Neural Correlates of Impulsive Aggressive Behavior in Subjects With a History of Alcohol Dependence

Abstract

Alcohol-related aggression is a complex and problematic phenomenon with profound public health consequences. We examined neural correlates potentially moderating the relationship between human aggressive behavior and chronic alcohol use. Thirteen subjects meeting DSM–IV criteria for past alcohol-dependence in remission (AD) and 13 matched healthy controls (CONT) participated in an fMRI study adapted from a laboratory model of human aggressive behavior (Point Subtraction Aggression Paradigm, or PSAP). Blood oxygen level dependent (BOLD) activation was measured during bouts of operationally defined aggressive behavior, during postprovocation periods, and during monetary-reinforced behavior. Whole brain voxelwise random-effects analyses found group differences in brain regions relevant to chronic alcohol use and aggressive behavior (e.g., emotional and behavioral control). Behaviorally, AD subjects responded on both the aggressive response and monetary response options at significantly higher rates than CONT. Whole brain voxelwise random-effects analyses revealed significant group differences in response to provocation (monetary subtractions), with CONT subjects showing greater activation in frontal and prefrontal cortex, thalamus, and hippocampus. Collapsing data across all subjects, regression analyses of postprovocation brain activation on aggressive response rate revealed significant positive regression slopes in precentral gyrus and parietal cortex; and significant negative regression slopes in orbitofrontal cortex, prefrontal cortex, caudate, thalamus, and middle temporal gyrus. In these collapsed analyses, response to provocation and aggressive behavior were associated with activation in brain regions subserving inhibitory and emotional control, sensorimotor integration, and goal directed motor activity.

Aggression is one of the most enduring, complex, and problematic forms of human social interaction. The consequences of human aggression exact a substantial toll on public health and criminal justice systems, communities, and individuals. Yet, even under conditions of intoxication or duress, most humans do not engage in excessive bouts of aggressive behavior, if they behave aggressively at all (Anderson & Bushman, 2002; Baron & Richardson, 1994; Bartol & Bartol, 2004; Cherek, Lane, & Pietras, 2003). However, some individuals repeatedly display exaggerated bursts of aggression that far exceed the form and intensity of provocation (Allen, Moeller, Rhoades, & Cherek, 1997; Bartol & Bartol, 2004; Cadoret, O’Gorman, Troughton, & Heywood, 1985; Cherek & Dougherty, 1997; Gerra et al., 1997; Reid & Gacono, 2000). Epidemiological and experimental data show that chronic alcohol dependence is related to an increased risk for assault and aggressive behavior (Boles & Miotto, 2003; Parrott & Giancola, 2006). There is a positive correlation between the quantity of alcohol consumed and the frequency of a wide variety of violent acts including sexual assault, child abuse, and homicide (Bushman, 1997; Cherek, Steinberg, Oliver, Mos, & Brain, 1987; Martin, 2001; Murdoch, Pihl, & Ross, 1990). Individuals who engage in aggressive behavior report a greater amount of alcohol consumption than those without such a history (Murdoch et al., 1990); (Leonard, Bromet, Parkinson, Day, & Ryan, 1985). By some estimates, alcohol is involved in 50% of all violent crimes and greater than 60% of intimate partner violence (Greenfeld & Henneberg, 2001; Martin, 2001; Stuart et al., 2006).

While many structural and functional imaging studies in alcohol-dependent subjects have revealed compromises in brain regions controlling affective and inhibitory processes (S. George, Rogers, & Duka, 2005; Sullivan & Pfefferbaum, 2005; Zhang et al., 2013), specific regions associated with aggression have not yet been elucidated in subjects with alcohol dependence. Several laboratory studies have demonstrated that poor response inhibition—which may be associated with aggression—is affected by both acute alcohol administration (Fillmore & Blackburn, 2002; Fillmore, Marczinski, & Bowman, 2005; Fillmore & Weafer, 2004; Marczinski & Fillmore, 2005) and chronic alcohol abuse and dependence (Endres, Donkin, & Finn, 2014; Hildebrandt, Brokate, Eling, & Lanz, 2004; Kamarajan et al., 2005; Nigg et al., 2006). Inhibitory control is subserved by limbic/thalamic and prefrontal/orbito-frontal cortical circuits, and impairments in this interconnected circuitry may underlie exaggerated aggressive responding (Kringelbach, 2005; Metzger et al., 2010).

Neuroimaging studies in individuals with alcohol dependence persistently find disruptions of function and integrity (e.g., white and gray matter volume reduction and reduced blood flow) in the orbital frontal cortex (OFC), dorsolateral prefrontal cortex (DLPFC), corpus callosum, thalamus, insula, and limbic regions (Durazzo, Gazdzinski, Yeh, & Meyerhoff, 2008; Hommer et al., 1997; Paul et al., 2008; Sullivan & Pfefferbaum, 2005; Volkow et al., 1992). Degradation of function in prefrontal cortex (PFC) following chronic alcohol use is related to impairments in attention, response inhibition, judgment, and affective processing (Fein, Bachman, Fisher, & Davenport, 1990; Oscar-Berman, Spencer, & Boren, 1990; Schulte et al., 2012; Sullivan & Pfefferbaum, 2005).

Several leading neurobiological theories suggest that aggression arises from the dysfunctional regulation of emotion in response to salient stimuli (Davidson, Putnam, & Larson, 2000; Siever, 2008). There is substantial evidence that emotional regulation is mediated by a network of brain regions involving PFC, particularly OFC and DLPFC, and connections with the limbic system, particularly the amygdala (Bechara, 2005; Davidson, Jackson, & Kalin, 2000; Kringelbach, 2005), thalamus, (Metzger et al., 2010), and insula (Potegal, 2012). Dysregulation of emotion and inhibitory control, in both healthy control and subjects with Axis II personality disorders, overlaps substantially with areas known to be affected by chronic alcohol use, such as limbic, dorsal striatum, insular cortex, OFC, PFC, and parietal cortex (Davidson, Putnam et al., 2000; Kringelbach, 2005; New et al., 2007; Raine, Lee, Yang, & Colletti, 2010; Siever, 2008).

DLPFC, OFC, and limbic brain regions are implicated in violent behavior, maladaptive decision-making, abnormal processing of emotional stimuli, and drug and alcohol addiction (Bechara, 2005; Brower & Price, 2001; Davidson, Putnam et al., 2000; Raine et al., 2010; Schoenbaum, Roesch, & Stalnaker, 2006). A review of the neurobiology of personality disorders concluded that aggression was characterized by abnormal functioning in amygdala, OFC, DLPFC, and anterior cingulate cortex (ACC) (Goodman, New, & Siever, 2004). Individuals with a criminal history of domestic violence who also met alcohol dependence criteria had lower metabolic activity in hypothalamus, thalamus, and OFC than both nonviolent alcoholics and healthy controls (D. T. George et al., 2004). History of domestic violence and alcoholism was correlated with reduced cortical and subcortical brain activity, and with reduced amygdala volumes (Zhang et al., 2013).

Delineation of the relationship between aggression and neuronal functioning can be furthered by examining brain function while subjects are engaging in operationally defined aggressive behavior under controlled laboratory conditions. The relationship between alcohol dependence and aggression has been well documented, yet few studies have examined brain function in humans during bouts of aggressive behavior during experimentally controlled conditions. Using a modified version of the Taylor Aggression Paradigm adapted for fMRI, Kramer and colleagues (Kramer, Jansma, Tempelmann, & Munte, 2007; Kramer, Riba, Richter, & Munte, 2011) examined neural correlates of human reactive aggression. Collectively, their results revealed increased activation in the middle and medial frontal gyrii, caudate, and insula during periods of provocation; and activation in the dorsal anterior cingulate cortex and insula during periods in which subjects chose to administer an aversive stimulus to a fictitious opponent (e.g., the operationally defined aggressive response). These two studies enrolled healthy normal adults; neural activation patterns related to aggression in clinical samples with documented levels of heightened aggression (i.e., alcohol use disorders) remain to be examined.

In the present study, we employed a well-validated laboratory task, the Point Subtraction Aggression Paradigm, or PSAP (Cherek, Bennett, & Grabowski, 1991; Cherek et al., 2003) that was adapted for use during fMRI. Several PSAP studies have established the validity and reliability of the task, including experiments in different laboratories and countries, in individuals with psychiatric disorders (ASPD, substance abuse), and in parolees with a history of violent behavior (Cherek et al., 2003; Cherek, Tcheremissine, Lane, & Nelson, 2006; Lane, Cherek, & Fishbein, 2000). The PSAP allows for control and manipulation of key independent variables, including the frequency of provocation. The timing and frequency of experimental events can be precisely controlled, allowing for examination of neural activity when individuals are provoked and during bouts of aggressive behavior. Based on previous work in the neurobiology of aggression and brain imaging studies of alcohol-dependent subjects, we hypothesized that, compared to control subjects, alcohol-dependent subjects would demonstrate (a) more aggressive responding on the PSAP; (b) reduced BOLD activation in frontal cortex (notably, OFC reflecting diminished activation of emotional regulatory circuitry); and (c) greater BOLD activation in the amygdala (a priori region of interest) following provocation and during aggressive responding. Additionally, exploratory analyses were conducted to examine activation patterns collapsed across groups. Given that this was the first fMRI study to utilize the PSAP methodology, the goal of the exploratory analyses was to uncover general, common patterns of activation related to the PSAP task components; specifically, provocation and aggressive responding.

Method

Procedures

This study was approved by the local IRB (Committee for the Protection of Human Subjects at UTHSC Houston) and in accordance with the Declaration of Helsinki and the APA ethical standards code of conduct. All subjects progressed through the following sequence of events: medical, psychiatric, cognitive, and drug/alcohol use screening (Days 1–2); three practice test sessions (i.e., “runs”) in a mock MRI scanning environment (Day 3); three actual fMRI test runs (Day 4). Details of each set of events are provided below.

Subjects and Screening

Male and female participants were recruited through local classified newspaper advertisements. Initial telephone screening was used to identify both individuals with a history of alcohol use and those who met control subject criteria. Potential subjects were brought into the laboratory for more extensive interviews and testing covering physical and mental health, drug and alcohol use history, blood chemistry, pregnancy test (females), and violent and criminal behavior. Exclusionary criteria included (a) current or past medical problems (e.g., seizures, diabetes, high blood pressure, renal or cardiovascular disease), (b) current use of any medications, (c) current illicit drug use or alcohol use (measured by daily urinalysis and breath alcohol testing), and (d) current or past history of an Axis I mood or psychotic disorder, as determined by the Structured Clinical Interview for the DSM–IV–TR using the SCID-I, version 2.0 (First, Gibbon, Spitzer, & Williams, 2002). All subjects with past alcohol dependence (N = 13) met DSM–IV criteria for alcohol dependence within the past 24 months, and were in early full remission, early partial remission, or sustained full remission. Control subjects (N = 13) did not met DSM–IV criteria for any current or past substance use disorder or psychiatric disorder.

At intake, subjects read and signed an informed consent document. Subjects were provided information about urine drug testing, breath alcohol testing, psychiatric evaluation, experimental procedures and compensation. Subjects were informed that they could earn money for participation and that bonuses would be awarded for drug-free breath and urine samples, arriving on time and completing the study. After consent, subjects provided urine samples for drug screen analysis using a one-step drug screen test card (Innovacon, Inc.), which tested for cocaine, stimulants, opiates, benzodiazepines, and marijuana. Temperature monitoring and creatinine level determinations were used to detect attempts to alter urine samples. Subjects also provided breath alcohol samples using an Alco-Sensor III (Intoximeters, Inc.), which required subjects to expire air for 10 seconds to measure alcohol content. No differences were observed in subject demographics, including gender [Fisher’s exact = 0.41, ns]; age [t(24) = 1.29, ns]; education level [t(24) = 0.51, ns]; number of smokers [6 in each group]; level of tobacco use [Mann–Whitney U, z = 0.78, ns]; Shipley scale of cognitive aptitude (Zachary, 1986) [t(24) = 0.75, ns]; or lifetime use of other drugs [Fisher’s exact = 0.24, ns]. Demographic information on the two groups is provided in Table 1.

Table 1.

Measures of Central Tendency and Variance on Demographic and Behavioral (PSAP) Variables for the Group of Past Alcohol-Dependent Subjects and the Group of Control Subjects

Variable	Group	Mean/N	Standard error	Between-group differences
Age (years)	Alcohol	37.77	1.98	nsa
	Control	33.19	2.91
Gender (count)	Alcohol	10 M, 3 F	—	nsb
	Control	7 M, 6 F	—
Education level (years)	Alcohol	13.85	0.59	nsa
	Control	14.3	0.67
Cognitive ability (Shipley)	Alcohol	54.31	2.87	nsa
	Control	57.31	2.82
Smoking level (cigarettes/wk)	Alcohol	5.46	2.49	nsc
	Control	1.25	2.3
Other drug use Lifetime (count)	Alcohol	9	—	nsb
	Control	5	—
Aggressive responses	Alcohol	157.01	17.55	0.023d
	Control	84.96	6.07
Aggressive responses/provocation	Alcohol	35	6.72	0.036d
	Control	17.41	1.35
Monetary responses	Alcohol	1014.64	151.75	0.032d
	Control	418.83	10.26
Earnings on PSAP (U.S. Dollars)	Alcohol	$8.42	0.7	nsd
	Control	$6.95	0.82
Provocations experienced	Alcohol	4.1	0.18	nsd
	Control	4.34	0.19

^aStudent’s t-test.

^bFisher’s exact test.

^cMann-Whitney U test.

^dAnalysis of variance.

Subjects provided breath and urine samples each day upon arrival, and were removed from the study if they provided two consecutive alcohol-positive breath or drug-positive urine samples. Two subjects were excluded from the study for consecutive positive THC samples. All other subjects were free of illicit drugs and alcohol on all testing days. At the end of each day, subjects were paid in cash the total amount earned for the day and those using public transportation were given bus passes.

Aggression Testing (PSAP)

Subjects completed a version of the Point Subtraction Aggression Paradigm, hereafter PSAP, (Cherek et al., 1991), a well-validated laboratory measure of human aggression that was adapted for fMRI. The paradigm utilized a computer-simulated social interaction in which subjects were paired with a fictitious other person, as established by instructional deception. Before the first test session, subjects were shown a diagram of the computer monitor and response panel and were read instructions (Cherek & Lane, 1999). There was no use of terms such as aggression, game, competition, or anything that might indicate the behavior of interest. Consistent with previous PSAP studies, participants were told that the object of the task was to earn money, and the object of the study was to measure mood and motor responding. During the task, subjects had two response options available, labeled A or B on the computer screen. The money-earning option (option A) added $2.00 to the subject’s earnings counter (shown near the top of the screen) after a fixed response ratio of 40 responses was completed. The aggressive option (option B) ostensibly subtracted $2.00 from the “other” person, at no gain to the subject, after a fixed ratio of 10 responses. At random intervals throughout the task, subjects were provoked via $2.00 subtractions from their earnings, and instructed that when the “other” person chose the B option s/he kept the $2.00 subtraction (providing an ostensible reason for subtractions). No provocations occurred when the counter was at $0. All provocations occurred at randomly scheduled intervals that occurred on average every 90 sec, and were scheduled only during bouts of pressing on the monetary option; this allowed independent assessment of neural activation patterns (a) during provocation prior to the initiation of aggressive responding, and (b) during aggressive responding. Switching to option B (aggressive) was only permitted after the completion of a ratio. However, there were no restrictions on the initiation of consecutive monetary ratios or aggressive ratios. Provocations did not occur during aggressive ratios. The time required to signal a provocation and subtract money from the counter was too short to allow provocations to be distributed throughout aggressive ratios (which lasted for only 10 paced responses). Provocations would have had to be programmed at nearly the same point during every aggressive interval. Provocations were assessed temporally independent of aggressive responding. This established a necessary dependency between monetary responding and provocation. However, the dependency between provocation and monetary responding was addressed in the analyses by contrasting periods of postprovocation monetary responding with periods of monetary responding in which provocations did not occur.

To complete the 40 responses on A, subjects were instructed to press the monetary (A) button when a blue dot appeared in the middle of the screen. The dot flashed on for 500 ms and off for 500 ms at a rate of 1 Hz. The function of the flashing dot was, as much as possible, to control the rate of motor output (button pressing) across subjects. Only responses made when the dot was on the screen counted toward the completion of monetary and aggressive ratios on the subject’s counter. All responses were recorded and analyzed in the behavioral data analyses; on-dot responses were utilized in the fMRI analyses to normalize across subjects. At the end of each day, subjects were given a questionnaire designed to assess the instructional deception, asking questions regarding how many individuals the subjects had been paired with, who had subtracted more money, and what strategies each person used. Any indication in the questionnaire that the subjects did not believe the deception, for example, “I think I was playing a computer,” would have resulted in the removal of the subject from the analysis. All 26 subjects reported being paired with other people during the task. The subtraction of money by the subject from the (fictitious) other person’s counter meets an established operational definition of aggression: the presentation of an aversive stimulus to another person (or organism) who would act to avoid it (Boles & Miotto, 2003). Subjects completed three 10-min PSAP practice sessions (“runs”) in a mock scanner on a separate test day preceding the actual fMRI test day. The mock scanner was provided by Philips and was identical to the Philips 3.0 T Intera system, with the exception of the electromagnetic fields. These practice sessions served to habituate the subjects to the fMRI environment (bore, head coil, noises) and familiarize them with the PSAP protocol and stabilize responding prior to the fMRI testing. On a subsequent day, subjects completed three 10-min PSAP fMRI test runs in the actual Philips 3.0 T Intera system.

fMRI Protocol

All subjects underwent scanning on the Philips 3.0 T system with SENSE head coil (Philips Medical Systems, Best, Netherlands). The Eloquence upgrade to Integrated Functional Imaging System-Stand Alone system (In vivo Corporation, Orlando, FL) was used for stimulus presentation and recording of performance data. Spin-echo Echo Planar Imaging (EPI) fMRI with a pulse sequence sensitive to the blood oxygen level dependent (BOLD) effect at 3.0 T was used in order to avoid signal losses caused by through-slice dephasing in regions near tissue/air interface (e.g., medial orbitofrontal cortex) (Kruger, Kastrup, & Glover, 2001; Norris, Zysset, Mildner, & Wiggins, 2002; Wang et al., 2004). Spin-echo EPI is sufficiently sensitive at 3.0 T to the BOLD effect to detect significant activation in cognitive fMRI studies (Ma et al., 2012; Moeller et al., 2012; Norris et al., 2002). The fMRI images were acquired in the transverse plane using single shot spin-echo EPI with SENSE acceleration factor 2.0, repetition time (TR) 2200 ms, echo time (TE) 75 ms, flip angle 90 degrees, number of slices 22, in-plane resolution 3.75 mm × 3.75 mm, slice thickness 3.75 mm, gap between slices 1.25 mm, 290 whole brain dynamic volumes per run, and run duration of 10 min 38 s. Each subject completed three runs with approximately 5 minutes rest between runs. A high-resolution T-1 weighted 3D-MPRAGE structural scan (0.94 mm × 0.94 mm × 0.94 mm) was acquired for coregistration with the fMRI scans.

Behavioral Data Analyses

All three PSAP fMRI runs were included in the analyses of the behavioral data. The primary dependent measures of interest were: total aggressive (B) responses, total monetary (A) responses, the ratio of aggressive responses to provocations (aggressive per provocation), number of provocations experienced, and total money earned. Each of these variables was examined using ANOVA models with a between-groups factor (Alcohol vs. Control) and a within-subject factor (PSAP Sessions 1–3).

fMRI Data Preprocessing and Analyses

Image preprocessing and analyses of BOLD activation included only PSAP sessions in which valid fMRI data were obtained, for example, free of artifacts or excessive head motion (details below), and with sufficient numbers of provocations and aggressive responses.

For each subject, one run was chosen to be included in the fMRI analysis based on the following criteria: the run must contain at least two monetary ratios, at least two aggressive ratios, and at least two provocation stimuli; the run must contain a more even spread over time for the occurrence of each of these conditions compared to other runs for that subject; the data from that run must be free of artifacts or excessive head motion. The purpose of the criterion for evenness of spread over time for the occurrence of each condition within a run was to reduce the influence of biased sampling of fluctuations in the BOLD signal due to low frequency noise. Evenness of spread over time for the occurrence of each condition was determined by visual inspection by coauthor JLS of the graph of each condition over time for the run. In order to judge evenness of spread, the time course of the run was visually estimated to be divided into the same number of equal subdivisions over time as the number of occurrences of that condition. When two runs (R1 and R2) had the same number of occurrences of a particular condition, R1 was judged to have more even spread of the condition than R2 for that subject if each occurrence of that condition in R1 fell into a different subdivision, whereas two or more occurrences of that condition in R2 fell into the same subdivision. In the case of a tie, then R1 was chosen over R2 if R1 showed less distance than R2 between the center of each occurrence of the condition and the center of each subdivision into which that occurrence fell. For some subjects, only one run out of three met all these criteria (but all had at least one run that met all criteria). Thus, we included only one run from every subject so as not to bias the results by including different numbers of runs across subjects. All decisions regarding which runs to include for each subject were made prior to conducting formal SPM analyses of BOLD activation on fMRI data for that subject. The quality criteria for the selection of blocks were specified before any analysis of the data allowing observation of results.

Preprocessing of the fMRI data for all subjects was conducted using SPM5 software (Wellcome Department of Cognitive Neurology, London, U.K.), implemented in Matlab (Mathworks Inc. Sherborn MA). After slice-timing correction, the fMRI series were realigned in order to correct head motion. Runs with head motion greater than 3.0 mm or 3.0 degrees of rotation were eliminated from the fMRI analysis. The 3D-MPRAGE image for each subject was coregistered to the mean fMRI image and then transformed to Montreal Neurological Institute atlas coordinates, MNI (Mazziotta et al., 2001) using the SPM5 Normalize module. This transformation was applied to the fMRI images to convert them to MNI coordinates and 2 mm isotropic voxels, and the fMRI images were then spatially smoothed with a Gaussian filter of 8 mm isotropic full width at half maximum.

Because of the fact that the fMRI data acquisition and preprocessing of this study began before the release of SPM8 software, preprocessing for all subjects was conducted using SPM5 for consistency. After all fMRI data were acquired and preprocessed for all subjects, the SPM8 upgrade to SPM5 was available, and thus the statistical analysis of the fMRI data was conducted using SPM8 for all subjects, since the preprocessed fMRI image files from SPM5 that are input into the statistical analysis are completely compatible with SPM8. At the first level of SPM8 analysis, each subject’s unique period of occurrence for each behavioral condition (i.e., monetary responding without any provocation stimuli, monetary responding after provocation but prior to the next ratio, and aggressive responding) were modeled by boxcar functions convolved with the SPM8 canonical hemodynamic response function. The fMRI time series was high-pass filtered with a cut-off period of 384 sec, determined by the Fourier transformation of each condition’s time model, in order to avoid cutting off the experimental signal. The proposed cut-off period for the high-pass filter was converted into an equivalent cut-off frequency using the standard formula: frequency (Hz) = 1/period (in seconds). The SPM-provided power versus frequency graph of the Fourier transformation was then inspected (using the time periods of each response type) to verify that the power from the experimental responding periods occurred at a frequency that was greater than the proposed cut-off frequency.

The parameters for each condition were estimated using the General Linear Model (Friston, Frith, Frackowiak, & Turner, 1995) at each voxel without global normalization. Based on these parameter estimates, the following contrast images were computed: (1) aggressive-monetary, defined as the parameter estimate for the BOLD signal during aggressive responding minus the parameter estimate for the BOLD signal during nonprovocation monetary responding (e.g., monetary responses that occurred within a 40-response block/ratio with no provocations). (2) provocation-monetary, defined as the parameter estimate for the BOLD signal during postprovocation monetary presses minus the parameter estimate for the BOLD signal during nonprovocation monetary presses. Both of these summary contrast images were output for each subject for entry into the SPM8 second-level analysis.

All second-level statistical analyses utilized SPM8 Random Effects models, with two-tailed family wise error (FWE) corrected cluster-level p < .05. Part 1 was the comparison of the provocation-monetary contrast between groups. Part 2 was the comparison of the aggressive-monetary contrast between groups. Part 3 was a whole brain voxelwise comparison between the two groups of the regression of PSAP BOLD activation on aggressive rate. The independent variable in this regression analysis was the behavioral measure aggressive rate (defined as the number of aggressive button presses divided by the number of monetary button presses during the fMRI run, in order to control for the overall response rate). The dependent variable in this regression analysis was the provocation-monetary activation. Part 4 was a whole brain voxelwise comparison between the two groups of the regression of aggression-monetary activation (as the dependent variable) on aggressive rate (as the independent variable). For the provocation-monetary and aggressive-monetary regression models (Parts 3 and 4), we observed only a few brain areas with significant differences in regression slopes between groups. Therefore, in Parts 5 and 6, both groups were combined (26 subjects) in whole brain voxelwise analyses. Part 5 examined brain regions that showed significant regression of provocation-monetary activation on aggressive rate for all subjects combined. Part 6 examined the regression of aggression-monetary activation on aggressive rate for all subjects combined. Similarly, for the between-groups comparisons of the provocation-monetary and aggressive-monetary activation contrasts in Parts 1 and 2, we observed only a few brain areas with significant differences between groups. Therefore, in Parts 7 and 8, both groups were combined (26 subjects) in whole brain voxelwise analyses. Part 7 examined the provocation-monetary activation contrast for all subjects combined. Part 8 examined the aggression-monetary activation contrast for all subjects combined.

The goal in Parts 5, 6, 7, and 8 was to examine the main effects in both groups combined. However, the results from Part 3 revealed a few brain regions in which the regression slopes were significantly different between the two groups, that is, there was a significant interaction of group × aggressive rate in those regions. Because the presence of significant interaction effects may confound the interpretation of main effects (Winer, Brown, & Michels, 1991; Rosnow & Rosenthal, 1989), in order to examine the overall main effects of regression slope for both groups combined in Part 5, we excluded (i.e., “masked-out”) those brain regions where the regression slopes had been found to be significantly different in Part 3. This approach is conservative, but proper uncoupling of interactions (e.g., Rosnow & Rosenthal, 1989) could further complicate an already complex level of analysis in the combined group analyses. There was a similar rationale for masking out the significant differences between groups from Part 1 when analyzing the main effects for the combined sample in Part 7. This masking-out procedure was not necessary for the combined group analyses in Parts 6 and 8 because there were no regions in Parts 4 and 2, respectively, which were significantly different between the two groups.

Finally, in Part 9, region of interest analyses were conducted based on the study hypotheses that provocation-monetary and aggressive-monetary activation would be positively related to aggressive rate in the amygdala. These hypotheses were tested for the regression of provocation-monetary and aggressive-monetary activation on aggressive rate within the left and right amygdala as a priori regions of interest (ROI) using SPM8 random effects small volume correction. The a priori left and right amygdala ROIs in MNI standard atlas coordinates were determined by the WFU PickAtlas toolbox (Maldjian, Laurienti, Kraft, & Burdette, 2003) implementation of Anatomical Automatic Labeling atlas (Tzourio-Mazoyer et al., 2002). Approximate anatomical locations of activated regions also used the Anatomical Automatic Labeling tool-box (Tzourio-Mazoyer et al., 2002).

Results

Behavioral Data

Table 1 provides measures of central tendency and variance for the demographic and PSAP behavioral data for each group. In the PSAP behavioral analyses, none of the ANOVA models produced a significant group × session interaction. Additionally, there were no significant main effects of session for any of the measures. Accordingly, only main effects of group will be described. Alcohol dependent subjects made significantly more total aggressive responses than controls across the three PSAP sessions, F(1, 48) = 5.90, p = .023; significantly more total monetary responses, F(1, 48) = 5.20, p = .032; and significantly more aggressive responses per provocation, F(1, 48) = 4.92, p = .036. Importantly, the two groups did not differ significantly in number of provocations experienced, F(1, 48) = 0.52, ns; or amount of money earned, F(1, 48) = 0.93, ns, suggesting that differences in aggressive and monetary responding were unlikely due to differential provocation levels or differential motivation for earnings.

fMRI Data

Part 1

The comparison of the provocation-monetary contrast between groups focused on brain activation following provocation (monetary subtraction) while subjects were still completing a monetary earning ratio and immediately prior to the opportunity to switch to the aggressive response option. The control group had significantly greater mean provocation-monetary activation than the alcohol group in two significant clusters. Cluster 1 had cluster extent of 1,211 voxels (FWE-corrected two-tailed cluster p < .002). Cluster 2 had cluster extent of 654 voxels (FWE-corrected two-tailed cluster p = .010). For Cluster 1, 34.10% of the voxels in the cluster were located in the left precentral gyrus, 16.52% in the left middle frontal gyrus (i.e., part of DLPFC), 13.96% in the left inferior frontal gyrus pars opercularis, 13.13% in the left postcentral gyrus, 11.40% in the left inferior frontal gyrus pars triangularis, and there were several other regions that each contained less than 10% of the cluster’s voxels. For Cluster 2, 41.28% of the cluster’s voxels were located in the right thalamus, 13.3% in the right hippocampus, 38.99% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. Table 2 presents the details of the statistical findings for this comparison. These clusters are depicted in Figure 1. There were no significant clusters in which alcohol subjects had greater provocation-monetary activation than control subjects.

Table 2.

Between-Group Comparison of Provocation-Monetary Activation (Provocation—Monetary Responding). Within Each Significant Cluster, the Three Relative Maximal Voxel t Values and Their Approximate Anatomical Locations Within 3-mm Radius Are Listed

Contrast	Cluster label	2-tailed pa	Mean differenceb	No. of voxels in cluster	t-valuec	X Y Zd	Regione
CONT > ALC	1	<.002	0.256 (0.078)	1,211	4.62	−48–24 28	L postcentral G
					4.25	−38–2 46	L precentral G
					4.16	−52–16 28	L postcentral G
	2	.01	0.317 (0.117)	654	4.04	24–30–2	R thalamus
					3.71	18–28–4	R thalamus
					3.65	30–40 10	Unlabeled
ALC > CONT		>.050	No signif clusters

^aFWE-corrected cluster p (residual field smoothness FWHM = [10.6 10.5 9.3] mm; search volume = 127,628 voxels = 908.2 resels; cluster defining threshold voxel t = 2.50).

^bMean difference between groups across all voxels in cluster; % whole brain BOLD (±90% confidence interval [CI]).

^cRelative maximal voxel t-value (degrees of freedom = 24).

^dX, Y, and Z = MNI standard atlas coordinates (mm).

^eL = left hemisphere; R = right hemisphere; G = gyrus.

Figure 1.

The main image is a montage of slices through the complete extent of two statistically significant clusters described in the text and in Table 2. Cluster 1 is shown in red; Cluster 2 in blue. The colored clusters depict regions in which the past alcohol-dependent group showed significantly less activation compared to the control group for the provocation-monetary contrast (post-provocation minus monetary earning). The left hemisphere of each brain slice is on the reader’s right-hand side. The slices progress from inferior to superior, starting with the most inferior in the upper right corner. The red and blue colors are arbitrary, and do not signify the extent or direction of activation. Both colors represent significant activation at the two-tailed FWE-corrected cluster p ≤ .01. The montage colors are overlaid on the MNI_152 T1 template in gray. The lower right-hand corner provides a cut-out view of portions of the same two significant clusters, with a portion of Cluster 1 shown in red, and a portion of Cluster 2 shown in blue. These colors are overlaid on the SPM8 MNI single-subject T1 template image in gray. The color labeling and significance level of the cut out is the same as in the slice montage.

Part 2

For the comparison of the aggressive-monetary contrast between groups, there were no regions of significantly different activation, either for Control > Alcohol or for Alcohol > Control.

Part 3

For the whole brain voxelwise regression analysis comparing between groups the regression slopes of provocation-monetary activation on aggressive rate (aggressive rate = aggressive responses per monetary responses), the control group had significantly greater regression slope than the alcohol group in one cluster (Control > Alcohol; two-tailed FWE corrected cluster p = .002; cluster extent = 941 voxels), in which 31.01% of the voxels were located in the left caudate, 11.16% in the right caudate, 52.39% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. Table 3 presents the details of the statistical findings for this comparison. There were no regions in the Part 3 analysis in which the alcohol group had statistically significant greater regression slopes than the control group.

Table 3.

Between-Group Comparison of the Regression Slopes of Provocation-Monetary Activation on Aggressive Rate (Aggressive Responses per Monetary Responses). Within Each Significant Cluster, the Three Relative Maximal Voxel t Values and Their Approximate Anatomical Locations Within 3-mm Radius Are Listed

Contrast	Cluster label	2-tailed pa	Mean differenceb	No. of voxels in cluster	t-valuec	X Y Zd	Regione
CONT > ALC	1	.002	4.119 (1.399)	941	4.54	2 12 2	R caudate
					4.35	− 12 20 0	L caudate
					4.35	0 0 16	Unlabeled
ALC > CONT		>.05	No signif clusters

^aFWE-corrected cluster p (smoothness of residual field FWHM = [10.9 10.8 9.5] mm; search volume = 127,628 voxels = 830.9 resels; cluster-defining threshold voxel t = 2.50).

^bMean difference in regression coefficient between groups, across all voxels in cluster (±90% confidence interval [CI]).

^cRelative maximal voxel t-value (degrees of freedom = 24).

^dX, Y, and Z = MNI standard atlas coordinates (mm).

^eL = left hemisphere; R = right hemisphere.

Part 4

For the comparison between groups of the regression slopes of aggressive-monetary activation on aggressive rate, there were no regions of significantly different regression slopes, either for Control > Alcohol or for Alcohol > Control.

Part 5

This whole brain voxelwise analysis examined the regression of provocation-monetary activation on aggressive rate (aggressive rate = aggressive responses per monetary responses) in both groups combined (N = 26 subjects), after excluding those brain regions that showed significantly different regression slopes between groups in Part 3. This combined group analysis showed significantly positive regression slopes in one cluster (two-tailed FWE corrected cluster p = .010; cluster extent = 679 voxels), in which 51.99% of the voxels were located in the left postcentral gyrus, 13.55% in the left supramarginal gyrus, 11.93% in the left precentral gyrus, and there were several other regions that each contained less than 10% of the cluster’s voxels. Significantly negative regression slopes were found in four clusters. For the first negative-slope cluster (two-tailed FWE corrected cluster p < .002; cluster extent = 1435 voxels), 24.39% of the voxels were located in the left inferior frontal gyrus pars orbitalis (i.e., a part of the lateral OFC), 12.13% in the left gyrus rectus, 11.01% in the right gyrus rectus, 11.85% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. For the second cluster (two-tailed FWE corrected cluster p = .004; cluster extent = 822 voxels), 17.27% of the voxels were located in the right inferior frontal gyrus pars opercularis, 13.50% in the left middle cingulate gyrus, 57.91% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. For the third cluster (two-tailed FWE corrected cluster p = .004; cluster extent = 766 voxels), 21.02% of the voxels were located in the left caudate, 19.58% in the left thalamus, 43.34% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. For the fourth cluster (two-tailed FWE corrected cluster p = .016; cluster extent = 639 voxels), 35.52% of the voxels were located in the left middle temporal gyrus, 53.36% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. Details of the Part 5 statistical results are presented in Table 4. There were no significant positive or negative regression slopes for the combined group within the brain regions that showed significantly different regression slopes between groups in Part 3.

Table 4.

Combined Group Regression Analysis of Provocation-Monetary Activation on Aggressive Rate (Aggressive Responses per Monetary Responses). Within Each Significant Cluster, the Three Relative Maximal Voxel t Values and Their Approximate Anatomical Locations Within 3-mm Radius Are Listed

Slope direction	Cluster label	2-tailed pa	Meanb	No. of voxels in cluster	t-valuec	X Y Zd	Regione
Positive	1	.01	0.825 (0.263)	679	5.13	−56–32–28	L inf temporal G
					5.01	−52–8 16	L postcentral G
					4.70	−38–22 50	L postcentral G
Negative	1	<.002	−2.555 (0.881)	1,435	5.75	28 22–20	R inf frontal G (pars orbitalis)
					5.38	−28 36–16	L mid orbital frontal G
					4.98	−20 32–22	L mid orbital frontal G
	2	.004	−0.865 (0.190)	822	7.20	28 8 28	R inf frontal G (pars opercularis)
					5.65	40 18 28	R inf frontal G (pars triangularis)
					5.48	−6 0 32	L mid cingulate G
	3	.004	−1.379 (0.507)	766	6.61	−18 4 18	L caudate
					4.56	−10–10 16	L thalamus
					3.67	−20–8 6	Unlabeled
	4	.016	−1.027 (0.267)	639	6.85	−50–34–8	L mid temporal G
					5.31	−42–44–4	Unlabeled
					3.63	−54–54–10	L inf temporal G

^aFWE-corrected cluster p (smoothness of residual field FWHM = [10.8 10.7 9.5] mm; search volume = 126,687 voxels = 849.4 resels; cluster-defining threshold voxel t = 2.50).

^bMean regression coefficient, across all voxels in cluster (±90% confidence interval [CI]).

^ct-value for relative maximal voxel (degrees of freedom = 24).

^dX, Y, and Z = MNI standard atlas coordinates (mm).

^eL = left hemisphere; R = right hemisphere; G = gyrus; inf = inferior; mid = middle.

Part 6

This whole brain voxelwise analysis examined the regression of aggressive-monetary activation on aggressive rate in both groups combined (N = 26 subjects). There were no significant clusters that showed either positive or negative regression slopes.

Part 7

This whole brain voxelwise analysis examined the contrast of provocation-monetary activation in both groups combined (N = 26 subjects), after excluding those brain regions that were significantly different between groups in Part 1. The analysis revealed a significant cluster where the BOLD signal following provocation (during monetary responding) was less than the BOLD signal during nonprovocation monetary responding. This cluster had a cluster extent of 1101 voxels (FWE-corrected two-tailed cluster p < .002). For this cluster, 13.99% of the voxels in this cluster were located in the right parahippocampal gyrus, 46.05% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. For both groups combined, there were no significant regions in which the BOLD signal following provocation was greater than the BOLD signal during nonprovocation monetary responding. Details of the Part 7 statistical results are presented in Table 5. There were no significant main effects of the provocation-monetary contrast for the combined group within the brain regions that were significantly different between groups in Part 1, except for 17 voxels in an unlabled area of posterior temporal lobe, where the BOLD signal following provocation (during monetary responding) was less than the BOLD signal during nonprovocation monetary responding.

Table 5.

Analysis of Provocation vs. Monetary Activation (i.e., Provocation vs. Nonprovocation Monetary Responding) for Both Groups Combined. Within Each Significant Cluster, the Three Relative Maximal Voxel t Values and Their Approximate Anatomical Locations Within 3-mm Radius Are Listed

Contrast	Cluster label	2-tailed pa	Meanb	No. of voxels in cluster	t-valuec	X Y Zd	Regione
Positive (provocation > monetary)	NS	>.050					No signifcant clusters
Negative (provocation < monetary)	1	<.002	−0.1801 (0.0633)	1,101	4.19	26–34–12	R parahippocampal G
					4.10	42–38–8	Unlabeled
					3.88	44–46 4	R sup temporal G

^aFWE-corrected cluster p (smoothness of residual field FWHM = [11.3 11.2 9.9] mm; search volume = 125,763 voxels = 733.1 resels; cluster-defining threshold voxel t = 2.50).

^bMean difference across all voxels in cluster (±90% confidence interval [CI]) [% whole brain BOLD].

^ct-value for relative maximal voxel (degrees of freedom = 25).

^dX, Y, and Z = MNI standard atlas coordinates (mm).

^eR = right hemisphere; G = gyrus; sup = superior.

Part 8

This whole brain voxelwise analysis examined the contrast of aggressive-monetary activation in both groups combined (N = 26 subjects). The analysis revealed a significant cluster where the BOLD signal during aggressive responding was less than the BOLD signal during nonprovocation monetary responding. This cluster had a cluster extent of 1193 voxels (FWE-corrected two-tailed cluster p < .002). For this cluster, 21.21% of the voxels were located in the left anterior cingulate gyrus, 16.18% in the left medial orbital frontal gyrus, 13.50% in the right medial orbital frontal gyrus, 10.81% in the right anterior cingulate gyrus, 10.23% in the left rectus gyrus, 12.15% were unlabeled, and there were several other regions that each contained less than 10% of the cluster’s voxels. For both groups combined, there were no significant regions in which the BOLD signal during aggressive responding was greater than the BOLD signal during nonprovocation monetary responding. Details of the Part 8 statistical results are presented in Table 6.

Table 6.

Analysis of Aggressive vs. Monetary Activation (i.e., Aggressive Responding vs. Nonprovocation Monetary Responding) for Both Groups Combined. Within Each Significant Cluster, the Three Relative Maximal Voxel t Values and Their Approximate Anatomical Locations Within 3-mm Radius Are Listed

Contrast	Cluster label	2-tailed pa	Meanb	No. of voxels in cluster	t-valuec	X Y Zd	Regione
Aggressive > Monetary	NS	>.050					No signifcant clusters
Aggressive < Monetary	1	<.002	−0.3084 (0.1209)	1193	5.27	−10 46–10	L medial orbital frontal G
					4.28	−4 32 8	L ant cingulate G
					4.06	14 44–12	R medial orbital frontal G

^aFWE-corrected cluster p (smoothness of residual field FWHM = [11.0 10.9 9.8] mm; search volume = 127,628 voxels = 802.7 resels; cluster-defining threshold voxel t = 2.50).

^bMean difference across all voxels in cluster (±90% confidence interval [CI]) [% whole brain BOLD].

^ct-value for relative maximal voxel (degrees of freedom = 25).

^dX, Y, and Z = MNI standard atlas coordinates (mm).

^eL = left hemisphere; R = right hemisphere; G = gyrus; ant = anterior.

Part 9

Finally, a priori planned ROI analyses using SPM8 small volume correction within the left and right amygdala showed that the Alcohol Group had a trend level greater regression slope compared to the Control Group for the regression of provocation-monetary activation on aggressive rate (Left amygdala: 1-tailed FWE-corrected cluster p = .089, cluster extent 6 voxels. Right amygdala: 1-tailed FWE-corrected cluster p = .091, cluster extent = 5 voxels). The Alcohol Group did not have a significantly or trend greater regression slope compared to the Control Group for the regression of aggressive-monetary activation on aggressive rate (1-tailed FWE-corrected cluster p > .10 for left amygdala and right amygdala).

Discussion

In experiments measuring aggression, most healthy adults respond at levels of retaliation equal to or less than the level of provocation, which is in line with real world behavior patterns (Baron & Richardson, 1994; Bartol & Bartol, 2004; Cherek & Dougherty, 1997; Cherek et al., 2003). In contrast, individuals with substance use disorders (and comorbid conditions such as ASPD, borderline, and bipolar disorders), often have problems with emotional control and impulse control, and demonstrate levels of retaliation well in excess of the level of provocation. We have replicated this observation in many prior studies using the PSAP (Alcorn et al., 2013; Allen et al., 1997; Cherek & Lane, 1999; Cherek, Schnapp, Moeller, & Dougherty, 1996; Moeller, Dougherty, Lane, Steinberg, & Cherek, 1998; Nouvion, Cherek, Lane, Tcheremissine, & Lieving, 2007). Here we utilized the PSAP with fMRI to better understand the neural correlates of reactive aggression in individuals with past alcohol dependence.

Alcohol-dependent subjects produced more aggressive responses per provocation than control subjects. This result, despite no group differences in provocations experienced, is consistent with heightened reactivity and reactive aggression on the PSAP. Because provocations are randomly scheduled and can vary in number across sessions (here, fMRI runs), aggressive responses per provocation can be the most sensitive measure of reactive aggression on the PSAP (Gowin, Swann, Moeller, & Lane, 2010). Alcohol-dependent subjects also made significantly more monetary responses. Subjects were instructed to respond only when the flashing dot was on the screen (in order to pace responding and control response rates), and only these responses counted toward the completion of response ratios to earn or subtract money. Thus, rates of monetary earnings and subtractions were similar even though alcohol-dependent subjects made more overall responses on both the monetary and aggressive options. This could reflect diminished inhibitory control over responding, greater problems organizing patterned motor movements, poor adherence to instructional control, or some combination thereof. In any case, this outcome leaves the specificity of the aggressive response unclear in the present context. In non-fMRI PSAP experiments in which the flashing dot is not present, subjects typically respond at equivalently high rates on the monetary option, making higher rates on the aggressive option operationally and conceptually clear (Lane et al., 2000). Further refinement of the response properties of the PSAP under fMRI conditions may aid interpretation.

Irrespective of overall response rates, analysis of BOLD activation following provocation revealed between-groups differences cortically in postcentral, middle frontal (including DLPFC), and inferior frontal gyrii, and subcortically in the thalamus and hippocampus. In all regions, subjects with past alcohol dependence showed less activation than controls. Broadly, these regions’ most well established functions include visual motor processing, and planning and coordinating complex movements (Meier, Aflalo, Kastner, & Graziano, 2008; Surmeier, Plotkin, & Shen, 2009). Thus the results may indicate neural differences in the alcohol group related to executing the externally paced responding (repetitive finger movements) after being provoked compared to not being provoked. However, the middle frontal gyrus is also implicated in go/no-go inhibitory control, decision making, and emotional regulation (Balleine, Delgado, & Hikosaka, 2007; Paulus, 2007; Potegal, 2012; Sabatinelli et al., 2011; Talati & Hirsch, 2005; Tomasi & Volkow, 2013)—functions more closely aligned with the psychological elements and operationally defined response functions of the PSAP (e.g., earning money, responding to provocations, making retaliatory aggressive responses).

In addition to the direct influence of the PSAP on brain activation, moderate and chronic alcohol consumption is associated with reduced activation in caudate and frontal cortex (Kubota et al., 2001; Li, Luo, Yan, Bergquist, & Sinha, 2009; Sullivan, 2007). It is therefore possible that alcohol-related deficits were uncovered by the combined motor and affective components of the PSAP. Notably, if the group differences in BOLD activation were due only to the greater overall response output by the alcohol group, then the more plausible predicted outcome would be greater activation in the alcohol-dependent group in primary motor cortex and prefrontal cortex. However, we observed the converse. Based on previous work in populations with alcohol dependence alone or with comorbid DSM Axis II disorders, greater BOLD activation would be expected in the limbic regions, with diminished activation in prefrontal cortex corresponding to dysfunctional circuitry, lack of top-down control, and overexpression of anger/emotion (Courtney, Ghahremani, & Ray, 2013; New et al., 2007; Siever, 2008). Chronic alcohol use may disrupt activation of the collective emotional/inhibitory circuitry and manifest as a general inhibitory control deficit (Endres et al., 2014; Fillmore et al., 2005; Li et al., 2009). Indirect behavioral evidence for this supposition comes from the greater number of responses generated on both the monetary earning and aggressive response options, despite instructions and contingencies intended to regulate response rates.

When all 26 subjects were combined across groups and the contrast of postprovocation to nonprovocation activation was regressed on aggressive response rate, we observed a cluster that showed a statistically significant positive regression slope in the left postcentral, supramarginal, and precentral gyrii. These regions are most well established in subserving sensorimotor functions, including visual and tactile information processing and coordination of motor movements (Hanakawa, Dimyan, & Hallett, 2008; Jastorff, Abdollahi, & Orban, 2012; Michels, Kleiser, de Lussanet, Seitz, & Lappe, 2009). The most conventional interpretation of this outcome is that this cluster was associated with the basic task demands of coordinating button pressing with onscreen events, such that these demands increased during postprovocation monetary responding (relative to nonprovocation monetary responding), in proportion to the increase in aggressive response rate.

When all 26 subjects were combined across groups, several clusters were observed that showed significant negative regression slopes of provocation-monetary activation on aggressive response rate. These clusters involved lateral OFC, inferior frontal gyrus, middle cingulate gyrus, middle temporal gyrus, caudate, and thalamus. The negative relationships suggest that the aggressive response rate was related to decreased activation in these regions during postprovocation monetary responding (relative to non-provocation monetary responding). In studies in which human subjects responded on options that provided either monetary reward or monetary loss (punishment), unique patterns of activation specific to the expectation and/or consequence of punishment have been shown in the ventral and dorsal striatum and OFC (Delgado, Nystrom, Fissell, Noll, & Fiez, 2000; Knutson, Adams, Fong, & Hommer, 2001; O’Doherty, Kringelbach, Rolls, Hornak, & Andrews, 2001). In addition, experiments involving punishment contingencies during human social interaction paradigms reported activation in the amygdala, the DLPFC, and the dorsal striatum (de Quervain et al., 2004; Fehr & Rockenbach, 2004; Kahn et al., 2002; Rilling & Sanfey, 2011). Neuroimaging studies that experimentally induced anger in humans demonstrated increased activation in the DLPFC and OFC (Carlson, Greenberg, & Mujica-Parodi, 2010; Kimbrell et al., 1999). Two fMRI studies that explicitly measured operationally defined aggressive behavior reported BOLD activation in the medial frontal gyrus and caudate (Kramer et al., 2007, 2011). These previous reports correspond somewhat with the present data with regard to regions of activation. However, methods of measuring BOLD activation patterns were not fully equivalent across studies, and some studies observed increased activation in these regions related to aggressive behavior, whereas the present study showed decreased activation during post-provocation monetary responding (i.e., prior to aggressive button pressing behavior), but no change in activation during aggressive button pressing behavior per se.

The present data suggest that the striatum may play a role in modulating response to aversive stimuli. More broadly, current theories propose that it is the interconnected network between the limbic, OFC, and DLPFC regions that primarily subserve the processing of emotional and goal driven behavior, and that damage or dysfunction in this network results in problems with the regulation of emotion and subsequent difficulties with response inhibition, decision making, and aggressive behavior (Bechara, 2005; Brower & Price, 2001; Davidson, Jackson et al., 2000; Kringelbach, 2005; Schoenbaum et al., 2006; Siever, 2008). Despite the limitations noted below, in this small sample of alcohol-dependent subjects who were able to abstain from alcohol, we conclude that the brain regions that showed the most distinct differences from controls were those associated with inhibitory control (Bari & Robbins, 2013). This observation requires replication but should be of interest for future research, including treatment studies. It is generally consistent with the observation of Schulte and colleagues (2012) who reported that inhibitory control deficits measured by the Stroop task were related to abnormal patterns of functional connectivity (relative to controls) in midbrain-OFC networks.

Limitations

While several significant clusters showed a relationship between aggressive response rate and postprovocation activation in the combined (N = 26) dataset, including those in the original between-groups hypotheses, we did not observe group differences in some of the predicted regions (OFC, amygdala, cingulate, insula) either following provocation or during bouts of aggressive responding. The lateral OFC notably showed a relationship between aggressive response rate and postprovocation activation when all subjects were combined to form a larger single group. Significantly lower activation postprovocation was found in DLPFC for the alcohol-dependent group compared to the control group, which was consistent with study hypotheses.

Generally, the small N of 13 per group may have resulted in lack of power and prevented the observation of significant between-groups effects. As detailed by Button and colleagues (2013), studies such as this one with small sample sizes and/or low power suffer from three problems: (a) low probability of identifying true effects; (b) low positive predictive value that an obtained significant effect is in fact a “true positive”; and (c) an exaggerated estimation of the magnitude of significant effects that are obtained. The present experimental results are vulnerable to all three of these problems. Ultimately, experimental validation of these findings and the PSAP fMRI protocol will best be established via both direct and systematic replication.

One set of analyses that could have provided potentially useful information is the contrast of postprovocation periods that were followed by retaliation versus those that were not followed by retaliation. Unfortunately, for most subjects postprovocation periods were followed by a period of aggressive responding, and thus there were not sufficient fMRI data to make this comparison. This limitation may be circumvented in subsequent studies by manipulating the duration of the postprovocation interval and introducing provocations more frequently during each fMRI run. Additionally, the combination of the experimental design and our criteria for selecting/rejecting runs led to considerable loss of data (only 1 of 3 runs per subject). This limitation was necessary given the current task structure. In more recent revisions to the PSAP fMRI task, these limitations were addressed by allowing the monetary and aggressive options to be free to vary within 18-s blocks, in order to insure a more even distribution of provocations and aggressive responses within each fMRI run.

The trend level outcomes in amygdala were in the expected directions, but may have failed to reach statistical significance due to small sample size, relatively coarse acquisition resolution of the MRI images (3.75 mm × 3.75 mm × 5 mm), and the small size of the amygdala leading to partial volume effects. This problem was likely magnified by the additional spatial smoothing of the fMRI images as required by the SPM preprocessing method (which was 8 mm FWHM in this case). Past alcohol-dependent subjects were recruited into the present study to avoid confounds that could arise from current dependence, including obtaining alcohol-free BAL on test days and withdrawal effects on brain and behavior. We constrained the duration and recency of past alcohol dependence to minimize variability in the study group, but inclusion of a broader range of lifetime exposures and capturing of more detailed quantifiable lifetime use patterns could potentially inform both the behavioral and neuroimaging data. Much of the data supporting an association between alcohol and violence pertains to those currently abusing alcohol. In selecting a population in remission, the neural and behavioral outcomes may not be generalizable to the potentially more relevant group of current alcohol-dependent patients, but may be informative to treatment studies focused on behavior patterns during abstinence. Notably, we did not a priori select those with a history of alcohol-related violence, as we have done in previous studies of alcohol, ASPD, and borderline (Cherek & Lane, 1999; Moeller et al., 1998). The results and the limitations of the present study call for continued research into the neurobiology of aggression in chronic alcohol abuse.