Effects of sex, pubertal stage, and pubertal hormones on frontal and limbic responses to fearful and happy faces during early adolescence

Abstract

Adolescence is a key stage of brain development marked by heightened emotional reactivity, with variability across individuals and sexes. Puberty onset also initiates a cascade of changes that influence how the brain responds to social-emotional information. To more fully characterize the role of sex and puberty in the brain’s response to social-emotional cues during early adolescence, we tested the main and interactive effects of sex, pubertal stage, and pubertal hormone levels on neural activation to negative and positive facial expressions. Combining baseline and 2-year follow-up data from the Adolescent Brain Cognitive Development^SM study (N = 13,182 observations at two timepoints; 53% male; ages 9–13), we examined neural activation to fearful and happy faces (vs. neutral faces) in the amygdala, anterior cingulate cortex (ACC), ventrolateral prefrontal cortex, ventromedial prefrontal cortex, and dorsolateral prefrontal cortex engaged during an emotional n-back fMRI task. Pubertal development was assessed via parent-report of the child’s pubertal stage and salivary hormone levels additionally indexed maturational changes from puberty beyond stage-based milestone achievements. A significant pubertal stage × estradiol interaction was found on ACC activation to happy faces among girls, indicating that estradiol level was positively associated with ACC activation in prepubertal girls but negatively in late-pubertal girls. No significant main effects of sex, puberty stage or hormone levels, or any other interactions withstood multiple comparison correction. These findings suggest estradiol’s effects on neural processing of positive social-emotional cues change across puberty among girls and may signify hormone-related shifts from emotional reactivity to regulation in the brain.

1. Introduction

Heightened reactivity to emotional information can influence adolescents' social interactions, impacting their ability to navigate increasingly complex social roles (Garcia & Scherf, 2015; Guyer et al., 2016). Starting in early adolescence, an uptick in emotionality coincides with continued maturation of brain regions involved in social-emotional processing (Guyer et al., 2016). In particular, the frontal-limbic network, including the amygdala, anterior cingulate cortex (ACC), and their connections with the prefrontal cortex (PFC), is central in social-emotional processing (Casey et al., 2019). Furthermore, puberty, marking the onset of adolescence, changes responses to social-emotional stimuli in these regions (Vijayakumar et al., 2018). Associations between markers of puberty and emotional reactivity further vary by sex (Lawrence et al., 2015) but less is known about these associations in relation to neural function. Understanding how sex and puberty influence frontal-limbic network function during social-emotional processing early in adolescence is critical for informing ways to address emotional challenges before they become more difficult to treat.

1.1. Pubertal stage

Puberty is a biological-driven developmental process marked by coordinated neuroendocrine events that reorganize multiple brain systems (Schulz & Sisk, 2016) and reshape adolescents’ complex social behaviors (Dai & Scherf, 2019). Converging evidence links advanced pubertal stages to heightened emotional sensitivity and altered responses to social reward and threat, patterns that likely reflect substantial changes in neural circuitry supporting emerging socio-emotional behavior during adolescence (Gur & Gur, 2016; Dai & Scherf, 2019). Emotional face stimuli provide an ecologically valid and reliable method for engaging brain regions involved in processing salient emotional information. Despite the importance of documenting sex and pubertal influences on frontal-limbic network activity during emotional face processing, the human findings are limited and mixed. One study reported a more advanced pubertal stage (controlling for age) was associated with increased amygdala reactivity to fearful versus other faces in adolescents aged 10–13 years old (Moore et al., 2012). Other studies observed mid-pubertal peaks (Vijayakumar et al., 2019) or no puberty-related differences in amygdala responses to fearful faces controlling for age-related maturation (Ferri et al., 2014; Forbes et al., 2011). Findings for amygdala response to happy faces are similarly inconsistent: some reported positive associations with pubertal development (Moore et al., 2012), whereas others found no relationship controlling for age (Vijayakumar et al., 2019). Using the Adolescent Brain Cognitive Development (ABCD) dataset, Morningstar and Burns (2025) found that youth more advanced in pubertal stage at baseline (8–11 years) showed less amygdala response to emotional faces (happy + fearful) concurrently and less change over two years.

Mixed findings have also been reported regarding pubertal stage and neural activation in frontal brain regions. More advanced pubertal stage was associated with decreased activity to fearful versus neutral faces in the ventrolateral PFC (vlPFC; Forbes et al., 2011), whereas it was associated with increased responses to happy faces in the vlPFC and dorsomedial PFC (dmPFC; Moore et al., 2012). Nonetheless, more work is needed to clarify the role of pubertal stage in frontal-limbic responses to emotional faces especially within a large, representative sample.

1.2. Pubertal hormones

Puberty is driven by two partially overlapping processes: adrenarche and gonadarche. Dehydroepiandrosterone (DHEA) is most closely linked to adrenarche, whereas estradiol is primarily associated with gonadarche and testosterone contributes to both processes. Evidence suggests these puberty-driving hormones are critical to both neural and physical maturation during adolescent development, extending beyond development of secondary sex characteristics (Goddings et al., 2019; Peper & Dahl, 2013). Examining DHEA, testosterone, and estradiol allows us to distinguish the potentially dissociable contributions of sex hormones to neural development, as these hormones vary in timing, function, and sex specificity. Much of the existing hormone-related evidence comes from studies conducted in gonadectomized rodents, primates (human and nonhuman) undergoing hormone treatments, or humans with hormone deficiency disorders (Schulz & Sisk, 2016). These findings cannot be simply generalized to inform brain development among healthy adolescents because ablation/replacement merely isolates hormone effects and the pubertal process culminates into more than the sum of its endocrine parts (Ruttle et al., 2024).

A long history of research in animal models converges on the conclusion that pubertal gonadal hormones exert organizational influences on adolescent brain reorganization, contributing to further sexual differentiation of neural systems (for review see Schulz & Sisk, 2016; Sisk & Zehr, 2005). Evidence indicates puberty-driving hormones are critical to both neural and physical maturation during adolescent development, extending beyond development of secondary sex characteristics (Goddings et al., 2019; Peper & Dahl, 2013). In contrast, parallel research in human adolescents remains comparatively nascent. Additionally, the brain regions that undergo significant change during human adolescence, especially those supporting complex social behaviors, reflect the combined influence of hormones and shifting social contexts, and do not have direct functional equivalents in animal models (Goddings et al., 2019).

Few human neuroimaging studies have investigated hormonal effects on neural response to emotional faces during normative pubertal development. Spielberg et al. (2015) found no effect of sex or testosterone on amygdala activation to fearful faces in adolescents aged 11–15 years old. In a longitudinal study of 9-to-18-year-olds, Vijayakumar et al. (2019) found that testosterone trajectories exhibited an inverted U-shaped relation with amygdala and hippocampal responses to fearful faces in females, beyond effects of pubertal stage.

In longitudinal work with adolescents aged 9–14, Whittle et al. (2015) found that dehydroepiandrosterone (DHEA) levels - controlling for pubertal stage and testosterone - were negatively correlated with ACC and dorsolateral PFC (dlPFC) response to viewing fearful faces in females. When viewing happy faces, DHEA was found to be positively associated with subgenual ACC and ventromedial PFC (vmPFC) activity in females after controlling for pubertal stage. Yet, little is known about how estradiol impacts neural response to emotional faces in healthy adolescents. Overall, these findings highlight that each hormone has unique contributions to neural responses to emotional faces, particularly in female adolescents, beyond the effects of reported physical pubertal development.

1.3. Present study

The divergent findings reported in the literature may be due to variability in study designs such as sample characteristics (e.g., age, sex), sample sizes, task paradigms (e.g., stimulus types, contrast conditions), puberty measures, and analytic approaches (e.g., linear vs. nonlinear models; see Table S1 for literature review). A larger and more diverse sample and the incorporation of multiple pubertal-related indices would aid in reconciling findings. Developing a clearer understanding of how pubertal processes shape neural responses to emotional faces is essential for identifying why adolescence marks both the expansion of sophisticated social functioning and a period of increased susceptibility to emotional and psychiatric disorders. Therefore, this study used ABCD data to investigate pubertal effects on brain activity to emotional faces and determine whether effects differ by sex¹.

Age and pubertal stage are often highly correlated during adolescence; however, neurodevelopment, particularly in pubertal hormone receptor-rich brain regions including the amygdala, may be more closely tied to puberty measures than to age (Goddings et al., 2019). Thus, we first examined the main and interactive effects of sex and pubertal stage on brain responses to emotional faces without age (puberty-only model) and with age as a covariate (puberty accounting for age effect model). This approach provided an opportunity to demonstrate whether brain responses in our regions of interest (ROIs) related to emotion processing and regulation would be specific to pubertal stage, or to the progression of age. Then, we examined the main and interactive effects of sex, pubertal stage, and each pubertal hormone (DHEA, testosterone, and estradiol) on neural responses to emotional faces, with age in one model and without age in the other. We aimed to investigate whether in early adolescence there is a dissociable or interactive effect of pubertal stage and pubertal hormones on brain responses. Table 1 shows a full list of the hypotheses and methods preregistered at https://osf.io/dw5ch.

Table 1.

Hypotheses in this study (registered in https://osf.io/dw5ch)

RQ1	Are there sex and pubertal stage effects on brain responses to fearful faces during early adolescence? Furthermore, is pubertal stage differentially associated with brain responses to fearful faces as a function of sex?
	Based on previous studies (Moore et al., 2012; Ferri et al., 2014; Vijayakumar et al., 2019), we expected that more advanced pubertal stage would be associated with greater amygdala, ACC, and vlPFC activation to fearful faces versus neutral faces. Based on work by Vijayakumar et al. (2019), we expected no main effect of sex on neural response during early adolescence, and an interaction effect whereby a stronger association between pubertal stage and amygdala, ACC, and vlPFC activation to fearful faces would be found among females than males.
RQ2	Are there sex and pubertal stage effects on brain responses to happy faces during early adolescence? Furthermore, is pubertal stage differentially associated with brain responses to happy faces as a function of sex?
	Based on previous work (e.g., Moore et al., 2012), we hypothesized that a more advanced pubertal stage would be associated with greater amygdala and vlPFC activation to happy faces versus neutral faces in both sexes. Sex differences were not found in adolescents (Vijayakumar et al., 2019; Moore et al., 2012), but Vijayakumar et al. (2019) reported the interaction effects between sex and the quadratic term of pubertal stage on brain responses to emotions. Therefore, we did not expect to see a main effect of sex on neural responses to happy faces. We did not have a priori expectations about the interactive effect of sex and pubertal stage on neural response to happy faces during early adolescence.
RQ3	What is the relation between DHEA levels and brain responses to fearful faces during early adolescence, and how does this vary by sex and pubertal stage?
	Based on previous studies (Whittle et al., 2015), we hypothesized that DHEA levels would be negatively correlated with ACC and dlPFC responses to fearful faces, independent of pubertal stage and age. We expected to find an interactive effect of sex and DHEA on neural response in these ROIs whereby females would show a stronger association than males. We did not expect to find significant effects for the other ROIs. We did not have a priori expectations about the other interactive effects (pubertal stage * DHEA; sex * pubertal stage * DHEA).
RQ4	What is the relation between testosterone levels and brain responses to fearful faces during early adolescence, and does this relationship vary by either sex or pubertal stage?
	Based on previous literature (Vijayakumar et al., 2019; Spielberg et al., 2014) which found no relation or an inverted U-shape association between testosterone levels and amygdala responses to fearful faces, we hypothesized that higher testosterone levels would be associated with minimal or weaker amygdala activation in response to viewing fearful faces during early adolescence. We expected to find an interactive effect of sex and testosterone levels on neural response whereby there would be a stronger association among females than males (Pagliaccio et al., 2015; Vijayakumar et al., 2019). We did not have an a priori expectation about the other interactive effects (pubertal stage * testosterone; sex * pubertal stage * testosterone). The association between testosterone and neural activity in the other four ROIs were treated as exploratory analyses.
RQ5	What is the relation between estradiol levels and brain responses to fearful faces during early adolescence among females?
	Given the prior study showing higher estradiol predicted lower brain activations in adult females (Rehbein et al., 2021), we hypothesized that estradiol levels would be negatively correlated with amygdala, ACC, dlPFC, and vmPFC activation to fearful faces in female adolescents. We did not expect to find significant results in the vlPFC. We did not have a priori expectations about the interactive effect of pubertal stage and estradiol on neural response.
RQ6	What is the relation between DHEA levels and brain responses to happy faces during early adolescence, and how does this vary by sex and pubertal stage?
	Based on a previous study (Whittle et al., 2015), we hypothesized that DHEA levels would be positively associated with ACC and vmPFC activation to happy faces, after controlling age and pubertal stage. We expected a stronger association would be found in females but not males, indicating the interactive effect of sex and DHEA level on neural response to happy faces. We did not expect to find significant effects for neural response in the other three ROIs. We did not have a priori expectations about the other interactive effects (puberty stage * DHEA; sex * puberty stage * DHEA).
RQ7	What is the relation between testosterone levels and brain responses to happy faces during early adolescence, and how does this vary by sex and pubertal stage?
	Based on previous literature (Vijayakumar et al., 2019), we expected no main effects of testosterone level on amygdala activation in response to viewing happy faces. Based on the same study, we did not expect sex differences or an interactive effect of sex and testosterone levels. We did not have an a priori expectation about the other interactive effects (pubertal stage * testosterone, sex * pubertal stage * testosterone). Additionally, we treated testing the association between testosterone and neural activity in the other four ROIs as exploratory analyses.
RQ8	What is the relation between estradiol levels and brain responses to happy faces during early adolescence?
	Based on findings of healthy female adults (Dubol et al., 2021; Sacher et al., 2013), we hypothesized that estradiol would be negatively correlated with neural activation in the amygdala, ACC, and dlPFC when viewing happy faces in female adolescents. We did not expect to find significant effects for the other ROIs. We did not have a priori expectations about the interactive effect of puberty stage and estradiol levels on neural response to happy faces.

2. Methods

2.1. Participants

Participants were enrolled at baseline (age 9–10) and 2-year follow-up (age 11–13) in the Adolescent Brain Cognitive Development (ABCD) Study^®, release 5.1 (https://abcdstudy.org/). Participants in the ABCD study include 11,878 children recruited at baseline at 21 sites throughout the United States (Karcher & Barch, 2021), starting at age 9–10 years and then annually for 10 years. Recruitment and sampling procedures for the ABCD study (Garavan et al., 2018) and ethical review and approval of the study (Clark et al., 2018) have been described in detail in previous publications. Centralized institutional review board (IRB) approval was obtained from the University of California, San Diego. In addition, study sites obtained approval from their local IRBs. Caregivers provided written informed consent, and the child provided written assent. Participants received monetary compensation for their study participation.

Data were preprocessed by the ABCD study staff according to standardized procedures (Casey et al., 2018; Chaarani et al., 2021). More detailed information about imaging protocol and quality control measures have been previously reported (Casey et al., 2018). Pre-specified inclusion/exclusion criteria specific to this study were pre-registered in https://osf.io/dw5ch. Intersex or transgender identity reported by child and/or parent at baseline and/or year 2 was an exclusion criterion. Participants were excluded if sex was not reported at baseline.

2.2. Measures

2.2.1. Pubertal stage

Physical markers of pubertal maturation were assessed at baseline and at 2-year follow up via self- and parent-report of the Pubertal Development Scale (PDS; Petersen et al., 1988), comprising 5 items assessing perceived pubertal maturation on a 4-point scale (not yet begun, barely started, definitely underway, seems complete). We converted PDS scores to Tanner Stages (ranging from 1 to 5) using procedures described in Shirtcliff et al. (2009). This coding system differentially captured gonadal and adrenal hormonal indicators of physical development. At both time points, we used caregiver reports considering the large number of “I don’t know” responses to several items from the child self-report; at these young ages, caregivers may have greater knowledge of where their child is in the process of change (Beltz et al., 2025; Herting et al., 2021; Dorn et al., 1990; Cheng et al., 2021).

2.2.2. Sex

Participant sex was determined by the child’s sex assigned at birth as reported by parents at the baseline assessment, when youth were aged 9–10 years old.

2.2.3. Hormones

We focused on the three main pubertal hormones that regulate sex differentiation and sexual maturation during pubertal development: DHEA, testosterone, and estradiol (available for girls only). These hormone levels were measured from a single saliva sample collected in the lab with the passive drool method and assayed in duplicate by Salimetrics (see Herting et al., 2021 for ABCD biospecimen methods) at baseline and year 2. Inclusion criteria for the hormonal data were based on a decision tree adapted from Herting et al. (2021), summarizing the steps taken to conduct a quality control assessment of each replicate in order to establish a single final estimate for DHEA and testosterone levels in both sexes, and estradiol levels in females only. The hormone data were cleaned and corrected for confounds by regressing the hormone level on all confounds and extracting the residuals for use by using multivariate fractional polynomial regression (Royston & Altman, 1994) in R’s mfp package (v1.5.2). Variable selection and identification of the best fractional polynomial transformation for each variable were evaluated at significance levels of 0.05 and done for both baseline and 2-year follow up data. We included the following confounds: time between waking and start of collection, collection duration (a proxy for flow rate), caffeine use (yes/no) and exercise (yes/no) in the 12 hours prior to collection, and glucocorticoid medication and oral contraceptives use in the past 2 weeks. Estradiol levels were additionally corrected for cycle regularity and menstrual cycle phase. Binary variables (1=yes, 0=no) were then created for each menstrual cycle phase and for irregular cycling (either self-reported or if the last period was over 36 days ago), which were added to the models for estradiol described above. Hormone values were log-transformed prior to analysis. Code and procedures for cleaning and correcting hormone levels were reported in prior work (Barendse et al., 2025) and available at https://github.com/marjolein15/Cleaning-ABCD-hormone-data/ (DOI: 10.5281/zenodo.10693410).

2.2.4. Emotional N-back fMRI task

We used the Emotional N-back fMRI task (EN-back task) to probe neural activity engaged during emotional face processing using a block design. Following practice in a simulated scanner environment, youth completed the EN-back task in an fMRI scanner. Participants were asked to identify whether an image in the series matched a given target in the N-back trial. A total of eight blocks varied in condition (0- or 2-back) and visual category (happy, fearful, or neutral faces, or places). Each block had 10 trials in which images were presented at the center of a grey screen for 2 s each, separated by fixation crosses presented for 500 ms. A full description of the task and imaging acquisition protocol is reported by Casey et al. (2018). Using criteria determined by the ABCD quality control team and pre-registered, the inclusion criteria were EN-back task images marked for inclusion (‘imgincl_nback_include’ =1), and an EN-back behavioral accuracy score greater than the guess rate (50%). Deviating from the preregistration, we added an inclusion criterion requiring no abnormal findings or variant of clinical significance on the youth’s MRI scans (‘mrif_score = 1 or 2’). Upon examination of variable distributions, we excluded five participants with erroneous beta magnitude (above ±3SD) for our contrasts of interest (reported in Supplemental Table S2).

2.3. Neuroimaging data processing

We used the processed and ROI-extracted neuroimaging data available on the National Institutes of Health Data Archive. These data were run through the abcd-hcp-pipeline (Hagler, 2019), created by the ABCD Data Analysis and Informatics Center. Scanning parameters, standard preprocessing, quality control, and first level analyses were reported in Casey et al. (2018) and Chaarani et al. (2021). Time points with framewise displacement > 0.9 mm were censored.

We focused on ten dependent variables in this study: beta weights from the EN-back task for the two contrasts of [fearful faces versus neutral faces] and [happy faces versus neutral faces], collapsed across the two EN-back conditions (0-back and 2-back) extracted from structurally-defined masks of five a priori ROIs: the amygdala, ACC, dlPFC, vmPFC, and vlPFC, respectively. We specifically focused on the neural processing of viewing fearful versus neutral faces, and happy versus neutral faces because each valence represents commonly encountered negative (e.g., threat) and positive (e.g., reward) stimuli (see the review: Goddings et al., 2019; Pozzi et al., 2021), and such probes have been widely used in studies of adolescents (Ferri et al., 2014; Forbes et al., 2011; Pagliaccio et al., 2015; Whittle et al., 2015) and adults (see meta-analysis: Filkowski et al., 2017; Stevens & Hamann, 2012). As in previous studies using these contrasts, we similarly used the more conservative method of contrasting response to an emotional face with a neutral face (rather than to a non-face cue) to establish a condition of emotional valence.

Activation in the left and right hemisphere was averaged for each ROI before entering them in statistical models. From the ABCD Study^® dataset, we included beta estimates derived using an amygdala mask segmented according to the Freesurfer subcortical atlas, and an anterior cingulate gyrus and sulcus mask (rostral ACC; anterior to aMCC and pMCC) based on the aparc2009 atlas (Destrieux et al., 2010). Based on previous literature, the dlPFC beta estimates were derived by combining beta estimates from the middle frontal gyrus and the inferior frontal sulcus (Chaarani et al., 2021). The vmPFC beta estimates were derived by combining beta estimates from the fronto-marginal gyrus and sulcus, transverse frontopolar gyri and sulci and lateral orbital sulcus (Fariña et al., 2021). The vlPFC beta estimates were derived by combining beta estimates from the triangular and opercular part of the inferior frontal gyrus following Kennedy et al. (2023). All cortical ROIs were parceled using the Destrieux atlas by the ABCD Study^® processing core.

2.4. Statistical analyses

Linear mixed effect modeling was the main analytic approach and was conducted using R Studio v4.2.2. The ABCD sample also contained a substantial number of siblings and twins and data were collected across 21 sites on 28 MRI scanners. Therefore, random intercepts by participant, family ID, and MRI scanner serial number (from 28 scanners) were included in all models to account for within participant, family and scanner correlation.

We first examined descriptive statistics for the demographic information at baseline and year-2 follow-up (sample size, age in years, race/ethnicity only at baseline, marital status, annual household income, highest education among parents; Table 2). Then, we examined the effects of sex and pubertal stage on ROI activation in response to fearful and happy faces respectively in the frontal-limbic system using linear mixed effect models controlling for covariates. Because puberty is moderately correlated with age, we conducted two sets of models with and without age as a covariate to aid in the interpretation of the results in the context of the existing literature. Specifically, we examined the main effects of sex and pubertal stage using the PDS and their interaction on brain reactivity to happy/fearful faces not controlling for age in the first set of models and controlling for age in the second set of models.

Table 2.

Demographic information of the participants at both visits

	Baseline (N=7,292)	Year 2 (N=5,890)	Total (N=13,182)
Sex at birth
Female	3504 (48.1%)	2694 (45.7%)	6198 (47.0%)
Male	3788 (51.9%)	3196 (54.3%)	6984 (53.0%)
Age (Years)
Mean (SD)	9.96 (0.63)	11.98 (0.65)	10.86 (1.19)
Range	8.92 – 11.00	10.58 – 13.83	8.92 – 13.83
Household income
N-Miss	0	5	5
<50K	1653 (22.7%)	1190 (20.2%)	2843 (21.6%)
>=50K and <100K	1972 (27.0%)	1542 (26.2%)	3514 (26.7%)
>=100K	3133 (43.0%)	2735 (46.5%)	5868 (44.5%)
DK/Refused	534 (7.3%)	418 (7.1%)	952 (7.2%)
Highest education
N-Miss	6	12	18
< HS Diploma	259 (3.6%)	232 (3.8%)	484 (3.7%)
HS Diploma/GED	546 (7.5%)	450 (7.7%)	996 (7.6%)
Some College	1748 (24.0%)	1827 (31.1%)	3575 (27.2%)
Bachelor	1961 (26.9%)	1314 (22.4%)	3275 (24.9%)
Post Graduate Degree	2772 (38.0%)	2170 (35.2%)	4834 (36.7%)
Race/ethnicity
White	4145 (56.8%)	3442 (58.4%)	7587 (57.6%)
Asian	163 (2.2%)	103 (1.7%)	266 (2.0%)
Black	822 (11.3%)	663 (11.3%)	1485 (11.3%)
Hispanic	1391 (19.1%)	1105 (18.8%)	2496 (18.9%)
Other	771 (10.6%)	577 (9.8%)	1348 (10.2%)
Marital status
N-Miss	48	35	83
Married	5245 (72.4%)	4202 (71.8%)	9447 (72.1%)
Divorced	633 (8.7%)	534 (9.1%)	1167 (8.9%)
Living with partner	357 (4.9%)	355 (6.1%)	712 (5.4%)
Never married	704 (9.7%)	519 (8.9%)	1223 (9.3%)
Separated	244 (3.4%)	190 (3.2%)	434 (3.3%)
Widowed	61 (0.8%)	55 (0.9%)	116 (0.9%)
Pubertal stage
N-Miss	457	417	874
Mean (SD)	1.81 (0.79)	2.69 (1.19)	2.20 (1.08)
Range	1 – 5	1 – 5	1 – 5
Testosterone (pg/ml)
N-Miss	421	1320	1741
Mean (SD)	33.42 (18.23)	51.23 (30.04)	40.54 (25.22)
Range	0.29 – 453.08	1.20 – 494.08	0.29 – 494.08
DHEA (pg/ml)
N-Miss	456	1325	1781
Mean (SD)	62.55 (48.13)	83.31 (65.62)	70.87 (56.71)
Range	0.00 – 426.63	0.00 – 897.69	0.00 – 897.69
Estradiol (pg/ml)
N-Miss	3993	3773	7766
Mean (SD)	1.01 (0.52)	1.03 (0.59)	1.02 (0.55)
Range	0.000 – 6.42	0.00 – 5.29	0.00 – 6.42

Note: DHEA = dehydroepiandrosterone

We further examined the effects of sex, pubertal stage, and hormones on neural responses to fearful faces and to happy faces. We examined whether including testosterone, DHEA, or estradiol improved model fit beyond pubertal stage in separate models per hormone by adding the main effect of hormones, and hormone-related interaction, controlling for covariates. Specifically, we examined the main effects of sex, pubertal stage, each hormone, and the sex-by-stage interaction, sex-by-hormone interaction, and sex-by-stage-by-hormone interaction (no sex-related interactions examined for estradiol which was only measured in girls) not controlling for age in one set of models and controlling for age in the other set of models. For each research question, each of the five ROIs was examined separately.

To conclude, there are 10 models for each research question, in which five models out of age were the primary analysis, while five models adding age were the sensitivity check whether age is a potential confound of pubertal metrics. We applied a Bonferroni correction to determine statistical significance across the five ROIs evaluated within each research question, leading to a significance threshold of 0.01 (two-sided) for each analysis. As a measure of effect size, we calculated marginal R-squared values of each model. We assessed model fit through evaluation of residual plots.

2.4.1. Covariates

We included several covariates in the models. We conducted analyses without age as a covariate to focus on the effect of pubertal stage and/or hormone on neural response. Then, we incorporated age in months as a continuous, time-varying covariate, since this is an important variable for the interpretation of pubertal stage-for-age and sex hormone effects (Vijayakumar et al., 2018). The other covariates included marital status, highest parental education level, household income at baseline and year-2 follow-up, as well as race/ethnicity and IQ at baseline. All of these variables, except IQ, were found to be related to pubertal development and hormone levels in the ABCD baseline sample (Herting et al., 2021). IQ was controlled for given its association with puberty status (Goddings et al., 2012) and emotion recognition ability (Lawrence et al., 2015). EN-back behavioral task accuracy score and head motion (specifically, average framewise displacement) in the fMRI scanner were also used as covariates in the models. Parent-report of children’s race and ethnicity at baseline (‘White’, ‘Asian’, ‘Black’, ‘Hispanic’, and ‘Other’) was included to account for effects related to race/ethnicity. Socioeconomic status (SES) was measured as caregivers’ combined household income at baseline and year-2 follow-up. Household income categories were categorized as <50k, between 50k and 100k, >100K, and Don’t know/Refused. IQ measured by WISC-V (Wechsler, 1991), EN-back task accuracy, and head motion index were modeled as continuous covariates. The other covariates were modeled as categorical variables and time-varying where applicable.

2.4.2. Approach to missing data

We anticipated missingness in the data due to restriction during COVID, attrition, and quality control. However, we did not expect this missingness to be related to our outcomes. Regression analyses used complete cases specific to each analysis.

2.4.3. Power analysis

Power analysis was conducted prior to analysis and reported in the preregistration (https://osf.io/dw5ch). To summarize, the predicted sample size was about 13,700 observations at baseline and year-2 follow-up in total. Given the sample size, we expected to have sufficient observations to detect small effects. Based on the small effect sizes reported in prior literature, we estimated that we achieved a minimum power level of 0.80 to detect an effect for each analysis (see the preregistration for the full report).

3. Results

3.1. Preliminary data analyses

The final sample included baseline data from 7,292 participants and 2-year follow-up data from 5,890 participants. Table 2 presents descriptive statistics overall and by each time point. Figure 1 illustrates the distribution of pubertal stage by sex and time point, and Figure 2 displays key variables by sex at each time point.

Figure 1.

Frequency plot of pubertal stage reported by parents separated by sex and visit. Girls depicted in green; boys depicted in orange.

Figure 2.

Age, pubertal stage, sex hormones, and neural activation in Emotional N-Back task in the regions of interest by sex and visit. Girls depicted in green; boys depicted in orange.

Supplementary Figure S1 shows correlations among key variables. Age, pubertal stage, and sex hormones were all significantly correlated. The correlation between age and pubertal stage across the two time points was moderate, r = 0.46, p < .001. At baseline, age and pubertal stage were moderately correlated, r = 0.18, p < .001, with a stronger correlation observed at the 2-year follow-up, r = 0.31, p < .001. At baseline, age showed modest correlations with various hormone levels (r = 0.15–0.20; all p < .001), and pubertal stage was similarly correlated with hormone levels (r = 0.13–0.30; all p < .001). These associations remained consistent at the 2-year follow-up, with age again modestly correlated with hormone levels (r = 0.15–0.20; all p < .001), and pubertal stage showing correlations ranging from r = 0.12 to 0.38 (all p < .001).

3.2. Sex and pubertal stage effects on neural response to emotional faces

There were no main effects of sex or pubertal stage on neural response to fearful vs. neutral or happy vs. neutral faces presented in the EN-back task in any ROI, regardless of whether age was controlled. Additionally, no interaction effects between sex and pubertal stage were observed. Full results with and without controlling for age are provided in Supplemental Tables S3–S12. Finally, age was not a significant predictor of neural responses to emotional faces in these models.

3.3. Pubertal hormones, sex, and pubertal stage effects on neural response to emotional faces

In the models examining the effects of DHEA, sex, and pubertal stage on responses to fearful versus neutral faces and happy versus neutral faces presented in the EN-back task in any ROI, we found no main effects of sex, pubertal stage, or DHEA, nor any interaction effects, with or without controlling for age. See Supplemental Tables S13 and S14 for detailed information.

In the models testing the effects of testosterone, sex, and pubertal stage on brain responses to fearful versus neutral faces, we found a negative effect of testosterone on vmPFC activation, raw p = 0.015 (Supplemental Table S15), controlling for covariates and age. No other main or interaction effects were observed. For brain responses to happy versus neutral faces, there were positive effects of sex on dlPFC activation, raw p = 0.050, and testosterone on vmPFC activation, raw p = 0.038, after controlling for covariates and age (Supplemental Table S16). Additionally, a three-way positive interaction between sex, pubertal stage, and testosterone levels was found, raw p = 0.049 (Supplemental Figure S3). However, none of these effects held at the Bonferroni correction for multiple comparisons (p < 0.01).

In the girls-only models examining the effects of estradiol and pubertal stage on brain responses to fearful versus neutral faces, we found a negative interaction between estradiol and pubertal stage on ACC activation (raw p = 0.038; Supplemental Table S17), controlling for covariates and age (Supplemental Figure S2a). For happy versus neutral faces, estradiol showed a positive main effect on ACC activation, raw p = 0.019, along with significant negative interactions between estradiol and pubertal stage on both ACC, raw p = 0.008, and dlPFC, raw p = 0.010, activation (Supplemental Figure S2b and S2c), controlling for covariates and age. Only the interaction effect between estradiol and pubertal stage on ACC activation to happy versus neutral faces survived Bonferroni correction (Supplemental Table S18), B = −.045, raw p = 0.008, ${\eta }_p^2=0.0014$, 95% CI [−0.078, −0.012]. Contrast analysis indicated a positive association between estradiol and ACC activation in prepubertal girls, but a negative association in late-pubertal girls (see Figure 3 and Table 3). All the results remained the same without controlling for age.

Figure 3.

Interaction between pubertal stage and estradiol on ACC response to happy versus neutral faces.

Table 3.

Pubertal stage and estradiol effects on ACC activation to happy faces versus neutral faces, and estradiol slope at each pubertal stage (with age as the covariate in the model)

Predictors	B	CI	Raw p	R 2
Pubertal stage (parent-report)	−0.001	−0.012 – 0.011	0.889	0.0000
Estradiol (corrected)	0.113	0.019 – 0.206	0.019*	0.0011
Pubertal stage × Estradiol	−0.045	−0.078 – −0.012	0.008 **	0.0014
Contrast Analysis: Pubertal stage × Estradiol
Pubertal stage	Estradiol slope	SE	Lower CL	Upper CL
Tanner Stage 1	0.068	0.033	0.003	0.132
Tanner Stage 2	0.022	0.021	−0.020	0.065
Tanner Stage 3	−0.023	0.020	−0.062	0.017
Tanner Stage 4	−0.068	0.030	−0.127	−0.008
Tanner Stage 5	−0.113	0.045	−0.200	−0.025

Note:

^*p < 0.05

^**p < 0.01

^***p < 0.001. Bold effect represents the significant effects which survive Bonferroni correction (** p < 0.01). ACC= anterior cingulate cortex. Shaded rows indicate significant effects.

4. Discussion

The current study investigated whether sex and physical indicators as well as hormonal drivers of pubertal maturation were associated with neural responses to emotional faces in 13,182 observations across ages 9–13 years. We fit linear mixed-effect models to examine the main and interactive effects of sex, pubertal stage (indexed by parent-report PDS), and hormone levels (including DHEA, testosterone, and estradiol) on brain activation to emotional faces in an EN-back task, focusing on the brain regions underlying emotional processing (amygdala, ACC, dlPFC, vmPFC, and vlPFC). Results indicated that estradiol levels among girls were differentially related to ACC activation to happy faces as a function of pubertal stage. No significant main effects of sex, puberty stage or hormone levels, or any other interactions withstood multiple comparison correction. This finding of a significant interaction between estradiol and pubertal stage on ACC activation to happy versus neutral faces is a novel contribution to the literature on brain development involved in emotional face processing.

4.1. Sex and pubertal stage effects on neural response to emotional faces

Our results showed no association between pubertal stage and neural response of any selected ROIs to emotional faces during early adolescence, with and without controlling for age. Our findings align with a report of no association between pubertal stage and brain activation to fearful faces versus neutral faces in girls (Ferri et al., 2014). Nonetheless, these were contradictory to our hypotheses and other previous studies (Forbes et al., 2011; Moore et al., 2012; Morningstar & Burns, 2025), which found positive or negative associations between pubertal stage and brain activation to emotions (see Table S1 for details). Morningstar and Burns (2025) found that more advanced pubertal stages were associated with decreased amygdala activation using the same ABCD baseline and year-2 dataset as our study. Different chosen contrasts (emotional faces versus fixation) and statistical analysis (bivariate latent change score models), which allowed for studying developmental changes over time, may contribute to the differing results, even within the same dataset. This comparison further highlights that differences in how pubertal development and neural responses to emotional faces are operationalized and analyzed can contribute to different findings.

Several reasons may explain the lack of association between pubertal stage and neural responses to emotional faces in this study. First, adolescents tend to process neutral facial expressions as emotionally-arousing stimuli (Rollins et al., 2021; Tottenham et al., 2013). Previous work using the ABCD study reported dlPFC activation when viewing neutral faces did not significantly differ from dlPFC response to fearful or happy faces in adolescents (see Figure 2 in Barendse et al., 2025), whereas viewing fearful or happy versus neutral faces evoked significant increased activation in the selected ROIs in adults in low cognitive load (Erk et al., 2007; Kesler et al., 2001). This suggests adolescents may appraise neutral faces differently than adults. Another possibility is that the relation between pubertal stage and neural responses to emotional faces is non-linear. Vijayakumar et al. (2019) provided such evidence by identifying an inverse U-shape relation between pubertal stage and amygdala and hippocampus activation to fearful faces (versus fixation) in girls only. Future studies should further investigate the mechanisms by which adolescents and adults vary in their neural activation to different emotions and should test for nonlinear developmental trajectories throughout adolescence.

The current study also investigated whether there were sex differences in brain responses to happy and fearful faces versus neutral faces, because most studies have used a sex-stratified method, but relatively few studies have investigated differences between sexes. After controlling for covariates, we did not find any sex differences. These results matched our hypotheses and a previous study finding no sex differences in the brain (Vijayakumar et al., 2019). There are some debates regarding whether secondary physical characteristics and hormones can be directly compared between sexes because the PDS items (e.g., breast development for girls vs. facial hair for boys) and the hormone sources (e.g., testosterone primarily from adrenal glands in girls vs. testes in boys) differ fundamentally, limiting measurement and conceptual equivalence across sexes.

4.2. Pubertal hormones, sex, and pubertal stage effects on neural response to emotional faces

The current study is the first to examine hormone effects (DHEA and testosterone for both sexes, and estradiol for girls), and their interactions with sex and pubertal stage on neural responses to emotional faces. The main finding of this study is the significant interaction effect between pubertal stage and estradiol on ACC responses to happy versus neutral faces (Figure 3). The ACC’s role in conflict monitoring is still developing during adolescence, which helps explain adolescents’ challenges in regulating emotions and behavior (Andrews-Hanna et al., 2011; Velanova et al., 2008). This result suggests estradiol may affect processing of positive emotional information during pubertal development. Some neuroimaging studies have also shown that estradiol levels are predictive of heightened brain activity in the ACC and dlPFC during response inhibition to positive stimuli in female adults (Amin et al., 2006) and of increased brain activity in the ACC to monetary reward processing in adolescents (Ladouceur, 2012; Poon et al., 2019).

Additionally, we found the association between estradiol and brain activation to positive emotions differed across various pubertal statuses: estradiol was related to heightened ACC activation to positive social cues for prepubertal girls, but blunted ACC activation for late-pubertal and fully matured girls (Table 3). We fully acknowledge that menstrual-cycle fluctuations in late-pubertal girls may confound this interaction effect, making direct comparison of estradiol levels across puberty stages challenging. Nevertheless, the observed pattern in late-pubertal girls aligns with adult findings (Dubol et al., 2024; Sacher et al., 2013), suggesting that higher estradiol levels are linked to reduced ACC responses to happy faces for females approaching or having reached sexual maturity. Our result further revealed that estradiol’s effects on ACC brain activity differs across puberty. Different from the other prefrontal and limbic regions examined in the current study, ACC is especially sensitive to social rejection and emotional pain during adolescence and its conflict monitoring function is still developing which may underlie difficulties adolescents can have with recognizing the need to regulate their emotions or behavior (Andrews-Hanna et al., 2011; Velanova et al., 2008). . Our findings may explain why social-emotional sensitivity peaks in adolescence, especially for girls (Bailen et al., 2019), when estradiol may amplify emotional responses to positive social cues via the ACC. It is important to note that although this interaction effect was statistically significant, the effect size was very small, with 0.14% variance explained by this effect, limiting its clinical significance in terms of understanding how well puberty-hormone interactions can explain individual differences in emotional sensitivity. The theoretical and methodological limitations in this study discussed below may explain this small effect size.

Although nonsignificant following multiple comparisons correction, similar patterns were noted for dlPFC activation to happy faces, and ACC activation to fearful faces (Supplemental Figure S2). These similar interaction patterns suggest that neural sensitivity may be shaped by hormonal exposure and developmental stage rather than either factor alone. The dissociation between estradiol and neural response across pubertal stages could be a mechanism by which emotional reactivity and regulation change over development rather than or in addition to the quadratic hormonal effects indicated by Vijayakumar et al. (2019). This finding also highlights the importance of examining hormone-brain relations within specific developmental windows rather than treating puberty as a uniform period. Future longitudinal work linking these patterns to mental health outcomes is needed.

Contrary to our hypotheses and past studies (Whittle et al., 2015) finding associations between DHEA or testosterone levels and neural activation to emotional faces, our results did not yield such associations. It is possible that our inclusion of a young sample of adolescents whose pubertal hormones may still be low and/or fluctuating markedly, especially for boys, account for our null findings. It is also plausible that the relation between androgens and neural responses to emotional faces is nonlinear, as suggested by Vijayakumar et al. (2019)’s findings, or vary as a function of pubertal stage, as we found for estradiol in this study. Future studies should consider a wider age range to allow for more variability in hormone patterns across adolescence.

4.3. Strengths and limitations

The current study had several strengths. First, by using the ABCD dataset, we drew on a large-scale national dataset to conduct robust and well-powered tests of the influence of sex and pubertal maturation on emotional face processing in the adolescent brain. Much of the existing literature has used relatively small samples. Given the large sample size of the current study, the null findings in this study are unlikely to reflect Type II errors. Second, we included data from multiple time points to increase variability in pubertal stage while also constraining the sample to the earlier phase of adolescence, allowing for more sensitive analyses. Third, of the studies that have examined puberty-related changes in regions of the frontal-limbic network in adolescence, few have examined the confluence of physical maturation and hormonal influences on the neural regions supporting emotional face processing, while also considering sex and age contributions. We investigated the potential puberty–hormone interaction effects on emotional face processing, which shed insight into the interplay between markers of pubertal development on neural responses to emotional faces.

Several limitations should also be acknowledged. First, there are measurement constraints inherent in the ABCD study (Cheng et al., 2021). Sex hormones were assessed using a single salivary sample. Despite our rigorous procedures to clean and correct for confounds, hormone levels are known to fluctuate across the diurnal and menstrual cycles (even before menarche), introducing variability that may compromise data validity. The questions on the PDS capture the self’s or parent’s perception of pubertal maturation rather than objective physical maturation (Cheng et al., 2021; Dorn et al., 2006). We used the parent-reported PDS in part due to a high level of missing youth-reported PDS data, especially at the baseline assessment. Although caregivers often have accurate knowledge about where their child is in the process of change at the earlier stages of puberty, parent-report may become less accurate as adolescents mature with age, possibly reducing our ability to detect associations across the pubertal stage range (Cheng et al., 2021). Additionally, although neuroimaging data were processed using the ABCD quality control pipeline, some participants still showed relatively high motion. Mean motion significantly predicted neural activation in most models, underscoring the need for stricter motion thresholds and quality control in future ABCD studies.

Second, the limited age range (9–13 years) restricted the representation of later stages of pubertal development, particularly in boys. Future studies would benefit from broader age ranges and/or longitudinal designs to capture a more complete picture of pubertal development. Such designs would also allow for the assessment of non-linear developmental trajectories over time.

Third, emotional face processing in this study occurred within an EN-back working memory task. Although we collapsed across the cognitive load conditions, the task was not optimized to probe neural response to emotional faces. Additionally, we used anatomically defined ROIs from ABCD FreeSurfer segmentation and parcellation. These larger ROIs may encompass heterogeneous subregions, potentially diluting activation magnitude due to signal averaging and/or constraining interpretation of the estradiol by pubertal stage interaction found here. Also, while we averaged brain activity across left and right hemispheres for each ROI in the absence of any functional lateralization hypotheses, it is plausible that this approach obscured hemisphere-specific activity. We used hypothesis-driven region selection in the present study and focused on five a priori defined ROIs; however, it is plausible that significant pubertal effects could emerge in a ROI not tested in our study.

Finally, there is ongoing debate about whether it is appropriate to directly compare brain–hormone associations across sexes. Vijayakumar et al. (2019) argued sex differences are qualitative as hormones may have different trajectories and mechanisms to drive puberty and neural development in girls and boys. Therefore, findings related to sex differences during puberty development should be interpreted with caution.

4.4. Conclusion

This study examined the associations among sex, pubertal stage, sex hormones, and neural responses to emotional faces during early adolescence using the ABCD dataset. Contrary to prior literature, we found no significant main effects of pubertal stage or pubertal hormones on activation within frontal-limbic regions involved in emotional processing. However, we found estradiol’s effects on processing positive emotional information varied across puberty in girls, indicating dissociable and interactive effects of physical maturation and hormones on brain maturation. These puberty–hormone interaction effects warrant further investigation with more precise hormone assessment, tasks specifically designed to probe emotional processing, and longitudinal analyses across additional timepoints. Overall, this study advances understanding of how puberty-related hormonal changes may influence neurobiological development underlying emotional reactivity and regulation during adolescence and suggests possible pathways to emotional challenges that often occur during this period of development.

Data Statement

Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development (ABCD) Study (https://abcdstudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children aged 9–10 and follow them over 10 years into early adulthood. The ABCD data used in this report came from release 5.1, DOI: http://dx.doi.org/10.15154/z563-zd24. DOIs can be found at https://nda.nih.gov/abcd/abcd-annual-releases.html