Personal insult disrupts regulatory brain networks in violent offenders

Abstract

The failure to adequately regulate negative emotions represents a prominent characteristic of violent offenders. In this functional magnetic resonance imaging study, we used technical, nonsocial frustration to elicit anger in violent offenders (n = 19) and then increased the provocation by adding personal insults (social provocation). The aim was to investigate neural connectivity patterns involved in anger processing, to detect the effect of increasing provocation by personal insult, and to compare anger-related connectivity patterns between offenders and noncriminal controls (n = 12). During technical frustration, the offenders showed increased neural connectivity between the amygdala and prefrontal cortex compared to the controls. Conversely, personal insults, and thus increased levels of provocation, resulted in a significant reduction of neural connectivity between regions involved in cognitive control in the offenders but not controls. We conclude that, when (nonsocially) frustrated, offenders were able to employ regulatory brain networks by displaying stronger connectivity between regulatory prefrontal and limbic regions than noncriminal controls. In addition, offenders seemed particularly sensitive to personal insults, which led to increased implicit aggression (by means of motoric responses) and reduced connectivity in networks involved in cognitive control (including dorsomedial prefrontal cortex, precuneus, middle/superior temporal regions).

Introduction

A 37-year-old Milwaukee man died in a shooting after a “spontaneous meeting and argument” with 2 people Monday afternoon on the south side, police said. “A great many of these incidents of violent crime are basically senseless and revolve around people being angry and not being able to control their anger,” an officer said (https://eu.jsonline.com/story/news/2020/09/21/milwaukee-shooting-36th-national-man-37-dies-police-arrest-2/5861485002/).

Such incidents occur on a daily basis. Spontaneous acts of violence are at the core of a large number of crimes with fatal consequences. But what exactly triggers anger in people that might culminate in uncontrolled violence? Which components of social encounters thwart behavioral control in some people? In order to understand the neurobiological and psychological dynamics of aggressive reactions, the multifaceted nature of anger and aggression must be acknowledged.

Impulsive outbursts are often preceded by negative stimuli such as aversive environmental conditions, threats, frustration, or provocation (Gilam and Hendler 2017; Bertsch et al. 2020), which trigger a negative internal state immediately, i.e. anger, and result in observable actions, i.e. aggression. Anger is accompanied by various autonomic physiological reactions, such as increased respiration, blood pressure, or skin conductance (Gilam and Hendler 2017), but also simultaneously activates manifested regulatory control mechanisms (Alia-Klein et al. 2020). If these regulatory mechanisms are successful, impulsive, aggressive behavior can be suppressed in most cases. However, if these regulatory mechanisms are ineffective, frustration or provocation might lead to uncontrollably expressed anger, i.e. aggression.

Numerous experiments have been conducted to systematically examine the various components of escalating anger and to determine its neurobiological correlates. Some previous meta-analyses compiled the results of these experiments and established a general neurological depiction of anger in the brain. For example, Vytal and Hamann (2010) reviewed 16 anger-related studies and found significant involvement of prefrontal regions in the experience of anger. Thus, bilaterally, the inferior frontal gyrus (IFG) and medial frontal gyri (medFG), the right middle frontal gyrus (MFG), and right anterior cingulate cortex (ACC) in concordance with the left amygdala, left thalamus, right parahippocampal gyrus, left superior temporal gyrus (STG), left fusiform gyrus (FFG), and right cerebellum were involved in anger. Later, in 2017, Kirby and Robinson (2017) performed a meta-analysis on 27 studies including 550 participants and reported again, anger-related involvement of the IFG and left MFG, but also activations bilaterally in the middle temporal gyri (MTG), right STG, left insula, bilateral substantia nigra, and left amygdala. These 2 meta-analyses revealed similar, yet not entirely consistent results, owing largely to the heterogeneity of the analyzed studies.

Attention to this inhomogeneity was noted in 2 recent meta-analyses. First, Alia-Klein et al. (2020) revisited the growing body of literature and found that anger experiences are associated with multiple neural networks, such as the mentalizing, salience, habitual, and self-regulatory networks. In addition, the authors found that internally induced anger (e.g. through the retrieval of autobiographical memories) is associated with different network components compared to externally induced anger (e.g. through unfair monetary offers). Thus, internally induced anger was found to be more associated with activity of the prefrontal cortex, such as the ventromedial and ventrolateral PFC, of insular and parietal regions, as well as subcortical regions, such as the striatum. In contrast, when anger is induced externally, neural responses are more prominent in prefrontal regions such as the ventromedial and dorsolateral PFC, the IFG, and the dorsal ACC. These results show that the type of emotional triggers determines its neuronal processing.

In a similar context, a subsequent meta-analysis (Sorella et al. 2021) examined neural correlates of anger perception on the one hand and anger experience on the other by discriminating paradigms, respectively. The authors claim the relevance of the different qualities of emotion processing. In this context, anger perception included the mere observation or detection of anger stimuli, whereas anger experience described the actual experience of anger. Again, the authors were able to reveal that different neuronal networks marked the perception and experience of anger. Thus, anger perception is linked to the activity of the right IFG, the right FFG, the right STG, and amygdalae, while anger experience, on the other hand, is characterized by activation of the insula, the ventrolateral PFC, and the right IFG. Interestingly, the right IFG was found to be involved in both types of anger processing and seems to play a crucial role in processing this emotion.

Reviewing these studies provides evidence that the perception and processing of anger are associated with the activation of a variety of subcortical and primarily prefrontal brain areas. Furthermore, it becomes evident that there are differences in the type of induction and experienced intensity of anger. For example, passive perception of negatively valenced stimuli (e.g. by viewing angry faces or movies; Coccaro et al. 2007; Lu et al. 2015; Tonnaer et al. 2017; da Cunha-Bang et al. 2019), being prevented from social inclusion or monetary reward (Herpertz et al. 2017; Radke et al. 2018), or receiving negative feedback during unfair offers (Gilam et al. 2015, 2017) provokes people to different extents. In addition, neuroscientific anger research employs a multitude of approaches which carry various problems. For example, often offline assessments of neural correlates “after” rather than during provocation were performed, or experiments in participants with low propensity for aggression (e.g. psychology students; Denson et al. 2009, 2013, 2014; Pawliczek et al. 2013) have been employed to study the neural correlates of anger processing.

While it is meaningful to investigate anger and aggression in subclinical groups, examining pathological aggression in, e.g., criminals has a large societal significance. The most common hypothesis about the escalation of anger is the frustration–aggression hypothesis (Dollard et al. 1939), according to which frustration is likely to be followed by aggression, with the extent of the aggressive reaction being typically determined by the strength of the frustration. This theory has been developed over the course of the decades (e.g. Berkowitz 1989, 1990) and extended to other concepts, such as provocation (Breuer 2017). By definition, frustration originates in disappointment and is often induced by frustrated non-reward, whereas provocation refers to a direct behavior of a person that is intended to elicit a response or reaction from the provoked person. However, it has been observed that individuals with high trait aggression, such as violent offenders, are particularly prone to frustration and provocation, which often results in severe anger and violent behavior.

This sensitivity has been shown to be often accompanied by increased activation of the limbic system. Recent neuroimaging research has demonstrated that aggressive behavior is determined by dysfunctional connectivity between the limbic system and the prefrontal cortex (Gilam and Hendler 2017; Ling et al. 2019; Romero-Martínez et al. 2020). The insufficient top-down inhibitory function of the prefrontal cortex may result in unrestrained emotional reactions in highly aggressive people due to insufficient (down) regulation of the limbic circuits. So far, few studies have been conducted on criminal offenders, which coherently found functional abnormalities during the processing of negative stimuli, by increased amygdala responsiveness (Prehn et al. 2013; da Cunha-Bang et al. 2017) while prefrontal activations were observed to be decreased (Volman et al. 2016). Reduced amygdala and increased prefrontal activation were also observed when offenders performed a theory-of-mind task, i.e. describing a person’s internal state by a photograph (Schiffer et al. 2017). Connectivity analyses revealed a decrease of limbic-prefrontal connectivity in criminal offenders during the processing of negative stimuli (Volman et al. 2016; da Cunha-Bang et al. 2017; Siep et al. 2019). Interestingly, similar observations could be made in patients suffering from Impulsive–Explosive Disorder, a disorder determined by the pathological inability to willingly regulate behavioral impulses resulting in aggressive behavior (McCloskey et al. 2016).

Measuring anger and aggression in violent offenders can be challenging; however, it displays the most naturalistic approach to tackle the question of whether and how brain function might contribute to the participant’s inability to control such behavior. To date, few studies have examined the neural correlates of escalating anger using direct personal insults and no study has yet been performed in violent, criminal offenders. In order to fill this gap, we established an experiment, in which we directly frustrated and provoked offenders and respective controls in a stepwise manner while simultaneously assessing their brain responses. We decided to explore the characteristics and peculiarities of anger-control networks in populations prone to violence as already suggested and illustrated by Alia-Klein et al. (2020). Therefore, it seemed appropriate that the neural correlates involved in anger processing should be studied beyond the scope of analyzing the prefrontal-limbic connectivity in a more holistic and comprehensive way. One atlas that is very suitable for that purpose is the social brain atlas (Alcalá-López et al. 2018). This map includes 36 brain regions that have been identified to be involved in social and emotional processes. Using a global atlas has the advantage of investigating large-scale brain networks in the context of anger processing, while many studies in the past have focused primarily on the connection between limbic and prefrontal regions during anger processing.

With this study, we aimed to identify neural networks that are involved in anger experience in general, and specifically in a forensic group. Furthermore, by inducing anger in a step-wise manner by frustration and provocation, we aimed to determine both, the effects of increasing levels of (nonsocial) frustration and the influence of personal insult (provocation) on experienced anger and its neural mechanisms. Both frustration and provocation can be seen as triggers for anger (as a negative feeling) and further (reactive) aggression (as an observable behavior). Another difference between the 2 mechanisms lies in the fact that frustration is often triggered by frustrated non-reward, in the case of this study technical failure, and provocation is more of an acute threat, in this case the insult by the experimenter. We expected to find that first, technical frustration would elicit anger and, that with increasing provocation by means of personal insult, the levels of anger should rise. We expected to find this mechanism to be orchestrated by an increase of implicit aggression (measured by motoric responses during a paradigm). Second, we proposed that the elicited anger would be stronger pronounced in offenders compared to controls. Our last 2 hypotheses dealt with the neural underpinnings of anger. Primarily, we expected to find differences between offenders and controls during both levels of frustration and provocation. Based on previous literature, we expected to find that offenders would show less connectivity between limbic (especially amygdala) and prefrontal regions compared to controls. Again, with increasing provocation, this result (i.e. the reduced connectivity in offenders) should be more pronounced.

Materials and methods

Participants

This study is part of a larger research project on the brain correlates of aggression in criminal offenders (Hofhansel, Regenbogen, et al. 2020a; Hofhansel, Weidler, et al. 2020b). Overall, this project recruited 67 male violent offenders (OF), between 18 and 55 years of age, convicted for at least one violent crime, from 3 different parole offices in Aachen, Germany. A noncriminal control group (HC) was recruited by local public advertisement. Participants, including offenders and controls, were excluded if they reported any history of seizures, acute mood, psychotic or anxiety disorder, consumption of any type of opiates in the past 12 months, or if contraindications for MRI measurements (e.g. metal implants) were present. After applying these criteria, 39 participants (HC = 17, OF = 22) were included in this part of the study. Further exclusion was caused by insufficient data quality (OF = 1), if participants indicated disbelief in the cover story, or requested preliminary termination of the MR measurement (HC = 5, OF = 2), resulting in 12 controls and 19 offenders being included in the statistical analyses. This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of the Medical Faculty, RWTH Aachen University (EK 337-15). All participants provided written informed consent.

Sample characteristics

To assess psychopathy in violent offenders, the German version of the Psychopathy Checklist—Revised (PCL-R; Mokros et al. 2017) was used. The PCL-R is based on a semi-structured interview and describes the 2 main dimensions of psychopathy, interpersonal problems (PCL-R factor 1), and deviant behavior (PCL-R factor 2). Within the first factor, a further differentiation can be made between interpersonal (facet 1) and affective problems (facet 2). Factor 2, on the other hand, consists of a lifestyle aspect (facet 3) and an antisocial one (facet 4). Further measures included crystallized intelligence quotient (IQ), including a verbal intelligence test (WST; Schmidt and Metzler 1992), trait aggression (Buss and Perry aggression questionnaire (AQ; Buss and Perry 1992), and the Reactive-Proactive Aggression Questionnaire (RPQ; Raine et al. 2006). In addition to a general score, both questionnaires allow the assessment of subfeatures of aggressive behavior (AQ: physical and verbal aggression, anger, and hostility; RPQ: proactive and reactive aggression).

Anger induction task

A modified version of the Technical Provocation Paradigm (Panagiotidis et al. 2017; Wagels et al. 2020) was used in this study and designed in Presentation® (Neurobehavioral Systems Inc., San Francisco, CA). This paradigm has successfully induced anger in noncriminal groups in previous studies. In this experiment, participants were instructed to aim a virtual ball into a barrel by pressing a button. To elicit anger, the program included manipulations in which the button press had no effect on the ball. This prevented the participants from hitting the barrel and thus collecting money (anticipated target), which triggered anger by technical frustration (see Panagiotidis et al. 2017; Wagels et al. 2020 for further details). Both studies could directly link motoric responses (usage of the response device) with self-reported anger; hence, we defined the number of relative button presses as a measure of implicit aggression in this study.

The following modifications of the paradigm were implemented for the current study: First, the response device was changed into a button box (instead of a joystick as in the previous 2 studies) to enable scanner compatibility. Second, a social component (personal insult) was included in the paradigm in order to provide an additional mode of provocation and thereby increase the anger response. Lastly, we decided to refrain from a valence evaluation after each trial (as was done in the previous validation studies of the paradigm) to increase the paradigm’s credibility and protect the cover story. The general procedure and task remained identical to those of Panagiotidis et al. (2017) and Wagels et al. (2020).

Procedure

Each participant was placed in the scanner and handed a button box. Their index finger was positioned on the response button and its functionality was tested before the start of the paradigm (Fig. 1: “button test”) with the experimenter asking the participant to press the button and informing them that a light outside of the scanner indicated if and how often the participant responded. Then, the task was explained. Participants were instructed to direct a virtual ball into a container by pressing a button to receive monetary reward. During a baseline run, they were able to practice this task, following which the participants were informed about the actual measure having started. To induce anger, the button was rendered unresponsive during some trials, making it impossible for participants to hit the container with the ball (technical frustration). After those fail trials, the prompt “Please press the button” (in German) appeared in large red letters on the screen. After that, the next trial started.

The experiment consisted of a total of 3 runs (Fig. 1), with 1 baseline run and 2 runs with increasing frustration and provocation. The baseline run consisted of 20 trials with no manipulations of the button. During the second run, and in order to induce technical, nonsocial frustration, 9 of the 18 trials were manipulated. Following the second run, the additional social provocation was induced by insulting the participants. The experimenter told the participant via intercom that the previous run had to be terminated due to the participant’s failure to play the game and press the button, which was observable from outside the scanner. The experimenter further accused the participant of lack of motivation, concentration, and ability to finish the task. The participant was again asked to fully focus and respond in order that valid data could be obtained. After that, the experimenter started the paradigm once again. During this third run, 8 out of 15 trials were manipulated. After this, and before the participants left the scanner, they were debriefed and informed about the nature of the paradigm.

As a result, technical frustration was applied throughout runs 2 and 3, while run 3 was further impacted by the preceding personal insult. We consciously decided to reduce the number of total trials from run to run to make it seem as if the premature termination of the runs was indeed due to the lack of response from the participants. The distribution of the manipulated trials was pseudorandomized and fixed for each participant. Implicit anger was defined by button presses per trial. Unlike in the previous study (Panagiotidis et al. 2017), we refrained from collecting self-report data on explicit aggression (e.g. emotional valence ratings during the task) as these measures might foil the cover story and make it less believable.

Data acquisition and processing

Behavioral data

The behavioral data were analyzed in SPSS (IBM SPSS Statistics for Windows, Version 25.0, released 2017, Armonk, NY: IBM Corp.). To test for group differences in the sample characteristics, Student’s t-tests were performed. Implicit aggression during the paradigm was assessed by the relative number of button presses per trial. Each participant’s relative response rate per trial (number of total presses/number of played trials) and run was entered into a repeated-measures general linear model (GLM), testing for the effects of group and runs. The rationale to account for not only the responses during the failed trials is based on the fact that the participants repeatedly pressed the button in between trials or when the trial was not manipulated.

FMRI data acquisition

Magnetic resonance imaging was performed using a 3-T Siemens PRISMA scanner (Siemens Medical Systems, Erlangen, Germany) with a 20-channel head coil. T1-weighted structural images were acquired by a 3-dimensional magnetization-prepared rapid acquisition gradient echo image sequence (voxel size: 1 × 1 × 1 mm³; 64 × 64 matrix; field of view [FoV]: 256 × 256 mm²; 176 volumes; time repetition [TR] = 2,300 ms; time echo [TE] = 2.98 ms; flip angle = 9°). The blood oxygen level-dependent signal during the paradigm was assessed by a T2^*-weighted, gradient-echo, echoplanar imaging sequence (voxel size: 3 × 3 × 3 mm³; 64 × 64 matrix; FoV: 192 × 192 mm²; 176 volumes; TR = 2,000 ms; TE = 28 ms; flip angle = 9°). During each run, the following number of volumes was acquired: run 1: 90 volumes (180 s), run 2: 80 volumes (160 s), and run 3: 70 volumes (140 s). In cases where the participants prematurely interrupted the measurement during provocation run 2 or 3, the data were rejected if less than 60 volumes (run 2) or 45 volumes (run 3) were acquired. The screen displaying the paradigm was placed at the head end of the scanner (52 × 33 cm/20.5 × 24″ with a resolution of 1,920 × 1,200 pixels, Full HD), visible to the participants via a mirror attached to the head coil. A handheld device by LUMItouch (Photon Control, Burnaby, Canada) recorded the participants’ responses.

FMRI data analyses

Functional connectivity analysis was performed using the CONN 2018a toolbox (Whitfield-Gabrieli and Nieto-Castanon 2012) in Statistical Parametric Mapping Software (SPM12) (https://www.fil.ion.ucl.ac.uk/spm) implemented in Matlab2015b (Mathworks, Inc., Natick, MA, United States). The CONN default preprocessing pipeline was used and involved the following steps. The raw functional images were realigned and unwarped and individual realignment parameters were saved for further implementation as first-level covariates. Then, these functional images were translated and slice time corrected. In a fourth step, a functional outlier detection (ART-repair) was performed to detect outliers, poor data quality, and excessive head movements. After that, functional images were segmented into white matter, gray matter, and cerebrospinal fluid and then normalized. Respective segmentation masks were saved for each participant. Following, the structural T1-images were segmented and skull-stripped and further normalized with the functional images. Both functional and structural images were translated into Montreal Neurological Institute space. Finally, functional images were smoothed with a kernel of 8 mm. A 2-step default denoising pipeline was implemented comprising first, linear regression and second, band-pass filtering. Next, the default denoising bandpass filter of 0.008–0.09 Hz was applied. Evaluated motion parameters were implemented into the analyses as regressors of no interest. Connectivity strengths across a priori-defined regions of interest were calculated using a regions of interest (ROI)-to-ROI approach. Second-level t-test results were thresholded at a false discovery rate of P_FDR = 0.05 to correct for multiple comparisons (Benjamini and Hochberg 1995; Rouam 2013). For maximum transparency, and to remain true to our explorative approach, we also report the results at P_uncorrected < 0.005 (Lieberman and Cunningham 2009).

Regions of interest

Despite anger and aggression have been linked with a number of specific brain regions as we presented in Section 1, the entire scope of involved brain regions remains rather unclear. Hence, we chose to use a rather exploratory approach and determine neural connectivity within a large network representing neural correlates most consistently involved in social behavior. The social brain atlas (Alcalá-López et al. 2018) is based on meta-analyses that extracted 36 brain regions with the highest involvement in social processes. We implemented all 36 regions in our analyses because, despite expecting disrupted connectivity between mainly the prefrontal and limbic structures, we wanted to use an exploratory approach and include regions found to be relevant to the full range of social behavior. Online supplements will provide complete results of connectives between those 36 ROIs.

Results

Sample characteristics

While controls and offenders did not differ in age (P = 0.850), controls had more years of education, higher verbal IQs, and suffered less frequently from substance use disorders (Table 1). In general, controls were less aggressive compared to offenders, with exception of the verbal aggression, anger, and hostility domains (of the Buss and Perry Aggression Questionnaire). Although we had a lot of dropouts, which were unevenly distributed on the side of the controls, both groups were comparable (Table 1).

Anger measure

This paradigm induced anger as anticipated. With increasing levels of frustration and provocation (i.e. runs), the relative number of button presses increased. Repeated measures GLM of the participants’ responses (button presses) revealed no main effect of group, but a significant main effect of run (F(2) = 37.54, P < 0.001). As shown in Fig. 2, subsequent within-group analyses revealed that in HC, the button press differences between baseline (M(SD) = 1.027(0.036)) and both nonsocial frustration (M(SD) = 3.028(0.448)) and social provocation (M(SD) = 3.483(0.578)) were significant (P < 0.001). Equally in OF, the response numbers during technical frustration (M(SD) = 2.798(0.341)) and social provocation runs (M(SD) = 3.709(0.440)) differed significantly from baseline (M(SD) = 1.066(0.028)). While there was no significant difference in the response numbers between the social and technical provocation runs in HC (P = 0.308), the increase of response from run 2 to 3 was significant (P = 0.011) for offenders.

Functional connectivity during provocation

Functional connectivity between all 36 ROIs defined as the social brain (Alcalá-López et al. 2018) was calculated during each run of the paradigm (baseline, technical frustration, social provocation) in both sample groups. T-values are illustrated in Fig. 3 and respective files can be found online.

(DOI 10.17605/OSF.IO/JG9X2; https://osf.io/jg9x2/?view_only=493cee32b85b48e8bd818fac9b1a1a18).

Effect of provocation within groups

Functional connectivity was compared between all 3 runs within each group separately in order to identify group specifics in the neural networks during provocation at P_FDR < 0.05. In controls, nonsocial frustration (run 2 > run 1) led to no significant results while after increasing provocation (run 3 > run 2) reduced neural connectivity was found between the left hippocampus and the right temporal pole (T(11) = −4.69, P_FDR = 0.02, P_(unc.) < 0.001). In offenders during nonsocial frustration, an increased connectivity between the left and right IFG (T(18) = 3.47, P_FDR = 0.05, P_(unc.) = 0.003, β = 0.23) and between the right IFG and left amygdala (T(18) = 4.64, P_FDR = 0.007, P_(unc.) = 0.0002, β = 0.27) as compared to the baseline was observed. A further increase of provocation by personal insult (run 3 > run 2) led to decreased connectivity of which none reached significance at the determined level. The interested reader can find illustrations and results with a lower significance threshold in the Supplemental Material S1 and full results can be obtained from the online source.

Effect of provocation between groups

To test for group differences, first, connectivity during each run was compared between both groups. At the proposed significance level of P_FDR < 0.05, there were no significant group differences during the baseline (run 1). During the nonsocial frustration (run 2), in offenders connectivity was increased between the left and right IPL (T(29) = 3.60; P_FDR = 0.041, β = 0.44), as well as between right anterior insula and left pSTS (T(29) = 3.52, P_FDR = 0.05, β = 0.35). During the social provocation (run 3) on the other hand, offenders had significant lower connectivity between the right hippocampus and the precuneus (T(29) = −4.05; P_FDR = 0.012, β = 0.53) and between the left TPJ and the left MTG (T(29) = −3.74, P_FDR = 0.028, β = 0.3). Connectivity between left TPJ and pMCC (T(29) = 3.48; P_FDR = 0.028, β = 0.33) was increased in offenders. Illustrations and results with a lower significance threshold can be found in the Supplemental Material S2 and full results can be obtained from the online source.

Secondly, the effect of group during provocation was analyzed by comparing the different effects of nonsocial frustration (run 2 > run 1) and personal insult (run 3 > run 2) between offenders and controls in terms of their neural connectivity (Fig. 4). Despite not finding any significant results below the threshold of P_FDR < 0.05, at a lower significance level (P_FDR < 0.1, P_{(uncorrected)} < 0.005) our results indicated that the effect of nonsocial frustration (run 2 > run 1) led to stronger connectivity in offenders compared to controls between the left amygdala and the right IFG (T(29) = 3.35, P_(unc.) = 0.0023, P_FDR = 0.0790, β = 0.33) while the vmPFC and the left FFA were less strongly connected (T(29) = −3.32, P_(unc.) = 0.0024, P_FDR = 0.0846, β = 0.46).

Comparing the 2 groups in terms of the effect of personal insult (run 3 > run 2), we observed considerably less connectivity in offenders. The connectivity between the dorsomedial prefrontal cortex (dmPFC) and the left MTG (T(29) = −3.91, P_(unc.) = 0.0005, P_FDR = 0.0178, β = 0.48) and between the right HCa and the PCu (T(29) = −3.83, P_(unc.) = 0.0006, P_FDR = 0.0223, β = 0.53) was decreased, and after lowering the threshold to P_{(uncorrected)} < 0.005, decreased connectivity in offenders between the right SMA and the right pSTS (T(29) = −3.12, P_(unc.) = 0.0040, P_FDR = 0.1413, β = 0.45) and between left and right IPL (T(29) = −3.06, P_(unc.) = 0.0048, P_FDR = 0.1674, β = 0.33) was found (see Fig. 4, full results can be obtained from the online source).

Discussion

Current neuroimaging studies on aggression are limited by the fact that they induce or measure anger indirectly or recruit subjects with a low propensity toward aggression. Here, we implemented a modified version of the TPP, which has been found to successfully induce anger directly by frustration and provocation (Panagiotidis et al. 2017; Wagels et al. 2020). Our study employed this task in the MR environment for the first time. Furthermore, with the aim to investigate neural networks during increasing anger, we explicitly focused on the effect of an additional personal insult on the regulatory networks in both criminal offenders and noncriminal controls.

First, we found that the provocation paradigm seemed to have worked and was able to induce anger. By using the number of button presses as a measure of implicit aggression (Panagiotidis et al. 2017; Wagels et al. 2020), we could determine an increase of relative button presses with increasing provocation. Secondly, we found indications that offenders responded more sensitive to the increase of provocation by means of an increased number of relative button presses. Hence, our paradigm successfully induced anger by (increasing) provocation and differences between offenders and controls could be determined on a behavioral level.

On a neural level, we also found striking differences between both groups during both levels of provocation. Interestingly, our hypothesis that offenders would show decreased neural connectivity between limbic and prefrontal regions during provocation could only be confirmed depending on the level of provocation. Both direct group comparisons and interaction analyses indicated that during the first level of the paradigm (technical frustration, run 2), offenders showed even stronger connectivity within the networks comprising the prefrontal and the limbic regions compared to the baseline measure (within-group), but also compared to controls (interaction analyses). On the other hand, after provocation was increased by means of personal insults, we could observe a drastic reduction in connectivity in behavioral control networks in offenders, but not controls.

In detail, we found that, frustrated by a technical failure, the offenders, compared to the controls, showed increased functional connectivity between the IFG and the amygdala. Interestingly, the connectivity between the right IFG and the amygdala was even stronger in the offenders compared to the controls. The right IFG has previously been reported to be crucially involved in anger processing and angry rumination (Fabiansson et al. 2012), as well as in cognitive control and response inhibition. Aron et al. (2004, 2014) found that the right, but not the left IFG, is frequently involved in motor inhibition tasks, such as stop-signal tasks, and concluded that the right IFG has a significant role in inhibiting both internally and externally triggered behaviors. This observation only partly corroborates our hypothesis. Based on the literature and the prefrontal hypothesis of escalating aggression (Dollard et al. 1939; Blair 2008), we expected, in the face of provocation, a failure of top-down inhibition in offenders in comparison with controls. Our results indicate that during technical provocation this hypothesis does not hold true, but does when provocation is reinforced by personal insult, as we found in offenders a decreased connectivity between the prefrontal and limbic regions.

Interestingly, with personal insult intensifying the provocation, the regulatory strategies that the offenders might have been able to employ during the nonsocial, technical frustration, seemed to collapse. This was indicated by the behavioral measure and paralleled by a significant decrease in connectivity within a large regulatory network comprising, e.g. the frontal and temporal poles, the SMA, both TPJs, the insula, the nucleus accumbens, and the hippocampus (see Supplemental Material S3). Thus, we conclude that this significant decrease in connectivity reflects the anger-triggered breakdown of regulatory mechanisms in violent offenders after personal insult. Notably, the control group was found to maintain or even increase neural connectivity strength within the regulatory regions after being insulted. This observation was corroborated by the behavioral data showing a significant increase in the measure of implicit aggression (button presses) from the first (technical) to the second (social) provocation run only in offenders. This indicates that while in general offenders may have both the capability and the willingness to control their behavior, they are unable to do so beyond a certain level of provocation. These results are coherent with the observation that provocation would lead to aggressive behavior when self-control, i.e. the ability to manage anger intensity and suppress angry thoughts and impulses, is reduced (Denson, Capper, et al. 2011a; Denson, Pedersen, et al. 2011b; DeWall et al. 2011).

These neuroimaging results shine a light on the malfunctioning of behavioral control in offenders. As it is often suggested in the literature (e.g. Blair 2010), emotional reactivity in offenders is explained by insufficient inhibitory qualities of the prefrontal cortex in response to heightened amygdala reactivity toward emotional stimuli. We can partly confirm this hypothesis by showing that the connectivity between the amygdala and the IFG determines behavioral control. Based on our observations, we can conclude that it is the quality and intensity of provocation that determines the involvement of particular neural networks and the evocation of the relevant behavioral response. Our results suggest that the breakdown of the (prefrontal) regulatory function occurs only after a personal insult, not yet during (technical) frustration.

While there are some paradigms that induce anger rather directly, e.g. by unfair offers or neglect of anticipated reward (e.g. the Taylor Aggression Paradigm), we state that the interpersonal provocation displays a more realistic and prompt induction of anger (Denson, Pedersen, et al. 2011b; Gilam et al. 2019; Beames et al. 2020). The group around Denson has contributed major work in the field of direct personal provocation. These authors developed an experimental set-up where participants were directly insulted by the experimenter for poor performance in a simple (but sometimes insolvable) task. After insulting noncriminal participants, Denson et al. (2009, 2013, 2014) found increased neural activation in primarily prefrontal (i.e. medial and lateral frontal gyrus, SFG, dmPFC, ACC), PCC, insula, amygdala hippocampus, and thalamus regions (Denson et al. 2009, 2013), as well as increased connectivity between amygdala and prefrontal regions (i.e. dlPFC, ACC, OFC) (Denson et al. 2013). In a similar study protocol, Gilam et al. (2015) determined increased connectivity between the vmPFC and the parahippocampal, fusiform, and lingual gyrus as well as the insula. Further, the insula was linked with the MFG, cingulate cortex, cuneus and precuneus, paracentral lobule, cerebellum, and thalamus during provocation in terms of unfair offers.

Due to the realistic study designs, these results are highly valuable in terms of aggression research. Yet, one major drawback of these studies is the fact that they were performed on students’ group with a low level of trait aggression. These studies indicate that personal insults profoundly induce anger, further accompanied by neural activation in many prefrontal and some limbic regions in noncriminal control groups. If we compare the results of our study on violent offenders using a similar study design, we find some similarities. Just as Denson et al. (2013), we determined a rather increased connectivity between the subcortical and among cortical areas, but only in controls. On offenders on the other hand, personal insult had a significant effect on neural connectivity by terms of a drastic reduction, as mentioned above. Peculiarities in offenders connectivity compared to controls have been reported before during other anger-inducing paradigms. da Cunha-Bang et al. (2019) found reduced connectivity between the amygdala and the SFG, as well as between the striatum and den medial OFC in offenders as compared to controls during a point-subtraction aggression paradigm. These results are coherent with our results by means of reduced connectivity in offenders during provocation. Anyhow, we must add, that during nonsocial, technical frustration, offenders establish an increase of connectivity between the amygdala and prefrontal areas. In order to place our findings in the existing body of research and deduce potential implications for therapeutic interventions in pathological aggression, we must give due consideration to the emotional instability and deficient cognitive capacities that offenders suffer from. Violent offenders are often impulsive, reactive-aggressive, and have trouble controlling their behavior in situations where people who are not particularly prone to violence are able to behave themselves. These behavioral problems often originate in a lifetime history of personal rejections, which eventually grow into low self-esteem and little perceived social support (Born et al. 1997; Huang et al. 2020). Additionally, offenders have difficulties trusting other people and are especially sensitive to personal rejections (Chester et al. 2019; Gao et al. 2021; Yang et al. 2019), tending to misinterpret ambiguous social situations as more hostile than they actually are (Lipsey et al. 2001). Due to a selection bias for aggressive or threatening cues, offenders’ social information processing is primarily threat-biased and restricted by a limited range of available ways of thinking and acting, causing them to misattribute others’ intent and misperceive them as hostile. Adequate response selection in a social situation requires a relatively high cognitive load to grasp possible alternatives and consequences. Due to the often lower cognitive flexibility, problems with decision-making, mental flexibility, planning, and anticipating the consequences of their behavior (Clark 2011), the choice of available responses is rather limited in offenders. In line with these observations, our results demonstrate that offenders fail to sustain the regulatory neural networks under high social pressure, i.e. personal provocation. Thus, one might argue that therapeutic work in offenders should focus more on improving their existing ability to employ emotion regulation mechanisms and strengthening their resilience in relation to perceived social threats and personal insults.

To our knowledge, this is the first study in which violent offenders were insulted directly during the online acquisition of neural correlates. Owing to the use of a control group, our experiment allows direct inferences about the differences in the processing of anger between criminal and noncriminal groups on a neural level. As most existing research targets specific regions of interest, usually based on the connectivity between the amygdala and the PFC regions, we sought to increase the number of regions of interest in our analyses, focusing on the entire social brain (Alcalá-López et al. 2018). We confirm that, besides the limbic-prefrontal connectivity, alterations in other neural networks are crucial for (social) information processing and that disturbances within large networks correlate with the expression of anger. Another specific advantage of this study is owed to the experimental setup, which enabled the identification of the exact circumstances under which anger escalated and the regulatory strategies collapsed in the group of offenders.

Despite the fact that we did not explicitly acquire self-report data on the participants’ perceived anger, we can infer through the behavioral results that the paradigm has effectively induced anger. The strongest indicator that the indirect anger measure reflects the participants’ perceived anger is derived from our previous studies (Panagiotidis et al. 2017; Wagels et al. 2020) in which the motor response to provocation (indirect measure of anger) was directly compared with self-report valence ratings during the paradigm (direct measure of anger), and it was found that the more intense the participants’ responses on the handheld device were, the angrier they reported being. Based on these results, we refrained from including a direct measure of anger in this experiment in order to preserve the credibility of the cover story (technical failure).

Since this is a pilot study, we must point out some limitations of both, the samples and the study design itself. First, our study suffers a small sample size. This restriction is on the one hand based on the fact that it is particularly challenging to access violent offenders who can participate in studies at research institutions. Second, we suffered a high number of dropouts in the control groups. About 30% of the participating controls mentioned a strong disbelieve in the cover story (in contrast to 10% in the offenders’ group). We will strive to reduce this number by improving the paradigm for future studies. More difficult tasks with more complex failures should possibly lead to a higher credibility of the paradigm. Third, we need to point out the differences between both groups in years of education and verbal intelligence. Both of these correlating measures are significantly decreased in the criminal group, which is considered to be characteristic for this type of sample (Hjalmarsson and Lochner 2012; Schwartz et al. 2015; Lochner 2020). Fourth, we found no significant difference in one utilized self-report aggression measure (i.e. the AQ Buss and Perry Aggression questionnaire) between offenders and controls, while they differed significantly in the RPQ. To explain this observation, one has to look at the specifics of both questionnaires. While the AQ asks how characteristic an (aggressive) trait is for the subjects, the RPQ asks whether or how often they showed (aggressive) behavior. It is reasonable to assume that the likelihood of answering untruthfully is greater when asking about personality attributes compared to being asked about actual events. One last drawback of our study sample is that it consists solely of male participants. In order to provide the greatest possible homogeneity of studied ample groups and with regard to the fact that there are remarkable differences between male and female aggression (Björkqvist 2018), we have refrained from admitting female participants to the study. Yet, additional gender research in this area is of great relevance and needs to be pursued in the future.

In addition to the shortcomings of our sample, this study suffers further limitations with regard to the design and paradigm, which with possible future modifications will be discussed in the following. First and foremost, we must state that the assessment of anger is based on indirect derivations of observed behavior and no direct self-report assessment of emotions. We decided to abstain from directly and repeatedly assess participants’ valences, for 2 main reasons. On the one hand, mood assessment after each run would have put the cover story at great risk and possibly exposed it. We already recorded high dropout rates without repeated assessment due to disbelief in the cover story, especially in the control group. A questionnaire after the completion of the entire paradigm would have only recorded the mood or aggressive tendencies after all 3 runs and an explicit assignment to the individual runs (provocation types) would have been impossible. On the other hand, assessing mood or aggressive thoughts in criminal offenders might be confounded by socially acceptable answering and should by all means be interpreted with caution.

Secondly, we must discuss the results of our study considering that we lack a paradigm-immanent control condition. Despite an (unprovoked) baseline and 2 phases of provocation serve as a basis for the analysis, it is not possible to finally determine whether the increased observed anger and its neuronal correlates are in fact a function of social interaction and not only of the time effect. To answer this question, an additional experimental condition would have to be implemented in the study in which subjects are exclusively frustrated in a nonsocial manner. However, this approach would require a much larger sample size. The purpose of this study was primarily to analyze and compare the 2 sample groups with regard to observed behavior and measured neural connectivity. Further studies could systematically compare the influence of social provocation with technical provocation to extract the value of personal insult.

Hence, despite those limitations and in order to draw a conclusion, we state that this pilot study has shown, for the first time, that violent offenders were able to employ behavior control networks between the amygdala and inferior frontal regions during frustration, but failed to maintain the strength of it during increasing provocation. One explanation for this observation might be that violent offenders are particularly sensitive to personal rejection and social threats, which might lead to significant disruptions in their cognitive and behavioral control networks.

Funding

This work was supported by the Faculty of Medicine, RWTH Aachen University (START program 130/15), the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—269953372/GRK2150, and by the Brain Imaging Facility of the Interdisciplinary Center for Clinical Research (IZKF) Aachen within the Faculty of Medicine at RWTH Aachen University.

Conflict of interest statement: None declared.

References