubertal events may be important mechanisms in these relations. One possibility is that changes in sex hormone levels occurring throughout puberty lead to altered neural structural impacts of stress. Sex hormones have been shown to interact with stress hormones to shape gray matter alterations resulting from stressful experiences in adult rodents (Garrett & Wellman, 2009), yet it is unknown how this occurs during puberty. Understanding the specific biological mechanisms driving the change in the neural correlates of stressors across puberty will be an important direction for future work.

Limitations and Future Directions

This study has limitations that are important to investigate in future work. These analyses are cross-sectional, preventing conclusions regarding directionality of effects. It will be important to try to differentiate stress generation versus neural sequelae of stressor mechanisms through longitudinal analyses. Additionally, despite the relatively large sample size, a recent study found that much larger samples may be necessary to detect robust brain-wide associations (Marek et al., 2022). However, in addition to widespread gender difference clusters in our cortex-wide exploratory analyses, we additionally found significant gender moderation of the link between dependent stressor frequency and whole-brain thickness. As this result does not rely on cortex-wide tests, it greatly strengthens the evidence for true gender differences in links between life stressor frequency and GM thickness. Future work should replicate these analyses in a larger, independent, sample to evaluate the replicability of clusters identified in the current study, including the precision of their location. As discussed earlier, it is possible that our observed effects are not location-specific but rather represent gender differences in links between dependent stress frequency and GM thickness across the cortex. The current study is also not highly-powered enough to investigate three-way interactions between age or pubertal development, gender, and stress. Our results highlighted the importance of looking at gender differences in relations between stress and GM–understanding how these interact with GM development will provide a better understanding of the mechanisms behind these effects.

Further, given neural developmental trajectories associated with increasing age (e.g., Giedd & Rapaport, 2010), the current study largely focused on age as an index of adolescent development. However, pubertal development is also associated with neural developmental changes (Herting & Sowell, 2017; Juraska & Willing, 2017), and may dissociate from age given individual differences in onset and trajectory of puberty. Indeed, pubertal development moderated the association between independent stressor frequency and mPFC SA in the current sample, highlighting pubertal changes as potential mechanisms shaping the impacts of stressor exposure during adolescence. The current study did not collect other indices of pubertal development such as sex hormone levels, which will be important for future investigations given evidence that stress hormones interact with sex hormones to impact neural structure (Garrett & Wellman, 2009). Further, a sample including younger participants than the current study would be better suited for a deeper investigation into how pubertal development may shape the effects of recent stressor exposure, as the current sample had already begun puberty at the time of the study (Table S3).

Despite these limitations, a strength of this study is that it investigated the relations between stressor frequency during adolescence and emerging adulthood and GM structure, which is an important step for understanding whether and how this developmental period acts as a sensitive period for the effects of stress, as many have proposed, but few have investigated (Romeo, 2017). Nonetheless, we did not investigate a comparable adult sample within these analyses, which will be important to fully understand how the correlates of life stress differ during adolescent and emerging adult development compared to adulthood, as others in the field have called for (Romeo, 2017; Fuhrmann et al., 2015).

Conclusion

Dependent stressor frequency was associated with GM thickness in salience and FPCN regions, and SA in DA and DMN regions, but in opposite directions for male and female youth. These results corroborate calls for greater focus on sex differences broadly in neuroscience research (Shanksy & Murphy, 2021) and specifically in work on neural effects of stressful experiences (Helpman et al., 2017). Further, pubertal development moderated the association between independent stressor frequency and mPFC SA. It will be crucial to better understand the importance of developmental changes on relations between life stress and brain structure, in particular during adolescence and emerging adulthood, a time posited to be a sensitive period for the effects of stress (Romeo, 2017; Fuhrmann et al., 2015). Future work should investigate these relations longitudinally and directly compare results in youth and adults to understand effects of life stress across development (Romeo, 2017; Fuhrmann et al., 2015). Last, these results highlight the importance of investigating dependent and independent life stressors separately to fully understand relations between life stress and the brain.

Data and Analysis: https://osf.io/p9xts/?view_only=0671799dd8c347cab1d7d3af2379d6d6

Public Significance Statement.

This study demonstrates that recent life stressor exposure is associated with opposite patterns of brain structure differences in female and male youth. This highlights the need for better understanding gender differences in how stress affects, and/or is affected by, brain structure during this developmental period.