Participants

Sixty-four mother-daughter dyads (adolescents age M = 11.45 years, SD = 0.77, 10–12 years) participated in the study. Female adolescent participants were included if they had elevated internalizing symptoms as determined by scores above the 50th percentile (raw scoresFootnote 1 > 44) on the self-reported Revised Children’s Anxiety and Depression Scale (RCADS) [24] at a screening assessment (raw score M = 66.84, SD = 17.15, Range = 45–115). Given emotion regulation difficulties are implicated in a range of internalizing symptoms, we examined broad internalizing symptoms rather than depression and anxiety symptoms separately. Exclusion criteria included: (1) current diagnosis of a developmental or intellectual disorder as reported by mothers; (2) current use of psychotropic medication; (3) any contraindications to MRI; (4) indications of claustrophobia; (5) history of head trauma or loss of consciousness for 5 min or more; (6) obesity (BMI > 30). Forty-six adolescents were reported by their mothers to be White/Caucasian (71.9%), followed by Mixed Heritage (14.1%) and Asian (7.8%). All participants provided verbal and written consent and were reimbursed $60 AUD for their time for participation. The study was approved by The Royal Children’s Hospital Human Research Ethics Committee (HREC 77,884). Participant demographics are presented in Table S1.

Measures

All measures have demonstrated good validity and reliability [24,25,26,27]. All Cronbach’s alphas (α) below are based on the current study sample.

Adolescent internalizing symptoms

Adolescent internalizing symptoms were assessed by adolescent report on the RCADS ( [24], 0 = Never to 3 = Always). The RCADS includes 47 items and six subscales: social phobia, panic disorder, major depression, separation anxiety, generalized anxiety, and obsessive-compulsive symptoms. Example items include “I feel sad and empty” and “I worry about making mistakes”. A total raw score was used, with higher scores indicating greater internalizing symptoms (α = 0.91).

Maternal emotion regulation

Maternal emotion regulation was assessed by mothers’ self-report on the Difficulties in Emotion Regulation Scale (DERS [25], 1 = Almost never to 5 = Almost always). The DERS includes 36 items comprising six subscales: lack of emotional awareness, lack of emotional clarity, difficulties controlling impulsive behaviors, difficulties engaging in goal-directed behavior, non-acceptance of negative emotional responses, and limited access to effective emotion regulation strategies. Example items include “When I’m upset, I become out of control” and “I have no idea how I am feeling”. A total score was used, with higher scores indicating greater difficulties (α = 0.8).

Maternal emotion socialization behaviors

Maternal emotion socialization behaviors were assessed by adolescent report on the Emotions as a Child (EAC [26]) scale (1 = Never to 5 = Very often). The EAC scale includes 45 items, measuring maternal emotion socialization of sadness, anger, and fear across five domains: reward, punish, override, neglect, and magnify. Six items were reverse scored. Items were summed into supportive (reward subscale, 3 items, α = 0.93) and unsupportive (neglect and punish subscales, 6 items, α = 0.89) emotion socialization subscales based on a previous study30. Example items include “When my child was sad, I comforted them” (reward), “When my child was sad, I did not pay attention to their sadness (neglect)”, and “When my child was sad, I told my child to stop being sad” (punish).

Emotional climate of the family

Emotional climate of the family was assessed by adolescent report on two subscales from the Parenting to Reduce Adolescent Depression and Anxiety Scale (PRADAS [27], 0 = Never to 3 = Often). The parent-child relationship subscale (6 items) assesses parental warmth, aversiveness, affection, and emotional availability. The home environment subscale (6 items) assesses family conflict, parental criticism, and parental modeling of conflict management. Example items include “My mom cares about my opinions” and “I hear my parents arguing with each other”. A total score of items across the two subscales was used, with higher scores indicating more positive family emotional climate (α = 0.79).

fMRI tasksAffect labeling task

We used a block-design affect labeling task to assess implicit emotion regulation [21]. During this task, participants were presented with negative (sad, fear, angry) faces from the NimStim set (available at http://www.macbrain.org) or shapes. In the ‘affect label’ condition, participants were instructed to label the target emotional face with one of the two emotional word labels below the face using a button box (Fig. 1a). In the ‘observe’ condition, participants were instructed to view the face and press their thumb against the button box to control for the confounding effect of motor activity [28]. In the ‘shape label’ condition, participants were asked to label the target shape with one of the two shape labels [21]. The ‘affect label’, ‘observe’, and ‘shape label’ conditions were presented randomly in a total of 10 blocks, with four blocks each for ‘affect label’ and ‘observe’, and two blocks for ‘shape label’. Each block included six trials, with a fixation of 12s in between blocks. A trial consisted of a face or shape stimulus presented for 4s, followed by a jittered fixation of 1.52–3.03 s. Faces were balanced for gender and race (two-thirds Caucasian and one-third Asian to reflect ethnic representation in Australia). Emotional labels included afraid, angry, sad, miserable, mad, and scared; and shape labels included triangle, rectangle, and oval.

Cognitive reappraisal task

We used a block-design cognitive reappraisal task to assess emotional reactivity and explicit emotion regulation [29]. Participants were presented with neutral or negative pictures and were asked to either regulate or observe them naturally. In the ‘reappraisal’ condition, participants were presented with negative pictures and instructed to think about the picture in a way that made them feel better about it. In the ‘look’ condition, participants were instructed to look at the picture and let themselves feel whatever the picture made them feel. There were 12 counterbalanced blocks and three conditions (reappraisal, look negative, look neutral). Each block began with a cue word ‘make it better’ or ‘look’ for 2s, followed by three picture stimuli each presented for 8s. Participants then rated how they were feeling on a scale from 1 (neutral) to 4 (very bad) using a button box (note that ratings were not included in the modeling of these conditions). The rating question was presented for 3.5s and there was an interstimulus interval of 0.5s between each picture stimulus and the rating (Fig. 1b).

All participants completed a practice before the task where they were given examples of reappraisal and were asked to report their reappraisals aloud. The pictures were comparable to previous studies using a cognitive reappraisal task in children and adolescents [29]. To reduce the risk of distress due to exposure to negative images, parents viewed all pictures and notified the research team if they wanted to swap out specific pictures. Two parents requested to replace pictures depicting violence with pictures of similar valence and arousal ratings. Picture stimulus sets and normative ratings are presented in Supplementary Table S2.

Fig. 1

Affect labeling and cognitive reappraisal fMRI tasks

fMRI acquisition and processingfMRI acquisition

Neuroimaging data were acquired on a 3T Siemens TIM Trio scanner with a 32-channel head coil at the Royal Children’s Hospital in Melbourne, Australia. Before the scan, participants were familiarized with the MRI in a mock scanner.

Structural T1-weighted images were acquired as follows: MPRAGE, slice thickness = 0.9 mm, repetition time = 2500ms, echo time = 1.72/3.45/5.18/6.91ms, flip angle = 8◦, field of view = 256 × 240 × 188 mm, matrix size = 284 × 266 × 208 mm, isotropic voxel size = 0.9 mm. The total sequence was 7.6 min. Functional images included T2*-weighted echoplanar images with the following parameters: 60 slices, slice thickness = 2.5 mm, repetition time = 1250ms, echo time = 30ms, flip angle = 90◦, field of view = 255 × 255 × 150 mm, isotropic voxel size = 2.5 mm. The total sequence was approximately 8 min for each task.

fMRI preprocessing and first-level analysis

Preprocessing was conducted using the ENIGMA Harmonized Analysis of Functional MRI pipeline (HALFpipe [30], https://github.com/HALFpipe). It performs preprocessing using fMRIPrep, which includes the following default settings: grand mean scaling with a mean of 10,000, spatial smoothing FWMH of 6 mm, ICA-AROMA denoising, temporal gaussian-weighted filter of 125s.

First-level analyses were performed in HALFpipe. Using a general linear model (GLM), predictors for each task condition were convolved with a double-gamma hemodynamic response function. The contrasts of interest were, for the affect labeling task, affect label > observe (implicit emotion regulation), and for the cognitive reappraisal task, look negative > look neutral (emotional reactivity) and reappraisal > look negative (explicit emotion regulation). Results of the affect label > shape label contrast are reported in Supplementary Table S3.

Regions-of-interest (ROIs) extraction

Two ROI masks were generated based on the Automated Anatomical Labeling (AAL) atlas [31] using the WFU PickAtlas Tool (version 3.0.5). The PFC ROI mask included precentral gyrus, superior frontal gyrus (SFG), middle frontal gyrus (MFG), and inferior frontal gyrus (IFG) orbital, IFG pars orbitalis, IFG triangular, gyrus rectus, and ACC. Because existing evidence on parenting and adolescent neural function during emotion regulation tasks are limited and inconsistent in terms of which specific PFC region is implicated, we utilized a large PFC mask that allowed us to investigate the specificity of PFC involvement in an exploratory fashion. The amygdala ROI mask included left and right amygdala as we had no a priori hypothesis on the differences between left and right amygdala.

Quality control

All preprocessed anatomical images were visually inspected for asymmetry, signal distortion or drop-out and artifacts. Participants with a mean framewise displacement > 0.5 mm were excluded [32] from analyses (N = 1 in the affect labeling task, N = 2 in the cognitive reappraisal task).

Statistical analysis

Second-level analyses were conducted using SPM12 implemented in MATLAB R2018b (Mathworks Inc). Separate GLMs were run to test the associations between each of the family and parenting factors and adolescent brain function for each of the task contrasts. Adolescent age was included as a covariate. We did not include race and family income as covariates given the limited variation in our sample. Both hypothesis-driven ROI (PFC, amygdala) and exploratory whole-brain analyses were performed. As such, 4 (parent/family predictors) x 3 (contrasts) x 3 (2 x ROI, whole-brain) analyses were performed. A small volume correction was applied for ROI analyses. For both ROI and whole-brain analyses, an uncorrected voxelwise correction of p < .001, and a cluster-level family-wise error (FWE) threshold of p < .0125 was applied. This threshold was applied to correct for multiple (i.e., 4) predictors (i.e., 0.05/4 based on a Bonferroni adjustment). Note that we did not additionally correct for the total number of contrasts given such a Bonferroni adjustment may be overly conservative [33].

Whole-brain mediation analyses were conducted using the CANlabMediation Toolbox (https://github.com/canlab/MediationToolbox). Bias-corrected bootstrapping with 1,000 resamples was conducted to test indirect effects of brain function in the relationship between family and parenting factors and adolescent internalizing symptoms. An FDR-corrected threshold of q < 0.05 was applied.