Lower Amygdala Volume in Men is Associated with Childhood Aggression, Early Psychopathic Traits and Future Violence

Amygdala Volume and Types of Aggression

The amygdala plays an important role in several aspects of emotion processing that have important implications for understanding the development of severe aggression (7). Individuals with damage to the amygdala often have problems recognizing distress cues in others and have difficulties establishing conditioned fear responses (7), and similar impairments have been observed in individuals with a high propensity for violence (1). Although structural studies have linked lower amygdala volume to aggressive behaviors (5, 6, 8), null findings have been reported (9–11). However, many prior studies did not identify males with an early emerging and persistent history of violence, a unique subgroup whose antisocial behavior may be driven in part by neurobiological abnormalities (12).

The inconsistent findings regarding amygdala volume may have also resulted from a failure to distinguish between aggression that is impulsive-affective-reactive (IAR) versus predatory-instrumental-proactive (PIP). IAR aggression is characterized by explosive anger outbursts in response to perceived threat or provocation, while PIP aggression involves goal-directed acts of violence typically committed with little emotional arousal (13). The amygdala is believed to play a role in both IAR and PIP aggression, albeit in different ways. Specifically, the amygdala is involved in the detection of emotionally salient information, particularly cues of potential threat/danger (7). Individuals with high IAR aggression over-interpret ambiguous social cues as hostile (14) and exhibit exaggerated amygdala responding when viewing emotionally ambiguous (6) and angry (15) facial expressions. Additionally, lower amygdala volume has been linked with higher IAR aggression (6) and trait anger (6, 16). In contrast, blunted amygdala reactivity to punishment and others’ distress are believed to drive PIP aggression (13). Although a recent study found that PIP aggression was not associated with amygdala reactivity to fearful distress in others, it was associated with lower amygdala gray matter concentration (6).

Amygdala Volume and Psychopathic Traits

Amygdala abnormalities may play an important role in understanding violent behavior among men with psychopathic traits (1). Psychopathy includes a set of interpersonal (e.g., deceitful, manipulative), affective (e.g., callous, unemotional), and impulsive (e.g., thrill-seeking, irresponsible) features that are associated with chronic antisocial behavior (17, 18). Psychopathic features have been associated with reduced amygdala reactivity to numerous emotional stimuli (e.g., punishment cues, others’ distress), and these deficits putatively underlie the interpersonal/affective features of psychopathy (1, 2, 19, 20). However, structural studies linking amygdala volume with psychopathic traits have been mixed, with some reporting reduced gray matter volume (3, 4, 21), and others reporting non-significant associations (6, 11, 22).

Current Study

There are several limitations in the literature regarding the association between amygdala volume and features of aggression and psychopathy. First, it is not clear if men with a longstanding history of violence and psychopathic features spanning from childhood to early adulthood have lower amygdala volume. Second, no studies have examined whether lower amygdala volume is associated with IAP and/or PIP types of aggression in both adolescence and adulthood. Third, no published studies have examined whether men with reduced amygdala volume are at risk for future violence and exhibiting persistent psychopathic features. The current study will be the first to link adult amygdala volume with aggression, violence, and psychopathic traits repeatedly assessed across a 22-year period from childhood to adulthood.

Participants

Participants were 56 men recruited from the youngest cohort of the Pittsburgh Youth Study (PYS). The youngest cohort of the PYS consists of 503 boys selected from 1^st graders attending the Pittsburgh public schools in 1986–87. PYS participants were recruited following an initial screening assessment that measured the boys’ antisocial behaviors using a combination of parent, teacher, and self-report instruments. Boys who scored within the upper 30% on the screening (n=256), as well as a roughly equal number of boys randomly selected from the remaining 70% of the distribution (n=247) were selected for longitudinal follow-up. The racial composition of the follow-up sample is predominately Caucasian (40.6%) and African-American (55.7%). The first assessment took place when boys averaged 7.46 years of age (SD=.55). Assessments occurred every six months for four years, with annual assessments occurring for the next nine years. Two additional assessments were conducted when the men averaged 25.78 (SD=.96) and 29.25 (SD=1.11) years of age. All procedures were approved by the University of Pittsburgh IRB, and informed consent was obtained from all adult participants (23).

Neuroimaging Substudy

At age 26, a subsample of 56 men from the PYS with a history of chronic serious violence (CSV; n=20), transient serious violence (TSV; n=16), and no serious violence (NSV; n=20) were recruited for a neuroimaging substudy. The groups were delineated based on self-report assessments and official criminal records. Items from the Self-Report of Delinquency (SRD)(24) were used to assess past-year serious violence (i.e., rape, robbery, gang fighting, attacking with a weapon or to seriously hurt/kill) across assessments spanning ages 10–19 and again at age 26, just prior to the imaging assessment. At age 26, a 15-item Violence History Questionnaire (VHQ) (25) collected more detailed information about serious violence in the past year (e.g., kidnapping). Criminal charges for violence (e.g., homicide, robbery) were collected using juvenile, Pennsylvania state, and federal records. Groups were delineated as follows: 1) CSV men self-reported or were charged with violence across 4+ years; 2) TSV men self-reported or were charged with violence between 1–3 years; and 3) NSV men had no history of serious violence. Exclusion criteria included: 1) a prior history of a psychotic disorder according to the Diagnostic Interview Schedule for DSM-IV (26); 2) use of psychotropic medications; 3) history of neurological disease, structural brain injury, post-concussive syndrome, and/or cardiovascular disease; 4) a full scale IQ below 70 on the Weschler Abbreviated Scale of Intelligence (27); and 5) current incarceration. Attempts were made to equate the groups on age, handedness, intelligence, and race.

Structural Data Acquisition

Participants completed a series of functional and structural neuroimaging scans using a 3.0 Tesla Siemens Allegra MRI scanner. The current study is based upon structural images obtained using high-resolution T1-weighted 3D gradient echo imaging with an SPGR sequence: axial plane, TR=1630ms, TE=2.48ms, flip angle=8°, number of excitations=1; bandwidth=210 Hz/pixel; echo spacing=6.8ms; matrix=256x256, FOV=204mm, 224 slices, .8mm isotropic thickness, 0mm gap.

Segmentation of Amygdala Volume

Segmentation and volumetric analysis of the amygdala was performed using FMRIB’s Integrated Registration and Segmentation Tool (FIRST) in FMRIB’s Software Library (FSL) version 4.1 (28). FIRST is a semi-automated subcortical segmentation tool that uses active shape and appearance models within a Bayesian framework based on information obtained from 336 manually-labeled T1 images (28). The 336 manually segmented and labeled T1-weighted brains have been parameterized as surface meshes and modeled as a point distribution model. FIRST searches through linear combinations of shape modes of variation for the most probable shape instance given the observed intensities in an individual’s T1 image. Deformable surfaces are used to automatically parameterize the volumetric labels in terms of meshes. These surfaces are constrained to preserve point correspondence across the training data. Volumetric labels are parameterized by a 3D deformation of a surface model based on multivariate Gaussian assumptions. FIRST segmentation of subcortical structures has demonstrated median Dice overlaps with manual tracings from .70–.90, which is comparable or better than other automated methods (28).

FIRST processing was conducted using standardized procedures (28). Initially, the subcortical regions are separated from the rest of the cortex and white matter using a two-stage affine registration process that aligns the whole-head from native space to a standard space template (MNI152). Next a subcortical mask within the MNI space is used to achieve a more refined and robust alignment with the subcortical structures. An inverse transformation then brings the image back into native space, allowing for subsequent segmentation to be conducted with the original voxel intensities. Subcortical structures are segmented based on a Bayesian Appearance Model derived from the manually-traced training images using 50 modes of variation. The modes of variation are optimized based on leave-one-out cross-validation on the training set. Boundary correction takes place for each structure, and voxels are classified as belonging to a structure based on a statistical probability (z-score > 3.00; p<.001).

All segmentations were visually checked by a study investigator (K.E.) blind to participant characteristics with approximately 10 years of experience in neuroanatomy and segmentation techniques of MRI data. These checks confirmed that each segmentation was adjacent to the lateral ventricle and temporal horn, located anterior and slightly dorsal to the hippocampus, and demarcated from the hippocampus by the alveus and subiculum (see Figure S1). Average left and right amygdala volumes were then extracted for each subject and were z-scored prior to analyses to place them on a standard metric (i.e., standard deviation units). To control for overall intracranial volume (ICV), the sum of gray, white, and cerebrospinal fluid volumes was calculated using FIRST.

Measures

Data Analysis Plan

First, an ANOVA was used to compare men with a history of CSV, TSV, or NSV in terms of left and right amygdala volume. Linear regression was then used to examine the association between amygdala volume and individual differences in aggressive/psychopathic features at the time of the scan using the entire sample.

Next, analyses examined the association between amygdala volume and early aggression/psychopathic features across all participants. Population-average generalized estimating equations (GEE) were used for the repeated measurements of aggression/psychopathic features in childhood. An exchangeable correlation structure among the repeated assessments was specified. Ordinary linear regression was used to examine the association between amygdala volume and aggression/psychopathic traits at the single assessment point in adolescence.

A final set of analyses using all participants examined amygdala volume as a predictor of post-scan outcomes. Linear regression models were used for the outcomes of aggression and psychopathic traits at follow-up. Levels of these features at the time of the scan were included as covariates. A logistic regression using a penalized likelihood estimator appropriate for small samples (44) was used to examine whether amygdala volume was associated with any post-scan violence. The number of years men had engaged in serious violence prior to the scan was included as a covariate in this model.

Several covariates of no interest were considered as potential confounds, including total intracranial volume, race, age, handedness, and IQ, as well as a prior history of childhood maltreatment, concussion, prior incarceration, psychotropic medication use, internalizing disorders, substance abuse, and substance dependence. Within each developmental period, scales assessing co-occurring depression/anxiety and inattention/hyperactivity were also considered as potential confounds. For analyses focusing on post-scan violence, the number of days that elapsed between the neuroimaging scan and the follow-up data collection was included as a covariate. Potential confounds were entered into each model prior to the inclusion of amygdala volume using a backward stepwise procedure, with a liberal threshold of retaining predictors with a p-value ≤ .15.

Behaviors Assessed Concurrent with Structural Scan

There were no significant differences in left amygdala volume between NSV (M=1597, SD=295), TSV (M=1489, SD= 302), and CSV (M=1587, SD=312) men (F[2,53]=.66, p=.52). Right amygdala volume also did not significantly differ between NSV (M=1806, SD=429), TSV (M=1710, SD= 362), and CSV (M=1668, SD=286) men (F[2,53]=.75, p=.48). However, dimensional analyses using all participants indicated that lower left and right amygdala volume was associated with higher ASR aggression and higher levels of premeditated aggression (see Table 2). Lower left amygdala volume was also associated with higher levels of IAR aggression. Amygdala volume was not significantly related to verbal/physical aggression measured using the AQ-SF and overall levels of psychopathic traits. Moreover, none of the facets of psychopathy were significantly associated with amygdala volume (ps > .20).

Behaviors Assessed in Childhood and Adolescence

Lower amygdala volume was significantly associated with measures of aggression and psychopathic features collected in childhood and adolescence (see Table 3). Lower left amygdala volume was associated with higher teacher-reported aggressive behaviors and interpersonal callousness across childhood. Lower right amygdala volume was associated with higher proactive aggression and overall psychopathic features in adolescence. In terms of the latter, right amygdala volume was significantly associated with the affective dimension (β=−.29, p<.05) of psychopathy, but not the interpersonal (β=−.16, p=.28) and lifestyle (β=−.20, p=.10) facets.

Post-Scan Outcomes

There were 21 men (37.5%) who engaged in violence across the post-scan follow-up, which included 5% of men in the NSV group (N=1), 37.5% of men in the TSV (N=6), and 70% of men in the CSV group (N=14). Lower left and right amygdala volumes were associated with an increased risk of committing violence, higher verbal/physical aggression, and increased psychopathic features at the follow-up, even after accounting for prior levels of these behaviors and potential confounds (see Table 4, pFigures 1–2). In terms of psychopathy facets, left amygdala volume was significantly associated with the lifestyle (β=−.24, <.05) dimension, but this effect only approached significance for the interpersonal (β=−.21, p=.05) and affective (β= −.25, p=.06) dimensions. In contrast, right amygdala volume was associated with the interpersonal dimension (β= −.22, p<.05), but not the affective (β= −.20, p=.11) and lifestyle (β= −.16, p=.14) facets.

Supplemental Analyses

When analyses were re-run excluding non-right handed men (N=5), the primary results were largely unchanged (see Table S2). To examine whether the findings exhibited evidence of anatomical specificity relative to nearby limbic regions, all analyses were re-run using right and left hippocampal volumes extracted using FIRST as predictors (see Table S3). Lower right hippocampal volume was associated with higher aggression on the ASR at the time of the scan. However, when right amygdala and hippocampal volumes were simultaneously entered as predictors of this outcome, amygdala volume remained a significant predictor (β= −.31, p<.05) and hippocampal volume became non-significant (β= −.08, p=.43).

Limitations

The results from this study must be interpreted cautiously due to some methodological limitations. The current sample consisted solely of men, making it unclear whether the results will generalize to women. Gray matter volume was also measured at a single assessment in early adulthood. As such, it is unclear whether low amygdala volume preceded the development of childhood and adolescent aggression and psychopathic features. The current study also focused exclusively on amygdala volume since this region has been consistently implicated in development of severe antisocial behavior (1, 4). Although the findings did not translate to a nearby limbic region in supplemental analyses (i.e., hippocampus), structural differences in other brain regions may also be related to the phenotypes examined here. Similarly, future studies should examine whether manual tracings and surface-based methods for delineating more fine-grained differences in amygdala nuclei morphology (e.g., basolateral complex) may help to better identify individuals at risk for developing aggression and psychopathic features (4).

While the current study involved theoretically focused analyses and controlled for numerous potential confounds, several outcomes were examined due to the longitudinal nature of the data. The use of multiple comparisons corrections is often debated, as these corrections increase the chance of making type II errors that minimize truly important findings and require the use of large samples (which are often prohibitively expensive in neuroimaging research) to detect modest effect sizes (51–54). Although the current results indicated a consistent negative association between amygdala volume and measures of aggression/psychopathic features, a strict Bonferroni correction would make the critical alpha level = .002 (.05/28). At this threshold, only the association between amygdala volume and aggressive behavior at the time of the scan would reach significance (see Table 2). As such, future studies with larger samples are needed to further validate these results.

Conclusions

This is the first study to demonstrate that lower amygdala volume is associated with features of aggression and psychopathy spanning from childhood to young adulthood. It is also the first investigation to demonstrate that lower amygdala volume is a significant risk factor for future violent behavior. Other investigators have found that children with deficient fear conditioning are at an increased risk for exhibiting adult criminal behavior, with abnormalities in the amygdala speculated to drive this association (55, 56). Although studies replicating these types of findings are needed, amygdala dysfunction may turn out to be an important biomarker for the development of severe and persistent aggression (56). However, multiple socio-contextual factors influence the development of antisocial behavior, and the relative influence that amygdala abnormalities play in the emergence and persistence of criminal behavior is in need of further study.