Regional Cerebral Blood Flow in the Amygdala and Medial PrefrontalCortex During Traumatic Imagery in Male and Female Vietnam Veterans With PTSD FREE

Several recent functional neuroimaging studies have investigated brainactivation during exposure to trauma-related stimuli in posttraumatic stressdisorder (PTSD). For example, the presentation of combat sights and soundsto male combat veterans (MCVs) with PTSD has been associated with relativelyincreased activation in the amygdala^1- 2 andcerebellum^3- 5 andrelatively decreased activation in subcallosal gyrus.³ Therecollection of personal traumatic events (via the script-driven imagery paradigm)in PTSD has been associated with activation in the amygdala,⁶ orbitofrontalcortex, anterior temporopolar cortex,^6- 7 andinsular cortex⁶ and relatively decreased activationin anterior cingulate gyrus,^7- 8 medialfrontal gyrus,⁸ and subcallosal gyrus.⁹

Recent functional magnetic resonance imaging studies using cognitiveactivation paradigms have further demonstrated the importance of the amygdalaand medial prefrontal regions in PTSD. Rauch et al¹⁰ demonstratedhyperresponsivity of the amygdala to masked fearful facial expressions inMCVs with PTSD. Shin et al¹¹ reported diminishedrecruitment of anterior cingulate cortex during the emotional counting Strooptask in MCVs with PTSD. Medial prefrontal structural abnormalities also havebeen reported in PTSD, including decreased volumes of pregenual anterior cingulatecortex and subcallosal cortex¹² and diminishedneuronal integrity in anterior cingulate cortex.¹³

The results of these neuroimaging studies are broadly consistent withthe hypotheses that, in PTSD, the amygdala is hyperresponsive and medial prefrontalregions are hyporesponsive, and that these regions are reciprocally related.^14- 15 Although researchers have hypothesizedsuch a reciprocal relationship between the amygdala and medial prefrontalregions in PTSD, no previous studies in the literature have provided correlationaldata in support of this hypothesis. The most relevant evidence to date comesfrom Semple et al,¹⁶ who reported higher regionalcerebral blood flow (rCBF) in the amygdala and lower rCBF in anterior cingulate/medialfrontal gyrus in patients with PTSD and substance abuse. However, correlationsbetween rCBF changes in the amygdala and medial frontal regions were not reported.

Most of the neuroimaging studies of PTSD to date have included eitherMCVs or women with histories of physical or sexual abuse. In contrast to maleVietnam veterans, women who served as nurses in Vietnam have received relativelylittle research attention.^17- 20 Femalenurse veterans (FNVs) were exposed to horrific war-related injuries, mutilatedbodies, death, and threats to personal safety and thus were also at risk ofdeveloping many negative outcomes, including PTSD.^17- 18

In the present research, we studied rCBF in 36 male and female Vietnamveterans using positron emission tomography (PET) and a script-driven imageryparadigm.^6- 7,21 Inseparate conditions, participants recalled and imagined personal traumatic(war-related) and neutral events. During traumatic imagery, compared withneutral imagery, we predicted that veterans with PTSD would exhibit (1) greateractivation in the amygdala, orbitofrontal cortex, temporopolar cortex, andinsular cortex and (2) diminished activation in medial prefrontal regions(including medial frontal gyrus, rostral anterior cingulate gyrus, and subcallosalgyrus) compared with veterans without PTSD. We also performed parallel analysesin MCVs and FNVs separately to determine whether patterns of brain activationin our regions of interest differed in these subgroups. Given the lack ofneuroimaging data on FNVs, we had no a priori hypotheses regarding the directionof such subgroup differences. In addition, we conducted correlational analysesto determine (1) the functional relationship between the amygdala and medialprefrontal regions and (2) the relationship between PTSD symptom severityand rCBF in the amygdala and medial prefrontal regions during the traumaticimagery condition. To demonstrate that participants achieved an emotionalstate during scanning, we also analyzed their subjective ratings and psychophysiologicdata.

Participants were 36 right-handed²² Vietnamveterans without a history of head injury, neurologic disorders, or othermajor medical conditions. Seventeen participants (7 men and 10 women) met DSM-IV diagnostic criteria for current PTSD (PTSD group)and 19 participants (9 men and 10 women) never had PTSD (control group) accordingto the Clinician-Administered PTSD Scale (CAPS),²³ astructured clinical interview. All of the male participants had served incombat, and all of the female participants had served as nurses in Vietnam.Urine drug screen results were negative for all participants. No participantwas taking psychotropic or cardiovascular medications at the time of the study.

Demographic and clinical data are given in Table 1. Age, education, and CAPS scores were analyzed using separate2 (diagnosis: PTSD vs control) × 2 (subgroup: MCVs vs FNVs) analysesof variance. For the sake of brevity, we list only the statistically significanteffects. A main effect of subgroup was observed for education (F_1,32 = 12.3; P = .002). Female nurse veterans hada greater mean number of years of education than MCVs, reflecting FNVs' nursingtraining (Table 1). A significantmain effect of diagnosis was observed for CAPS scores (F_1,32 =205.1; P<.001) and for mean depression subscalescores on the Symptom Checklist-90–Revised²⁴ (F_1,32 = 17.4; P<.001). The PTSD group hadhigher scores on these measures than the control group.

The presence of other Axis I mental disorders was assessed using the Structured Clinical Interview for DSM-IV.²⁵ Participantsin the PTSD group met diagnostic criteria for the following current comorbiddiagnoses: major depression (3 MCVs and 5 FNVs), panic disorder (2 MCVs and1 FNV), social phobia (1 MCV and 1 FNV), specific phobia (2 FNVs), binge eatingdisorder (1 FNV), and somatoform disorder (1 FNV). Participants in the controlgroup met diagnostic criteria for dysthymia (2 MCVs), specific phobia (1 FNV),and somatoform disorder (1 MCV).

This study was approved by the institutional review boards of the MassachusettsGeneral Hospital, Boston, and the Veterans Affairs Medical Center, Manchester.Written informed consent was obtained from each participant.

The design and procedures of the script-driven imagery task were identicalto those reported elsewhere.^6- 7 Beforethe PET session, participants provided written descriptions of 2 neutral and2 Vietnam-related traumatic autobiographical events. After describing eachevent, participants examined a list of bodily responses (eg, "heart races"and "labored breathing") and circled those responses (if any) that they experiencedduring each event. Later, one of us (M.A.C. or M.L.M.) composed scripts describingeach event in the second person and the present tense, including up to 5 ofthe bodily responses that each participant selected.²¹ Thescripts were tape-recorded in a neutral voice for playback in the PET scanner.

Each participant was studied in 2 conditions (neutral and traumatic)with 2 scans (ie, replicates) per condition. (Participants completed othertasks in the larger scanning session, and those data will be reported separately.)During each scan, participants recalled and imagined the contents of a neutralor a traumatic script. The order of conditions was counterbalanced acrossparticipants.

Before each scan, participants were instructed to close their eyes,listen carefully to the script, and imagine the described event as vividlyas possible, as if they were actually participating in it. The PET camerawas turned on when the script started playing. Thirty seconds later, the scriptended and oxygen-15–labeled carbon dioxide administration began. Duringthe next 60 seconds, participants continued to recall and imagine the eventwhile PET data were acquired. Then, oxygen-15–labeled carbon dioxideadministration and PET data acquisition were terminated, and participantswere instructed to stop imagining the event. After a 30-second "recovery"period, participants gave ratings of their emotional state. The PET scanswere separated by at least 10 minutes to allow for radiation decay.

Participants' heart rate, skin conductance, and left lateral frontaliselectromyographic (EMG) responses were measured via a modular instrument system(Coulbourn Instruments, Allentown, Pa) in the PET laboratory (MassachusettsGeneral Hospital, Boston) according to established procedures.^21,26- 27

Psychophysiologic measurements were recorded for 30 seconds before eachPET scan (baseline), for 60 seconds during each PET scan (imagery), and for30 seconds immediately after each PET scan (recovery). Within the baselineand imagery periods (for each scan), readings were averaged. For each scan,the mean value during the baseline period was subtracted from the mean valueduring the imagery period, yielding "response" (ie, change) scores.

Immediately after each scan, participants rated the intensity of severalemotions using separate visual analog scales (0 = absent and 12 = maximal).^7,21,26 The rated emotionsincluded happiness, sadness, anger, fear, disgust, surprise, and guilt. Participantsalso rated their arousal level, vividness of imagery, awareness of presentsurroundings, and degree to which they felt that the visualized event washappening again.

The PET equipment and procedures have been described in previous studies.^6- 7,28 Briefly, PET data weregathered using a 15-slice, whole-body tomograph (Scanditronix PC4096; GeneralElectric Medical Systems, Milwaukee, Wis). The camera produced contiguousslices 6.5 mm apart, with axial resolution at 6.0-mm full-width half maximum(axial field, 97.5 mm). Images were reconstructed using a measured attenuationcorrection and a Hanning-weighted reconstruction filter set to allow for 8-mmin-plane spatial resolution (full-width half maximum).

After entering the scanner, each participant was fitted with a thermoplasticcustom-molded face mask, an overlying face mask attached to a vacuum, andnasal cannulae, which delivered the oxygen-15–labeled carbon dioxide.The concentration of oxygen-15–labeled carbon dioxide was 80 mCi/L (2960MBq/L); the flow rate was 2 L/min. Each participant's head was aligned inthe scanner relative to the canthomeatal line, and transmission measurementswere made using an orbiting pin source.

Statistical parametric mapping analysis of the PET data was conductedusing a computer software package (SPM99; Wellcome Department of CognitiveNeurology, London, England).²⁹ Within SPM99,all images were corrected for interscan movement using sinc interpolationand then were transformed into a standard stereotactic space using bilinearinterpolation. Images were then smoothed using a 2-dimensional Gaussian filterwith a width of 10-mm full-width half maximum. At each voxel, the PET datawere normalized using the global mean and fit to a linear statistical modelusing the method of least squares. Hypotheses were tested as contrasts inwhich linear compounds of the model parameters were evaluated using t statistics, which were then transformed to z scores.

We assessed our predictions with (1) separate voxelwise traumatic vsneutral contrasts in each diagnostic group and (2) a voxelwise test of thecondition × diagnosis interaction. Separate parallel statistical parametricmapping analyses were conducted for each subgroup. We chose to conduct thetraumatic vs neutral contrasts within a fixed-effects model because this procedureminimizes type II error. Although fixed-effects analyses limit our abilityto generalize from the study sample to the larger population of patients withPTSD, the present findings in the amygdala and medial prefrontal cortex (seethe "Results" section) are similar to those of previous studies^3,6- 7,9 usingfixed-effects models. Furthermore, random-effects analyses of the presentdata set revealed similar findings in the amygdala and medial prefrontal cortex.

To determine whether the rCBF changes in medial prefrontal regions wererelated to rCBF changes in the amygdala in PTSD, we (1) defined a functionalregion of interest (diameter, 4 mm) around the deactivation in medial frontalgyrus in the traumatic vs neutral contrast in the PTSD group (Montreal NeurologicalInstitute [MNI] coordinates, +10, +52, +2), (2) extracted rCBF values percondition per participant from that region of interest, (3) calculated thetraumatic vs neutral rCBF change score per participant, and (4) determinedwhether those change scores were associated with rCBF changes in other brainareas in the traumatic vs neutral comparison (using individual participant"con" images) via a voxelwise correlational analysis. Finally, within SPM99,voxelwise "covariates only" analyses were conducted to determine the relationshipbetween PTSD symptom severity (CAPS) scores and rCBF in the amygdala and medialprefrontal regions for the traumatic condition. All of these correlationalprocedures were performed on the entire PTSD group and separately for theMCV and FNV subgroups with PTSD. For the sake of brevity, we focused specificallyon the amygdala and medial prefrontal regions in the correlational analyses.

The statistical parametric maps resulting from the analyses describedherein were inspected for activations in our a priori regions of interest.Given our strong, directional hypotheses, we used a significance thresholdof P<.001, uncorrected (z≥3.09)for activations in these a priori regions. Most of the key activations occurringin the amygdala and medial prefrontal cortex would remain significant evenif we used an extremely conservative Bonferroni volume-corrected P = .002 (z≥2.88, 24 voxels of 6 ×6 × 6 mm) for the amygdala and P = .0007 (z≥3.19, 75 voxels) for medial prefrontal cortex. Becausethe procedure of correcting P values based on regionsize is biased toward finding statistical significance in small structures,we used the previously stated constant significance threshold. For regionsabout which we had no a priori prediction, we used a more conservative constantsignificance threshold of P<.00001, uncorrected(z≥4.27).

The statistical analyses of the psychophysiologic and subjective ratingdata were conducted not to present these results as findings in their ownright (which has already been done in numerous publications with essentiallythe same results) but rather to demonstrate that the predicted emotional activationsand associated group and condition differences, on which the validity of theneuroimaging findings depends, had been achieved. For this reason, correctionsfor multiple comparisons were not performed for the results of these analyses.

Heart rate, skin conductance, and EMG responses to traumatic and neutralimagery scripts were averaged across scans in each condition and were analyzedusing separate 2 (diagnosis: PTSD vs control) × 2 (subgroup: MCVs vsFNVs) × 2 (condition: neutral vs traumatic) analyses of variance (Table 2).

Regarding heart rate responses, the only significant effect was a maineffect of condition (F_1,32 = 22.7; P<.001).Heart rate responses were greater in the traumatic condition than in the neutralcondition across both groups (Table 2).

Regarding skin conductance responses, there were significant main effectsof diagnosis (F_1,32 = 4.1; P = .05) andcondition (F_1,32 = 19.2; P<.001). Thecondition × diagnosis interaction was also significant (F_1,32 =7.7; P = .01). Increases in skin conductance responsesfrom the neutral to the traumatic condition were greater in the PTSD groupvs the control group (Table 2).Also significant was the condition × diagnosis × subgroup interaction(F_1,32 = 10.7; P = .003). Examinationof the means indicated that the difference in skin conductance response changesbetween PTSD and control participants was greater in FNVs than in MCVs.

Finally, regarding EMG responses, significant main effects of diagnosis(F_1,32 = 14.0; P<.001), subgroup (F_1,32 = 4.3; P = .05), and condition (F_1,32 = 30.3; P<.001) were observed. Thecondition × diagnosis interaction was also significant (F_1,32 =11.5; P = .002). The EMG response increases fromthe neutral to the traumatic condition were greater in the PTSD group thanin the control group (Table 2).A significant condition × subgroup interaction was also found (F_1,32 = 4.3; P = .05). Inspection of the meansdemonstrated that EMG response increases from the neutral to the traumaticcondition were greater in FNVs than in MCVs.

Ratings were averaged across scans in each condition and were submittedto separate 2 (diagnosis: PTSD vs control) × 2 (subgroup: MCVs vs FNVs)× 2 (condition: neutral vs traumatic) analyses of variance. Significantmain effects of condition were observed for happiness, sadness, anger, fear,disgust, surprise, guilt, arousal, and vividness of imagery (for all: F_1,32≥20.4; P<.001). Compared with theneutral condition, the traumatic condition was associated with lower ratingsof happiness and higher ratings on all the other scales. Significant maineffects of condition also were observed for ratings of awareness of presentsurroundings and the degree to which participants felt that the visualizedevent was happening again (for all: F_1,30≥7.7; P≤.01). All participants reported feeling less aware of their surroundingsand more like the visualized event was happening again during the traumaticcondition relative to the neutral condition.

Furthermore, significant main effects of diagnosis were found for fear,arousal, disgust, and guilt (for all: F_1,32≥4.4; P≤.05). Ratings on these scales were higher in the PTSD group comparedwith the control group. Significant condition × diagnosis interactionswere observed for fear, arousal, guilt, and surprise (for all: F_1,32≥4.6; P≤.05). The PTSD group had greaterincreases on these scales than the control group. A significant condition× diagnosis × subgroup interaction was observed for ratings ofsurprise (F_1,32 = 4.9; P = .04). The increasein surprise ratings between conditions was stronger for MCVs with PTSD thanfor FNVs with PTSD. Finally, a significant condition × diagnosis interactionwas observed for the degree to which participants felt that the visualizedevent was happening again (F_1,30 = 5.6; P =.03); the PTSD group had greater increases on this scale compared with thecontrol group.

In the PTSD group (n = 17), the traumatic vs neutral comparison yieldedno statistically significant rCBF increases. Significant rCBF decreases occurredin medial frontal gyrus (z = 4.70; MNI coordinates,+10, +52, +2) (Figure 1 and Figure 2). Nonpredicted regions with significantrCBF decreases included superior frontal gyrus (z =4.35; MNI coordinates, −28, +62, +2), middle temporal gyrus (z = 4.61; MNI coordinates, −52, −10, −16; and z = 4.40; MNI coordinates, +58, −4, −26), inferiorparietal cortex (z = 5.00; MNI coordinates, −42,−50, +40), and occipital cortex (z = 5.20;MNI coordinates, −34, −90, +2).

In the control group, the traumatic vs neutral comparison yielded nostatistically significant rCBF increases or decreases. For completeness, wenote that a nonsignificant rCBF increase occurred in medial frontal gyrus(z = 2.60; MNI coordinates, +16, +58, +20).

The diagnosis × condition interaction revealed no regions withgreater rCBF increases in the PTSD group or greater decreases in the controlgroup. Regions with greater increases in the control group or greater decreasesin the PTSD group included medial frontal gyrus (z =3.56; MNI coordinates, +6, +58, +2) and occipital cortex (z = 4.79; MNI coordinates, −34, −88, +4).

Table 3 gives regions ofrCBF increases in MCVs and FNVs with PTSD. Increases in rCBF in the amygdalawere observed in MCVs with PTSD only. Table4 gives regions of rCBF increases in MCVs and FNVs without PTSD.Finally, diagnosis × condition interactions in the MCV and FNV subgroupsdemonstrated that amygdala activation in the traumatic vs neutral comparisonwas greater in MCVs with PTSD than in MCVs without PTSD (Table 5). No such differential activation in the amygdala was observedin the analogous interaction among FNVs.

In all of the PTSD participants, rCBF changes in medial frontal gyrus(MNI coordinates, +10, +52, +2) were negatively correlated with rCBF changesin the left amygdala (z = 3.23; MNI coordinates,−26, +2, −14) and the right amygdala/periamygdaloid cortex (z = 3.48; MNI coordinates, +18, +6, −22). In otherwords, smaller rCBF responses in medial frontal gyrus were associated withlarger rCBF responses in the amygdala and periamygdaloid cortex. These correlationsremained when participants with current major depression were removed fromthe analyses.

In addition, this relationship was observed in each PTSD subgroup separately.In MCVs with PTSD, rCBF changes in medial frontal gyrus were negatively correlatedwith rCBF changes in the left amygdala (z = 3.51;MNI coordinates, −16, +2, −12) and right periamygdaloid cortex(z = 2.88; MNI coordinates, +22, +6, −26).In FNVs with PTSD, rCBF changes in medial frontal gyrus were negatively correlatedwith rCBF changes in the left amygdala (z = 3.54;MNI coordinates, −28, +4, −18) and the right amygdala/periamygdaloidcortex (z = 3.38; MNI coordinates, +16, +4, −18).

To assess the specificity of this inverse functional relationship betweenmedial frontal gyrus and amygdala, analogous correlational analyses were performedusing rCBF change data in another prefrontal region that demonstrated rCBFdecreases in the traumatic vs neutral comparison in the PTSD group; rCBF changesin superior frontal gyrus were not correlated with rCBF changes in the amygdala.In addition, no other regions were statistically significantly correlatedwith rCBF changes in medial frontal gyrus except in MCVs with PTSD. In thatsubgroup, we also observed positive correlations with the left hippocampus(z = 4.37; MNI coordinates, −28, −12,−12) and fusiform gyrus (z = 4.28; MNI coordinates,+38, −46, −8) and negative correlations with the cerebellum (z = 4.68; MNI coordinates, +20, −30, −16) andorbitofrontal cortex (z = 4.42; MNI coordinates,+26, +42, −14).

In the PTSD group (n = 17), CAPS scores were positively associated withrCBF in the traumatic condition in the right amygdala (z = 3.83; MNI coordinates, +28, +4, −14) (although centered onthe anterior/lateral margin of the amygdala, this activation extended posteriorlyand medially to include the entire right amygdala and the anterior right hippocampus[z = 4.01; MNI coordinates, +22, −10, −16]).In MCVs with PTSD, CAPS scores were positively associated with rCBF in thetraumatic condition in the right amygdala (z = 2.93;MNI coordinates, +24, +4, −16) extending posteriorly to the right hippocampus(z = 3.07; MNI coordinates, 20, −12, −16)and were negatively associated with rCBF in medial frontal gyrus (z = 3.29; MNI coordinates, −2, +52, +6) and anterior cingulategyrus (z = 3.46; MNI coordinates, −4, +20,+28). In FNVs with PTSD, CAPS scores were also positively associated withthe right amygdala/hippocampus (z = 2.63; MNI coordinates,+24, −6, −14) and negatively associated with medial frontal gyrus(z = 4.40; MNI coordinates, +10, +70, −6).These relationships remained even after controlling for depression severityscores.

During the recollection and imagery of traumatic vs neutral personalevents, MCVs and FNVs with PTSD exhibited rCBF decreases in medial frontalgyrus. Only MCVs showed rCBF increases in the left amygdala. However, forboth subgroups with PTSD, rCBF changes in medial frontal gyrus were inverselycorrelated with rCBF changes in the left amygdala and the right amygdala/periamygdaloidcortex. Furthermore, in the traumatic condition, for both subgroups with PTSD,symptom severity was positively related to rCBF in the right amygdala andnegatively related to rCBF in medial frontal gyrus.

Our finding of decreased activation of medial frontal gyrus during thesymptomatic state in PTSD is consistent with the results of previous research^3,7- 9 and withmodels of PTSD that hypothesize abnormal function of medial prefrontal structures.However, not all studies have reported such decreased activation of medialprefrontal regions in PTSD. For example, using single-photon emission computedtomography, Zubieta et al³⁰ reported relativelyincreased medial prefrontal cortex blood flow in a combat sounds conditionrelative to a white noise condition. The reason for such a disparate resultis unclear but may be attributed to the different neuroimaging techniques(single-photon emission computed tomography) implemented in that study. Laniuset al^8,31 found that functionalmagnetic resonance imaging signal changes in medial frontal gyrus and anteriorcingulate gyrus may vary depending on the dissociative state of participants;specifically, patients with PTSD who dissociated during scanning showed increasedactivation in these regions, whereas those who did not dissociate showed lessactivation in these regions. Thus, perhaps discrepant findings in the literatureregarding medial prefrontal regions can be explained by variability in thedissociative state of participants across (and within) studies.

In participants with PTSD, blood flow changes in medial frontal gyruswere inversely correlated with blood flow changes in the amygdala. This findingresonates with those of a previous study¹⁶ reportingdecreased blood flow in anterior cingulate gyrus/medial frontal gyrus andincreased blood flow in the amygdala in patients with PTSD. Although the directionof causality cannot be inferred from correlational analyses, the relationshipbetween the amygdala and medial prefrontal cortex is likely to be reciprocal.Medial prefrontal regions send projections to the amygdala in primates,^32- 35 andthey may play an important role in the process of extinction of fear conditioning.^36- 39 Conversely,in rodents, prefrontal neurons show decreases in spontaneous activity in thepresence of a conditioned aversive tone as a function of amygdala activity,suggesting that the amygdala may modulate prefrontal neuronal activity.⁴⁰ Longitudinal and twin studies of PTSD may help elucidatethe precise functional relationship between these structures and determinewhether the more primary abnormality involves medial prefrontal regions orthe amygdala.

In the PTSD group, PTSD symptom severity was positively correlated withrCBF in the right amygdala and negatively correlated with rCBF in medial frontalgyrus during traumatic imagery. Positive correlations between subjective ratingsof distress and right amygdala activity have been reported previously in PTSD² and in social anxiety.⁴¹ Thepresent results supplement these previous findings.

In the present study, left amygdala activation in the traumatic vs neutralcomparison was found among MCVs with PTSD but not among FNVs with PTSD. Thisdifference between PTSD subgroups may be driven by differences in trauma type(eg, directly experiencing danger vs witnessing frightening situations) orsex or may have reflected a type II error in FNVs with PTSD. However, regardingrCBF findings in our a priori regions of interest, these 2 subgroups exhibitedmore commonalities than differences.

In contrast to the results of previous research, we found no evidenceof rCBF increases in orbitofrontal cortex, temporopolar cortex, or insularcortex during traumatic imagery in the PTSD group. An explanation for thisremains elusive, especially given that the methods implemented in this studywere similar to those of previous studies.^6- 7 Inaddition, previous studies have reported increased activation in medial frontalregions in control participants without PTSD.^3,7,9 Inthe present study, FNVs without PTSD exhibited significant activation in medialfrontal gyrus (Table 4), but similaractivations in MCVs without PTSD, and in the control group as a whole, fellbelow our significance threshold. The reason for subthreshold medial frontalactivation in the control group is unclear.

This study cannot (and was never intended to) directly assess sex differencesin neural responses to traumatic imagery because the men and women in thisstudy experienced different types of trauma. Thus, any rCBF differences betweenMCVs and FNVs may be attributable to sex, trauma type, or a combination ofthe two. In addition, although the number of participants in our PTSD groupwas relatively large by functional neuroimaging standards (n = 17), analyzingdata in MCVs and FNVs separately resulted in a relative loss of power andan increased risk of type II error for the subgroup analyses. However, despitethe small numbers per subgroup, the rCBF results in a priori regions of interestwere strikingly similar in MCVs and FNVs with PTSD. The present study is limitedby the presence of comorbidity in the PTSD group, and future neuroimagingstudies of PTSD should use psychiatric control groups. It should be noted,however, that the key results from the present study remained even after controllingfor depression. Finally, the present design did not include a low-level baselinecondition as a secondary comparison condition, which could have helped determinethe patterns of activation associated with the neutral and traumatic conditionsseparately and might have further clarified activation differences betweengroups.

In summary, veterans with PTSD exhibited rCBF decreases in medial frontalgyrus during traumatic vs neutral script-driven imagery. These rCBF changeswere inversely correlated with rCBF changes in the left amygdala and the rightamygdala/periamygdaloid cortex. Furthermore, in the traumatic condition, symptomseverity was positively related to rCBF in the right amygdala and negativelyrelated to rCBF in medial frontal gyrus. These results are consistent withfunctional neuroanatomic models of PTSD that posit a reciprocal relationshipbetween medial prefrontal cortex and amygdala in PTSD.

Corresponding author: Lisa M. Shin, PhD, Department of Psychology,Tufts University, 490 Boston Ave, Medford, MA 02155 (e-mail: lisa.shin@tufts.edu).

Submitted for publication March 27, 2003; final revision received July2, 2003; accepted July 10, 2003.

This study was supported by merit review grants from the Veterans AffairsMedical Research Service (Drs Pitman, Carson, and Orr); a Young InvestigatorAward from the National Alliance for Research on Schizophrenia and Depression,Great Neck, NY (Dr Shin); and grant MH-60219 from the National Institute ofMental Health, Bethesda, Md (Dr Rauch).

We thank Paul Whalen, PhD, Cary Savage, PhD, Christopher Chabris, PhD,and Christopher Wright, MD, PhD, for their comments on this manuscript; theindividuals who served as research participants; and Sandra Barrow, BS, AvisLoring, RTN, and Steve Weise, BS, for their technical assistance.