Ventromedial Prefrontal Cortex and Amygdala Dysfunction During an AngerInduction Positron Emission Tomography Study in Patients With Major DepressiveDisorder With Anger Attacks FREE

Major depressive disorder (MDD) is a clinical syndrome characterizedby affect dysregulation, neurovegetative symptoms, autonomic disturbances,and endocrine abnormalities. Freud¹ theorizedthat depression resulted from anger turned inward. In fact, numerous studies^2- 8 havedemonstrated that depressed patients have higher rates of anger and aggressionthan controls. One study⁹ even found that thedegree of anger expressed inward in depressed patients correlated with theseverity of depressive symptoms.

The concept of "anger attacks" in patients with MDD was introduced byFava and colleagues¹⁰ in 1990. These angerattacks are characterized by sudden spells of anger that are inappropriateto the situation in which they occur and are uncharacteristic of the patient'susual behavior. In addition, clinical characterization of this depressivesubtype reveals that patients with MDD with anger attacks (MDD + A) have higherscores on measures of hostility, anxiety, and somatization than patients withMDD without anger attacks (MDD − A).¹¹ Theprevalence of anger attacks in depressed patients is approximately 30% to40%,^{4- 5,11- 14} andthe attacks resolve after successful treatment of the depressive episode.^5,11- 12

Studies¹⁵ have demonstrated that symptomsof anger or aggression may have a prevalence as high as 50% in outpatientpsychiatric populations, indicating that anger or aggression may be as commonas symptoms of depression and anxiety in this population. Anger and aggressionare especially common in patients with diagnoses of MDD, bipolar disorder,intermittent explosive disorder (IED), and cluster B personality disorders.Despite the high prevalence of anger and aggression in psychiatric populationsand the obvious public health impact of these symptoms, the role of angerand aggression in psychiatric illness has been understudied. In addition,anger and hostility are associated with higher rates of coronary artery disease,^16- 18 myocardial infarction,^19- 22 andabnormal glucose metabolism.^23- 27 Giventhat depression itself is a significant risk factor for heart disease^28- 30 and diabetes mellitus,^31- 32 it would seem that individuals withMDD + A would be especially at risk for medical consequences. Thus, the societalcosts of anger and aggression in psychiatric illness stem from the consequencesof violence and the medical sequelae associated with anger and hostility.In patients with MDD + A, these costs to society are above and beyond thealready staggering financial burden of depression.³³ Thesefindings underscore the importance of elucidating the pathophysiology of angerand aggression in psychiatric illness and of conducting clinical trials inthe search for effective treatments.

Converging data implicate a network of brain regions in the pathophysiologyof MDD. These regions include, but are not limited to, territories of theprefrontal cortex (PFC), anterior cingulate cortex, and medial temporal lobestructures, including the amygdala and hippocampus.^34- 38 Currentconceptualization of the underlying neuroanatomy of anger and aggression implicatesthe amygdala and related temporolimbic structures, the hypothalamus, the anteriorcingulate cortex, and the PFC (most notably the ventral PFC) as brain regionsinvolved in mediating aggression.^39- 41 Althougha variety of functional neuroimaging studies^42- 46 haveused emotion induction pdigms to investigate the neural basis of angerin control subjects, no functional neuroimaging studies have been conductedin individuals who are diagnosed as having MDD and who have a predilectionfor anger, aggressive behavior, or both. The present study uses positron emissiontomography (PET) and autobiographical narrative scripts to study the neuralbasis of anger, particularly the relationship between the ventral PFC andthe amygdala, in unmedicated patients with MDD + A, unmedicated patients withMDD − A, and control subjects. There is especially strong evidence thatanger and aggression seen in a multitude of diagnoses are associated withhypofunctionality of the ventral PFC and amygdala.^47- 52 Incontrast, numerous studies^36,53- 54 havedemonstrated greater activity in the ventral PFC and amygdala in patientswith MDD compared with control subjects. These results strongly suggest thatventral PFC and amygdala function should differentiate patients with MDD +A and patients with MDD − A from each other and from control subjectsand form the basis for our a priori hypotheses. Specifically, we predicted(1) that, based on previous results from our laboratory⁴⁴ andothers,^{42- 43,45- 46} thecontrol group would exhibit activation of the ventral PFC with correspondingdeactivation of the amygdala during anger induction, (2) that the MDD + Agroup would exhibit diminished activation in the ventral PFC and amygdaladuring anger induction relative to the control group and the MDD − Agroup, and (3) that the MDD − A group would have greater activationin these same brain regions during anger induction relative to the controlgroup and the MDD + A group.

The study sample was composed of 30 individuals divided evenly among3 study groups: MDD + A, MDD − A, and controls. All 3 study groups werematched for age and sex; the 2 MDD groups were also matched for depressionseverity (Table 1). The studywas conducted in accordance with the guidelines of the Human Subjects in ResearchCommittee of the Massachusetts General Hospital. Written informed consentwas obtained from each participant. All participants were right-handed (EdinburghInventory⁵⁵) and had normal hearing and normal/corrected-to-normalvision. Exclusion criteria included pregnancy, a history of a major medicalor neurologic disorder, a history of head injury, a history of seizure disorder,and current use of psychotropic medications.

All patients participating in this study were recruited through theDepression Clinical and Research Program at Massachusetts General Hospital.All patients underwent comprehensive evaluation by the Depression Clinicaland Research Program staff. A full medical and psychiatric history was performedby a study psychiatrist. During the screening visit, the patients were administeredthe Structured Clinical Interview for DSM-IV Disorders,⁵⁶ the Anger AttacksQuestionnaire,⁴ and the 17-item Hamilton DepressionRating Scale.⁵⁷ Inclusion criteria were a DSM-IV diagnosis of major depression, single or recurrent,of at least 4 weeks' duration at the time of the screening visit. Exclusioncriteria included current or past Axis I diagnoses other than MDD and a historyof mood congruent or mood incongruent psychotic features.

In addition, patients had a diagnosis of MDD + A subtype based on theAnger Attacks Questionnaire,⁴ a 7-item self-ratinginstrument designed to assess the presence or absence during the previousmonth of anger attacks, defined as spells of anger inappropriate to the situation.

Controls were recruited by advertisements in the community. Subjectsparticipated in a screening, including administration of the Structured Clinical Interview for DSM-IV Disorders,⁵⁶ to ascertain their relevant psychiatric,medical, and neurologic history. None of the controls had a history of majorneurologic, medical, or psychiatric disorders.

Scripts of participants' past personal events were prepared accordingto a previously published procedure.^44,58- 61 Eachparticipant provided a written description of the 2 life events correspondingto when they were the most and second most angry. Two autobiographical neutralscripts (eg, going for a walk and cooking dinner) were likewise developed.After describing each event, the participant examined a list of bodily responses(eg, "heart racing" and "labored breathing") and circled those responses (ifany) that they experienced at the time. Based on the material furnished bythe participants, an investigator (D.D.D.) composed a script in the secondperson, present tense and then audiotaped it in a neutral voice for playbackin the laboratory. All scripts were 30 to 40 seconds in duration.

After habituation to the PET suite environment, participants were scanned8 times as part of a larger study. Two scans corresponded to the anger condition,2 scans corresponded to the neutral condition, and 4 scans corresponded toother script-induced emotions. Neutral conditions were performed first andlast, whereas the order of the remaining 3 conditions (anger and the otherinduced emotions) was counterbalanced across participants. Before each scan,the participant was instructed as follows: "Close your eyes, listen carefullyto the script, and imagine the event portrayed as vividly as possible, asif you are actually participating in the event rather than just ‘watchingyourself' in it." Then the audiotape was played. During the 60 seconds immediatelyafter the script audiotape, as per instructions, participants continued torecall and imagine the event while PET data were acquired. The ¹⁵O–carbondioxide administration and PET data acquisition were then terminated, andthe participant was instructed to stop imagining the event. Positron emissiontomographic scans were septed by at least 10 minutes to allow for radiationdecay to negligible levels. In addition, psychophysiologic measures were requiredto return to within 10% of baseline values before beginning the next PET scan.

After scanning, the participants rated their emotional responses (ie,happiness, sadness, anger, fear, disgust, surprise, guilt, and shame) to eachscript on septe subjective 0- to 10-point analog scales,^44,59- 61 where0 indicated the "complete absence of a response" and 10 indicated the "maximumpossible response" for the specified emotion. The participants also completedsimilar analog scales for difficulty recalling the event, vividness of imagery,and strength of visual, auditory, tactile, olfactory, and gustatory imagery.Paired t tests were used to compare differences inanalog scale scores between conditions.

Psychophysiologic assessment was performed during the PET study usingequipment from ADInstruments (Sydney, Australia). Measured pmeters includedheart rate and galvanic skin response (GSR). Psychophysiologic pmeterswere recorded continuously during the PET study. For purposes of data analyses,data were calculated during 2 epochs associated with each scan: 30 secondsbefore the reading of the script (baseline) and 1 minute during each scan(imagery). Within the baseline and imagery periods (within each scan), heartrate values were averaged, whereas GSR values were calculated using area underthe curve (AUC) methods. For each scan, the values of the baseline periodwere subtracted from the values of the imagery period. Paired t tests, analyses of variance, and independent t tests were used, where appropriate, for psychophysiologic data analyses.

A 15-slice whole-body tomograph (model PC4096; Scanditronix/GeneralElectric Medical Systems, Milwaukee, Wis) was used in its stationary modeto acquire the PET data.⁶² The slice geometryconsists of contiguous slices with center-to-center distance of 6.5 mm (axialfield equal to 97.5 mm) and axial resolution of 6.0-mm full width at halfmaximum. Image reconstruction was performed using a computed attenuation correctionand a Hanning-weighted reconstruction filter set to yield 8.0-mm in-planespatial resolution full width at half maximum. Additional corrections weremade in the reconstruction process to account for scattered radiation, randomcoincidences, and counting losses due to dead time in the camera electronics.

Head alignment was made relative to the canthomeatal line using projectedlaser lines whose positions were known with respect to the slice positionsof the scanner. An individually molded thermoplastic mask was used to minimizehead motion. Once the head was in place, the patient was fitted with a pairof nasal cannulae and an overlying face mask, which were attached to radiolabeledgas inflow and vacuum, respectively.

The participants were studied while continuously inhaling tracer quantitiesof ¹⁵O–carbon dioxide mixed with room air. The concentrationof the delivered gas was 2960 MBq/L (80 mCi/L), with a flow rate of 2 L/min,further diluted by free mixture with room air within the face mask, resultingin a rapidly rising count rate in the brain, reaching terminal count ratesof 100 000 to 200 000 events per second. Previous work at MassachusettsGeneral Hospital using radial artery cannulation has demonstrated that theintegrated counts over inhalation periods up to 90 seconds are a linear functionover the flow range of 0 to 130 mL/min per 100 g (N.M.A., unpublished data,1991). Therefore, for data to be produced with units of flow relative to thewhole brain, no arterial access was necessary.

Statistical analysis of the PET data was conducted following the theoryof statistical pmetric mapping.^63- 64 Datawere analyzed using a software package (SPM99; Wellcome Department of CognitiveNeurology, London, England). Positron emission tomographic images were motioncorrected, spatially normalized to the standardized normalized space establishedby the Montreal Neurological Institute (MNI) (available at: http://www.bic.mni.mcgill.ca), and smoothed to 10-mm full width at half maximum. At each voxel,the PET data were normalized by the global mean and fit to a linear statisticalmodel by the method of least squares. Planned contrasts at each voxel wereconducted; this method fits a linear statistical model, voxel by voxel, tothe data, and hypotheses were tested as contrasts in which linear compoundsof the model pmeters were evaluated using t statistics,which were then transformed to z scores. Region ofinterest (ROI) definition for interregional correlation analyses (describedin the "VMPFC ROI-Based Interregional Correlation Analyses" subsection) wasconducted using MarsBaR software.⁶⁵

We report regions containing foci of activation with z scores ≥3.09 (corresponding to P≤.001[1-tailed], uncorrected for multiple comparisons). Note that the data wereinspected in a hierarchical manner: first, regions from the a priori hypotheseswere inspected, then the entire brain volume was inspected, and post hoc findingsare reported using a compble threshold to obviate bias.

All values are reported as mean ± SD.

Analyses of the self-report data revealed that compared with the neutralcondition, the anger condition was associated with a higher rating of angerin all 3 groups. The mean difference in anger between the anger and neutralconditions was 7.06 ± 1.85 (t₁₇ =16.18) for patients with MDD + A, 7.11 ± 2.32 (t₁₉ = 13.70) for patients with MDD − A, and 6.91 ±2.57 (t₁₉ = 12.04) for control subjects(P<.001 for all). In addition, the anger self-reportscore difference between the anger and neutral conditions was significantlylarger than any of the other emotional self-report score differences (t₅₅ = 3.36; P = .02compared with disgust, the self-report score with the next largest differencebetween the anger and neutral conditions). All patients confirmed that theemotional state achieved was reflective of an anger state and that visualand auditory modes represented its most prominent imagery components.

Psychophysiologic data were successfully collected in 7 patients withMDD + A, 7 patients with MDD − A, and 10 control subjects; missing datawere attributable to technical difficulties.

The average change in heart rate for patients with MDD + A from theneutral condition to the anger condition was 3.53 ± 5.72 bpm, whichwas a significant increase (t₁₂ = 2.23; P = .046). These patients also experienced an increasein GSR AUC of 72.69 ± 112.75 microsiemens during the anger condition,which was significant (t₁₂ = 2.32; P = .04).

In patients with MDD − A, the average change in heart rate fromthe neutral condition to the anger condition was 0.50 ± 8.24 bpm, whichwas not significant (t₁₃ = 0.23; P = .82). However, these patients experienced a significantdecrease in GSR AUC of −55.27 ± 62.71 microsiemens during theanger condition (t₁₃ = −3.30; P = .006).

The average change in heart rate for control subjects from the neutralcondition to the anger condition was 5.83 ± 7.57 bpm, which was a significantincrease (t₁₉ = 3.44; P = .003). These subjects also experienced an increase in GSR AUC of81.80 ± 96.99 microsiemens during the anger condition, which was significant(t₁₉ = 3.77; P<.001).

Analysis of variance revealed that a significant difference exists betweengroups in GSR AUC (F_2,44 = 10.114; P<.001)but not in heart rate (F_2,44 = 2.173; P =.13). Further analyses confirmed that although no difference existed betweenthe control and MDD + A groups in GSR response (t₃₁ = 0.27; P = .96), a significant differencewas detected between patients with MDD − A and control subjects (t₃₂ = −4.22; P<.001)and between patients with MDD − A and those with MDD + A (t₂₅ = −3.56; P = .003).

In the control group, the anger vs neutral comparison demonstrated increasedregional cerebral blood flow (rCBF) in the left ventromedial PFC (VMPFC) (Table 2). Regarding the a priori territoriesof interest, no significant activations were found in within-group analysesinvolving the MDD + A and MDD − A groups.

Regarding a priori hypotheses, the between-group analyses revealed greaterrCBF increases in the control group than in the MDD + A group in the leftVMPFC during the anger vs neutral comparison (Table 3 and Figure1). These differences were not present in other between-groupanalyses. Last, there were no between-group differences in rCBF changes inthe amygdala during the anger vs neutral comparison.

Interregional correlation analyses examining the relationship betweenrCBF responses in the left VMPFC and those in the rest of the brain were conductedin each group for the anger vs neutral comparison (Table 4 and Figure 2). We defined a functional ROI in the left VMPFC(MNI coordinates = −8, 62, −10) based on the anger vs neutralcomparison in the control group. We then extracted rCBF values from the ROIand conducted a correlational analysis between these ROI values and whole-brain,voxelwise rCBF changes in the anger vs neutral comparison for each group.Based on known bidirectional connections between the PFC and the amygdalaand evidence that these 2 structures are mutually inhibitory, we hypothesizedthat these interregional correlation analyses would demonstrate an inversecorrelation of rCBF changes during anger induction between the left VMPFCand the left amygdala in the control group. The analysis confirmed this hypothesis(Table 4 and Figure 2). Identical interregional correlation analyses of rCBFchanges during anger induction did not demonstrate any significant relationshipbetween the left VMPFC and the left amygdala in the MDD − A group (Table 4). However, interregional correlationanalyses of rCBF changes during anger induction in the MDD + A group revealeda positive correlation between the left VMPFC and the left amygdala (Table 4 and Figure 2).

Last, to perform a statistical comparison of these correlations betweengroups, we defined functional ROIs in the left amygdala based on the interregionalcorrelation analyses. One functional ROI corresponded to the left amygdalalocus from the interregional correlation analyses in the control group (MNIcoordinates = −22, 2, −12) (Table 4), and the other functional ROI corresponded to the leftamygdala locus from the interregional correlation analyses in the MDD + Agroup (MNI coordinates = −22, −12, −22) (Table 4). Then, septe within-group analyses were conducted todetermine the degree of correlation between rCBF values from the left VMPFCROI and the 2 amygdala ROIs. As expected, there was a significant inversecorrelation between left VMPFC ROI rCBF values and rCBF values from the leftamygdala ROI derived from the control group interregional correlation analysesin the control group (r = −0.87; P<.001) but not in the MDD + A (r = −0.08; P = .83) and MDD − A (r =−0.07; P = .85) groups. As would also be expected,there was a significant positive correlation between left VMPFC ROI rCBF valuesand rCBF values from the left amygdala ROI derived from the MDD + A groupinterregional correlation analyses in the MDD + A group (r = 0.90; P<.001) but not in the MDD −A (r = −0.14; P =.69) and control (r = 0.31; P =.38) groups. Fisher z transformation of the within-groupcorrelation coefficients was used to perform a statistical comparison of thesecorrelations between groups. The correlation coefficient arising from thecomparison of left VMPFC ROI rCBF values and rCBF values from the left amygdalaROI derived from the control group interregional correlation analyses in thecontrol group (r = −0.87) differed significantlyfrom the identical comparisons in the MDD − A and MDD + A groups (P = .02 for both), whereas the MDD − A and MDD +A groups did not differ significantly from one another (P = .99). The correlation coefficient arising from the comparison ofleft VMPFC ROI rCBF values and rCBF values from the left amygdala ROI derivedfrom the MDD + A group interregional correlation analyses in the MDD + A group(r = 0.90) differed significantly from the identicalcomparisons in the MDD − A (P = .002) and control(P = .03) groups, whereas the MDD − A and controlgroups did not differ significantly from one another (P = .38).

Previous functional neuroimaging studies conducted with individualspredisposed to anger or aggression have principally used neutral-state orpharmacologic challenge studies. In contrast, the present study representsan initial symptom provocation PET study in patients with MDD predisposedto anger or aggression and yields several important findings. First, thisstudy replicated findings from our laboratory⁴⁴ andothers^{42- 43,45- 46} ofincreased ventral PFC (specifically, the left VMPFC in this study) rCBF duringanger induction in controls. Second, the control subjects demonstrated statisticallysignificantly greater left VMPFC rCBF increases than the patients with MDD+ A during anger induction. There was no corresponding difference in leftVMPFC rCBF during anger induction when comparing the patients with MDD −A with either the patients with MDD + A or the control subjects. Third, inthe control subjects, interregional correlation analyses found an inversecorrelation between left VMPFC rCBF changes and left amygdala rCBF changesduring anger induction. However, patients with MDD + A demonstrated rCBF changesin the left VMPFC and the left amygdala in the same direction during angerinduction, suggesting an aberrant functional relationship between these brainregions in the MDD + A group. Last, whereas patients with MDD − A demonstratedblunted autonomic responses during anger induction, those with MDD + A hadautonomic responses that were compble to the responses of controls. Thus,autonomic response during anger induction clearly differentiates the MDD +A subtype from the MDD − A subtype.

Electroencephalographic^66- 67 andfunctional neuroimaging^42- 46 studieshave used emotion induction pdigms to investigate the neural basis of angerin control subjects. Although these studies used different techniques to induceanger, all of them demonstrated the involvement of common anterior plimbicstructures during anger states. All of the studies found that the ventralPFC was recruited during anger states. This finding makes sense in the contextof multiple lines of evidence indicating that the ventral PFC plays a crucialrole in constraining impulsive outbursts.⁴⁰ Thus,it is proposed that individuals who exhibit excessive impulsive behavior (includingaggression) do so because they cannot mobilize the ventral PFC in this manner.In fact, many studies^68- 81 usingneuropsychologic tests to assess frontal functional integrity have found deficitsin functions mediated by the frontal lobes in violent and antisocial personalitydisordered (APD) individuals. One structural magnetic resonance imaging (MRI)study⁸² demonstrated that a group of patientswith APD and a history of violent crimes had an 11% reduction in PFC graymatter volume compared with a control group, and another⁸³ foundthat patients with temporal lobe epilepsy and IED had a 17% reduction in PFCgray matter compared with patients with temporal lobe epilepsy without IED.A growing number of functional neuroimaging studies^84- 93 ofindividuals with a predisposition to anger and aggression have found thatthese individuals (groups have included murderers, violent offenders, andthose with APD) exhibit decreased activity in the PFC compared with controlsubjects. Postmortem studies of individuals completing violent suicide haverevealed a variety of serotonergic abnormalities in the PFC.^94- 96 Inaddition, recent [¹⁸F]fluorodeoxyglucose PET studies have demonstratedthat, unlike controls, patients with impulsive aggression do not show activationof the left VMPFC in response to administration of fenfluramine^49- 50 ormeta-chlorophenylpiperazine.⁵¹ Taken together,these studies provide strong evidence that dysfunction of the PFC, particularlythe ventral PFC, is common to the pathophysiology of impulsive aggressionseen in a multitude of diagnoses.

Consistent with our hypothesis, correlational analyses examining therelationship between a functionally defined left VMPFC ROI and the rest ofthe brain in controls demonstrated a negative correlation with the ipsilateral(left) amygdala. A growing literature suggests that the amygdala plays a prominentrole in antisocial behavior.⁴⁷ Patients withAPD show reduced potentiation of the startle reflex following exposure tothreatening visual stimuli⁹⁷ and impaired aversiveconditioning.^98- 99 These impairmentsare also found in patients with amygdala lesions.^100- 102 Patientswith amygdala lesions and those with APD also exhibit impairments in the processingof fearful (and possibly sad) facial expressions.^103- 107 Neuropsychologicstudies¹⁰⁸ have shown similar deficits in decisionmaking in patients with amygdala and VMPFC lesions. In addition, one structuralMRI study¹⁰⁹ of patients with temporal lobeepilepsy found that those with comorbid IED had substantially higher ratesof amygdala atrophy or amygdala lesions than those without comorbid IED, andanother structural MRI study¹¹⁰ found thatlevels of antisocial behavior in violent offenders was inversely correlatedwith amygdala volume. These findings are especially relevant given the roleof the amygdala in one model of antisocial behavior, the violence inhibitionmechanism model, which suggests that antisocial individuals are less likelyto activate the violence inhibition mechanism in the context of fearful andsad facial expressions of others.^42,105,111 Afunctional MRI study¹¹² using a memory taskdemonstrated that participants who scored higher on a scale of antisocialbehavior demonstrated reduced amygdala activation while processing negativelyvalenced words compared with individuals who scored lower on the scale. Anotherrecent functional MRI study¹¹³ of patientswith APD using a differential aversive-delay conditioning task found bluntedactivation of the orbitofrontal cortex, anterior cingulate cortex, insula,and amygdala compared with control subjects. Thus, multiple lines of evidencesuggest that in addition to dysfunction of the PFC, amygdala dysfunction isalso common to the pathophysiology of impulsive aggression seen in a varietyof diagnoses.

It has been suggested that APD and IED may be associated with "dualbrain pathology" in which abnormalities in the amygdala result in dysfunctionalarousal states and those in the PFC result in dyscontrol states.¹¹⁴ Thereare known bidirectional connections between the PFC (especially the medialPFC) and the amygdala,^115- 120 andthere is evidence^121- 122 thatin control subjects these 2 structures are mutually inhibitory in that increasedactivity in one structure inhibits activity in the other structure. Becausethe controls in the present study demonstrated a reciprocal (or inverse) relationshipbetween left VMPFC and left amygdala rCBF during the anger vs neutral comparison,we examined this relationship in the MDD groups. We did not demonstrate anystatistically significant relationship for rCBF changes during anger inductionbetween the left VMPFC and left amygdala in the MDD − A group. In contrast,for the MDD + A group the rCBF changes during anger induction in the leftVMPFC and the left amygdala were in the same direction (ie, they demonstrateda positive correlation). This suggests that, at least in the context of angerinduction, the normal (inverse) functional relationship between the VMPFCand the amygdala is absent in the MDD − A group and is reversed in theMDD + A group. Therefore, this profile of VMPFC and amygdala activity andtheir interactions may distinguish MDD + A and may be responsible for theunique clinical presentation of patients with this subtype of MDD. Last, thedifferential abnormalities in VMPFC and amygdala function in the MDD + A groupare consistent with the dysfunction of both of these regions found in otherimpulsively aggressive patient populations.

There are some limitations of the present study. First, this study wasnot designed to assess sex differences during anger induction. Second, formalassessment for Axis II diagnoses was not performed in the study populations.Future studies that address these issues would be desirable. Last, structuralMRIs were not used for anatomic localization of significant rCBF changes.Instead, the MNI atlas, a spatially normalized composite of 152 MRIs of ahealthy brain, was used for localization purposes. Concerns regarding anatomiclocalization were further mitigated by the fact that we had concise, evidence-based,a priori hypotheses for the present study involving the ventral PFC and amygdala.

Patients with MDD + A experience a remission of anger attacks in concertwith remission of their depressive symptoms after successful treatment, whereasother patient populations that frequently exhibit impulsive aggression typicallyexhibit a more chronic, treatment-refractory clinical course. For these reasons,future studies that include assessments of MDD + A patients before and aftertreatment may provide valuable insight into the brain mechanisms underlyingthe resolution of these symptoms.

Correspondence: Darin D. Dougherty, MD, MSc, Massachusetts GeneralHospital–East, CNY-2612, Bldg 149, 13th Street, Charlestown, MA 02129(ddougherty@partners.org).

Submitted for publication July 7, 2003; final revision received January9, 2004; accepted February 17, 2004.

This study was supported by Mentored Patient-Oriented Research CareerDevelopment Award MH01735 from the National Institute of Mental Health, Bethesda,Md (Dr Dougherty).

We thank the individuals who served as research participants and SandraBarrow, BS, and Steve Weise, BS, for technical assistance.