A normative model of brain responses to social scenarios reflects the maturity of children and adolescents’ social–emotional abilities

Abstract

The rapid brain maturation in childhood and adolescence accompanies the development of socio-emotional functioning. However, it is unclear how the maturation of the neural activity drives the development of socio-emotional functioning and individual differences. This study aimed to reflect the age dependence of inter-individual differences in brain responses to socio-emotional scenarios and to develop naturalistic imaging indicators to assess the maturity of socio-emotional ability at the individual level. Using three independent naturalistic imaging datasets containing healthy participants (n = 111, 21 and 122), we found and validated that age-modulated inter-individual concordance of brain responses to socio-emotional movies in specific brain regions. The similarity of an individual’s brain response to the average response of older participants was defined as response typicality, which predicted an individual’s emotion regulation strategies in adolescence and theory of mind (ToM) in childhood. Its predictive power was not superseded by age, sex, cognitive performance or executive function. We further showed that the movie’s valence and arousal ratings grounded the response typicality. The findings highlight that forming typical brain response patterns may be a neural phenotype underlying the maturation of socio-emotional ability. The proposed response typicality represents a neuroimaging approach to measure individuals’ maturity of cognitive reappraisal and ToM.

Introduction

Socio-emotional functioning is the ability to process mixed social and emotional information (McLaughlin et al., 2015; Campbell et al., 2016). As children and adolescents confront an increasingly complex and influential social environment, their socio-emotional functioning develops rapidly at this stage of life (Shulman et al., 2016; Hall et al., 2021). The development of socio-emotional functioning is manifested in enhanced mentalizing and emotion regulation (Pfeifer and Peake, 2012). On the one hand, the mentalizing ability facilitates inferring others’ mental states, which is quantified by the performance in the theory of mind (ToM) task (Dumontheil et al., 2010). Childhood is a pivotal period for developing ToM, which constantly develops during adolescence (Dumontheil et al., 2010; Symeonidou et al., 2020). On the other hand, emotion regulation manages emotional reactivity and plays a critical role in adolescents facing stress (Young et al., 2019). Emotion regulation undergoes profound changes during adolescence, reflected in the increasing use of sophisticated emotion regulation strategies such as cognitive reappraisal (Gullone et al., 2010; McRae et al., 2012; Theurel and Gentaz, 2018).

The development of socio-emotional functioning is accompanied by systematic brain functional changes. As mentalizing abilities mature from childhood to adulthood, core brain regions of the mentalizing network, such as the temporal lobe, superior parietal lobe and medial prefrontal cortex, are more extensively activated during the ToM task (Blakemore, 2008). In addition, adolescents specifically activate the middle/inferior temporal cortex (Fehlbaum et al., 2022). Besides, the maturation of the prefrontal cortex, a key component of the cognitive control system, improves the efficiency of cognitive reappraisal by enhancing top-down regulation (Casey and Jones, 2013; Stephanou et al., 2016). More globally, progressively dominant cortical connections enable adolescents to respond more similarly to adults in socio-emotional situations (Casey et al., 2019).

Socio-emotional development among children and adolescents embodies considerable individual variation due to genetic background and experience (Blakemore, 2012). However, it is unclear how the neural activity of children’s and adolescents’ brains reflects individual differences in their socio-emotional abilities. Although studies have revealed brain activation and functional connectivity networks exhibiting age differences (Cui et al., 2020), especially in emotion regulation tasks at the group level (McRae et al., 2012; Silvers et al., 2017), numerous factors beyond chronological age, such as childhood experience, culture, socioeconomic statuses and peer relationships, contribute to children and adolescents’ individual differences in socio-emotional maturity (Foulkes and Blakemore, 2018). Studies focusing on narrower age groups have linked individual differences in emotion regulation to neurocognitive bases at certain ages (Ahmed et al., 2015; Desatnik et al., 2021).

Nonetheless, we still lack neurological indicators to reflect the functional maturation of the brain underlying the development and individual differences of socio-emotional functioning in children and adolescents. Furthermore, the vast majority of research has used simple tasks to reflect the socio-emotional brain function of adolescents, such as repetitive facial expressions and affective pictures (Vink et al., 2014; Vetter et al., 2015; Wu et al., 2016). Although these paradigms have helped shape and validate various theories of socio-emotional maturity, they generally limit the vision to a localized subset of the neurological sources of individual socio-emotional differences. In contrast, individual differences in socio-emotional functioning may be reflected in rich and vivid scenarios that trigger participants to speculate and understand characters’ mental states (Nummenmaa et al., 2012, 2014; Schlochtermeier et al., 2017; Kaltwasser et al., 2019).

We propose to combine the normative model (Marquand et al., 2019) with naturalistic functional magnetic resonance imaging (fMRI) (Hasson et al., 2004) to characterize an individual’s brain activity relevant to social–emotional maturity. Different from the case–control group comparisons that examine mean differences between populations, the normative model attempts to make individualized inferences via quantifying the degree to which an individual’s brain deviates from a reference population (i.e. typicality), allowing for larger inter-subject differences (Marquand et al., 2016; Wolfers et al., 2018; Gruskin et al., 2020; Gruskin and Patel, 2022). The naturalistic fMRI is a parallel approach to resting-state fMRI, in which subjects are often told to watch movies that could evoke various emotional states and thoughts, allowing monitoring of brain response to close-to-real-life conditions (Hasson et al., 2004; Hasson and Honey, 2012; Redcay and Moraczewski, 2020; Saarimäki, 2021). Relative to the traditional paradigms that focus on specific condition contrasts naturalistic fMRI could better engage the participants and improve the validity of experiments for underage participants (Centeno et al., 2016; Moraczewski et al., 2018; Redcay and Moraczewski, 2020). With some advantages in test–retest reliability (Sonkusare et al., 2019; Gao et al., 2020) and head motion control (Greene et al., 2018), naturalistic fMRI has recently been applied to detect functional brain deficits of mental disorders (Guo et al., 2015; Rosenblau et al., 2017; Tu et al., 2019; Gruskin et al., 2020). Combining naturalistic fMRI with normative models, we can reflect the deviation of individual brain activity from a reference group when confronted with socio-emotional situations, thereby measuring the maturity of an individual’s socio-emotional processing-related brain activity throughout adolescence. Our previous studies have demonstrated the ability of this combination to identify individuals with mental disorders in highly heterogeneous samples (Yang et al., 2020; Zhang et al., 2022).

Previous naturalistic fMRI studies have shown that individual variability in brain activity in adulthood is significantly smaller than in childhood (Moraczewski et al., 2018; Richardson et al., 2018). Additionally, as adolescents’ age, there is an increase in concordance of activity in the sensory cortex among individuals (Cohen et al., 2022). This can be attributed to the accumulation of experiences that children and adolescents gain from social norms and education. It forms an emotional experience that is shared as they age (Verhoeven et al., 2019). Their evolving perspective-taking and mentalizing abilities yield a similar understanding of social context (Crone and Dahl, 2012; Blakemore and Mills, 2014). Meanwhile, inter-individual synchronization of brain responses reflects emotional states and mutual understanding (Nummenmaa et al., 2012, 2018; Dziura et al., 2021). Thus, if the maturation of socio-emotional function improves the concordance of brain responses to social scenarios between individuals, we can characterize the development of individual socio-emotional function by the similarity of the average brain activity pattern between individuals and a relatively mature group (Richardson et al., 2018; Richardson, 2019). Given these rationales, we expect that the inter-individual concordance of brain responses to socio-emotional situations increases with age during childhood and adolescence, and relatively high levels of individual concordance appear in late adolescence. Accordingly, we could use older participants as a reference group for the brain’s socio-emotional responses and apply a normative model to derive an individualized measure of brain response typicality for younger individuals. We hypothesize that the brain response typicality could reflect individual differences in socio-emotional functions.

To examine the hypothesis, we first identified brain regions with significant age–relation in inter-individual concordance for adolescents (n = 110, 8.0- to 17.9-years old) when watching the socio-emotional naturalistic stimuli. Then we used two independent datasets to validate the findings in age-dependent inter-individual concordance. The changes in the inter-individual concordance of these regions during adolescence were also examined, showing the age-related transitions from heterogeneous to common response patterns. We then proposed a response typicality index based on the normative model and examined whether the response typicality reflects the development of socio-emotional functioning by testing its effectiveness in evaluating emotion regulation strategies and ToM. As these socio-emotional abilities develop with age, we compared the predictive performance of age and response typicality on emotion regulation strategies and ToM. We further compared the effectiveness with cognitive measures, such as the performance of working memory and attention tasks, since those cognitive functions also accompany the maturation of cognitive regulation of emotion (Sanger and Dorjee, 2015; Malagoli and Usai, 2018). Furthermore, there is a parallel developmental progress for ToM and executive function, lying in mutual transfer between those skills (Perner et al., 2002; Arslan et al., 2017). Therefore, we compared the ToM predictive capability between response typicality and executive function. The latter was measured by the dimensional change card sort (DCCS) task (Zelazo et al., 1996). Finally, we used a dynamic typicality analysis to verify that the individual differences in the brain response patterns are associated with the stimuli’s emotional contents (valence and arousal).

Method

Figure 1 presents a workflow of this study. This study included three independent datasets. The findings from the main dataset were replicated using two validation datasets. Sections from ‘Participants in the main dataset’ to ‘Image acquisition’ describe the main dataset’s technical details, and Section ‘Independent validation datasets’ presents the validation datasets. Sections ‘Image preprocessing’ to ‘Dynamic typicality of brain responses and emotional relevance’ depict image processing and statistical analyses.

Participants in the main dataset

The main dataset contained 110 healthy adolescents (aged 8.0–17.9 years) who completed brain scans. The participants and at least one of their primary caregivers were assessed separately with the child and parent versions of the Mini-International Neuropsychiatric Interview for Children and Adolescents (MINI Kid) to exclude participants’ past or present mental disorders. For those participants who were judged to have no mental disorder or have a suspected diagnosis in the MINI-Kid interview, a senior psychiatrist further interviewed the participants and their primary caregivers to exclude any of the DSM-5 diagnoses of mental disorders. The participants with organic brain disorders (such as brain tumors, traumatic brain injuries and neurodegenerative diseases), physical disabilities, severe visual or hearing impairments, a history of severe head trauma, substance and/or alcohol addiction or severe suicidal tendency were excluded. The Institutional Review Board (IRB) of the Shanghai Mental Health Center approved the study (2018–46), and all participants provided written informed consent.

Emotion regulation strategies and cognitive measures

Ninety adolescents (50 females, mean age = 13.85 years) completed the Emotion Regulation Questionnaire (ERQ; Gross and John, 2003), which consists of two subscales: cognitive reappraisal (CR, Cronbach’s α = 0.79) and expressive suppression (ES, Cronbach’s α = 0.73). Higher scores on each subscale represent more frequent use of the corresponding strategy.

We measured attention and working memory functions by the N-back task and the attention network task (ANT) (Fan et al., 2002; Jaeggi et al., 2010). Fifty-six adolescents (30 females, mean age = 14.57 years) participated in both tasks. We used the reaction time (RT) of one-back and two-back conditions in the N-back task and the RT of altering, orienting and executive control components in ANT as measures of cognitive performance. Due to the high accuracy with tiny individual differences [mean accuracy (std): one-back = 0.95 (0.032), two-back = 0.905 (0.093), ANT = 0.977 (0.035)], the accuracy of N-back and ANT were not treated as performance measures. The RT of N-back and ANT were used to reflect individual variations in cognitive processes (working memory and attention capacity) (Koppen et al., 2011; Forns et al., 2014).

Video stimuli

We made a 7ʹ50”-long visual-perceptive socio-emotional video by concatenating materials from four public-interest advertisements and a segment of the cartoon ‘inside-out’. The material selection criteria were: evoking positive or negative emotions, understandable without sound and subtitles and representing engaging contexts. The video presents social interactions with parents and peers that activate perspective-taking and ToM functions. Adverse events such as war and the harshness of life in the video can induce strong negative emotions, which also involve emotional regulation. Further, the video depicts social challenges that adolescents frequently encounter, such as adjusting to a new environment and integrating into peer groups. Overall, these themes elicited intense emotional and social experiences and reflections in children and adolescents, stimulating rich socio-emotional functioning. More information regarding the video contents is presented in Supplementary Table S1. The audio track and subtitles of the video were removed, considering the influence of language on participants’ understandings and potential interference in speech processing (Hasson et al., 2008; Honey et al., 2012).

The emotional dimensions (valence and arousal) of the video were rated by 33 college students. Details of the rating procedure are described in Supplementary Methods 1.1 and Figures S1–3. The Cronbach’s α coefficients of internal consistency for the ratings were 0.91 and 0.98 for valence and arousal, respectively, indicating that the ratings were reliable. After the participants viewed the video during the fMRI scan, we performed a post-scan test on most participants to assess their engagement and understanding of the movie. The test consisted of a recall task, which asked the participants to judge whether a plot has appeared in the movie, and a question list to ask about the details of the stories. Details of the test are presented in Supplementary Methods 1.2. The mean accuracies were 89% and 86% for the two parts of the test, supporting that participants engaged in the video and understood the content.

Image acquisition

The video was presented to the participants during the fMRI scan. The participants were instructed to watch the video carefully and keep their heads still. The video was played without sound.

The scans were acquired on a 3.0 Tesla Siemens Trio system at the Institute of Neuroscience, Chinese Academy of Sciences. Foam padding was provided for the participants to minimize head movement and reduce scan noise. T1-weighted anatomical data were collected using a magnetization-prepared rapid acquisition gradient-echo sequence (repetition time = 2530 ms, echo time = 3.65 ms, inverse time = 1100 ms, flip angle = 7°, number of slices = 224, slice thickness = 1.0 mm, acquisition matrix = 256 × 256 and voxel size = 1.0 × 1.0 × 1.0 mm³). An experienced radiologist reviewed the images to exclude clinical abnormalities. The Blood oxygen level–dependent scan was acquired with an echo-planar imaging (EPI) sequence, repetition time = 2000 ms, echo time = 30 ms, flip angle = 77°, FOV = 220 mm × 220 mm, slice thickness = 3 mm, voxel size = 2.973 × 2.973 × 3.0 mm³, and 245 volumes.

Independent validation datasets

We included two independent datasets as validation sets. Validation dataset 1 included 21 healthy adolescent participants aged 9.0–17.9 years (13 males, average age = 11.98 years). The inclusion and exclusion criteria, as well as the video stimuli were the same as in the main dataset, but the fMRI data were collected using a Siemens Verio 3T MRI scanner at Shanghai Mental Health Center. The image acquisition parameters were the same as in the main dataset.

Validation dataset 2 was acquired from the OpenfMRI database (accession number: ds000228), originally collected by Richardson et al. (2018) and Kamps et al. (2022). This dataset includes 3- to 12-year-old children (n = 122) and adults (n = 33). All participants were scanned in MRI while viewing an emotional animation, ‘Partly Cloudy’. The participants and data collection details can be found in Richardson et al. (2018) and Kamps et al. (2022). The acquisition parameters are provided in Supplementary Methods 1.3. All children completed an explicit ToM task. The task consisted of listening to a story told by an experimenter and answering questions that required understanding and reasoning about the mental states of the characters in the story. The score of ToM was the average accuracy of all items. Sixty-two children participated in DCCS task, which measured executive function, an aspect of cognitive control.

Image preprocessing

The main dataset, validation dataset 1 and validation dataset 2 were organized and preprocessed on the INCloud platform (Li et al., 2021) using the Phi-Pipe procedure (Version 1.0.0, Hu et al., 2023). Specifically, the T1-weighted structural images and movie fMRI images were preprocessed using FreeSurfer (Version 6.0, Dale et al., 1999), Analysis of Functional NeuroImages (AFNI, Version 18.3.03, Cox, 1996), FMRIB Software Library (Version 6.0.0, Jenkinson et al., 2012), Advanced Normalization Tools (ANTs, Version 2.2.0, Tustison et al., 2014) and Convert3D (Version 1.0.0, Yushkevich et al., 2006). The T1-weighted images were brain-extracted and tissue-segmented using FreeSurfer. The brain image was then non-linearly registered into the MNI152 standard template using ANTs. For fMRI data, the main procedures included: (i) removing the first five volumes; (ii) head motion correction; (iii) rigid registration between median volume and the structural image using a boundary-based approach (Greve and Fischl, 2009); (iv) estimating the whole-brain, white matter and ventricular masks; (v) motion outlier detection and interpolation (i.e. motion censoring). Volumes with frame-wise displacement (FD) > 0.5 mm (Power et al., 2012) were treated as motion outliers and interpolated by neighboring volumes; (vi) nuisance regression. The nuisance signals included mean signals of white matter and ventricles, Friston’s 24-parameter head motion model; (vii) high-pass filtering at 0.01 Hz. Motion censoring, nuisance regression and high-pass filtering were performed using AFNI’s 3dTproject; (viii) grand mean scaling to 10 000; (ix) transformation into MNI152 space by combining T1-MNI152 and fMRI-T1 registration results and resampling into 3×3×3 mm³.

The brain extraction, tissue segmentation and spatial registration results were visually checked. The participants with mean FD > 0.5 mm and outlier ratio > 0.2 were excluded. The outlier ratio was the ratio between the motion outliers and the total volumes. Seven participants in the main dataset and 30 participants in validation dataset 2 were excluded due to excessive head motion. The final sample sizes were: 103 (47 males, mean age = 13.57) for the main dataset; 21 (13 males, mean age = 12.0) for validation dataset 1; and 93 children (43 males, mean age = 6.8) and 32 adults (12 males, mean age = 24.5) for validation dataset 2. Supplementary Figure S4 shows the age distribution of the three datasets.

Age effect on ISC

To examine the age effect on the concordance of brain activity response to the socio-emotional scenario, we applied an ISC analysis using the 3dISC program in AFNI (Chen et al., 2017), which implements a univariate linear mixed-effect model to identify the brain region modulated by age (Figure 1A). The correlation of brain activity in each voxel within participant pairs (ISC) served as the independent variable, while age (the sum of dyad’s centralizing age) served as the dependent variable. We included each participant pair as random effects terms (represented as subject i, subject j). We used a false discovery rate threshold (FDR, q < 0.05) to control multiple comparison errors in the resultant effect map. Considering that spatial clustering is beneficial to further reduce type I error and alleviate the impact of noise on significant results (Loring et al., 2002; Lieberman and Cunningham, 2009), we applied an additional cluster size threshold of 20 voxels to identify regions of interest (ROIs) for further analyses (Kober et al., 2019; Yan et al., 2019; Giannì et al., 2021). Besides, since head motion negatively correlates with age (Meissner et al., 2020), we conducted an analysis on the identified ROIs to rule out the potential impact of head motion level (Lund et al., 2005). The mean FD was regressed out from both ISC and age, and the correlation coefficients between the residual ISC and residual age matrices were calculated. The correlation coefficients were further examined using permutation tests that randomized the order of the participants (1000 repetitions).

Validation of age effects on the ISC

We validated the age dependence of ISC using validation datasets 1 and 2. We focused on the ROIs showing a significant age effect on ISC in the main dataset and extracted the time courses of these ROIs in both validation datasets. In validation dataset 1, we tested the relationship between age and ISC through the same procedure mentioned in Section ‘Age effect on ISC’. Due to a large age gap between children and adults in dataset 2, we compared the age-group difference (child versus adult group) on ISC using the 3dISC program in AFNI.

Associations between brain response typicality and emotion regulation strategies

We constructed a normative model to investigate whether the similarity of individuals’ brain responses to a reference group reflects individual differences in emotion regulation strategies. The similarity was quantified as typicality, i.e. the Fisher z-transformed Pearson correlation coefficient between an individual’s brain response and the mean brain response of the reference group at a given ROI (Figure 1B).

This analysis focuses on the ROIs identified in Section ‘Age effect on ISC’. We averaged the time series over all voxels within the ROI to generate the ROI-wise representative time series for each participant. We defined a reference group of adolescents (16.5–17.9 years, n = 20), since they are relatively older and most adult-like compared to younger adolescents in this sample. In addition, the above analysis showed that the ISC of most ROIs reached relatively high levels among older adolescents (see Figure 3). The time series were averaged across participants in the reference group to obtain the template time series. The response typicality for each ROI of the participant (under 16.5 years old, n = 83) was computed by correlating their time series to the corresponding template time series from the reference group.

We investigated whether response typicality explained emotion regulation strategies by stepwise regression. Of the 83 adolescents, 14 lacked CR scores and 13 lacked ES scores, and they were excluded from the CR and ES analyses, respectively. There were no group differences in brain response typicality between the participants with and without ERQ scores (Supplementary Table S2). In stepwise regressions with age and sex as covariates, response typicality of ROIs were used as independent variables (all stepwise regressions below included age and sex), and CR and ES were used as dependent variables, respectively. Using the Akaike information criterion criterion, we selected the best model to fit CR and ES.

We further validated the predictive power of response typicality for CR/ES using support vector regression (SVR, linear kernel) and leave-one-out cross-validation (LOOCV). Independent variables were response typicality of 20 ROIs, and the dependent variable was CR/ES. All SVR models included age and sex as covariates. Correlations between predicted and observed values were calculated to indicate the performance of the models. We conducted a permutation test with 1000 repetitions to examine the significance of the correlations.

Since adolescents’ emotion regulation strategies vary with age, we further explored whether response typicality could explain emotion regulation strategies beyond age. We used likelihood ratio tests to compare two models: (i) the full model, using response typicality, age and sex as predictors; (ii) the nested model, only using age and sex as predictors. We further trained SVR models corresponding to the above models and compared the LOOCV performance.

We further compared the effectiveness of brain response typicality and cognitive measures in predicting emotion regulation strategies, as cognitive functions, such as working memory and attention, may also contribute to predicting emotion regulation strategies. For this analysis, only those who completed the N-back and ANT tasks (n = 29 for CR, n = 30 for ES) were included, and they covered a wide age range of the main dataset (Supplementary Figure S5). These 29 and the other participants did not show significant differences in any item or region (CR: t = 0.136, P = 0.892; ES: t = −1.01, P = 0.316; see Supplementary Table S3 for statistics on response typicality). Considering that regressing a large number of features with a relatively small sample size may lead to overfitting, we used principal component analysis (PCA) to compress response typicality measures of ROIs identified in Section ‘Validation of age effects on the ISC’ into PCA-typicality measures. Then, we conducted stepwise regressions to examine the contributions of the PCA-typicality in interpreting the CR and ES. Furthermore, we examined the predictive power of response typicality by comparing two models: (i) the full model included PCA-typicality and cognitive measures (RT of one-back, two-back, altering, orienting and executive control) as independent variables, age and sex as covariates, and (ii) the nested model only included cognitive measures as independent variables, age and sex as covariates. The likelihood ratio test was used to compare the full and nested models. We further trained SVR models corresponding to the full and nested models to predict CR and ES. The performance of the two SVR models was compared.

Associations between brain response typicality and ToM

To assess whether response typicality reflects another critical aspect of social–emotional functioning, ToM, we applied stepwise regression, SVR models, and likelihood ratio tests in the validation dataset 2. To avoid a ceiling effect in older children, which distorts the effect of age on ToM scores, we removed 19 participants who gained a full score in the ToM task (Supplementary Figure S6). Seventy-three children and thirty-two adults remained for response typicality analysis. As earlier, we focused on the response typicality of the significant ROIs obtained in Section ‘Validation of age effects on the ISC’. In each ROI, we calculated the response typicality of children by correlating their brain responses with the average brain responses of the adult group. Stepwise regression examined the contribution of response typicality to ToM (age and sex as covariates). We then used SVR models to assess the predictive performance of response typicality on ToM (age and sex as covariates). Likelihood ratio tests compared the goodness of fitting of the models with and without response typicality. In addition, we tested whether response typicality improved the goodness of fitting (likelihood ratio test) and predictive performance (SVR model) of the models that included the executive function as reflected by the DCCS scores. Same as in Section ‘Associations between brain response typicality and emotion regulation strategies’, we used a PCA approach to compress the ROI-wise response typicality due to the limited sample with DCCS scores (n = 48).

Dynamic typicality of brain responses and emotional relevance

We then examined the associations between the dynamic typicality and the emotional relevance of the video (Figure 1C). We observed low intra-class correlation (ICC) values for all ROIs, as indicated in Supplementary Table S4. The low ICC values suggest that a smaller proportion of the total variance is attributable to between-subject differences (Luke, 2004). These findings support the use of the general linear model (GLM) in our analysis. Therefore, we conducted a GLM with dynamic typicality as the dependent variable and valence/arousal of each time window (window width = 10 TR, increment = 5 TR, consistent with the sampling frequency of emotional ratings) as the independent variables. Age, sex and head movement (FD) were included as covariates. The effects of valence and arousal on the dynamical typicality were modeled separately. All variables included in the models were standardized. Finally, the main effects and interaction effects were corrected for multiple comparisons using FDR.

Results

Age effect of brain activity concordance

Age modulated the brain activity concordance across adolescents who watched the same socio-emotional video. Figure 2 and Table 1 show significant age modulation of ISC in 16 ROIs distributed across the occipital, temporal and parietal cortices, anterior cingulate cortex (ACC), precentral gyrus and insula. Age positively modulated the ISC of all ROIs except the one at the middle temporal gyrus. Supplementary Figure S7 presents pair-wise ISC matrices for each ROI.

Figure 3 reflects the ISC’s increasing trends with age in most ROIs, except ROI 16 (R.MTG_2). As ROIs 1 and 2 covered a wide range of anatomical locations, for further analyses, we partitioned them into 7 ROIs according to the AAL atlas (Rolls et al., 2020). When adjusting for mean FD, the correlation between ISC and age patterns were still significant for all ROIs (P < 0.001; Supplementary Table S5).

Validation of age effects on brain activity concordance

We validated the moderating effect of age on ISC for the 21 ROIs found in the main dataset using validation datasets 1 and 2. In Dataset 1, 12 ROIs had a significant positive age effect on ISC (Supplementary Figure S8); in Dataset 2, 17 ROIs showed significantly higher ISC in the adult group than in the child group (Supplementary Figure S9). The results of the two validation datasets overlapped in 10 ROIs, supporting the robustness of the age-moderating effect on ISC for most ROIs (Supplementary Table S6).

Brain response typicality predicts emotion regulation strategies

Based on the above findings that suggest that the inter-subject concordance achieved a relatively higher level among older adolescents, we used the average brain responses of adolescents > 16.5 years of age as a reference. This group corresponds to the last age-window in the sliding age-window analysis and represents the oldest participants in this dataset. With this reference, we calculated the typicality of brain responses for each ROI for individuals < 16.5 years of age (n = 83). We removed the R.MTG_2 (ROI16) from further analyses because the age effect on the ISC of this ROI was negative, and defining response typicality for this ROI is not proper.

Adolescents’ CR significantly increased with age (F = 4.606, P = 0.013, Supplementary Figure S10). Additionally, older adolescents have a smaller variation in the score of use in cognitive reappraisal. To predict an age-relative ability, it is necessary to compare the statistical power between age and response typicality. When modeling the CR, the stepwise regression retained the contribution of 11 ROIs (Supplementary Table S7) and significantly explained the individual differences in CR (F = 3.980, P < 0.001). In addition to age and sex, response typicality made a distinct contribution to the interpretation of CR, as reflected in the significance of the likelihood ratio test (χ² = 39.844, P = 0.005, Table 2). Similarly, the LOOCV showed that the SVR model with response typicality measures could predict the CR scores (r = 0.380, P < 0.001, Figure 4B), which outperformed the SVR model only including age and sex as predictors (r = 0.176, P = 0.01, Figure 4A). To gain further insight into the relationship between response typicality and CR, we have examined the correlation between response typicality and cognitive reappraisal for each ROI. We found the positive correlation between response typicality and cognitive reappraisal in multiple ROIs (Supplementary Figure S11).

Although the performance on the cognitive task explained and predicted CR scores in adolescents (F = 3.404, P = 0.016, Supplementary Table S8), response typicality contributed significantly to explanatory power and could improve CR prediction accuracy. The explanatory power of the model that added PCA-typicality measures (F = 4.647, P = 0.003, Supplementary Figure S12 and Table S9) as independent variables was significantly superior to the model that only included N-back and ANT performance (χ² = 10.358, P = 0.006, Table 2), according to the likelihood ratio tests. In the same way, incorporating PCA-typicality into the SVR model improved CR’s prediction accuracy (Figure 4; N-back and ANT: r = 0.346, P = 0.039; N-back, ANT and PCA-typicality: r = 0.492, P = 0.002).

When modeling the ES, the stepwise regression of response typicality did not achieve significant goodness of fitting (F = 2.11, P = 0.129, Supplementary Table S10). The SVR models failed to predict the ES scores with significant performance (age and sex as predictors: r = −0.367, P = 0.996; age, sex and response typicality: r = −0.149, P = 0.870). Response typicality did not outperform the age model on ES in the likelihood ratio test (χ² = 17.475, P = 0.622, Table 2). The cognitive measures did not predict ES either (Supplementary Table S11).

We tested whether changes in the reference group would affect the effectiveness of response typicality of CR. We varied the number of participants (n = 14, 25 and 37) to define three reference groups consisting of the oldest adolescents and repeated the above statistical analysis. We obtained similar results that the new response typicality significantly explained and predicted CR. Its model fit was specific to age, sex and cognitive performance. As shown in Supplementary Results 2.1 and Table S12.

To investigate the relationship between response typicality and other age-related variables, we conducted additional analyses to examine the association between response typicality and cognitive measures, specifically working memory and attention. The results of these analyses indicated that there was no significant correlation between response typicality and these cognitive variables (Supplementary Results 2.2). This suggests that response typicality is more specifically related to socio-emotional competence rather than general cognitive abilities.

We used Inter-Subject Representational Similarity Analysis to estimate whether the similarity of brain activity related to the CR (Supplementary Results 2.3). The results showed that most of the ISC matrices showed positive correlations with the CR matrix (Supplementary Figure S13), which further demonstrates the relevance of neural similarity to cognitive reappraisal.

Brain response typicality predicts ToM

We further examined the explanatory and predictive power of response typicality for ToM based on the 17 ROIs that showed a significant positive correlation between ISC and age. Stepwise regression showed that 2 ROIs’ response typicality significantly explained ToM, and 6 ROIs reported a marginally significant contribution (F = 7.841, P < 0.001, Supplementary Table S13). The SVR model with response typicality measures as predictors was significant in predicting ToM (r = 0.606, P < 0.001, Figure 4F), surpassing the SVR model with age and sex only in predicting ToM (r = 0.537, P < 0.001, Figure 4E). In addition, response typicality significantly enhanced the explanatory power of the ToM (likelihood ratio test: χ² = 35.179, P = 0.006, Table 2). These results suggest that the response typicality can reflect individual differences in ToM. The results without removing the 19 children with full ToM scores are presented in Supplementary Results 2.4. We also examined the relationship between response typicality and ToM scores (Supplementary Figure S15), which showed that higher response typicality across multiple ROIs were associated with higher ToM capability.

Compared to the executive function reflected in DCCS (r = 0.324, P = 0.003, Figure 4G; F = 3.865, P = 0.015, Supplementary Table S15), PCA-typicality (Supplementary Figure S16) exhibited a distinct contribution to the prediction of ToM, as reflected by the superior performance in the SVR model that combined PCA-typicality and DCCS (r = 0.427, P < 0.001, Figure 4H; F = 7.581, P = 0.001, Supplementary Table S16). The likelihood ratio test also showed trending significance when comparing the DCCS + PCA-typicality model with the DCCS-only model (χ² = 4.741, P = 0.192, Table 2).

In IS-RSA, positive correlations were observed between the ISC matrices and the ToM matrix (Supplementary Figure S17), suggesting that shared neural activities are associated with the development of ToM ability.

The brain response typicality is relevant to emotional contents

We further examined whether the typicality was related to the emotional contents of the stimuli. As shown in Figure 5A, significant age-by-valence interaction was found in the dynamic typicality of the right superior temporal gyrus (R.STG_1, F = 10.341, P = 0.028; Supplementary Table S17). The positive valence contents evoked higher dynamic typicality in the R.STG in older adolescents compared to younger adolescents. As presented in Figure 5B, the main effect analyses showed that most ROIs’ dynamic typicality was significantly related to valence. The number of ROIs for positive and negative valence was relatively balanced (Supplementary Table S18).

The dynamic typicality of most ROIs showed significant main effects for arousal, suggesting that higher arousal elicited higher dynamic typicality of brain responses (Figure 5C, Supplementary Table S19). We did not observe a significant age-by-arousal interaction effect (Supplementary Table S20). We obtained similar results when valence and arousal were included in a regression model (Supplementary Tables S21–24). These findings suggest that the dynamic response typicality is relevant to emotional contents with some age modulation, supporting that the response typicality is driven by socio-emotional processing.

To address the potential confounding of valence and arousal ratings, we reanalyzed the relationship between valence and response typicality by transforming valence ratings to a 3-point scale. We obtained similar results to the original results, confirming the robustness of the association. Further details and a visual representation of these results can be found in Supplementary Results 2.5 and Figure S18.

Discussion

The present study investigated whether individual differences in brain activity in response to social–emotional scenarios reflect the maturity of children’s and adolescents’ socio-emotional abilities. We first found that inter-individual concordance increased with age in multiple brain regions when adolescents watched socio-emotional videos. Then, we replicated the findings in two independent datasets. This finding suggests that as children and adolescents grow up, they develop common patterns of brain activity for processing socio-emotional information, which may enable them to feel shared experiences and form a social consensus.

We applied a normative model of typical brain response and defined a ‘response typicality’ metric, which characterizes the similarity between an individual’s brain response to a video and the average brain response of older adolescents or adults. We show that response typicality predicts individuals’ cognitive reappraisal and ToM and its contribution cannot be accounted for by age, sex, or cognitive benchmarks. These findings indicate that individual differences in brain responses to socio-emotional scenarios reflect the two critical components of socio-emotional functioning (CR and ToM). Thus, this study reveals a generalized socio-emotional brain network associated with multifaceted socio-emotional ability, which exhibits high plasticity during childhood and adolescence. Based on the positive relationship between response typicality and CR, a lower response typicality may indicate an ineffective or less efficient cognitive reappraisal strategy or the use of different strategies by individuals.

The three independent datasets reflected ten overlapping ROIs, supporting the robustness of the age modulation of the ISC in children and adolescents in general. However, when considering brain indicators of the specific socio-emotional functions (ToM or emotion regulation strategies) that exhibit large individual variabilities in different developmental stages (childhood or adolescence), including all brain regions associated with the specific age stages helps increase the predictive power.

We further demonstrated that the response typicality is associated with the contents’ emotional dimensions (valence and arousal). This evidence confirms that the neural processing of emotional information drives the response typicality. Previous studies have verified the reliability in reflecting individual differences (Gao et al., 2020) and the sensitivity of the naturalistic stimuli paradigm (Lahnakoski et al., 2014; Trost et al., 2015; Adebimpe et al., 2019). Compared to resting state and repetitive tasks, this paradigm could vividly convey social and emotional information. Combining naturalistic imaging with the normative model that avoids assuming homogeneity across individuals, the methodology used in this study provides an ecological and practical approach to characterize adolescents’ common and individualized brain activity patterns (Yang et al., 2020).

A novel theoretical contribution of this study is revealing the development of the neurological basis of the shared mindset in childhood and adolescence. The shared mindset is a concept in sociology and management (Haas and Mortensen, 2016), which means common beliefs and values informed through the social norm and accumulative experience (Verhoeven et al., 2019). The shared mindset is manifested as shared neural processes elicited by the same external input, generating similar emotional experiences and cognitive interpretations (Yeshurun et al., 2021). It catalyzes interpersonal communication, teamwork (Schwarz and Bennett, 2021) and the evolution of human culture and society (Gupta and Govindarajan, 2002). Older adolescents are more likely to achieve effective interpersonal information transmission in complex socio-emotional situations and generate shared emotion, cognition and behavioral performance, resulting from social learning (Pan et al., 2022). Echoing the ‘shared mindset’ theory, our findings reveal that the neural circuits relevant to socio-emotional processing also mature by converging on adults. This tendency of brain response conforms to the law of organizational socialization and cultural awareness. During childhood and adolescence, an individual gradually meets social expectations and becomes a person capable of cooperating (Özdemir and Ergun, 2015).

The brain response concordance increasing with age further reveals that ‘all roads lead to Rome’ in the maturation pathway of social–emotional functioning. While previous studies have shown differences in inter-subject concordance between children and adults (Moraczewski et al., 2018; Vanderwal et al., 2019), the present research locates specific brain regions associated with the maturation of socio-emotional functioning and conveys a full picture of the convergence of individual differences in these brain regions during development. The age-sensitive regions identified in this study are the visual network, dorsal attention network, ventral attention network, default mode network and sensorimotor network. These networks have been proposed to underlay social cognition components, such as empathy, mentalizing and mirror imitation (Redcay and Warnell, 2018). Particularly, ACC and insula have an essential role in perceptual representations of emotional experience and social information, underpinning the shared socio-emotional processing (Adolphs, 1999; Dolan, 2002; Salmi et al., 2013). The changes in the inter-individual concordance in these regions link to the gradual maturity of mental ability (Paulus et al., 2015; Schurz et al., 2021; Winters et al., 2021) that support observing, processing and applying social information in the context of interpersonal interaction. The older adolescents’ higher concordance of brain response in the visual network reflects that visual processing is still developing in childhood and adolescence when the individual is involved in socio-emotional processing (Conner et al., 2004). The maturation of socio-emotional functioning is based on stable information integration in the high-level visual cortex (Nelson et al., 2016). In addition, the performance of visual development impacts the socio-emotional outcomes (Tonks et al., 2009).

In previous studies, the neural processes of emotion regulation and ToM were researched separately, with limited exploration of their potential shared neural patterns. Our study focuses on exploring the shared neural patterns that may underlie these skills. The generalized socio-emotional brain network associating with both CR and ToM may be linked to various functions: (i) Inhibitory control, through which individuals achieve goal-directed behavior by suppressing impulses and resisting interference (i.e. successful CR and ToM). This process is associated with the SPG and MTG (Leslie et al., 2004; Ahmed et al., 2015; Gil et al., 2022); (ii) Mental attribution, which allows individuals to rethink or reason about each other’s emotional states. This process is related to activation of the ACC and superior temporal sulcus (STS, ROI 14 in this study) (Abu-Akel, 2003; McRae et al., 2012); (iii) Attentional deployment, through which individuals focus their cognitive resources on the task at hand. This process involves the extensive occipital cortex (Ma et al., 2017; Morris et al., 2014; Sasaki et al., 2001); and (iv) Semantic representation, which allows individuals to retrieve semantic information relevant to the task. This process recruits the STG and MTG (Ochsner et al., 2012). In addition, the insula is involved in both CR and ToM, but may perform different functions. In the CR task, the insula plays a crucial role in the interpretation and regulation of emotions (Zhang et al., 2020b); in the ToM task, the insula, together with the STS, SPG and PreCG, provide social information about gestures and body movements, which is their important function as the mirror neuron system (Lawrence et al., 2006; Emmorey et al., 2011; Redcay and Warnell, 2018; Di Cesare et al., 2021).

The predictive power achieved by the naturalistic normative model sheds light on an objective evaluation of socio-emotional functions in children and adolescents and the screening of altered brain activity in mental disorders. Insufficient socio-emotional functioning increases the severity of emotional dysfunction and the risk of mental illness (Whitney et al., 2013). The impaired performance on ToM indicates a high risk of autism spectrum disorder, with apparent difficulty in mentalization in childhood (Hoogenhout and Malcolm-Smith, 2017). Relative to self-report questionnaires, watching movies and constructing corresponding response typicality may be a more effective and objective measurement. Previous studies have supported the possibility of inferring cognitive development using child-to-adult brain concordance (or synchronization) (Cantlon and Li, 2013; Moraczewski et al., 2018, 2020; Richardson et al., 2018; Richardson, 2019).

There are a few limitations in this study. First, the cross-sectional design limited the ability to characterize the trajectories of the maturation process. A longitudinal method should be employed in future work to explore the developmental trajectories of socio-emotional processing. Second, our behavior analysis focused on two common emotion regulation strategies: cognitive reappraisal and expressive suppression. Emotion regulation models include situational selection, modification, attentional deployment, cognitive change and response modulation (Gross, 2015). Other essential emotion regulation strategies for different developmental stages include acceptance, behavior avoidance, distraction, experiential avoidance, mindfulness, problem-solving and rumination (Gross, 1998; Naragon-gainey et al., 2017). Future research could consider the diversity of adolescents’ emotion regulation strategies and explore a more comprehensive indicator for socio-emotional development. Third, previous studies have shown that individual neurological differences could distinguish individuals with mental illness from healthy control (Yang et al., 2020; Zhang et al., 2020a; Liu et al., 2021). Further studies should examine whether the response typicality could help reflect mental problems (Mikulincer and Shaver, 2019; Gruskin et al., 2020). Fourth, the limited sample in cognition analyses may reduce statistical power, so a larger sample should be used to further validate the contribution of the predictive ability of brain responses as distinct from cognitive ability. Fifth, our study did not examine low-level video features, including brightness, vibrance and motion, which may have the potential to affect intersubject synchronization (Kim et al., 2016). Future research should consider incorporating the analysis of these physical properties to gain a more comprehensive understanding of the underlying mechanisms involved in the synchronization of responses to audiovisual stimuli (McNamara et al., 2017). Sixth, extending the age range of participants in future studies would be beneficial. Including data from adult participants would indeed enhance the reliability and generalizability of the reference group.

Conclusions

We identified distributed brain regions that exhibit age-dependent inter-individual differences in response to socio-emotional scenarios in adolescents. The response in these regions becomes more concordant in older adolescence than in early adolescence, indicating that a common emotion-related brain response pattern is consolidated as adolescents mature. With a combination of naturalistic imaging and a normative model, we revealed a general socio-emotional network predictive of multiple perspectives of socio-emotional capabilities and exhibits high plasticity during childhood and adolescence. Furthermore, the response typicality of this network reflects a child or adolescent’s socio-emotional capability.

Supplementary data

Supplementary data is available at SCAN online.

Funding

This work was supported by STI 2030 (Major Projects-2022ZD0209101); the National Natural Science Foundation of China (81971682, 81571756); Natural Science Foundation of Shanghai (20ZR1472800); Medical guidance project of Shanghai Science and Technology Commission (20Y11906700); Clinical Research Project of Shanghai Mental Health Center (CRC2018DSJ01-5, CRC2019ZD04); Shanghai Municipal Commission of Education (Gaofeng Clinical Medicine-20171929); Shanghai Science and Technology Commission (18JC1420305); Shanghai Municipal Health Commission (2019ZB0201, 2018BR17); Shanghai Clinical Research Center for Mental Health (19MC1911100); and Research Funds from Shanghai Mental Health Center (13dz2260500, 2018-YJ-02, 2018-YJ-05).

Conflicts of interest

The authors declared that they had no conflict of interest with respect to their authorship or the publication of this article.

Acknowledgements

The authors are grateful to the clinicians who referred patients to the study, and to Shuaichi Wu, MS, of the Fudan University, Zhen, Liu, MS and Hanshu, Yang, MS, of Shanghai Jiao Tong University for their help with participant recruitment and testing. The children and parents who participated in the project are greatly appreciated.

References

Author notes