Structural and functional biomarkers of the insula subregions predict sex differences in aggression subscales

Abstract Aggression is a common and complex social behavior that is associated with violence and mental diseases. Although sex differences were observed in aggression, the neural mechanism for the effect of sex on aggression behaviors remains unclear, especially in specific subscales of aggression. In this study, we investigated the effects of sex on aggression subscales, gray matter volume (GMV), and functional connectivity (FC) of each insula subregion as well as the correlation of aggression subscales with GMV and FC. This study found that sex significantly influenced (a) physical aggression, anger, and hostility; (b) the GMV of all insula subregions; and (c) the FC of the dorsal agranular insula (dIa), dorsal dysgranular insula (dId), and ventral dysgranular and granular insula (vId_vIg). Additionally, mediation analysis revealed that the GMV of bilateral dIa mediates the association between sex and physical aggression, and left dId–left medial orbital superior frontal gyrus FC mediates the relationship between sex and anger. These findings revealed the neural mechanism underlying the sex differences in aggression subscales and the important role of the insula in aggression differences between males and females. This finding could potentially explain sexual dimorphism in neuropsychiatric disorders and improve dysregulated aggressive behavior.


| INTRODUCTION
Aggression is a general behavior in social interactions. Individuals with aggression may intend to cause physical and mental injury to others, especially when there is a conflict of interest between two or more individuals (Moyer, 1971;Nelson & Trainor, 2007). Aggression is a natural disposition of both humans and animals and is an adaptive trait during the early stages of human evolution (McCall & Shields, 2008).
However, out-of-control aggression is pathological in modern society, while aggressive clinical symptoms are common in multiple neuropsychiatric disorders (Connor et al., 2019;Jensen et al., 2007). Therefore, aggression imposes a huge burden on individuals and society (Connor et al., 2019;Jensen et al., 2007;Nelson & Trainor, 2007).
The insula is an important brain region involved in emotional processing and aggression (Blair, 2016;Nelson & Trainor, 2007;Phillips et al., 2004). Previous studies have demonstrated the effects of sex on the volume and functional connectivity (FC) of the insula, most of which showed that men have greater volume and FC of the insula than women (Dai et al., 2018;Hong et al., 2014;Jin et al., 2019;Klabunde, Weems, Raman, & Carrion, 2017;Lotze et al., 2019;Oz et al., 2021;Ruigrok et al., 2014;Sie, Chen, Shiau, & Chu, 2019;Wierenga, Langen, Oranje, & Durston, 2014). More interestingly, recent studies on sex classification have shown that the insula is one of the most discernable features to discriminate males from females (Brennan, Wu, & Fan, 2021;Weis et al., 2020). On the other hand, the insula was implicated in aggression. Reactive aggression provoked by other persons or the social environment was associated with insula-related networks (Kramer, Riba, Richter, & Munte, 2011;Repple et al., 2017;White, Brislin, Sinclair, & Blair, 2014) and may cause a series of negative effects. A previous study investigated the neural mechanisms of reactive aggression and failed inhibition and found that the anterior insula played an important role in self-control (Dambacher et al., 2015). Three related studies showed that cognitive regulation affected the activation and network of the insula in aggression (Abram et al., 2015;Achterberg, van Duijvenvoorde, Bakermans-Kranenburg, & Crone, 2016;Jiang, Hou, Wang, & Li, 2018). Some studies investigated the influence of the social environment on aggression and found the activated anterior insula and ACC in response to social feedback and rejection (Achterberg et al., 2017;Achterberg, van Duijvenvoorde, van der Meulen, Bakermans-Kranenburg, & Crone, 2018;Chester et al., 2014). Taken together, the insula plays an important role in aggression-related networks. Based on histological or connectivity features, the insula can be divided into different subregions involved in various physiological functions (Cauda et al., 2012;Fan et al., 2016;Gordon et al., 2016;Kelly et al., 2012). Although previous studies have demonstrated that both the anterior insula and mid-posterior insula are associated with aggression (Beames, Gilam, Schofield, Schira, & Denson, 2020;Cope et al., 2014;Cupaioli et al., 2021;Dambacher et al., 2015;Krauch et al., 2018), the effect of specific subregions within the insula on aggression subscales remains largely unclear. Therefore, it is of great interest to evaluate the neural mechanism for sex differences in aggression subscales by examining the association between the insula subregion and aggression subscales.
Since associations among sex, aggression subscales and insula were observed, the structure and function of the insula may underlie the neurobiological mechanism of the effect of sex on aggression subscales. However, the association among sex, insula, and aggression is still unclear, especially for different insular subregions. Therefore, this study aims to elucidate the role of insula subregions on the relationship between sex and aggression subscales in a large onefold Chinese young sample. We first subdivided insula into six subregions using the Brainnetome atlas (Fan et al., 2016) and further calculated each subregion's gray matter volume (GMV) and FC. The subtraits of aggression for each individual were assessed by the Buss-Warren Aggression Questionnaire. We further investigated the association between sex, aggression, GMV, and the FC map of each subregion in a large healthy Chinese sample. Demographic information, including sex and age, was measured.

| Subjects
The aggression of each participant was assessed by the Chinese version of the Buss-Warren Aggression Questionnaire (AQ-CV), including five subscales: physical aggression, verbal aggression, anger, hostility, and indirect aggression.

| MRI data acquisition
All participants were scanned by a 3.0 Tesla GE MRI scanner (General Electric, Milwaukee, Wisconsin) and instructed to keep still and keep their eyes closed without falling asleep. After scanning, the subjects were asked to ensure that they were awake during the experiment. The structural MRI data were acquired by using the T1-weighted brain volume (BRAVO) MRI sequence, and the param-

| Resting-state fMRI data preprocessing
The resting-state fMRI data were preprocessed by using a pipeline BRAinNetome Toolkit (BRANT, https://github.com/YongLiuLab/ brant-stable). The specific preprocessing steps included (a) discarding the first 10 volumes to maintain magnetization equilibrium; (b) slice timing; (c) head motion correction; (d) normalizing the EPI images to Montreal Neurological Institute (MNI) standard space with resampling to 3 Â 3 Â 3 mm; (e) regressing out head motion parameters, linear trends, white matter and cerebrospinal fluid signals; (f) temporal bandpass filtering between 0.01 and 0.08 Hz; and (g) smoothing with a Gaussian kernel of 6 mm full-width at half maximum. The global brain signal variable is not included as a covariate in the regression model because this may exaggerate the negative correlation Saad et al., 2013). We excluded 28 participants since their maximum displacements were greater than 2 mm in any cardinal direction, such as x, y, z, or their maximum rotations were greater than 2 about any of the x-z axes.

| GMV of insula subregions
The GMV map of each participant was produced using SPM12 (http://www.fil.ion.ucl.ac.uk/spm) which was executed in MATLAB R2016a (MathWorks, Natick, Massachusetts). First, the standard unified segmentation model was used to segment structural MRI images into gray matter and white matter. Furthermore, the gray matter template was constructed from the entire structural image dataset by the diffeomorphic nonlinear registration algorithm (DARTEL). Subsequently, the gray matter template was affinely registered to the tissue probability map in MNI standard space. Afterward, the gray matter image of each participant was nonlinearly normalized to the gray matter template in MNI space. Finally, modulation of the normalized gray matter images was carried out to obtain normalized and modulated gray matter map for each subject (termed GMV map).
To divide the insula into subregions, the Brainnetome atlas (Fan et al., 2016) was implemented, which defines fine-grained brain subregions using connection patterns. According to this atlas, the insula was segmented into six subregions in each hemisphere ( Figure 1), including the hypergranular insula (G), ventral agranular insula (vIa), dorsal agranular insula (dIa), ventral dysgranular and granular insula (vId/vIg), dorsal granular insula (dIg), and dorsal dysgranular insula (dId). The above 12 subregions were extracted and resampled to 1.5 Â 1.5 Â 1.5 mm in MNI standard space. Afterward, the average GMV of each subregion was calculated for each participant.

| Insula subregions FC analyses
The 12 subregions of the insula from the Brainnetome atlas were selected as the seed regions and resampled to 3 Â 3 Â 3 mm in MNI standard space. For each participant, Pearson correlation coefficients F I G U R E 1 The anatomical location of 12 insula subregions. dIa, dorsal agranular insula; dId, dorsal dysgranular insula; dIg, dorsal granular insula; G, hypergranular insula; vIa, ventral agranular insula; vId_vIg, ventral dysgranular and granular insula between the mean time series of each insula subregion and the time series of each voxel in the whole brain were calculated, resulting in 12 correlation maps. A Fisher r-to-z transformation was used to normalize the correlation coefficient (r value) to z scores. Subsequently, a one-sample t-test with a gray matter mask in SPM was carried out on whole z maps of all subjects for each insula subregion to test the significance of the connectivity between the seed region and the voxels of the whole brain. The clusters that survived at a threshold of p < .001 at a voxelwise false discovery rate (FDR) correction were determined to be significant and were used as binary masks for further analysis.

| Statistical analysis
A two-sample t-test was performed to assess the effect of sex on age.
This test was also used to evaluate sex differences in aggression subscales, such as physical aggression, verbal aggression, anger, hostility, and indirect aggression, with age as the confounding variable. The Bonferroni correction was used to control for the false discovery rate.
All the above analyses were completed using SPSS24.
Sex differences in GMV of the 12 insula subregions were tested by two-sample t-tests with age as the confounding variable under Bonferroni correction. Then, correlation analysis was carried out to evaluate the association between GMV and aggression subscales while controlling for age. Third, the classic mediation model was used to determine whether the GMV of the insula subregions can serve as a potential mediator of the association between sex and aggression subscales, with sex as the independent variable, significantly correlated GMV as the mediator and significantly correlated aggression subscales as dependent variables, with age as the confounding variable.
The analysis process for the FC of insula subregions was similar to that for GMV. First, a two-sample t-test was used to assess the effect of sex on FC of each insula subregion in a voxel-wise manner, while age was regarded as a nuisance covariate. The significance of sex differences was determined under a voxel-level FDR corrected threshold of p < .001. The regions with significant sex differences (after FDR correction) were used for further analysis. The mean FC of each identified region was calculated. Then, the correlation between the mean FC of the identified insula subregions and aggression subscales was assessed with age as the nuisance covariate. Third, the mediation effect of FC of insula subregions on the association between sex and aggression subscales was estimated by the classic mediation model. In this model, sex was considered an independent variable, FC of insula subregions showing the significant correlation with aggression subscales was considered a mediator, and aggression subscales were considered dependent variables, while age was used as a confounding variable. The mediation analyses were conducted using the PROCESS v3.3 macro in SPSS24 with 5,000 bootstrap tests and 95% confidence interval (CI) (Hayes, 2012;Preacher & Hayes, 2008).

| Demographics and behavioral characteristics
Statistical analysis showed no significant difference in age between males and females (Table 1). Significant effects of sex on physical aggression (p = 6.28 Â 10 À7 ), anger (p = .000001), and hostility (p = .009) were observed under Bonferroni correction. Specifically, males showed higher physical aggression and hostility scores but lower anger scores than females. Additionally, there was a trend of sex difference in verbal aggression (p = .02), in which a higher score was observed in the males than in the females (Table 1).

| Sex effect on GMVs of insula subregions
This study demonstrates that sex significantly influenced the GMV of all insula subregions under the strict Bonferroni correction, as shown in Table 2. For all insula subregions, the males had higher GMV than the females. The physical aggression scores were significantly corre-

| Sex effect on the FC of insula subregions
Our study found the significant effect of sex on the FC of bilateral dIa, dId, and right vId_vIg (Table 5, Figures 3 and 4). Specifically, the

| DISCUSSION
This study mainly investigated the association between sex, structure, and function of the insula and aggression subscales. We identified significant effects of sex on physical aggression, anger and hostility. Sex also influenced the GMV and FC of insula subregions. Even more striking, the GMV of the left dIa and right dIa mediated the association between sex and physical aggression, and the left dId-left ORBsupmed FC mediated the association between sex and anger, which may reveal the underlying neural mechanism of sex differences in aggression subscales.
The observed significant effect of sex on the physical aggression is consistent with previous studies that showed males tended to take physical aggression action more than females (Archer, 2004;Gerevich et al., 2007;Harris & Knight-Bohnhoff, 1996;Kalmoe, 2015;Sadiq & Shafiq, 2020). Additionally, we also found sex difference in hostility, that is, males having higher hostility than females, which is in line with a previous finding that males showed more hostility than females in Spanish and Japanese samples (Ramirez et al., 2001). These findings indicated that physical aggression may be associated with hostility. In addition, the higher anger scores in females than in males were also similar to the results in Isanzu and Buryats (Butovskaya et al., 2020).
Additionally, an fMRI experiment of facial expressions also showed higher anger recognition levels in females than in males (Dores, Barbosa, Queiros, Carvalho, & Griffiths, 2020), which indicates their emotional dysregulation.  Step dorsal anterior insula, a ventral anterior insula, a central region, a more ventral region, and two posterior subregions (Fan et al., 2016). In brief, the dorsal anterior insula is associated with cognitive function, the ventral anterior insula is associated with social-emotional tasks, and the mid-posterior insula is related to interoception, perception, somatosensation, pain, and motor (Chang, Yarkoni, Khaw, & Sanfey, 2013;Kelly et al., 2012;Uddin, Nomi, Hebert-Seropian, Ghaziri, & Boucher, 2017). Previous studies found that males exhibited significantly larger volumes across many cortex regions, including the insula, than females (Oz et al., 2021;Wierenga et al., 2014), which indicated sex differences in cortical development. Other studies have shown that sex affected the volumes of insula subregions. A related study demonstrated that males with posttraumatic stress symptoms had a larger volume of ventral anterior insula than females with posttraumatic stress symptoms, but this difference was not significant in control subjects (Klabunde et al., 2017). Another study found the larger GMV of the posterior insula in females than in males (Lotze et al., 2019). The inconsistent results of previous studies may be associated with different contexts of subjects and different locations of the insula. Therefore, our study investigated more fine-grained insula subregions based on the Brainnetome atlas and found that males showed the larger GMV of each insula subregion than females, which was partly consistent with previous studies and revealed sex difference in brain maturation, with cortex volume decreasing more in females than males during puberty (Vijayakumar et al., 2016;Wierenga et al., 2014).
In addition, we also found the mediation of bilateral dIa GMV on the association between sex and physical aggression. The dIa belongs to the anterior insula and is related to cognitive tasks, decision making, and awareness (Craig, 2009;Deen, Pitskel, & Pelphrey, 2011). The anterior insula, which is involved in the salience network, is associated with social cognition and evaluation, and is sensitive to social saliency (Achterberg et al., 2016;Achterberg et al., 2018;Cacioppo et al., 2013). In addition, prior studies found that 19% of the variance in callous-unemotional traits was explained by the GMV of the anterior insula in males, and callous-unemotional traits were related to physical aggression (Raschle et al., 2018;Wright, Hill, Pickles, & Sharp, 2019). The fMRI studies also showed an association between anterior insula and reactive aggression and motor impulsivity (Chester et al., 2014;Dambacher et al., 2015;Werhahn et al., 2020). Therefore, compared with females, males received more social salience stimuli because of greater GMVs of bilateral dIa, which led to more physical aggression.
On the other hand, this study investigated the effect of sex on the FC of insula subregions. First, we found males showing greater FC between dIa and some prefrontal and parietal cortex, such as ORBinf, ORBsupmed, PCUN, and PUT, which was similar to previous studies (Hong et al., 2014;Sie et al., 2019 the insula as a whole and found significant sex differences in FC between the insula and prefrontal cortex and sensorimotor cortex, where men showed increased FC in the insula than women (Jin et al., 2019). The effect of sex on FC mainly focuses on the relationship between the insula and brain regions in the default mode network (DMN) (Buckner, Andrews-Hanna, & Schacter, 2008;Liu et al., 2010;Smith et al., 2009). Overall, compared with females, males demonstrated a stronger modulation of insula subregions in the DMN, while the modulation of vId_vIg on Cbe9 and CUN was weaker in males than in females for the compensation mechanism.
Moreover, correlation analysis and mediation analysis revealed the important role of left dId-left ORBsupmed FC in mediating the relationship between sex and anger. The dId belongs to the middle insula and is related to interoception, sensory perception, and somatosensation (Kelly et al., 2012;. The functional experiment showed that middle insula activity was associated with tolerance of anger expression (de Greck et al., 2012).
Anger is a common experience during interpersonal communication, and some interpersonal conflict behaviors, such as unfair treatment and personal insults, may arouse anger (Averill, 1983;Baumeister, Stillwell, & Wotman, 1990;Gilam & Hendler, 2017). Moreover, anger is associated with emotion underregulation, and ORBsupmed is an important region of the emotion regulation network (Gilam & Hendler, 2017  Previous studies have shown that cultural background may influence aggression behavior (Butovskaya et al., 2020;Hyde, 2014;Ramirez et al., 2001). Therefore, further studies in different populations are needed to clarify the effect of sex on aggression in different populations. Second, our study only investigated the effect of sex on aggression and related neural mechanisms. In fact, aggression is a very complex social behavior that is influenced by multiple factors, such as genetic or environmental factors. Thus, further studies are needed to assess the effects of other factors on aggression. Advanced studies using machine learning models are also needed to predict aggression based on images and behavior measures. Third, although the Brainnetome atlas has been validated to effectively define more finegrained brain subregions and is consistent with other brain parcellation atlases (Fan et al., 2016), the impact of the potential interindividual variability of the insular subregions should also be investigated in further studies.

| CONCLUSION
In conclusion, the current study assessed the neural mechanism of sex differences in aggression subscales based on the structure and function of insula subregions. We found that sex significantly influenced physical aggression, anger, and hostility. The GMV of all insula subregions and FC of dIa, dId, and vId_vIg were significantly different between males and females. More interestingly, this study found that the GMV of bilateral dIa mediated the association between sex and physical aggression, and the FC between the left dId and left ORBsupmed mediated the relationship between sex and anger. These findings reveal the neural mechanism underlying the sex differences in aggression subscales and could potentially be used to improve dysregulated aggressive behavior.
T A B L E 6 Significant mediation effect of FC between dId.L and ORBsupmed.L on the association between sex and anger with age as a control variable Step Note: *p < .05, **p < .01, ***p < .001, two tailed. Abbreviations: Boot SE, bootstrapping standard error; FC, functional connectivity.