Healthy individuals and individuals having “anger issues” were recruited from the general population, through newspaper advertisements and word-of-mouth. All individuals provided written informed consent prior to study participation in accordance with Stony Brook University's Institutional Review Board.

All participants underwent psychiatric interviewing including the Structured Clinical Interview (Ventura et al., 1998) to assess DSM-IV Axis-I psychiatric disorders and Axis-II Cluster-B personality disorders including antisocial and borderline personality disorder (First et al., 1997). Significant behavioral features consistent with IED were assessed based on an interview (Coccaro et al., 2004; Coccaro, 2012). In line with the “Research Domain Criteria” approach for functional domains such as negative valence systems (e.g., reactivity to provocation) (Morris and Cuthbert, 2012), we included RA participants with full (IED-interview score > = 15) and subclinical IED (operationalized in the current study as an IED-interview score = 13–14) to acknowledge the continuous nature of aggressive behavior. The State-Trait Anger Expression Inventory-2 (STAXI-2, Spielberger, 1988) was used to validate the grouping of participants on trait anger and anger expression.

Eleven male RA individuals (n = 6 full IED, n = 5 subclinical IED), and 14 healthy male controls without any DSM-IV psychiatric disorder performed four runs of the fMRI-PSAP. Of those, 18 participants (9 RA, as confirmed by elevated trait anger and anger expression scores on the STAXI relative to 9 non-aggressive controls, Table 1) were included in the final fMRI analysis (see Data analysis for fMRI inclusion criteria). Groups were matched on age, race, handedness, years of education, and estimates of verbal intelligence (Reading scale, Wide Range Achievement Test-III, Wilkinson, 1993), but not on non-verbal intelligence (Matrix reasoning scale, Wechsler Abbreviated Scale of Intelligence, Wechsler, 1987) and current depression symptoms (Beck and Steer, 1984; Table 1). No group differences were observed in current, past or occasional alcohol, cigarette or marijuana use based on the Structured Clinical Interview (Pearson Chi-square tests, p > 0.10). Four RA individuals reported comorbid disorders which are common for IED (Coccaro et al., 2005; Kessler et al., 2006): full IED, current general anxiety disorder and a single/remitted major depressive episode (n = 1), remitted cannabis abuse (n = 1); subclinical IED, remitted alcohol abuse (n = 1), current & lifetime antisocial personality disorder (n = 1).

Exclusion criteria for study participation were any (a) neurological disease including seizures, history of head trauma with loss of consciousness (>30 min); (b) major medical conditions (e.g., cardiovascular, endocrinological, oncological, or autoimmune diseases); (c) major psychiatric disorder with psychosis (e.g., schizophrenia, bipolar disorder); (d) use of psychoactive medication within 6-months prior to study date; (e) positive urine screens for psychoactive drugs or their metabolites (amphetamine/methamphetamine, cocaine, phencyclidine, benzodiazepines, cannabis, opiates, barbiturates, inhalants); and (f) MRI contraindications.

Participants were told that our study examined mood, social interaction, and motor coordination (avoiding explicitly the terms “aggression” or “competition”). Participants were instructed that they and another male “person” (located in a different building) could earn as much money as possible by pressing buttons. The other person had the opportunity to subtract $1 from the participant's earnings at any time during the test and add this money to his own earnings; thereby provoking retaliatory behavior. In contrast, participants could subtract $1 from the other person but could not add this money to their earnings (thus choosing to forgo monetary gain).

Each fMRI-PSAP run lasted for 363 s including 18-s trials and 2-s inter-trial intervals. At the beginning of each trial, the accumulated earnings were presented and participants had the choice between increasing their earnings by $0.4 by pressing the A-button 50 consecutive times with their left thumb (1 monetary ratio = 50 A-button presses), or subtracting $1 from their opponent's earnings by pressing the B-button 40 consecutive times with their right thumb (1 retaliatory ratio = 40 B-button presses; see Figure 1). Within each 18-s reward or retaliation trial, participants could press the respective button chosen at the beginning of the trial as often as possible to complete as many monetary or retaliatory ratios; they could not switch buttons within a trial. Participants were unaware that the opponent was fictitious and that the subtractions of money (provocations) occurred at random intervals (6–60 s between provocations) in the monetary trials (compare, Kose et al., 2015). There were no provocations in retaliation trials.

Compared to previous neuroimaging and behavioral PSAP studies (Cherek et al., 1991; New et al., 2009; Perez-Rodriguez et al., 2012; Schluter et al., 2013; Kose et al., 2015), we adjusted the effort of obtaining a monetary reward with achieving retaliation (i.e., 50 A- vs. 40 B-button presses to complete a ratio) to keep both effort and motor-related brain activation comparable across reward and retaliation trials. In agreement with the original behavioral PSAP literature and the Kose et al. (2015) study, which used a comparable fMRI-adapted PSAP, we used the number of monetary and retaliatory responses (A- and B-button presses) as our main outcome variables as they are most meaningful in indicating how hard individuals worked to complete the respective ratios. Additionally, the “number of provocations experienced” and total “earnings” were assessed. Other PSAP aggression measures (e.g., #retaliatory responses/#provocations, or #retaliatory responses/#monetary responses) are not reported here, but yielded similar results.

Before and after the fMRI-PSAP, all participants rated their positive and negative affect using the Positive and Negative Affective Schedule (Watson et al., 1988). Moreover, we asked participants after each run how they felt about their opponent and about their strategy at the end of the PSAP. Based on these self-report measures, all participants believed in the deception, were engaged in the game, were competitive and reported to be annoyed by the subtractions of money.

In the single fMRI run, there were no group differences in number of monetary and retaliatory responses (Figure 2 left), earnings, and provocations (independent t-tests, p_s> 0.30). Across all 4 runs, the 4 × 2 repeated-measures ANOVAs revealed that RA participants pressed significantly more buttons for the monetary reward than controls [main effect of group, F_{(1, 15)} = 7.62, p = 0.015, partial η² = 0.34], but the number of retaliatory button presses did not differ between groups [F_{(1, 15)} = 0.15, p = 0.701, partial η² = 0.01; Figure 2 right]. There was no significant effect of run on either the number of monetary [F_{(3, 45)} = 1.35, p = 0.27, partial η² = 0.08] or retaliatory responses [F_{(3, 45)} = 1.38, p = 0.261, partial η² = 0.08], and no significant run × group interaction [reward: F_{(3, 45)} = 2.61, p = 0.063, partial η² = 0.15; retaliation: F_{(3, 45)} = 2.35, p = 0.085, partial η² = 0.14]. While RA individuals did not earn overall more money than controls [F_{(1, 15)} = 1.01, p = 0.33, partial η² = 0.06; no effect of run], a significant run × group interaction indicated a rise in earnings over the four runs in RA participants, and a decrease in controls [F_{(3, 45)} = 3.28, p = 0.03, partial η² = 0.18]. However, post-hoc independent t-tests did not reveal any significant difference between groups for earnings in any one of the runs (p_s > 0.05). Additionally, RA individuals experienced significantly more provocations across runs [F_{(1, 15)} = 7.53, p = 0.015, partial η² = 0.33; no effect of run or run × group interaction], possibly because RA individuals chose more often the monetary option (provocations occurred in monetary trials only).

There were no differences in positive and negative affect pre and post the fMRI-PSAP across all participants (Wilcoxon Signed Rank tests; p_s > 0.046, corrected α-level: p < 0.0025, 20 scales), and between groups (for pre and post scores separately, and pre-post difference scores, Mann Whitney-U tests; p_s > 0.11).

Relative to controls, when working for reward vs. retaliation, RA individuals exhibited increased brain acitivity in the mesocorticolimbic salience network (bilateral: anterior insula, putamen; Left: inferior frontal operculum; RA>Controls, Figure 3A, Table 2) and decreased activation in areas belonging to the default-mode network (bilateral ventro-medial [vm] PFC, right precuneus; RA<Controls, Figure 3B, Table 2). Across groups, brain responses were increased for reward vs. retaliation in the left cuneus (Reward>Retaliation) and decreased in the precuneus, the superior frontal gyrus, and the occipital cortex (Reward<Retaliation, Table 2). None of the ROI analyses revealed significant differences between groups nor between reward and retaliation.

The whole-brain voxelwise regression analysis (Table 3) revealed significant positive correlations in the left ventral/dorsal striatum and the anterior PFC (Figure 4), and significant negative correlations in the precuneus, the supramarginal gyrus, and medial and lateral PFC for reward-related brain responses (“Reward>Retaliation”) of the single fMRI run and monetary responses across all four runs. ROI analyses supported the whole-brain results in the striatum, revealing a significant positive correlation in the left ventral striatum (MNI coordinates: x = 20, y = 14, z = −10; cluster size = 7 voxels; T = 3.81; family-wise error corrected: p_peak = 0.041, p_cluster = 0.031). There was no difference in correlations between groups.

Increased activation of the anterior insula and the putamen during a single run, along with pronounced reward-seeking behavior in RA individuals across all runs likely points to enhanced salience attribution to effortful-behavior with the goal of obtaining money (Liu et al., 2011) vs. achieving retaliation. The anterior insula and putamen belong to the mesocorticolimbic salience network and are anatomically and functionally closely connected (Postuma and Dagher, 2006). Both areas respond to negative and positive rewards suggesting their role in salience- rather than value-processing (Liu et al., 2011; Bartra et al., 2013). Being part of an intrinsic connectivity network that links cognition, emotion, and interoception (Laird et al., 2011), the anterior insula plays a prominent role in salience processing in attention-demanding situations (Seeley et al., 2007; Craig, 2009, 2011; Touroutoglou et al., 2012). For the putamen, we observed the increased reward-related activation for “RA>controls” in the pre-commissural part of the dorsal striatum (Draganski et al., 2008), a region implicated in processing the salience of stimuli (Bromberg-Martin et al., 2010; Bartra et al., 2013), in stimulus-action-outcome learning, and action selection based on reward expectations (Haruno and Kawato, 2006; Balleine et al., 2007; Bromberg-Martin et al., 2010). In the current fMRI-PSAP, the contingencies of stimulus-action-outcome learning were simple; participants could either opt for the reward or the retaliation, with approximately equal effort for each option. Interestingly, whole-brain and ROI regression analyses showed that pronounced reward-related activity in the ventral/dorsal striatum (and the anterior PFC), observed in the single run, was associated with overall increased reward-seeking behavior across all runs and all participants. Our findings are consistent with previous reports that increased reactivity of the mesocorticolimbic system during the anticipation of rewards predicts the average effort invested to obtain monetary rewards (Bühler et al., 2010), and support the notion that increased reactivity of the brain's reward system might be a neural marker for impulsive RA (Buckholtz et al., 2010; Kose et al., 2015; Stahl, 2015). Thus, increased reward-related striatal activity in RA individuals likely reflects enhanced salience attribution to monetary rewards, and possibly early habit formation to seek rewards instead of retaliation in our specific experimental situation. In line with Buckholtz et al. (2010), our data indirectly support Stahl's theory that impulsive retaliatory aggression might be a “compulsive habit” by showing that RA individuals share enhanced reward sensitivity with other “compulsive” problem behaviors such as drug and behavioral addiction (Stahl, 2015). However, due to task design limitations (see below), we could not test group differences separately for reward-seeking and retaliatory behavior, which would have been interesting as both conditions may share appetitive features indicated by striatal responsivity observed for the anticipation of monetary rewards (Liu et al., 2011) and the engagement in retaliation/aggression (Krämer et al., 2011; Beyer et al., 2014; Gan et al., 2015). Thus, it remains to be elucidated if recurrent impulsive-RA develops due to a shift from ventral striatum activity triggering goal-directed behavior to more habit-like stimulus-directed behavior controlled by the dorsal striatum (Stahl, 2015), as previously proposed for compulsive drug use behavior (Everitt and Robbins, 2005).

Interestingly, we observed decreased precuneus and vmPFC activation for reward vs. retaliation in the RA group compared to controls. Down-regulation of the precuneus and vmPFC in RA individuals during reward-seeking vs. retaliation-driven behavior could indicate increased suppression of the default-mode network which is deactivated during attention-demanding cognitive tasks and active at rest (Raichle et al., 2001; Fox et al., 2005; Utevsky et al., 2014). Previously, we proposed that increased metabolic activity of the default-mode network at rest in aggressive vs. non-aggressive individuals might be a neural marker of increased self-referential processing (Alia-Klein et al., 2014). Here, we suggest that the down-regulation of the default-mode network during reward-seeking behavior in RA individuals might indicate that they were more engaged in the reward condition than controls. As previously shown for neuropsychiatric disorders (e.g., schizophrenia and biploar disorder, Whitfield-Gabrieli and Ford, 2012), there might be compromised functioning of the default-mode network in human RA.

It is important to note that the group differences observed in the mesocorticolimbic areas including the putamen and the anterior insula and in areas of the default-mode network including the precuneus and the vmPFC in the single fMRI run occurred in the absence of group differences in behavior and number of provocations in this single run. Thus, group differences in brain activation cannot be explained by an imbalance in the number of modeled events in the first-level GLMs between groups.

Pronounced reward-seeking behavior and up-regulation of the mesocorticolimbic salience network during reward-seeking behavior in RA individuals imply that the use of positive reinforcement (e.g., contingency management) could be beneficial in anger management treatments. According to a previously proposed theory on “the behavioral economics of violence” (Rachlin, 2004), violent behavior is regulated because it is costly in the long run (e.g., dropping out of school, lower education and lower income jobs; health-related problems). However, Rachlin argues that the short-term benefits (e.g., releasing anger) can outweigh the short-term costs of violent behavior (e.g., physical effort, injuries), a mechanism potentially involved in severe RA. In real life, the long-term benefits of not being aggressive (i.e., more money, stable relationships) are typically delayed and have a low salience. Thus, to reinforce the salience of those long-term benefits, anger management treatments could offer education/social support or even the opportunity to earn money as incentives for aggressive individuals to refrain from aggressive behavior.

Our findings in a small sample of RA men, of whom half reported significant behavioral features but did not meet the criteria for full IED, require caution in generalizing results to a broader population of IED patients. An important limitation of this study is that only men were studied, and thus we do not know whether similar effects would be seen in women. Future neuroimaging studies including both men and women are needed to further investigate how potential sex differences in hormones and brain structure affect the neural circuitry of reactive aggressive behavior (e.g., de Almeida et al., 2015). Nevertheless, there is reason to speculate that women scoring high on RA would show comparable patterns of enhanced reward-seeking behavior and reward-related brain responses in the fMRI-PSAP as shown for men in the current study. For instance, although men have slightly higher odds ratios to be diagnosed with IED once in their lifetime than women (Kessler et al., 2006), the lifetime prevalence for IED in women is 5.6% (compared to 9.3% in men), and experimental evidence from the behavioral PSAP suggests that women are as likely to show retaliatory behavior as men when they are provoked (Bjork et al., 1997; Dougherty et al., 1999). Moreover, the hypersensitivity of the mesocorticolimbic reward system has been observed for men as well as women with pronounced impulsive anti-social traits (Buckholtz et al., 2010).

In contrast to previous PSAP studies in borderline and alcohol-dependent patients with a propensity for disproportionate RA (New et al., 2009; Kose et al., 2015), the RA group did not differ from controls in retaliatory behavior in the current fMRI-PSAP study. Although, this might seem counterintuitive at first blush, healthy individuals were less likely to retaliate when the effort to achieve retaliation was adapted to the effort to obtain money in the PSAP (Cherek et al., 1992). A systematic manipulation of effort ratios and larger sample sizes could help to elucidate if there is an “effort threshold” needed to motivate RA individuals to choose a non-aggressive reinforcement over a retaliatory option.

In the current fMRI-PSAP, one main limitation is that we could only analyze brain responses for a single run because the number of reward and retaliation events was determined by the participants' choices and turned out to be too small in some runs to effectively model brain responses. Future fMRI-PSAP studies should consider using pre-determined rates of reward and retaliation blocks as well as provocations to maximize the number of events per experimental conditions. However, such experimental modifications might ultimately change the behavior one aims to measure with the PSAP as one unique feature of the PSAP is that individuals are free to decide in each trial if they want to earn money or want to retaliate. Here, to allow the freedom to not retaliate it might be recommended to modify the retaliation condition by giving participants the option to adjust the amount they would subtract from their opponent (from $0 to a maximum amount of about $2). Alternatively, researchers might want to increase the intensity of provocations (e.g., subtracting higher amounts of money, and display “malicious joy” by their opponent) to reach the threshold for self-report of anger and to provoke more retaliatory behavior. Moreover, including an adequate baseline (e.g., sustained button pressing for neither a reward nor retaliation) into the fMRI-PSAP would enable the separate comparison of reward and retaliation blocks. When optimizing the fMRI-PSAP in future investigations, it might also be interesting to incorporate the measurement of behavioral contrast effects (Crespi, 1942) to assess how the motivation to seek rewards is altered in RA individuals when the magnitude of the incentive is varied throughout the experiment.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.