A fronto-insular network underlies individual variations in anger expression and control

Anger is a primary emotion typically characterized by discomfort serving to mobilize resources and enact change in response to provocation, hurt, or threat (Sorella et al., 2021, 2022; Videbeck, 2006). Its constructive role encompasses facilitating goal achievements, overcoming obstacles, and maintaining interpersonal boundaries (Grecucci, Giorgetta, Brambilla, et al., 2013; Grecucci et al., 2013b; Sorella et al., 2021, 2022). Nonetheless, anger regulation is challenging due to the intense physiological reactions associated with the fight-or-flight response, activated to safeguard oneself from the provoking circumstances (Lazarus, 1991). Difficulties in regulating anger may serve as precursors of aggressive behaviours (Lochman et al., 2010), with consequent interpersonal difficulties and social maladaptation (Baron et al., 2006; Heilbron & Prinstein, 2008). And anger dysregulation is a core feature of psychiatric disorders such as borderline personality disorder (De Panfilis et al., 2019), antisocial personality disorder (Kolla et al., 2017), and intermittent explosive disorder (Coccaro et al., 2014). Therefore, a comprehensive understanding of anger regulation is imperative. A widely recognized psychological framework aimed at elucidating the nature and components of anger regulation has been proposed by Spielberger and operationalized in the State and Trait Anger Expression Inventory (STAXI-2; Spielberger, 1999). This model recognizes anger expression and anger control as two main dimensions of anger regulation. Anger externalization (Anger Expression—Out) refers to an individual’s propensity to externalize or openly express their anger, through questions involving behaviours, actions, or reactions in anger-inducing situations, in contrast to anger internalization (Anger Expression—In) that regards the tendency of individuals to suppress or internalize anger. Anger control refers to people’s ability to control the physical or verbal expressions of anger (Anger Control-In) and to relax, calm down, and reduce angry feelings before they get out of control (Anger Control—Out). Among such dimensions of anger regulation, externalization and control represent independent but complementary components of anger regulation. Indeed, anger control can be viewed as the ability to restrain anger expression through control and outcome monitoring (Wilkowski & Robinson, 2008, 2010). If on the one hand externalizing anger in same cases may play a positive role (e.g., maintenance of boundaries) (Grecucci, Giorgetta, Brambilla, et al., 2013; Grecucci et al., 2013b; Sorella et al., 2021, 2022), on the other hand, when not counterbalanced by adequate anger control may have negative consequences. For instance, the combination between anger externalization and low anger control garners attention because both are considered precursors of hostility (Bridewell & Chang, 1997) and related social interaction issues (Baron et al., 2006; Messina et al., 2023) and aggressive behaviours (Birkley & Eckhardt, 2015; Lochman et al., 2010; Roberton et al., 2015). Similarly, it has been shown that people with poor control of anger have a greater propensity to angry externalization, fostering aggressive behaviours (Bettencourt et al., 2006; Dodge & Coie, 1987; Mattevi et al., 2019). From a clinical perspective, these dimensions of anger regulation require tailored interventions focused on externalizing psychological problems (distinct from those targeting internalized anger expression, more typical of internalizing problems) (Achenbach et al., 2016; Bjureberg et al., 2023; Manfredi & Taglietti, 2022; Pascual-Leone et al., 2013). Despite the importance of individual variances in regulating anger for mental well-being, there exists limited evidence regarding the neural foundations of these individual differences.

Anger control involves the ability to manage, calm, and monitor the outcomes of anger. The Integrative Cognitive Model (Wilkowski & Robinson, 2010) examines the link between trait anger and the capacity for anger control. Although the role of specific brain areas in anger regulation requires further investigation, the involvement of the prefrontal cortex has been proposed in a high-road model adapted from fear studies. According to Alia-Klein et al. (2020), uncontrollable anger triggers a low-road brain activation, where higher-order cognition (i.e., the prefrontal cortex) plays a minimal role, leading to aggression with greater automaticity. Conversely, higher-order cognition is crucial in regulating anger; for instance, regulation through reappraisal (an emotion regulation strategy that changes the interpretation of a situation) involves the activation of prefrontal regions, such as the inferior frontal gyrus (Grecucci et al., 2013a, 2013b). Stimulation of the medial prefrontal cortex can reduce anger and aggression (Gilam et al., 2018). Both medial and lateral prefrontal areas are part of the default mode network, linked to anger control (Sorella et al., 2022) and internalization (Grecucci et al., 2022). This aligns with the neural model of emotion regulation, where implicit regulation involves the medial prefrontal cortex, while explicit regulation also engages the lateral prefrontal and parietal cortices (Etkin et al., 2015). Thus, anger may be automatically managed by the medial prefrontal cortex, while conscious regulation involves the right inferior frontal gyrus and other lateral prefrontal regions, especially during strategies such as reappraisal (Grecucci et al., 2013a, 2013b). Beside the prefrontal cortex, other regions may play a role in anger control. At a structural level, Sorella et al. (2021) found that the concentration of grey matter in a network comprising ventromedial temporal areas, posterior cingulate, fusiform gyrus, and cerebellum correlated with trait anger. From a network perspective, previous functional studies have also found altered resting state activity (Sorella et al., 2021) and functional connectivity (Fulwiler et al., 2012) in the Default Mode Network (DMN), which includes frontal and temporal areas, confirming previous observations. Last but not the least, an effort to control anger may increase the connectivity between the amygdala and prefrontal cortices, which are responsible for top–down control (Denson et al., 2013). Accordingly, Fulwiler et al. (2012) showed a positive correlation between anger control and the functional connectivity (FC) of the amygdala with the contralateral orbitofrontal cortex. In line with this finding, it has been reported that violent offenders’ resting state activity after being provoked into anger showed increased amygdala–paralimbic connectivity and decreased amygdala–medial prefrontal cortex (mPFC) connectivity, suggesting that an inability of regulation inside the mPFC can lead to reactive aggression (Siep et al., 2018). The amygdala and prefrontal cortices, which oversee top–down behaviour, become more connected when individuals make an effort to control their anger in response to an insult (which is simulated in experimental settings using anger provocation paradigms) (Siep et al., 2018). Anger management and the functional connectivity (FC) of the amygdala with the oblique orbitofrontal cortex were found to be positively correlated (Fulwiler et al., 2012).

More related to externalization, some studies tried to understand the neural bases of aggression, a dysregulated form of externalization. Researchers found activations in frontal regions, in the insula, and in the striatum to be related with aggression tendencies (Dambacher et al., 2015; Repple et al., 2018; Skibsted et al., 2017). In another study, aggression was related with both the left superior frontal gyrus and the left middle temporal gyrus (Gong et al., 2022). Additionally, trait impulsivity that may be related with externalizing anger was linked to prefrontal, temporal, and parietal cortices (Pan et al., 2021). From a structural point of view, a recent study used supervised machine learning to identify grey matter features related to the individual differences in externalizing anger (Consolini et al., 2022), revealing that the medial, lateral, and orbitofrontal regions, the temporal and parietal regions (temporal poles, insula, fusiform and angular gyrus, posterior cingulate), the basal ganglia, and parts of the cerebellum were found to be involved in the structural network that predicted anger externalization.

Thus, the first aim of the present study is to test the hypothesis that there is a negative relationship between the expression and control of anger. We predict that this relationship stands true: the more the individuals can control anger, the less they externalize it. If this hypothesis is true (negative relationship between anger control and anger expression), someone can expect that the same neural circuit is involved in both externalizing and controlling anger.

The second aim of this study is to test the hypothesis that the same GM–WM circuit related to anger externalization may be related with anger control. To investigate the existence of a common neural circuitry for “anger expression” and “control,” we employed for the first time a data fusion unsupervised machine learning approach known as tIVA to decompose the brain into covarying GM–WM neural networks, and a Backward Regression approach to predict expression (STAXI Anger-outward) and anger control (STAXI Anger-control). In line with the previous evidence on these topics, we expect that an affective network including subcortical structures such as the amygdala, the basal ganglia, as well as the frontal network and temporal cortices may be included in this network. We also expect the cerebellum in having a role on this network. The cerebellum has been linked to many cognitive and affective functions (Sorella et al., 2022). Finally, we expect that WM regions related to the connections between these areas should be in the same way related to anger externalization and control.

Behavioural and structural MRI data from a cohort of 212 healthy participants (mean age: 26.06 ± 4.14 years, 131 M, 81 F) were obtained from the MPI-Leipzig Mind Brain-Body dataset available at OpenNeuro Dataset, http://openneuro.org, RRID:SCR_005031, accession number ds000221 (Babayan et al., 2020). This project was approved by the ethics committee of the University of Leipzig (Ethics Committee Approval Number: 097/15-ff). Participants were selected for the project with the following exclusion criteria: pregnancy, metallic implants, braces, non-removable piercings, tattoos, claustrophobia or tinnitus, and surgical operation in the past 3 months. Individuals with any history of psychiatric diseases that required inpatient treatment for longer than 2 weeks within the past 10 years, or history of neurological disorders (including multiple sclerosis, stroke, epilepsy, brain tumour, meningoencephalitis, severe concussion) were excluded too. Moreover, individuals with intake of active drugs, beta- and alpha-blocker, cortisol, and any chemotherapeutic or psychopharmacological medication were excluded. Finally, individuals with positive drug anamnesis (extensive alcohol, MDMA, amphetamines, cocaine, opiates, benzodiazepine, cannabis) were excluded too. They also met the MRI safety requirements of the MPI-CBS (Mendes et al., 2019) and provided informed consent before the experimental sessions. The final sample was determined based on specific criteria, including the availability of structural images, ages ranging from 20 to 45 years, and the availability of scores from STAXI. This age range was selected to exclude potential effect of ageing on brain networks. Previous studies have shown that GM and WM concentrations inside macro networks detected by ICA- and IVA-based approaches are subjected to effect age (see, e.g., Baggio et al., 2023). Adding individuals outside our age range would have added noise to the analyses.

The State and Trait Anger Expression Inventory-2 (STAXI-2; Spielberger, 1988) was considered to assess anger facets. In line with the aims of the present study, and to test the hypothesis of a negative relationship between externalization and control, and that they rely on the same brain circuit, we considered participants’ score on the subscale Anger-Expression-Out that refers to the extent to which people express their anger outwardly in a poorly controlled manner (i.e., the externalization of anger; eight items, e.g., “I do things like slam doors”). We also considered scores on the Anger-Control subscales, which assess individuals’ ability to monitor and control their emotions, by calming down (Anger-Control-In; 8 items, e.g., “I control my angry feelings”) and avoiding anger externalization through physical and verbal expressions (Anger-Control-Out; eight items, e.g., “I control my behaviour”). The STAXI scales have demonstrated adequate convergent and discriminant validity (Deffenbacher et al., 1996), internal consistency (Fuqua et al., 1991; Spielberger, 1996), test–retest reliability (Jacobs, Latham, & Brown, 1988), and a stable factor structure (Forgays, Forgays, & Spielberger, 1997; Fuqua et al., 1991). Indeed, the STAXI has been defined as an instrument with strong psychometric properties (Foley et al., 2002). Preliminary research has also supported the validity of the more recently developed Anger-Control subscale (Spielberger et al., 1985). In a more recent validation, the STAXI-2 scales showed to be reliable in terms of both internal consistency and test–retest reliability. Indeed, the internal consistency of the STAXI-2 was adequate, with alpha coefficients for the STAXI-2 scales all above 0.70, and a test–retest reliabilities fairly stable (Lievaart et al., 2016). Of note, the STAXI-2 used in this study was the German validated questionnaire that proved the goodness of the original scale (Tibubos et al., 2020). In the present study, participants had a mean of 11.929 (SD = 3.33) for anger externalization and of 22.726 (SD = 3.78) for anger control.

The MPI-Leipzig Mind Brain-Body dataset comprises anatomical, functional, and resting-state data collected at the Day Clinic for Cognitive Neurology, University of Leipzig, utilizing a 3 T Siemens Magnetom Verio scanner (Magnetom Verio, Siemens Healthcare, Erlangen, Germany) equipped with a 32-channel Siemens head coil. Each participant underwent a comprehensive imaging protocol, which included a high-resolution structural scan, four resting-state fMRI scans, two gradient echo field maps, two pairs of spin echo images with reversed phase encoding direction, and a low-resolution structural image acquired using a Fluid Attenuated Inversion Recovery (FLAIR) sequence, typically employed in clinical protocols (Mendes et al., 2019). For the purposes of our research, we exclusively utilized the structural images available within the MPI-Leipzig Mind Brain-Body database. These structural images were acquired using the MP2 RAGE sequence (Marques et al., 2010) and were characterized by the following parameters: TR = 5,000 ms; TE = 2.92 ms; TI1 = 700 ms; TI2 = 2,500 ms; flip angle 1 = 4°; flip angle 2 = 5°; voxel size = 1.0 mm isotropic; FOV = 256 × 240 × 176 mm; bandwidth = 240 Hz/Px; GRAPPA acceleration with an iPAT factor of 3 (32 reference lines); prescan normalization; and a total acquisition time of 8.22 minutes.

All structural MRIs (sMRI) underwent preprocessing using SPM12 (SPM, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/, RRID:SCR_007037). Initially, we conducted a comprehensive data quality check to identify and address potential distortions, including issues such as head motion or artefacts. Subsequently, we executed the reorientation procedure to align the images according to a common reference point and conducted image segmentation to delineate grey matter, white matter, and cerebrospinal fluid using the CAT12 toolbox (Computational Anatomy Toolbox for SPM, http://www.neuro.uni-jena.de/cat/, RRID:SCR_019184). For the current research, both grey and white matter images were utilized in our analyses. To facilitate registration, we employed the Diffeomorphic Anatomical Registration using Exponential Lie algebra (DARTEL) tools for SPM12 (https://github.com/scanUCLA/spm12-dartel). Finally, we conducted image normalization to the MNI (Montreal Neurological Institute) space, followed by spatial smoothing using a Gaussian kernel with a full width at half maximum (FWHM) of 12 following the suggestions by Chen and Calhoun (2018).

To investigate the neural underpinnings of the anger constructs of interest, we employed the unsupervised machine learning method known as Transposed Independent Vector Analysis (tIVA) (Adali et al., 2015). Transposed Independent Vector Analysis (tIVA) is a blind source separation method (BSS), which provides a fully multivariate approach and enables fusion of data from multiple modalities, such as GM and WM and then decomposing the brain into joint GM–WM profiles (Adali et al., 2015). The tIVA method is an extension of Independent Component Analysis (ICA) that exports statistical independence and generalizes ICA to multiple datasets by analyzing data across datasets, enabling multimodal fusion. tIVA was applied to structural data by using the Fusion ICA Toolbox (FIT, http://mialab.mrn.org/software/fit) (Calhoun & Adali, 2006) in the MATLAB 2018a environment (https://it.mathworks.com/products/matlab.html) (MATLAB (R2018a)). The number of components for both modalities was estimated via the minimum norm criterion. To investigate the reliability of each modality, the ICASSO (Himberg & Hyvarinen, 2003; Himberg, Hyvarinen, & Esposito, 2004) and the Infomax algorithms were used. ICASSO repeats the analysis multiple times with different random initializations and subsequently quantifies the consistency of the outcomes (Wei et al., 2022). The following parameters were selected: one group with 212 subjects, 2 features (GM and WM images), number of components 8, normalization in z-scores, RandInit mode (randomizing different initial values), PCA standard type, default mask, ICA options: learning rate = 0.0072, max steps = 512, annealing 0.9000, annealdeg = 60, posact=on, sphering=off, bias=on, verbose=on. The resulting output consisted of a matrix with the number of subjects (rows) and the loading coefficients for each component (columns). Loading coefficients represent how each component is expressed for every participant in terms of GM and WM concentration (density of values in each voxel). Subsequently, the independent components were translated into Talairach coordinates. Finally, the significant networks were plotted in Surf Ice (https://www.nitrc.org/projects/surfice/).

To explore the relationship between anger externalization and control, we performed a correlation analysis, which revealed a significant negative association between the two scales (rho = -0.418, p = 0.001; Fig. 1A). To assess eventual effects of gender, we conducted two sample t-tests to explore the potential influence of gender on anger externalization (t = 2.968; p = 0.003) and anger control (t = -4.436; p = 0.001) which confirmed gender differences. Additionally, we performed a correlation analysis to assess the impact of age on both anger externalization and control. Anger externalization was not correlated with age (rho = 0.046, p = 0.502) as well as anger control rho = -0.005, p = 0.940; Fig. 1B, C).

The Information Theoretic Criteria (Wax & Kailath, 1985) estimated eight networks of covarying grey matter (GM) and white matter (WM) that were subsequently estimated via tIVA. Each component included an estimated GM component and a corresponding estimated WM component with a similar pattern of concentration across subjects. Positive values indicated increased concentration of GM/WM, while negative values indicated a decreased concentration. The loading coefficients of these eight components were entered into two backward regression analyses, one to predict anger externalization and one to predict anger control. For anger externalization, the final model was significant (R = 0.243, R² = 0.059, adjusted R² = 0.050, RMSE = 3.249, F = 6.557, p = 0.002), and included tIVA5 (beta = 52.000, p = 0.042), gender (beta = -1.161, p = 0.015), and the intercept (beta = 4.658, p = 0.245. tIVA5 indicates five out of eight networks estimated by the tIVA algorithm. The higher the externalization, the higher the GM–WM concentration inside this network. For anger control, the final model was significant (R = 0.326, R² = 0.106, adjusted R² = 0.098, RMSE = 3.595, F = 12.435, p < 0.001), and included again tIVA5 (beta = -61.951, p = 0.029), gender (beta = 2.021, p < 0.001), and the intercept (beta = 30.995, p < 0.001). The higher the control, the lower the GM–WM concentration inside this network. See Tables 1 and 2 for a description of the areas included in the network, and Figures 2 and 3, for a visual representation of GM and WM as well as the residual plots. To better understand the direction of the effect of gender, we computed a t-test and we found females having higher GM–WM concentration than men (t = 3.287, p = 0.001). Although the regression did not return significant effect of age, we additionally computed a correlation between tIVA5 and age, and we found significant negative effect of age for tIVA5 (rho = -0.296, p < 0.001), meaning that the older the participants, the lesser the GM–WM concentration.

The aim of this study was to test two hypotheses. The first hypothesis proposed a negative relationship between anger externalization and anger control. The second hypothesis suggested that both the expression and control of anger could be predicted, at the neural level, by the same GM–WM network. Specifically, a frontal control network may be involved in controlling and externalizing anger. To test these hypotheses, we took into consideration behavioural scores from the STAXI-2 subscales of anger out and anger control of 212 healthy participants, as well as their GM and WM images. We found a significant negative correlation between anger externalization and control as predicted. Moreover, by applying an unsupervised machine learning method known as Transposed Independent Vector Analysis, we found that one specific GM–WM network was able to predict both the externalization and the control of anger. Departing from previous studies, we considered both GM and WM features in the same model. This allowed us to capture comprehensive information about both aspects of brain structure, as certain psychological processes rely on both tissue types (Baggio et al., 2023). In the following paragraphs, we provide details of these results.

At a behavioural level, we confirmed our hypothesis of a negative correlation between anger externalization and anger control. Thus, the individual tendency in directing anger outside (e.g., through aggressive behaviours) and the difficulty in controlling, calming down, and monitoring the outcomes of anger are often concomitant. It follows that anger externalization and control are two manifestations of a common core difficulty in regulating anger. This result provides an empirical confirmation of a clinical intuition but also aligns with the Integrative Cognitive Model (Wilkowski & Robinson, 2008, 2010). According to this model, controlling anger requires a form of effortful control, and individual differences in providing such cognitive effort can importantly influence anger externalization, mitigating aggression and reactivity in the presence of more effortful control resources. This model could also explain some aspects of anger dysregulation in many psychopathologies such as borderline personality disorder (De Panfilis et al., 2019) and antisocial personality disorder (Kolla et al., 2017), characterized by this combination of high anger externalization and low anger control.

Beyond self-report data, in the present study we provided, for the first time, a neurobiological explanation of this association, with the identification of a GM–WM circuit (tIVA5) that predicted both variables, with higher concentrations associated with higher anger externalization and lower anger control. Conceptually, one can interpret this finding as a manifestation of a possible “anger regulation continuum” between externalization and control that is reflected in the GM–WM features of this network. The fact that the same network covaries with both anger facets may indicate that anger control and externalization lie at the extremes of the same continuum. Higher concentration in this anger regulation circuit corresponds on one extreme to high externalization/low control, whereas lower concentration reflects low externalization/high anger control. The placement of single individuals on this continuum can offer a useful perspective for both clinical and research reflections concerning individual anger management, for the identification of potential tailored interventions, and assessment points.

The observation of the specific brain areas predicting individual differences in anger regulation may also shed light on psychological mechanisms that influence such individual differences. The most extended component of the circuit responsible for individual differences in anger regulation was in supplemental motor areas (SMA) in the paracentral lobule, with anterior extension towards the executive areas in medial frontal gyrus and superior frontal gyrus. The importance of emotional motor control, and motor planning have relevant implications for anger regulation (Friedman & Robbins, 2022). Consistently, the involvement of SMA may explain individual differences in anger regulation in terms of differences in threshold for action. Coherently, a previous study found that trait anger modulates the brain connectivity between the bilateral supplementary motor areas and the right frontal pole (Kim et al., 2022). The authors hypothesized that trait anger may be characterized by action readiness (supplementary motor area), more influenced by self-referential and somatomotor information (hyperconnectivity with the default mode and somatomotor networks). According to Kohn et al. (2014), the SMA should be involved in execution of regulation initiated by frontal areas, also detected in the present study, and recognized in emotion regulation literature (for meta-analyses see Kohn et al., 2014; Messina et al., 2015).

For example, it has been affirmed that medial prefrontal regions should play a major role in anger regulation evaluating potential outcomes and directing behaviours towards anger-eliciting stimuli (Gilam & Hendler, 2015). Interestingly, a previous study coherently found that the electrical stimulation of the medial prefrontal cortex with transcranial direct current stimulation reduced anger reactions during an interpersonal game (Gilam et al., 2018). Finally, the involvement of prefrontal medial regions (including the cingulate) was found by a recent supervised machine learning study to predict individual differences in anger externalization from GM feature only (Grecucci et al., 2022).

Among prefrontal areas, also the inferior frontal gyrus has been detected in the present study. One recent meta-analysis by Sorella et al. (2021) showed that the right inferior frontal gyrus is active for both the perception of angry stimuli and the subjective experience of anger. This may indicate that the higher the GM concentration in this circuit, the higher the anger experiences (and thus the externalization). From a functional point of view, anger externalization has been linked to brain connectivity in the prefrontal cortex, specifically between the inferior frontal gyrus and other cortical and subcortical regions. This suggests a clear role modulating anger in the decision to externalize it (Sorella & Grecucci, 2023).

Another important region in our study was the cingulate cortex. The cingulate, especially the anterior part, as well as the medial prefrontal area, has been associated with angry rumination (Denson et al., 2009), which characterizes the internalization of anger (Consolini et al., 2022). Indeed, in one study, authors found that after the presentation of angry faces, the negative connectivity of the ventral anterior cingulate cortex with the amygdala was reduced in individuals with high appetitive motivation (associated with aggression) (Passamonti et al., 2008). The involvement in the anterior cingulate in rumination and internalization of anger may appear contradictory to our findings, which show a positive association between the cingulate and externalization. However, in our study, the cingulate area was more centrally located and posterior compared with the anterior cingulate observed in previous studies on rumination and internalization (Consolini et al., 2022; Denson et al., 2009). In another functional study, after the presentation of angry faces, the negative connectivity of the anterior cingulate cortex with the amygdala was altered, especially in individuals with aggression (Passamonti et al., 2008). This study highlights a role of the cingulate in anger perception and in the modulation of other brain areas.

Another interesting region found inside the tIVA5 network was the precuneus. Although the precuneus serves numerous functions, it is recognized to be the main hub of the posterior default mode network. In a recent study, it was found that anger control was associated with the default mode network (Sorella et al., 2022). The role of DMN in anger reaction/control could be responsible for hostile bias and reactivity in front of anger eliciting stimuli. In particular, the default mode network is responsible for self-referential processes and rumination that are implicated in the experience of anger.

We also found a role for the insula. This is not surprising because the insula and other limbic and subcortical areas are all involved in the emotional experience (Picó-Pérez et al., 2018). For example, the insula has been found in angry reactions towards unfair behaviours in decision-making tasks (Grecucci et al., 2013a, 2013b). Also, according to Alia-Klein et al. (2020), uncontrollable anger relies on a subcortical low road including the insula and the amygdala. In another study, it has been found that in response to prohibitive language, individuals with specific genotypes are more likely to experience anger, and that this relies on insula and right hippocampus activity (Alia-Klein et al., 2009). Moreover, the activity of the insula is involved not only in anger and aggression (Dambacher et al., 2015; Emmerling et al., 2016; Skibsted et al., 2017), but also in anger rumination (Denson et al., 2009).

For what concerns the white matter side of the tIVA5 network, results indicate a diffuse effect of white matter portions adjacent to the relative main GM regions detected by the algorithm. Several WM portions have been found close to frontal and temporal regions, thus supporting the functionality of the GM regions they communicate with. Of note, the method used in this study (tIVA) does not provide information on white matter fibres integrity as usually DTI does. WM was used in a similar way as GM was, with the idea of detecting WM density and its association with anger externalization and control (see Baggio et al., 2023; Grecucci, Sorella, et al., 2023; Grecucci et al., 2024; Jornkokgoud et al., 2024 for similar methods).

Another intriguing finding of the present study was that anger externalization and control displayed were different in females and males, and coherently there is a difference in GM–WM concentration between females and males. In our sample, females were found to have higher externalization and coherently lower control compared with men. Coherently, the anger regulation circuit presented loading coefficients were higher in women than in men (higher GM–WM concentration in females than in males). This result can be somewhat counter-intuitive because we know from the literature that the tendency for men to aggress is more pronounced in man than in women (at least for aggression that produced pain or physical injury), whereas perceived aggressivity (e.g., perception that enacting a behaviour would produce harm to the target, guilt, and anxiety in oneself) is higher in females (Bettencourt & Miller, 1996; Eagly & Steffen, 1986). There are some possible explanations for why some women may experience higher levels of anger than men. One possibility is that women are more sensitive to social injustices, and experience gender stereotypes, injustice, role conflicts, lack of decision-making power, and even harassment or abuse, and social and familial pressures (Thomas, 2005). Alternatively, it is possible that female participants in the present study are not representative of the general population. Finally, we should consider that self-reported anger expression/control does not coincide with behavioural manifestation of anger. In order to draw clear conclusions regarding gender differences in anger management and the related brain correlates, future studies should investigate such differences on the basis of observed behaviours. The effect of gender was also visible in the regression analyses to predict both anger control and externalization. Indeed, additional analyses confirmed a clear effect of gender for tIVA5 network. Specifically, women had higher GM–WM concentration inside tIVA5 network. Since we found a positive correlation between tIVA5 and anger externalization (and negative correlation between tIVA5 and anger control), the fact that women had higher GM–WM concentration is coherent with the fact that women externalize anger more than men.

When anger is excessively externalized, individuals tend to express their anger through aggressive behaviours, which can lead to negative interpersonal outcomes. Conversely, excessive suppression of anger is associated with adverse traits such as hypertension, low self-esteem, maladaptive thoughts and behaviours, including rumination, and psychopathological conditions (Stimmel et al., 2005). Therefore, we suggest that an optimal balance between the two is advisable. From a translational perspective, intervening in the functionality of the GM–WM circuit reported in this study may be an effective method to modulate excesses in both aspects of anger. Altering the functionality of this circuit in the desired direction (e.g., reducing it in the case of excessive externalization or enhancing it in the case of excessive suppression) could be beneficial for individuals with anger issues. Neurostimulation methods such as transcranial direct current stimulation (tDCS) or transcranial magnetic stimulation (TMS) can be valuable tools to achieve this goal. Personality disorders characterized by difficulties in anger regulation may be particularly good candidates for such interventions, especially borderline personalities (De Panfilis et al., 2019; Grecucci et al., 2022; Grecucci, Dadomo, et al., 2023), and antisocial personalities (Kolla et al., 2017). Beside clinical implications, the combination of neuromodulation techniques with functional magnetic resonance imaging can provide causal information on the role of the brain circuit found in this study in modulating behavioural outcomes of anger externalization and control.

This study expands on previous research on the neural bases of anger, by adding new evidence that a network of covarying GM–WM predicted anger externalization and control. These findings may help future treatment of people suffering from anger dysregulation. Several psychopathologies are associated with excessive anger externalization or lack of control, including borderline personality disorder, antisocial personality disorder, intermittent explosive disorder, and anxiety disorders (Coccaro et al., 2014; De Panfilis et al., 2019; Kolla et al., 2017). However, excessive anger control characterizes anxiety disorders and dependent personality disorder (Grecucci et al., 2020). In this context, this study reveals potential clinical value, as it could be used for diagnostic or predictive purposes regarding the onset of anger-related disorders. This circuit could be considered a potential target of neurostimulation treatment to reduce anger externalization and increase anger control.

This study does not come without limitations. We investigated individual differences in anger regulation on the basis of self-reported questionnaires. Although STAXI is a widely recognized questionnaire whose reliability has been confirmed by several independent studies (Deffenbacher et al., 1996; Foley et al., 2002; Forgays et al., 1997; Fuqua et al., 1991; Spielberger, 1996; Tibubos et al., 2020), it must be acknowledged that cultural differences may exist in how we express and control anger across different populations (Alcázar Olán et al., 2015). Our results may, therefore, be limited to the European culture, and future studies may want to confirm these results across other cultures.

Another limitation is that this study focused only on structural analysis, leaving us without knowledge of activation patterns related to anger externalization/control. Future studies may want to explore the possibility of fusing structural and functional MRI data to expand these results. While further research is necessary, we believe that these findings could provide valuable insights for understanding and in the future predicting anger-related problems. The available data could ultimately assist healthcare practitioners in developing specific psychological treatments, such as targeted interventions involving brain stimulation/modulation or pharmacological techniques, for individuals experiencing anger-related issues.

Behavioural and structural MRI data were selected from the open-access MPI-Leipzig Mind Brain-Body dataset (OpenNeuro Dataset, http://openneuro.org, RRID:SCR_005031, accession number ds000221).

A.G., F.G., and E.M. conceptualized the study, analyzed the data, wrote the manuscript; X.Y., G.S., and I.M. wrote and reviewed the manuscript; I.M. and G.S. conceptualized the study

Authors declare no competing interests.