Abstract
Despite advances in understanding the psychological and social consequences of peer victimization, the immediate effects of bullying on the central nervous system remain elusive. Here we mapped the neural, affective, and attentional responses to simulated bullying in adolescents and adults and tested whether these responses are associated with real-life victimization experiences. Fifty-one adolescents (29 females, 22 males) aged 11–14 years and 47 adults (29 females, 18 males) underwent a functional MRI (fMRI) while watching first-person videos of bullying (victimization) in the school environment, as well as neutral and positive social interactions in a similar setting. Additionally, 57 adults (36 females, 21 males) watched the same videos during an eye tracking experiment. Exposure to bullying versus positive social interaction engaged the socioemotional and threat response systems, as well as regions related to social cognition, sensory and interoceptive processing, and motor control. These responses were consistent across adolescents and adults and were associated with the current and past victimization experiences of the participants. This large-scale activation of neural systems subserving socioemotional, somatosensory, and interoceptive processing was in line with the amplified emotional and attentional responses revealed by larger pupil size and higher fixation frequency during simulated bullying in the eye tracking experiment. Altogether these results highlight how peer victimization evokes a state of stress and alarm in the central nervous system.
Significance Statement
Victimization by bullying is associated with serious mental, somatic, and social problems, but little is known about how the brain reacts to bullying. Here, we used functional magnetic resonance imaging to investigate the brain responses to natural, simulated bullying and positive social interaction. We also compared these responses between adolescents and adults. Exposure to bullying activated the socioemotional distress system, as well as networks processing social and sensory information, bodily sensations, and motor actions. These responses were consistent across adolescents and adults. Our findings reveal how bullying induces a state of stress or alarm in the central nervous system, highlighting the adverse and threatening nature of bullying.
Introduction
School bullying, defined as repeated aggression against a peer in a more vulnerable position, is prevalent worldwide, with almost one-third of students reporting being bullied by their peers at least once in the past month (UNESCO, 2019). Victimization by bullying is a precursor of psychiatric problems, including anxiety and depression (Christina et al., 2021), somatic symptoms such as headache and abdominal pain (Gini and Pozzoli, 2013), as well as suicidal thoughts (Van Geel et al., 2022). On the neuronal level, childhood psychological stress can alter brain networks related to emotion processing and executive functioning (Palamarchuk and Vaillancourt, 2022). However, it remains unresolved (1) how these networks acutely respond to peer victimization during adolescence and (2) whether adolescents’ response profiles differ from those of adults.
Bullying can take numerous forms, ranging from social exclusion to physical aggression and persistent verbal harassment. Such phenomena are difficult to investigate in the laboratory, and consequently, neuroimaging studies on life-like peer victimization are scarce. Previous studies have utilized primarily simplified experimental paradigms focusing on mapping neural responses to social exclusion and rejection. Coordinate-based meta-analyses of the widely used Cyberball game (Mwilambwe-Tshilobo and Spreng, 2021) and other social exclusion and rejection paradigms (Vijayakumar et al., 2017) have found consistent activation of the frontocortical emotion regulation system (lateral prefrontal cortex), social cognition and self-evaluative system (ventral anterior cingulate cortex, posterior cingulate cortex, medial prefrontal cortex), and the affective system (orbitofrontal cortex) in response to simulated social exclusion, suggesting that social exclusion elicits a distress response in the brain. However, simplistic experimental paradigms of social rejection and exclusion do not capture the full complexity of the social interactions involved in peer victimization. To understand the neural responses to bullying, it is necessary to go beyond tightly controlled yet simplistic and artificial paradigms and map the brain responses in naturalistic settings that resemble the complex dynamics of adolescents’ everyday social life (Adolphs et al., 2016).
Cross-sectional studies suggest that chronic peer victimization and rejection could lead to hypersensitivity to future peer rejection via stronger activation of the threat response system. Specifically, real-life victimization and peer rejection have consistently been associated with increased amygdala and fusiform gyrus activation in response to social exclusion and negative interpersonal feedback (Lee et al., 2014; Rudolph et al., 2016; McIver et al., 2018). Neural responses to peer victimization could also differ between adolescents and adults, as the priority of perceived status among peers peaks in early adolescence (LaFontana and Cillessen, 2010). However, meta-analytic evidence for such age-dependent sensitivity is weak, at least for the simulated social exclusion tasks, where surprisingly, only ventral striatal responses to exclusion were stronger in younger versus older subjects (Vijayakumar et al., 2017). Consequently, the developmental time course of the distress response to bullying remains poorly characterized.
The current study
Here, we mapped the functional neural and attentional responses to simulated bullying. We measured hemodynamic brain activity while participants viewed first-person video clips representing realistic peer interactions in a school environment and modeled the neural responses to the moment-to-moment experience of bullying and positive social interaction simulated by the videos. We compared the brain responses of adolescents (11–14 years old) with those of adults to understand how adolescents and adults process the experience of victimization. Additionally, we tested how real-life victimization experiences are related to brain responses to simulated bullying, controlling for internalizing symptoms for both age groups and workplace victimization experiences for adults. Finally, we studied attentional patterns during bullying simulation in adults using eye tracking. We found that exposure to (simulated) bullying engaged the socioemotional and sensory processing brain regions consistently in adolescents and adults, and the neural responses were associated with the real-life victimization experiences of the participants. These results were in line with the amplified emotional and attentional responses in adults revealed by larger pupil size and higher fixation frequency in the eye tracking experiment during bullying simulation.
Materials and Methods
Participants
The participants were 53 Finnish-speaking adolescents and 49 adults with normal or corrected to normal vision, with no current self-reported psychiatric conditions or medication affecting the central nervous system. Two participants from each group had to be excluded because of technical problems resulting in incomplete fMRI data. Hence, data from 51 adolescents (29 females, 22 males; age range, 11–14 years; mean age, 12.20 ± 1.02) and 47 adults (29 females, 18 males; age range, 19–39 years; mean age, 24.02 ± 4.38) were included in the sample. Sexes were matched between adolescent and adult samples (χ2(1) = 0.08, p = 0.78), and ages were matched between the sexes in the adolescent (d = 0.02, Welch's T(46.276) = 0.09, p = 0.93) and adult samples (d = 0.09, Welch's T(41.46) = 0.31, p = 0.76). Adolescents were recruited through their parents using social media, flyers, and University and University Hospital social media and mailing lists. Adult participants were students and personnel from the University of Turku. Adolescent participants received four movie tickets and a 3D print of their brain, and adults received 50 euros as compensation. All participants and the guardians of under-aged participants signed a written informed consent. The ethics board of the Hospital District of Southwest Finland approved the protocol, and the study was conducted in accordance with the Declaration of Helsinki.
Self-report measures
Participants completed the following self-report questionnaires before the scan: Sum of the multidimensional Peer-Victimization scale (Mynard and Joseph, 2000) was used to measure peer victimization in adolescents. The scale includes 16 self-reported items and 4 subscales (physical victimization, verbal victimization, social manipulation, and attacks on property) and has acceptable to excellent reliability (Cronbach's α = 0.74–0.96; Joseph and Stockton, 2018). Participants were asked how often another pupil had done different adverse things to them over the last school year, and responses were scored on a three-point Likert-scale (0 = “Not at all”, 1 = “Once”, 2 = “More than once”). For adults, both retrospective victimization and current workplace victimization were measured. Participants were asked to report the duration of their victimization at school and outside of school before adulthood, scored from 0 (“Never”) to 4 (“Throughout my school years”), and the higher score out of these two measures was used as the retrospective victimization score. For current workplace victimization, adults reported how often they had been the target of 20 different offending acts at their workplace or at their current community (0 = “Never”, 4 = “Once a week or more often”). Adolescent internalizing symptoms were measured as the sum score of the shortened 25-item version of the Revised Child Anxiety and Depression Scale (RCADS-25; Ebesutani et al., 2012). Both subscales have acceptable reliability in school-based samples (Total Anxiety α = 0.86, Total Depression α = 0.79; Ebesutani et al., 2012). For adults, the sum of depression and anxiety subscales from the Depression, Anxiety and Stress Scale-21 Items (DASS-21; Lovibond and Lovibond, 1995) was used as a measure of internalizing symptoms. DASS-21 subscales have acceptable reliability (α ≥ 0.74) under the bifactor structure (Lee et al., 2019).
Experimental design for fMRI
We used naturalistic first-person video stimuli depicting various degrees of bullying, neutral, and positive social interaction targeted at the viewer. The videos were filmed in a school with child actors. Such engaging social and emotional content presented in a naturalistic fashion makes them ideal for modeling life-like victimization experiences in the laboratory (Adolphs et al., 2016; Santavirta et al., 2024). Bullying behavior consisted of physical, verbal, and relational victimization, such as being called names by peers or being left out of a peer group. We note that the textbook definition of bullying includes repetitive behavior and imbalance in power dynamics, yet the current stimuli provide only singular “snapshot” of the canonical bullying experience. We nevertheless refer to them as bullying because they are used for modeling the real-life bullying scenarios within the limits of the imaging laboratory context. Moreover, majority of adolescents define bullying as negative behavior without including criteria of repetitive nature and power imbalance (Vaillancourt et al., 2008); thus, the stimuli adhere well with the subject population's typical definition of bullying. Each video lasted from 20 to 87 s (seven bullying videos and five positive social interaction videos), and each of the positive social interaction videos was matched with a bullying video filmed in the same location and with same actors, and only the content of the interaction was different. A confirmatory analysis requested by the reviewers confirmed that there were no differences in low-level features between the video categories (see Text S1 for more details).
The intensity of bullying and positive social interaction in the videos was rated by a separate pool of 271 Finnish-speaking adults in an online experiment. Each participant rated the intensity of either bullying (offensive behavior) or positive social interaction (kind behavior) for a subset of four videos using a dynamic response slider for the whole duration of the videos, ranging from “None” to “Very high.” After removing bad quality data (technical problems, etc.), data from 235 participants was used. This yielded ratings of 30–39 participants for each video at each time point. Ratings were recorded on a 10 Hz frequency. To match the ratings with the fMRI data, mean values for bullying and positive social interaction were downsampled to the temporal resolution of the EPI data (3s). These ratings were used as regressors in the analysis of the fMRI data to model the neural responses to bullying. Additionally, to validate that the bullying videos evoked more negative emotions than the positive social interaction videos, participants were asked to rate the emotional content of the videos (see Text S2 and Fig. S1 for results).
In the fMRI experiment, participants were instructed to watch the first-person videos as if they were the person experiencing the events but to refrain from reacting by moving or talking. Altogether, the video stimulus lasted for 9 min, and videos were presented in a fixed pseudorandomized order without breaks to enable brain synchronization analyses and to avoid repetition of video categories and settings. The video presentation was controlled with Presentation software (version 23.0, Neurobehavioral Systems; www.neurobs.com). Visual stimuli were shown from a screen viewed by the participant via a mirror fixed to the head coil. Sensimetrics S14 insert earphones (Sensimetrics) were used to deliver the video sounds, and sound level was adjusted individually for each participant. To verify that the fMRI participants perceived bullying and positive social interaction in the videos as intended, they were asked to rate the bullying and positive social interaction content of the videos after the fMRI scan. They were shown a representative 7–14 s clip of each video, after which they were asked to report the perceived amount of bullying (offensive behavior) and positive social interaction (kind behavior) with a slider ranging from “Not at all” to “Very much.”
Neuroimaging data acquisition and preprocessing
MR imaging was conducted at Turku PET Centre. Data were acquired using GE Signa 3T PET/MR scanner. A specially designed silent sequence was used for the T2 MRI to get the participants used to the scanner noise before the T1 MRI and fMRI, and participants could choose to watch either animal videos or an animated film during the structural MRIs to further increase comfort. High-resolution structural images were obtained with a T1-weighted (T1w) BRAVO sequence (1 mm3 resolution; TR, 7.9 ms; TE, 3.4 ms; flip angle, 10°; 228 mm FOV; 256 × 256 reconstruction matrix). A total of 192 functional volumes (9 min 51 s) were acquired for the experiment with a T2∗-weighted echoplanar imaging sequence sensitive to the blood oxygen level-dependent (BOLD) signal contrast (TR, 3000 ms; TE, 30 ms; 90° flip angle; 256 mm FOV; 128 × 128 reconstruction matrix; 250 kHz bandwidth; 2.7 mm slice thickness; 51 axial slices acquired sequentially in an ascending order). Caregivers joined the debriefing of the experiment for most of the adolescent participants, and for two adolescents the caregiver joined them in the scanner room for the duration of the structural MRIs to ensure comfort. Structural brain abnormalities that are clinically relevant or could bias the analyses were checked by a consultant neuroradiologist and no subjects had to be excluded from the sample.
The anatomical and functional imaging data were preprocessed with fMRIPrep (v21.0.0rc2; Esteban et al., 2019; Markiewicz et al., 2024), which is based on Nipype 1.6.1 (Esteban et al., 2019; Markiewicz et al., 2024). T1w reference image volumes were corrected for intensity nonuniformity using N4BiasFieldCorrection (ANTs 2.3.3; Tustison et al., 2010) and skull-stripped using antsBrainExtraction.sh workflow and OASIS30ANT as template. Brain tissue segmentation of cerebrospinal fluid, white matter, and gray matter was performed on the brain-extracted T1w image using FAST (Zhang et al., 2001; FSL v5.0.11). Brain surfaces were reconstructed using recon-all (FreeSurfer 6.0.1; Dale et al., 1999) and the brain mask estimated previously was refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of the cortical gray matter of Mindboggle (Klein et al., 2017). Spatial normalization to the ICBM 152 Nonlinear Asymmetrical template version 2009c (MNI152NLin2009cAsym; Fonov et al., 2009) and FSL's MNI ICBM 152 nonlinear 6th Generation Asymmetric Average Brain Stereotaxic Registration Model (MNI152NLin6Asym; Evans et al., 2012) was performed through nonlinear registration with the antsRegistration (ANTs v2.3.3; Avants et al., 2008), using brain-extracted versions of both T1w volume and template. Normalization to the adult template was done for both adults and children to allow direct comparisons between the samples.
Functional data were preprocessed as follows: First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. BOLD runs were slice time-corrected using 3dTshift from AFNI (Cox, 1996) and motion-corrected using MCFLIRT (Jenkinson et al., 2002; FSL v5.0.11). The preprocessed BOLD was then coregistered to the T1w image using bbregister (FreeSurfer v6.0.1) for boundary-based registration (Greve and Fischl, 2009) with six degrees of freedom. All volumetric transformations were applied in a single step using antsApplyTransforms (ANTs) and Lanczos interpolation. Independent component analysis-based Automatic Removal Of Motion Artifacts (ICA-AROMA) was performed on the preprocessed BOLD time series to denoise the data nonaggressively after removal of non-steady-state volumes and spatial smoothing with 6-mm Gaussian kernel (Pruim et al., 2015). The first two and last nine functional volumes were discarded to exclude the time points before and after the stimulus.
Whole-brain GLM data analysis
The fMRI data were analyzed using SPM12 (Wellcome Trust Center for Imaging; http://www.fil.ion.ucl.ac.uk/spm). A general linear model (GLM) was fitted to the data to identify the brain regions activated, in a parametric fashion, by bullying and positive social interaction of the videos. In the first level analysis, standardized dynamic mean ratings for bullying and positive social interaction [resampled to one repetition time (TR) and convolved with the canonical Haemodynamic Response Function (HRF)] were entered as regressors into a single design matrix. The signal threshold was set to 10% of the global signal, and the MNI152NLin6Asym mask was used to exclude signals outside the brain. For each participant, voxel-wise contrast images were generated for the main effects of bullying and positive social interaction, as well as for the bullying-minus-positive social interaction contrast.
These contrast images were subjected to group-level analysis with one-sample t test to identify the brain regions where the relationship between the intensity of bullying or positive social interaction and the hemodynamic activity was consistent across participants. A similar group-level analysis was conducted for the bullying-minus-positive social interaction contrast to identify the regions where activation was different between bullying and positive social interaction. The analyses were run separately for adolescents and adults. To investigate the possible differences in brain responses between adolescents and adults, a two-sample t test between the groups was conducted. Clusters surviving FDR correction (Benjamini and Hochberg, 1995; q < 0.05) after voxel-level thresholding with a p value of 0.001 are reported for the main analyses.
Region of interest (ROI) data analysis
For the ROI analysis, anatomical ROIs were extracted using AAL3v1 atlas (Rolls et al., 2020). A subset of ROIs involved in socioemotional processing (Saarimäki et al., 2016; Santavirta et al., 2023), including amygdala, anterior cingulate cortex (ACC), caudate, dorsomedial prefrontal cortex (dmPFC), insula, mid-cingulate cortex (MCC), posterior cingulate cortex (PCC), putamen, thalamus, ventrolateral prefrontal cortex (vlPFC), and ventromedial prefrontal cortex (vmPFC) were chosen due to these regions' role in processing of emotions and social exclusion (see details in Table S1). Bilateral ROIs were used, but unilateral models were also run for the main analyses, and the results are reported in Text S3 and Figure S2.
Subject-specific mean β values for each ROI were obtained from the first-level whole-brain contrast images. On the group level, one-sample t test was used to derive 95% confidence intervals for the beta values within each ROI and predictor for adolescents and adults separately. FDR correction for p values was applied to correct for testing of multiple ROIs within age groups and predictors. In addition, paired and FDR corrected t tests were used to compare the mean β values between bullying and positive social interaction within each ROI. Effect sizes are reported as Cohen's d. Similar analyses but with unpaired t tests were conducted to compare regional differences between adolescents and adults for each predictor.
Associations of real-life peer victimization and brain responses
The association between real-life peer victimization and brain responses to bullying and positive social interaction was studied using explorative whole-brain GLM analyses. Victimization score was added as a predictor in the GLM in the second-level analyses separately for adolescents and adults. Current self-reported peer victimization at school was used for adolescents and retrospective measure of victimization duration during school years for adults, as the latter allows for studying long-term neural correlates of childhood victimization. Given that internalizing symptoms (Gunther Moor et al., 2010; Masten et al., 2011; Silk et al., 2014; Rudolph et al., 2016) may affect neural processing of simulated victimization experiences, internalizing symptoms were included as covariates for both groups. Additionally, to separate the long-term effects of victimization from the effects of current victimization experiences, workplace victimization was included as a covariate for adults. A complementary analysis was also conducted without any covariates to avoid possible removal of variance of key interest. Before deciding on the final model, effects of age and sex (within adolescent and adult groups) were tested in a separate analysis. These factors were not found to affect the brain responses to bullying or positive social interaction, apart from minor differences between adult males and females for the bullying-minus-positive social interaction contrast (Fig. S3). Hence, age and sex were not included in the final models. Clusters surviving FDR correction (Benjamini and Hochberg, 1995; q < 0.05) after voxel-level thresholding with p values of 0.001 and 0.05 are reported. A supplementary analysis was also conducted at the ROI level using linear regression (FDR corrected within a predictor) in the preselected ROIs. In the adult sample, two participants had missing values in the DASS-21 questionnaire, and hence only 45 adult participants were used in the GLM and ROI analysis for studying the effects of retrospective victimization.
Intersubject correlation analysis
To examine temporal dynamics of synchronization of brain activation across individuals during bullying and positive social interaction, intersubject correlation (ISC) analysis was performed. See Text S4 and Figure S4 for more details and results.
Eye tracking experiment
Attentional patterns during bullying exposure were studied in a separate eye tracking experiment. The videos used in fMRI were shown in a randomized order to 58 adult participants, while their gaze patterns were recorded using an eye tracking camera. Data from one participant was lost because of technical problems, leaving 57 participants (36 females, 21 males; age range, 18–44 years; mean age, 26.88 ± 7.27). Exclusion criteria were the same as for the fMRI participants. The eye tracker was calibrated and validated using a 5-point calibration, and validation was repeated with one point before each video. Validation was successful if gaze position error was below 1°. The stimuli were presented with a 27″ Retina 5K monitor at 68 cm distance from the eyes. The eye tracking data was collected with SR EyeLink 1000 Plus eye tracker (SR Research; v5.15 Jan 24, 2018; eyes, right; file filtering level, extra; pupil tracking algorithm, centroid).
Fixations shorter than 80 ms were considered unreliable and were therefore excluded from the data. Mean number of fixations per 10 s, number of blinks per 10 s, gaze duration (in seconds) per 10 s, and standardized mean pupil size (centered to participant's mean over all videos) were obtained for bullying and positive social interaction videos for each participant, and bullying videos were compared with the positive social interaction videos using nonparametric Wilcoxon paired signed-rank test.
Results
Subjective ratings of bullying and positive social interaction of the videos
Results from the online experiment (adult sample) revealed that the videos contained behaviors perceived as intense bullying (offensive behavior) and positive social interaction (kind behavior), which varied over time (Fig. 1B). Presence of bullying and positive social interaction were negatively correlated [Pearson r = −0.60, p < 0.001, variance inflation factor (VIF) = 1.56], and these behaviors had unique time series (Fig. 1B).
Experimental design. A, Twelve first-person videos (total duration nine minutes) were used to simulate experiences of bullying during fMRI. B, Mean dynamic intensity ratings with 95% confidence intervals for bullying (offensive behavior) and positive social interaction (kind behaviour) for each video (nraters = 235). These ratings were subsequently used as regressors in the fMRI experiment.
Self-reports
Self-report measures and differences between males and females are reported in Table 1. There were no statistically significant differences in the self-report measures between males and females in adolescents or adults. No correlation between age and real-life victimization was observed (rAdolescents = 0.00, pAdolescents = 0.98; rAdultsRetro = 0.03, pAdultsRetro = 0.86; rAdultsCurrent = −0.03, pAdultsCurrent = 0.84). Internalizing symptoms had very weak to moderate correlations with real-life victimization (rAdolescents = 0.35, pAdolescents = 0.01; rAdultsRetro = 0.03, pAdultsRetro = 0.86; rAdultsCurrent = 0.34, pAdultsCurrent = 0.02), and adult retrospective and current victimization were weakly correlated (r = 0.25, p = 0.09).
Demographics and self-report scores
Behavioral results for the fMRI experiment
Overall, the ratings of the fMRI participants for the videos were concordant across adolescents and adults, as indicated by Pearson’s correlations between adolescent and adult ratings (rbullying = 1.00, pbullying < 0.0001; rpositive = 0.99, ppositive < 0.0001). However, adults rated the amount of bullying (offensive behavior) slightly higher (Mdn = 96.43) than adolescents (Mdn = 91.43) in the videos categorized as bullying videos (Mann–Whitney U test U = 831, effect size r = 0.26, FDR corrected q = 0.04). No other differences in ratings between groups were found. Ratings are reported in Figure 2.
Bullying (offensive behavior) and positive social interaction (kind behavior) of the stimulus videos as evaluated by the fMRI participants. The horizontal line indicates the median, the lower and upper ends of the boxes indicate the lower and upper quartiles, and error bars indicate the 1.5 interquartile range. Nonparametric Mann–Whitney U test was used for comparing ratings between adolescents and adults. r = effect size, *FDR corrected q < 0.05.
Hemodynamic responses to bullying and positive social interaction
Whole-brain analysis
Whole-brain result maps are available on NeuroVault (https://neurovault.org/collections/BJNYACTH/). The whole-brain GLM revealed that exposure to bullying engaged large-scale limbic and paralimbic networks in adolescents and adults (Fig. 3). Overall, the responses were stronger and more widespread for bullying versus positive social interaction. Subcortically, bullying led to increased activity in the amygdala, caudate, hippocampus, insula, and putamen in both age groups. Cortically, the effect of bullying was particularly consistent in areas processing visual and auditory information such as lateral parts of the occipital cortex and superior temporal cortex, as well as in the fusiform gyrus. Dorsomedial and right ventrolateral parts of prefrontal cortex (dmPFC, vlPFC) were activated for both groups in response to bullying, and for adolescents, the activation extended to parts of the anterior and mid-cingulate cortex (ACC, MCC) and dorsal orbitofrontal cortex (dOFC). In contrast, a decrease in mid- and posterior cingulate cortex (MCC, PCC) activation was observed in adults in response to bullying. In adolescents, viewing bullying was also linked to activation in the precuneus and somatosensory and motor areas: primary somatosensory cortex (S1), supramarginal gyrus (SMG), supplementary motor area (SMA), and parts of primary motor cortex (M1). Similar but spatially more restricted patterns were revealed for adults, apart from S1, where activation in response to bullying was not observed.
Hemodynamic responses to bullying and positive social interaction for (A) adolescents and (B) adults. Color code indicates the t statistic range for activations (hot colors) and deactivations (cool colors). The data are thresholded at p < 0.001 at the voxel level and FDR corrected at q < 0.05 at the cluster level. L, left hemisphere; R, right hemisphere; ACC, anterior cingulate cortex; Amy, amygdala; Cau, caudate; dlPFC, dorsolateral prefrontal cortex; dmPFC, dorsomedial prefrontal cortex; FS, fusiform gyrus; Hip, hippocampus; Ins, insula; M1, primary motor cortex; MCC, mid-cingulate cortex; OC, occipital cortex; OFC, orbitofrontal cortex; PCC, posterior cingulate cortex; PH, parahippocampal gyrus; Prec, precuneus; Put, putamen; S1, primary somatosensory cortex; SMA, supplementary motor area; SMG, supramarginal gyrus; STG, superior temporal gyrus; Tha, thalamus; vlPFC, ventrolateral prefrontal cortex.
For positive social interactions, both activation and deactivation were observed for both age groups. Activation increased in the precuneus, superior temporal gyrus, lateral occipital cortex, and restricted parts of the dmPFC and vlPFC. In contrast, deactivation was observed in the superior and medial occipital cortex, fusiform gyrus, parahippocampus, and primary motor cortex in both age groups, as well as in the left dorsolateral PFC (dlPFC) and ACC/MCC in adults.
When responses to bullying and positive social interaction were contrasted directly with each other, responses to bullying were stronger in broadly similar regions than for bullying alone, but the activation clusters were more concise. Subcortical activation was increased for bullying versus positive social interaction in the hippocampus, insula, and putamen in both groups. Amygdala activation remained significantly increased for bullying only in the adolescent group. Cortical activation was increased in the dorsal ACC and MCC, medial and superior occipital cortex, fusiform gyrus, superior temporal gyrus, SMG, SMA, M1, and parts of the lateral PFC in both age groups, as well as S1 in adolescents.
ROI analysis
The overall pattern of results in the whole-brain analysis was mostly replicated in the ROI analysis, where amygdala, insula, putamen, caudate, and vlPFC showed statistically significant responses to bullying in both age groups (FDR corrected q < 0.001; Fig. 4). Additional responses were observed in the thalamus, ACC, MCC, and dmPFC in adolescents but not in adults. No significant responses to positive social interaction were observed in adolescents, whereas in adults, deactivation was observed in response to positive social interaction in the ACC and MCC.
Regional effects (mean beta weights and 95% confidence intervals) for bullying and positive social interaction for adolescents and adults. Color of the dot indicates the FDR corrected p value for one-sample t test against the null hypothesis that the mean of beta values in a specific ROI equals to zero (black, q < 0.001; white, q > 0.001; FDR corrected for multiple ROIs within age group and predictor). Asterisks denote significance levels for a paired t test between bullying and positive social interaction at each ROI after FDR correction for multiple ROIs within age group, and d value indicates effect size (Cohen's d) for the same test. ACC, anterior cingulate cortex; dmPFC, dorsomedial prefrontal cortex; MCC, mid-cingulate cortex; PCC, posterior cingulate cortex; vlPFC, ventrolateral prefrontal cortex; vmPFC, ventromedial prefrontal cortex. *FDR corrected q < 0.05, **q < 0.01, ***q < 0.001, ****q < 0.0001.
For bullying versus positive social interaction, cortical patterns were different for ROI and whole-brain analysis. The ROI analysis revealed increased activation in the vlPFC for bullying versus positive social interaction for both age groups, and an increase in dmPFC activation for adults, while in the whole-brain analysis widespread vlPFC activation was only seen in adolescents, and no dmPFC activation was observed. Additionally, ROI analysis indicated increased activation in the amygdala for adults for bullying versus positive social interaction, as well as increased activation in the caudate and thalamus for both groups, whereas in the whole-brain analysis amygdala and thalamus activation was increased only in adolescents, and caudate activation was spatially very restricted in both groups. For putamen, insula, and the cingulate cortex, the results were in line with the whole-brain analysis; activation in response to bullying was higher than in response to positive social interaction in the forementioned subcortical regions and ACC and MCC for both age groups.
Comparison between adolescents and adults
Most responses to viewing bullying and positive social interaction were consistent across adolescents and adults, with spatial Pearson’s correlations between adolescent and adult result maps exceeding r = 0.72 (rBullying = 0.72, pBullying = 0.00; rPositive = 0.80, pPositive = 0.00; rBullyingVsPositive = 0.74, pBullyingVsPositive = 0.00). Viewing bullying elicited stronger brain responses in adolescents in comparison with adults in restricted parts of ACC/dorsal PFC, MCC, fusiform gyrus, M1, S1, superior parietal cortex, and SMG, as well as areas processing visual information (cuneus, calcarine sulcus, and lingual gyrus; Fig. 5, p < 0.001 at voxel level, FDR corrected at q < 0.05 at cluster level). For positive social interaction, adolescents’ responses were stronger than adults’ in the restricted parts of dlPFC and PCC. The only difference between the age groups for bullying versus positive social interaction was observed in the crossing of M1/S1, in which the difference between bullying and positive social interaction was larger in adolescents in comparison with adults.
Differences in hemodynamic responses to bullying, positive social interaction, and bullying versus positive social interaction between adolescents and adults. The activation maps show t values from two-sample t test, thresholded at p < 0.001 at voxel level, and FDR corrected at q < 0.05 at cluster level. Positive values indicate stronger responses for adolescents in comparison with adults. L, left hemisphere; R, right hemisphere; ACC, anterior cingulate cortex; Cal, calcarine sulcus; Cun, cuneus; dlPFC, dorsolateral prefrontal cortex; dmPFC, dorsomedial prefrontal cortex; Fs, fusiform gyrus; Hip, hippocampus; LG, lingual gyrus; M1, primary motor cortex; MCC, mid-cingulate cortex; PCC, posterior cingulate cortex; S1, primary somatosensory cortex; SMG, supramarginal gyrus; SP, superior parietal cortex.
ROI-level analysis revealed stronger responses to bullying for adolescents in comparison with adults in the insula, putamen, thalamus, ACC, MCC, and PCC (FDR corrected q < 0.05; Table S2). In response to positive social interaction, a significant deactivation was observed in ACC in adults in comparison with adolescents. Higher activation in response to bullying versus positive social interaction was observed in adolescents in comparison with adults in the thalamus.
Associations of real-life peer victimization and brain responses
Using the standard threshold of p < 0.001 at voxel level, FDR corrected at q < 0.05 at cluster level, no statistically significant associations between real-life victimization experiences and brain responses to bullying were observed in adolescents. However, using a more lenient threshold, p < 0.05 at voxel level, FDR corrected at q < 0.05 at cluster level, both positive and negative associations between hemodynamic responses to bullying and self-reported peer victimization at school were observed (Fig. 6A, Fig. S5A). A positive association between peer victimization and brain responses to bullying was observed in the bilateral ACC, anterior MCC, vmPFC, OFC, and SMA, right dmPFC, anterior insula, and lateral PFC, as well as left ventral striatum. In contrast, victimization experiences were related to decreased responses to bullying in parts of the bilateral posterior MCC, left dorsal striatum, insula, and STG, in addition to right precuneus and angular gyrus. In the positive social interaction contrast, activation in parts of bilateral superior parietal cortex, paracentral lobule, and precuneus had a negative association with peer victimization (Fig. S5A, p < 0.05 at voxel level, FDR corrected at q < 0.05 at cluster level). Victimization was associated with stronger responses to bullying in contrast to positive social interaction in the right anterior insula and vlPFC, as well as bilateral SMA and anterior MCC (Fig. S5A; p < 0.05 at voxel level, FDR corrected at q < 0.05 at cluster level).
Associations between (A) peer victimization in adolescents and (B) retrospective peer victimization in adults and brain responses to bullying. The activation maps show t values for one-sample t test for the victimization scores for adolescents and adults separately, thresholded at p < 0.05 at voxel level, and FDR corrected at q < 0.05 at cluster level. Hot colors indicate positive relationship between victimization and hemodynamic responses to bullying, and cool colors indicate negative associations. Model covariates included internalizing symptoms for both groups and workplace victimization for adults. L, left hemisphere; R, right hemisphere; ACC, anterior cingulate cortex; Cau, caudate; dlPFC, dorsolateral prefrontal cortex; dmPFC, dorsomedial prefrontal cortex; Ins, insula; MCC, mid-cingulate cortex; OFC, orbitofrontal cortex; Prec, precuneus; Put, putamen; SMA, supplementary motor area; Tha, thalamus; vlPFC, ventrolateral prefrontal cortex.
In the adult sample, retrospective victimization was negatively associated with the hemodynamic responses to bullying in the left dlPFC (p < 0.001 at voxel level, FDR corrected at q < 0.05 at cluster level). With a more lenient threshold (p < 0.05 at voxel level, q < 0.05 at cluster level), negative associations were observed also in anterior parts of the left insula, dorsal striatum, thalamus, M1, and S1, and bilateral ACC, dmPFC, dlPFC, precuneus, and superior parietal cortex (Fig. 6B, Fig. S5B). Retrospective victimization was found to be associated with stronger activation in response to bullying in contrast to positive social interaction in the left posterior insula, superior temporal gyrus, and supramarginal gyrus, as well as right thalamus (Fig. S5B; p < 0.05 at voxel level, q < 0.05 at cluster level). For the ROI analysis results and GLM results without covariates, see Text S5 and Figure S6.
Attentional patterns during viewing of bullying and positive social interaction
The eye tracking data revealed larger pupil sizes for bullying videos (Mdn = 0.15) in comparison with positive social interaction videos (Mdn = −0.22), with large effect size r = 0.63, p < 0.0001 (Fig. 7A). Participants also made moderately more fixations per 10 s when watching bullying videos (Mdn = 22.50) in comparison with positive social interaction videos (Mdn = 21.90), r = 0.41, p = 0.002 (Fig. 7B). However, participants spent more time watching the positive social interaction videos (Mdn = 8.95 s per 10 s) in comparison with the bullying videos (Mdn = 8.85 s per 10 s), with large effect size r = 0.72, p < 0.0001 (Fig. 7C). There were no differences in number of blinks per 10 s between the conditions (MdnBullying = 1.97, MdnPositive = 1.99, r = 0.11, p = 0.42; Fig. 7D).
Differences in (A) standardized pupil sizes, (B) number of fixations per 10 s, (C) dwelling time (s) per 10 s, and (D) number of blinks per 10 s between the bullying and positive social interaction videos. Data was obtained in an eye tracking experiment on a separate sample of adults (n = 57). The horizontal line indicates the median, the lower and upper ends of the boxes indicate the lower and upper quartiles, and error bars indicate the 1.5 interquartile range. Nonparametric Wilcoxon paired signed-rank test was used for comparing attentional patterns between bullying and positive social interaction videos. r = effect size, **FDR corrected q < 0.01, ****q < 0.0001.
Discussion
Our main finding was that naturalistic simulated bullying activates large-scale limbic and paralimbic socioemotional distress circuits, as well as somatosensory and interoceptive cortices. Activation was also observed in the temporoparietal and frontal regions that extract diverse and complex information from social interactions (Santavirta et al., 2023). These responses were observed in both adolescent and adult participants. Altogether, these data indicate a shift in the social and affective processing in the brain during the threatening and adverse social interactions involving bullying. This is in contrast with the relatively focused activation patterns in response to social exclusion and rejection in simplified and artificial laboratory tasks, where effects are mostly observed around the cortical midline regions and in the default mode network (Vijayakumar et al., 2017; Mwilambwe-Tshilobo and Spreng, 2021).
Brain responses to bullying
We observed large-scale limbic and paralimbic activation, including the amygdala, dorsal striatum, insula, hippocampus, and orbitofrontal cortex, when the participants were exposed to simulated bullying versus positive social interaction. This activation pattern indicates the recruitment of the fear and affective circuits during victimization (Saarimäki et al., 2016). Anterior insula activity together with posterior sensorimotor regions of the insula and somatosensory association cortex region (SMG) in both age groups, as well as primary somatosensory cortex (S1) in adolescents, suggest that these affective experiences related to victimization also have a strong corporal and visceral component.
Activation in the precuneus increased in response to bullying and positive social interaction in both age groups. Precuneus has been suggested to encode others' intentions, and its activation likely reflects the encoding of the intended actions of the perpetrators (Vijayakumar et al., 2017). Activation of this region was stronger in response to positive social interaction versus bullying in adults, suggesting that experiencing victimization may acutely decrease the mental effort used for understanding the minds of others, possibly due to guiding cognitive capacities toward survival functions and emotion regulation. Increased emotion regulation in response to bullying was supported for both groups as indicated by the increased recruitment of the vlPFC in the bullying versus positive social interaction contrast in the ROI analysis. This is in line with meta-analysis on social exclusion (Vijayakumar et al., 2017), although vlPFC activation was not observed in adults in the whole-brain analysis.
Viewing bullying versus positive social interaction also activated regions involved in social perception, mainly the fusiform gyrus, superior temporal cortex, and occipital cortex (Santavirta et al., 2023). Apart from the occipital pole, these regions have not been consistently observed in meta-analytic studies on simulated social exclusion (Vijayakumar et al., 2017; Mwilambwe-Tshilobo and Spreng, 2021). These differences in activation patterns highlight the importance of using stimulus videos that are broadly representative of real-life bullying scenarios and contain detailed social information. Finally, motor cortical (M1 and SMA) activity increased in both groups in response to bullying versus positive social interaction, suggesting that simulated victimization induces initiation of motor actions possibly related to escape or initiation of counter-aggression in the threatening situation.
Consistent responses to bullying in adolescents and adults
In general, the responses to bullying were consistent across adolescents and adults, yet the activation patterns were more widespread in adolescents. The only differences that remained between the groups when contrasting bullying with positive social interaction were observed in the thalamus (ROI analysis) and the crossing of M1 and S1 (whole-brain analysis). These results suggest that simulated victimization may have been a more bodily or visceral experience for adolescents than for adults. The absence of more widespread differences between the age groups indicates however that the acute effects of victimization remain consistent from adolescence to adulthood. A prior meta-analysis on the role of ACC in social rejection found ACC to be more activated in adults than in children in response to social rejection versus inclusion (Rotge et al., 2015). However, we did not observe differences in the cingulate activation between adolescents and adults when comparing bullying with positive social interaction. Yet, the ROI analysis revealed that this lack of difference resulted from the fact that in adolescents the ACC response increased for bullying while remaining at zero for positive social interaction, whereas in adults the ACC response decreased for positive social interaction while remaining at zero for bullying. All in all, our results thus suggest the involvement of ACC during bullying in adolescents, whereas this region becomes disengaged in adults during potential social conflict, potentially reflecting the higher affective salience of social threats during adolescence.
Real-life peer victimization experiences predict neural responses to bullying
Most consistent associations between past victimization experiences and brain responses to bullying were observed in adults, where negative associations were observed in the left dlPFC (GLM analysis) and dmPFC (ROI analysis). This suggests that toward adulthood the continuous victimization experiences in childhood may decrease emotion regulation during victimization experiences (Ochsner et al., 2012). No other statistically significant associations were observed using the conventional threshold level in the GLM or in the ROI analysis for any of the contrasts for either group. Results obtained using a more lenient threshold nevertheless suggest a relationship between real-life victimization and alterations in the affective and emotion regulation system (ACC, MCC, medial and lateral PFC, insula, striatum) also in adolescents in response to bullying. Additionally, victimization-related increase in activation was observed for bullying in comparison with positive social interaction in both groups in the emotion and social cognition processing areas (anterior insula, vlPFC, and anterior MCC for adolescents, posterior insula, thalamus, STG, and SMG for adults). In line with prior studies done in the context of social rejection (Lee et al., 2014; Rudolph et al., 2016; McIver et al., 2018), these findings suggest that real-life victimization may sensitize the circuits subserving emotions. However, studies with bigger sample sizes and longitudinal data should be conducted to better understand the short and long-term associations between real-life victimization and brain responses to bullying.
Amplified emotional and attentional responses to bullying
The results from the eye tracking experiment confirmed that bullying evoked stronger emotional arousal than positive social interaction, as indicated by larger pupil sizes during watching the bullying videos in comparison with watching positive social interaction videos. Participants also searched for information more vigilantly in the bullying context, reflected by higher fixation frequency for the bullying videos. These patterns were observed despite the lack of low-level visual differences between the video categories and use of same locations and actors for the bullying and positive social interaction videos. Number of blinks did not differ between the conditions, but the decreased dwelling times for the bullying videos in comparison with positive social interaction suggest that the threatening nature of bullying may have caused avoidance of looking at the videos. These results are in line with the fMRI data, indicating that the increased brain responses in the areas processing social and sensory information and activation of the limbic system in response to bullying reflect higher emotional arousal and attention allocation to the threatening events of bullying.
Limitations
Our stimuli depicted events from the school environment and were tailored for adolescent rather than adult participants. Despite this, adolescent and adult static ratings for bullying were highly correlated in the fMRI experiment, and the overall pattern of brain responses to bullying was consistent across age groups, indicating that the videos were sufficient in evoking an experience of victimization in adults. The dynamic regressors used in the fMRI data analyses were based on the ratings of adults to avoid extra exposure of adolescents for simulated victimization. However, given the high correlation between adolescent and adult static ratings, and the scaling of the dynamic regressors, adult ratings were considered as sufficient for studying the neural responses to bullying in both age groups. To allow comparisons across groups, all functional data were normalized into common adult stereotactic space. Despite anatomical differences between developmental groups, registration of adolescent brains in the common adult space is unlikely a source of bias on the functional level (Burgund et al., 2002; Ghosh et al., 2010). Finally, real-life victimization experiences were measured using self-report questionnaires, introducing a degree of uncertainty due to possible inaccuracy in recalling the actual experiences. The self-reported victimization levels were low to moderate in our adolescent sample, and the associations between real-life victimization and brain responses to bullying could be more robust or even different in more regularly victimized participants.
Conclusions
We conclude that exposure to life-like, naturalistic bullying acutely engages the socioemotional distress system and social processing regions of the brain in adolescents and adults. These responses are accompanied by increased activity in the somatosensory and interoceptive cortices, indicative of strong visceral and corporal components in the bullying experience. Altogether, this large-scale activation of neural systems subserving socioemotional, somatosensory, and interoceptive processing highlights the adverse and threatening nature of bullying and reveals how it evokes a state of stress or alarm in the central nervous system. Future longitudinal imaging studies should address how different risk and protective factors affect the way the brain reacts and adapts to sustained victimization.
Footnotes
We thank all the participants and their families for participating in this study; Jinglu Chen, Taru Puustjärvi, and Turku PET Centre staff for assistance in data collection; and Mika Kurkilahti from A1 Media for the stimulus videos. This work was supported by the Eino Jutikkala Fund, Finnish Governmental Research Funding (VTR) for Turku University Hospital and for the Western Finland collaborative area, Finnish Brain Foundation, Olvi Foundation, and The Paulo Foundation grants to B.P., ERC Advanced Grant (2019/#884434) to C.S., and ERC Advanced Grant (#101141656) to L.N.
The authors declare no competing financial interests.
This paper contains supplemental material available at: https://doi.org/10.1523/JNEUROSCI.0738-25.2025
- Correspondence should be addressed to Birgitta Paranko at bspara{at}utu.fi.
