Neuropsychologia 49 (2011) 1771–1778

Contents lists available at ScienceDirect

Neuropsychologia

journa l homepage: www.e lsev ier .com/ locate /neuropsychologia

Neural correlates of recovery from post-traumatic stress disorder:A longitudinal fMRI investigation of memory encoding

Erin W. Dickiea, Alain Bruneta,b, Vivian Akeriba, Jorge L. Armonya,b,!

a Douglas Mental Health University Institute, 6875 LaSalle Boulevard, F.B.C. Pavillon, Verdun, QC H4H 1R3, Canadab Department of Psychiatry, McGill University, Montreal, Quebec, Canada

a r t i c l e i n f o

Article history:Received 15 December 2010Received in revised form 11 February 2011Accepted 27 February 2011Available online 5 March 2011

Keywords:Post-traumatic stress disorderFunctional magnetic resonance imagingRecoveryAmygdalaMedial prefrontal cortexHippocampus

a b s t r a c t

Post-traumatic stress disorder (PTSD) is characterized by a failure of psychological recovery from a trau-matic experience. At a neural level, it is associated with abnormalities of the areas of the neural system thatprocess threatening information, including the amygdala and medial-prefrontal cortex, as well as of thatinvolved in episodic memory, including the hippocampus. However, little is known about how the func-tion of these regions may change as one recovers from the disorder. In this investigation, PTSD patientsunderwent two functional magnetic resonance imaging (fMRI) scans, 6–9 months apart, while viewingfearful and neutral faces in preparation for a memory test (administered outside the scanner). At Time2, 65% of patients were in remission. Current symptom levels correlated positively with memory-relatedfMRI activity in the amygdala and ventral-medial prefrontal cortex (vmPFC). In addition, the change inactivity within the hippocampus and the subgenual anterior cingulate cortex (sgACC) was associated withthe degree of symptom improvement (n = 18). These results suggest differential involvement of structureswithin the fear network in symptom manifestation and in recovery from PTSD: whereas activity withinthe amygdala and vmPFC appeared to be a marker of current symptom severity, functional changes inthe hippocampus and sgACC reflected recovery. These results underscore the importance of longitudinalinvestigations for the identification of the differential neural structures associated with the expressionand remission of anxiety disorders.

© 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Post-traumatic stress disorder (PTSD) is an anxiety disorder aris-ing as a result of exposure to a traumatic event (Diagnostic andStatistical Manual of Mental Disorders, 4th ed., Text Revision (DSM-IV-TR), 2000). It is associated with a variety of symptoms, includingintrusions (e.g., nightmares and flashbacks of the trauma), avoid-ance of trauma reminders and hyperarousal. Neuroimaging studieshave shown decreased activity in the medial prefrontal cortex(mPFC) and, less frequently, increased activity of the amygdalain PTSD individuals exposed to trauma reminders, compared tocontrols (Francati, Vermetten, & Bremner, 2007; Rauch, Shin, &Phelps, 2006). Importantly, similar patterns of activity, partic-ularly an exaggerated amygdala response, have been observedwith trauma unrelated threat-relevant stimuli (e.g., fearful faces)

! Corresponding author at: Douglas Mental Health University Institute,6875 LaSalle Boulevard, F.B.C. Pavillon, Verdun, QC H4H 1R3, Canada.Tel.: +1 514 761 6131x3360; fax: +1 514 888 4099.

E-mail addresses: erin.dickie@mail.mcgill.ca (E.W. Dickie),alain.brunet@douglas.mcgill.ca (A. Brunet), vakerib@videotron.ca (V. Akerib),jorge.armony@mcgill.ca (J.L. Armony).

(Armony, Corbo, Clement, & Brunet, 2005; Rauch et al., 2000; Shinet al., 2005), suggesting that PTSD is associated with a generaldysfunction of the brain’s networks involved in the detection ofthreatening information (Armony & LeDoux, 1997; Rauch et al.,2006). In addition, PTSD is also thought to involve a dysfunctionof the memory system, as some of its symptoms are directly asso-ciated with the traumatic memory itself (Ehlers, Hackmann, &Michael, 2004), and as patients consistently perform more poorlythan controls on memory tests (Brewin, Kleiner, Vasterling, & Field,2007; Isaac, Cushway, & Jones, 2006).

Notably, only a fraction of trauma-exposed individuals developPTSD and, of these, a large proportion recover with time (Breslau,2001; Breslau et al., 1998; North, Smith, & Spitznagel, 1997). Ithas therefore been proposed that chronic PTSD be considered theresult of two types of pathogenic processes, the first characterizedby exaggerated responses to a stressful event, and the second byinadequate mechanisms of recovery (Yehuda & LeDoux, 2007).

While most previous studies have focused, directly or indirectly,on the first question, by exploring the neural correlates of currentPTSD, little is known about the brain processes involved in thesuccessful recovery from the disorder. As a first step in addressingthis issue, we conducted a longitudinal fMRI study on a group ofindividuals suffering from PTSD using an emotional face memory

0028-3932/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.doi:10.1016/j.neuropsychologia.2011.02.055

1772 E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778

paradigm. We chose this paradigm (adapted from Sergerie, Lepage,& Armony, 2005) as it can be considered a point of convergenceof the two main processes thought to be affected in PTSD, namelythreat-detection and memory. We tested subjects twice, oncewhen they were highly symptomatic and again 6–9 months later,when most, but not all, of the participants had seen their symp-toms diminish substantially. Results from data obtained at Time1 were previously reported (Dickie, Brunet, Akerib, & Armony,2008) and showed that PTSD symptom severity correlated withoverall memory encoding-related activity in the ventral medialprefrontal cortex and with the successful encoding of fearful, butnot neutral, faces in the amygdala. Here, we directly comparedactivity between the two sessions and therefore were able todirectly identify (1) the neural correlates of current PTSD symptomseverity (i.e., those regions showing a significant correlation withcurrent PTSD symptom severity at both time points), and (2) thebrain regions associated with the recovery process, that is, areaswhere change in neural activity between testing times correlatedwith the improvement of PTSD symptom severity.

2. Methods

2.1. Participants

Thirty-two individuals (age range: 20–60 years) with a primary diagnosis ofPTSD were recruited from two clinics in Montreal (Traumatys Clinic and CharlesLeMoyne Hospital). PTSD diagnosis was verified in all participants using theClinician-Administered PTSD Scale (CAPS; Blake et al., 1995). All participants hada CAPS score greater than 45 at initial assessment and no history of neurological,learning or psychotic disorders. Psychiatric comorbidity and depressive symptomswere assessed with the MINI International Neuropsychiatric Interview (Sheehanet al., 1998) and Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock,& Erbaugh, 1961), respectively. Most patients underwent psychotherapy betweenfirst (Time 1) and second (Time 2) testing sessions (confirmed by all but two partic-ipants contacted at follow-up). As we were not able to fully characterize the precisenature of this treatment (beyond prescribed medications), we did not include itin the analysis as an independent variable of interest. Participants provided writ-ten informed consent and received financial compensation for their time and travelexpenses. All procedures were approved by the ethical review boards of the DouglasMental Health University Institute, the Faculty of Medicine of McGill University andthe Charles Lemoyne University Hospital.

Of the 32 participants recruited, complete data was available for 27 participantsat Time 1 (Dickie et al., 2008). For 18 of these participants, complete data was alsoavailable at Time 2, 6–9 months following the first scan [exclusions: dropouts (N = 6),excessive movement during the scan (N = 2), technical malfunction (N = 1)]. Theirdemographic and clinical information is listed in Table 1. Two additional subjectswere included in the Time 2 analysis (cf. Section 3.3.1) even though their data atTime 1 was incomplete due to no responses for one of the stimulus categories (seeSection 2.3 below).

2.2. Experimental design

The stimuli and experimental procedure used have been previously described(Dickie et al., 2008) and are depicted in Fig. 1. Briefly, 224 face stimuli (112 fearful,112 neutral; half female) were selected from several validated databases, modifiedto remove gender-typical features and pseudo-randomly divided into four equiv-alent sets of 56 faces. Faces from each set were used as either encoding stimuli orlures at either Time 1 or Time 2 (counterbalanced across subjects). Consequently, theprocedures for Time 1 and Time 2 testing were identical and no face was repeatedbetween testing sessions. Fifty-six indoor and outdoor scenes were also presented,intermixed with the faces during the fMRI scanning (data not reported here). Dur-ing the fMRI scans (encoding session), one set of faces was presented twice and 56scenes were presented once for 1500 ms each, preceded by a 1000 ms cue instruct-ing subjects to indicate, by pressing one of two buttons, whether the face presentedwas male or female, or whether the scene corresponded to an interior or an exte-rior location. In addition, subjects were told to memorize all images in preparationfor a memory test (the longitudinal aspect of the study precluded the use of a sur-prise recognition test). The inter-stimulus interval was 2000 ms on average; longerintervals (so-called null events) were also included in order to obtain an adequateestimate of baseline activity (Josephs, Turner, & Friston, 1997).

Following the scanning session, participants completed a face recognition mem-ory test. The encoding set and a new one (lures) were presented on a PC laptop screenwhile participants performed an old/new task. Faces were presented for 1500 mseach with a 2000 ms inter-stimulus interval.

Table 1Demographic and clinical data (n = 18).

Mean years of age (SD) 36.8 (11.8)Years of education (SD) 12.9 (1.9)Number right handed (%) 15 (83%)Number female (%) 13 (72%)Trauma typeMotor vehicle accident 9 (50%)Physical assault or death threat 2 (11%)Sexual assault 1 (6%)Witness to physical assault or death threat 6 (33%)Peri-traumatic distress inventory (Brunet et al.,

2001)30.1 (12.1)

Peri-traumatic dissociative experiencequestionnaire (Birmes et al., 2005)

28.2 (9.61)

Mean weeks since trauma (SD) 17.8 (13.0)Mean time 1 to time 2 weeks (SD) 31.0 (4.4)Number recovered at T2 (%) 11 (61%)

Repeated measures Time 1 Time 2

Clinical measure means (SD)Clinician administered PTSD scale (CAPS) 80.6 (15.9) 44.7 (29.9)Beck depression inventory (BDI)a 9.9 (5.45) 7.7 (6.3)a

Number with comorbid disorder (%) 14 (78%) 7 (39%)Major depression 9 (50%) 4 (22%)Other anxiety disorder 10 (56%) 6 (33%)Bulimia 1 (6%) 1 (6%)Number taking medication (%)a 8 (44%) 7 (39%)Anti-depressants 7 (39%) 7 (39%)Benzodiazepines 5 (28%) 1 (6%)Anti-psychotic 1 (6%) 2 (11%)

a Missing data from one participant.

Fig. 1. Schematic of the memory encoding task. During the scan, 28 fearful and28 neutral faces were presented, as well as 56 scenes, in a pseudo-random order.Depending on the preceding cue, participants were instructed to indicate whetherthe face presented was male or female (M/F), or whether the scene corresponded toan interior or an exterior location (I/E). Immediately following the encoding scanningsession, the encoding set and an equal set of new faces (lures) were presented inpseudo-random order on a PC laptop. Participants indicated whether the face wasseen during the preceding scan (old) or never seen before (new).

E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778 1773

2.3. fMRI acquisition and analysis

Scanning was conducted at the Montreal Neurological Institute (MNI) using a1.5 Tesla Siemens Sonata whole-body system equipped with a standard head coil.Stimuli were generated by a PC laptop computer running E-PRIME (Psychology Soft-ware Tools, Pittsburgh, PA) and displayed using an LCD projector, a screen and amirror system. An optical mouse collected the participants’ responses. FunctionalT2*-weighted images were collected with blood oxygen level dependent (BOLD)contrast (360 volumes, TR = 2450 ms, TE = 50 ms, Flip angle = 90" , FOV = 256 mm,matrix = 64 # 64), covering the entire brain (30 interleaved slices, 4 mm thickness,parallel to the anterior–posterior commissural plane). Prior to the functional scan aT1-weighted anatomical volume was collected using a gradient echo pulse sequence(TR = 22 ms, TE = 9.2ms, Flip angle = 30" , voxel size 1 mm3 isotropic).

Functional data were analyzed using SPM2 as in our previous study (Dickie et al.,2008). Briefly, images were time-corrected, realigned to the first volume, spatiallynormalized to MNI space (Evans, Kamber, Collins, & MacDonald, 1994) (final voxelsize: 2 # 2 # 2 mm3) and smoothed with an isotropic 8 mm FWHM Gaussian kernel.A high-pass filter (cut off 128 s) was applied to remove low-frequency temporaldrifts in the signal. Individual events were modeled by a synthetic hemodynamicresponse and its temporal derivative. The experiment was analyzed as a 2 # 2 facto-rial design with memory and emotional expression as the main factors. Therefore,four event types were defined based on the stimulus type and each participant’sresponses in the memory recognition test: subsequently remembered fearful (FR)and neutral (NR) faces, and subsequently forgotten fearful (FF) and neutral (NF) faces.In addition, scenes and the movement parameters obtained during the realignmentprocedure were also included in the model. Following our previous analysis (Dickieet al., 2008) and current hypotheses, two first-level linear contrasts were calculatedcorresponding to the main effect of overall subsequent memory success (remem-bered minus forgotten: [FR + NR] $ [FF + NF]) and successful memory for fearful faces(memory-by-emotion interaction: [FR $ FF] $ [NR $ NF]) (Dolcos, LaBar, & Cabeza,2004). Brain correlates of current symptom severity at Time 2 were explored byperforming a second-level (random-effects) regression analysis between each ofthese two contrasts and CAPS scores, both measured at Time 2 (see Section 3.3.1).Small volume correction (SVC) for multiple comparisons was applied to functionallydefined regions-of-interest within the amygdala and vmPFC based on the activa-tions obtained for the same contrasts at Time 1 (Dickie et al., 2008), thresholded atp = 0.05 (uncorrected). This analysis allowed us to answer our first question, namelywhether there were any common areas associated with current symptom severity,as a function of overall and/or emotional memory, at both time points.

A separate analysis was performed to address our second and main objective,namely to identify regions where changes in memory-related activity between Time1 and Time 2 significantly correlated with symptom severity changes. To do so,difference maps were created, for each subject and for each of the two first-level con-trasts of interest defined above, by subtracting the corresponding images obtainedat Time 1 from those obtained at Time 2 (i.e., [Time 2 minus Time 1], so that posi-tive values represented an increase in signal over time). These contrast images werethen entered in second-level whole-brain random-effects regression analysis withsymptom improvement, defined as CAPS1 minus CAPS2, as the explanatory variable(cf. Section 3.3.2). Threshold for whole-brain analyses was set at p < 0.05 correctedfor multiple comparisons (false discovery rate; Genovese, Lazar, & Nichols, 2002). Inaddition, because of their hypothesized involvement in PTSD, small volume correc-tion was applied, for the second analysis, to anatomically defined regions-of-interestfor the amygdala, vmPFC, hippocampus and the anterior cingulate cortex, createdusing the WFU pickatlas (Maldjian, Laurienti, Kraft, & Burdette, 2003), according tothe labeling of Tzourio-Mazoyer et al. (2002).

Single-subject, first-level parameter estimates (effect sizes) for the conditions ofinterest were extracted from the peak voxels of the significant activations obtainedin the different analyses for plotting and post hoc analyses.

3. Results

3.1. Clinical scores

In all but one of the participants, CAPS scores diminished fromTime 1 to Time 2. Sixty-five percent of the 20 participants scannedat Time 2 no longer met the diagnostic criteria for PTSD (Recoveredgroup). CAPS scores at Time 1 and Time 2 were not significantly cor-related (r = 0.32, p = 0.17). Symptom Improvement (CAPS1-CAPS2)was significantly correlated with CAPS2 (r = $0.85, p < 0.001) butnot with CAPS1 (r = 0.20, p = 0.43). As expected, CAPS and depres-sive symptoms (BDI) scores significantly correlated at both Time 1(r = 0.53, p < 0.05) and Time 2 (r = 0.55, p < 0.05). In addition, therewas a trend-level correlation (r = 0.44, p = 0.06) between the changein BDI and CAPS scores across testing sessions. No other clinicalor demographic variables were significantly correlated with thereduction in PTSD symptom severity.

Table 2Means and standard deviations for memory performance (hits-false alarms, HFA) atboth time points.

Memory performance means (SD) HFA fearful HFA neutral

Time 1 0.25(0.13) 0.24(0.18)Time 2 0.26(0.14) 0.23(0.15)

All values are significantly greater than chance performance (0).

3.2. Memory performance

Hit minus False Alarm rates for fearful and neutral faces atboth time points are shown in Table 2. A repeated-measuresanalysis of covariance on memory performance, with time andfacial expression as the within-subject factors, and the degree ofrecovery (i.e., CAPS1-CAPS2) as an explanatory variable revealed atime-by-change-in-CAPS interaction (F(1, 18) = 5.71, p < 0.05). Thisinteraction reflects an improvement in overall memory perfor-mance as a function of symptom severity reduction from Time 1 toTime 2 (r = 0.49, p < 0.05; see Fig. 2). This relationship was reducedto a trend if only those participants with intact neuroimaging dataare included (r = 0.44, p = 0.07).

3.3. fMRI analyses

3.3.1. Correlates of current symptom severityResults obtained in the analysis of the relation between

memory-related activity and current symptom severity at Time 2replicated those previously reported for Time 1 (Dickie et al., 2008).Specifically, there was a significant positive correlation betweenoverall subsequent memory success ([FR + NR] $ [FF + NF])T2 andcurrent CAPS scores within the vmPFC (xyz = [2 54 $16], Z = 3.27,pSVC = 0.04; see Fig. 3a and b), as well as a positive correlationbetween memory-by-emotion interaction ([FR $ FF] $ [NR $ NF])T2and CAPS scores within the left amygdala (xyz = [$16 $2 $16],Z = 3.03, pSVC = 0.03; see Fig. 3a and c). These activation clustersoverlapped with those obtained for the corresponding contrastsat Time 1 (see Section 2). There was a positive correlation betweenresponses of the peak amygdala and vmPFC voxels associated withthe successful encoding of fearful faces ([FR $ FF]) at both times

Fig. 2. Scatterplot depicting the correlation (r = 0.49, p = 0.02) between improve-ment in memory performance [(H-FA)T2 $ (H-FA)T1] and diminishment in PTSDsymptoms (CAPS1–CAPS2; n = 23). Open circles represent those participantsincluded only in the behavioural analyses (see Section 2 for fMRI analyses exclusioncriteria).

1774 E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778

Fig. 3. (a) Statistical parametric map depicting the clusters within the right ventral medial prefrontal cortex (vmPFC) and left amygdala where activity related to successfuloverall and emotional memory encoding, respectively, correlated with current symptom severity at Time 2 (CAPS2). Regions outlined in red correspond to the areas wherethe same contrasts were significant at Time 1. Scatterplots depicting the significant correlation between (b) vmPFC and (c) amygdala activity and CAPS2 for the Rememberedminus Forgotten contrasts for fearful faces [(FR $ FF)T2] (Top) and neutral faces [(NR $ NF)T2] (Bottom).

(Time 1: Z = 3.17, p < 0.005; Time 2: Z = 3.30, p < 0.005). Importantly,no significant correlation was observed for the contrast involv-ing only neutral faces ([NR $ NF]; Time 1: Z = 1.02, p = 0.3; Time 2:Z = 0.38, p = 0.7).

3.3.2. Correlates of recoveryAnalysis of brain activity associated with symptom

severity changes across time (see Methods) revealed asignificant negative correlation for the successful encod-ing of faces, regardless of emotional expression (i.e.,([FR + NR] $ [FF + NF])T2 $ ([FR + NR] $ [FF + NF])T1) and CAPS scores

difference (CAPS1–CAPS2) in the right head of the hippocampus(xyz = [32 $12 $18], Z = 3.55, pSVC = 0.04; Fig. 4a–c). Further analysisof this activation revealed that this effect was mainly due toa significant positive correlation between BOLD activity for thiscontrast at Time 1 and degree of recovery (Z = 3.17, p < 0.001; Fig. 4dand e). No correlations between BOLD activity and current CAPSscores were observed in this cluster for either Time 1 (Z = 1.04,p = 0.3, see Fig. 4f) or Time 2 (Z = 1.42, p = 0.2).

In the case of changes across time for the emotion-by-memory interaction contrast (i.e., ([FR $ FF] $ [NR $ NF])T2 $([FR $ FF] $ [NR $ NF])T1), a significant negative correlation with

E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778 1775

Fig. 4. (a) Statistical Parametric Maps depicting coronal and sagittal views of the cluster within the right hippocampus head where the change in activity associated withoverall successful memory encoding, ([FR + NR] $ [FF + NF])T2 $ ([FR + NR]–[FF + NF])T1, correlated with symptom improvement (CAPS1–CAPS2). (b) Scatterplot depicting theparameter estimates for this correlation. (c) For illustrative purposes, the mean (±SE) change in memory encoding activity for Recovered and Non-Recovered groups areplotted separately. (d) Scatterplot of the parameter estimates for successful memory encoding activity at Time 1 as a function of change in CAPS scores (CAPS1–CAPS2). (e)Mean (±SE) memory encoding contrast estimates at Time 1 for Recovered and Non-Recovered groups. (f) Scatterplot of the parameter estimates for the memory encodingcontrast at Time 1 as a function of current symptoms (CAPS1).

changes in CAPS scores was observed in the left subgenual ante-rior cingulate cortex (sgACC) (xyz = [$4 26 $8], Z = 4.21, pSVC = 0.02;Fig. 5a–c). Post hoc analyses of data from the peak voxel of thiscluster showed that activity at Time 1 was significantly correlatedwith the degree of recovery (i.e., CAPS1–CAPS2, Z = 2.01, p < 0.05,see Fig. 5d and e) but not current symptoms (Z = 0.05, p = 0.5, seeFig. 5f).

No other regions were significant in the whole-brain analysis(p < 0.05, FDR). For exploratory purposes, activations at a lowerthreshold (p < 0.001, uncorrected) are reported in SupplementalTable S2.

The relationship between the change in activity in the hip-pocampus and sgACC and the change in CAPS scores remainedsignificant after including several potential confounding regressorsin the analysis (time elapsed since trauma, time between testingsessions, participants’ age, change in BDI, the presence of a co-morbid affective disorder and the use of prescribed anxiolytic oranti-depressant medication; see Supplemental Figs. S1 and S2).

4. Discussion

This study investigated the neural correlates of recovery fromPTSD using an emotional face memory encoding task in a longi-tudinal fMRI design. Replicating our previous report (Dickie et al.,2008), current PTSD symptoms significantly correlated with activ-ity in the vmPFC as a function of memory and the amygdala as afunction of emotional memory, suggesting that these regions may

act as functional markers of current symptom severity. In contrast,changes in activity of the hippocampus (as a function of memory)and sgACC (as a function of emotional memory) correlated withimprovement in PTSD symptoms, suggesting that activity in theseareas may be associated with recovery.

4.1. Neural correlates of current symptomatology

Numerous studies, using a variety of tasks and stimuli, havereported significant positive correlations between PTSD symptomseverity and amygdala activity (Armony et al., 2005; Rauch et al.,2000) and both positive (Bryant, Kemp, Felmingham et al., 2008)and negative (Britton, Phan, Taylor, Fig, & Liberzon, 2005; Shin,Orr, et al., 2004; Shin et al., 2005; Williams et al., 2006) correla-tions with activity of the vmPFC. Importantly, these correlationshave been observed within one month (Armony et al., 2005), oneyear (Williams et al., 2006) and decades (Rauch et al., 2000; Shin,Orr, et al., 2004) after trauma exposure, suggesting that they cor-respond to an individual’s current state rather than being a markerof vulnerability or resiliency for the disorder. Our present resultsprovide direct support for this notion: we observed a correlationbetween activation in these areas and current symptom severityin the same subjects across time, even though most patients hadexhibited some recovery in the time that elapsed between test-ing sessions. Importantly, the correlates of PTSD symptom severitywithin the amygdala and the vmPFC were found using separatecontrasts. It is thus probable that our task probes more than one

1776 E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778

Fig. 5. (a) Statistical parametric maps depicting the cluster within the subgenual anterior cingulate (sgACC) where the change in activity associated with the interactionof successful memory and emotional expression ([FR $ FF] $ [NR $ NF])T2 $ ([FR $ FF] $ [NR $ NF])T1) correlated with the reduction of symptom severity (CAPS1–CAPS2). (b)Scatterplot depicting this correlation. (c) For illustrative purposes, the mean (±SE) change in memory encoding activity for Fearful and Neutral Faces for Recovered andNon-Recovered groups are plotted separately. (d) Scatterplot of the parameter estimates for the interaction of successful memory and emotional expression at Time 1 asa function of change in CAPS scores (CAPS1–CAPS2). (e) Mean (±SE) remembered minus forgotten contrast estimates for fearful and neutral faces at Time 1 for Recoveredand Non-Recovered groups. (f) Scatterplot of the parameter estimates for the interaction of successful memory and emotional expression at Time 1 as a function of currentsymptoms (CAPS1).

neural system related to PTSD symptom severity: one for episodicmemory, including the vmPFC, and the other for the modulationof memory by emotion, with the amygdala playing a key role.Nonetheless, these two processes are likely related, as evidenced bythe significant positive correlation between amygdala and vmPFCactivity for successfully remembered fearful, but not neutral, facesat both time points.

4.2. Neural correlates of the recovery process

While there is a consensus that hippocampal volume in PTSDpatients is reduced (Karl et al., 2006), studies of hippocampal func-tion in this population have yielded mixed results. Reports of bothincreased (Osuch et al., 2001; Piefke et al., 2007) and decreased(Bremner et al., 1997, 1999) hippocampal activity during symptomprovocation exist. Studies that tried to elicit hippocampal activityusing memory tasks are equally inconsistent, with increased (Shin,Shin et al., 2004; Thomaes et al., 2009; Werner et al., 2009) anddecreased (Bremner et al., 2003a, 2003b; Geuze, Vermetten, Ruf,de Kloet, & Westenberg, 2008; Sheehan et al., 1998) activationsreported. One possible explanation for these discrepancies involvesimportant methodological differences between experiments. Ourresults highlight another possible relevant factor: the time of thescan. Hippocampal activity during our memory task changed overtime, in a manner that was proportional to the degree of recov-ery. Importantly, hippocampal activity at our initial time point,when most patients were commencing treatment, was stronglyrelated with later recovery, and not current symptom levels. There-

fore, it appears that memory encoding related hippocampal activitywas “turned-on” in those who were undergoing early processesof successful recovery and then decreased over time. It has beenproposed that the hippocampus’ role in recovery from PTSD mayact through the processes of contextualization and reinterpreta-tion of the traumatic experience (Yehuda & LeDoux, 2007), whichis likely to be particularly important in the early stages of remission.This is also consistent with the finding that early (pre-treatment)memory recall task performance in PTSD patients predicts recov-ery following Cognitive-Behavioural Therapy (Wild & Gur, 2008).Thus, hippocampal memory-related activity appears to be an earlymarker for recovery potential from PTSD. The precise relationbetween this activity and the recovery process, however, remainsto be determined.

Our results also add to the existing literature suggesting acritical role of the subgenual anterior cingulate (sgACC) in recoveryfrom PTSD. Increased ACC gray matter density has been asso-ciated with better treatment response to cognitive behaviourtherapy (Bryant, Felmingham, Whitford, et al., 2008). In termsof function, Bryant, Felmingham, Kemp, et al. (2008) observedthat pre-treatment sgACC responses to masked fearful faces wereincreased in those individuals who failed to respond to cognitivebehavioural therapy. This finding suggests that non-consciousperception of fearful faces, a potential probe of “bottom-up” inputfrom fear centers, was increased in those with the poorest prog-noses. Interestingly, the same group showed, in a small sampleof 8 patients, that responses to overtly presented fearful facesincreased over time in a manner that was correlated with symptom

E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778 1777

improvement (Felmingham et al., 2007). This finding was inter-preted as a resolution of overt fear perception related hypoactivityin this region, which has been hypothesized to contribute topoorer “top-down” inhibition of fear centres (Rauch et al., 2006).Our present finding extends the relationship between sgACCactivity and PTSD recovery to the neural processes associated withmemory encoding of threat-related stimuli. Specifically, activity inthis area changed over time in a manner that correlated with theamount of symptom improvement. Unlike what has been reportedwith overt fear perception (Shin et al., 2005; Williams et al., 2006),however, we did not observe a correlation between sgACC activityduring emotional memory encoding and current symptoms whenall patients were symptomatic (Time 1). Instead, early duringrecovery, we observed a greater interaction between emotion andmemory encoding in those with the poorest prognosis (see Fig. 5band c), together with an increase in memory encoding relatedactivity for fearful faces across time. Taken together, these resultsseem to suggest that, in contrast to the hippocampus, activity inthe sgACC may be an index of poor recovery potential.

In contrast to the activity pattern observed in sgACC, activityin a more anterior portion of the vmPFC, as a function of mem-ory encoding, varied in accordance with current symptom severityand, for the case of fearful faces, correlated with amygdala activ-ity. Therefore it is possible that dissociable processes contributingto PTSD pathology and recovery within the orbital-medial pre-frontal network may be localized in anterior and posterior portionsof this large cortical region. Indeed, a growing body of evidencefor functional heterogeneity within the medial prefrontal cortex isdeveloping. For instance, the infralimbic subregion in rodents isthought to be involved in extinction of conditioned fear (Quirk,Russo, Barron, & Lebron, 2000), whereas the adjacent prelimbicsubregion and anterior medial orbital frontal cortex have recentlybeen implicated in the expression of learned fear (Corcoran &Quirk, 2007) and anxiety-like behaviours (Wall, Blanchard, Yang,& Blanchard, 2004) respectively. These rodent findings are con-sistent with the recent observation that, contrary to the modelpositing that vmPFC protects from PTSD development, Vietnamveterans with lesions predominantly in the anterior vmPFC wereless likely to develop PTSD than their peers (Koenigs et al., 2008).Interestingly, the location of greatest lesion overlap in this sampleof “PTSD protected” veterans was in a similar location to the vmPFCcluster which we found to correlate with current symptom sever-ity. However, the limited spatial resolution of our fMRI study doesnot allow us to unambiguously identify the underlying anatomi-cal structures where our activation clusters were located. Furtherstudies are therefore necessary to explore this question.

Interestingly, the literature associating changes in sgACC activ-ity with a psychological recovery process extends beyond PTSDpathology. Indeed, there is considerable evidence that sgACC activ-ity predicts, and changes in accordance with, positive treatmentresponses to various interventions for major depressive disor-der (Chen et al., 2007; Davidson, Irwin, Anderle, & Kalin, 2003;Goldapple et al., 2004; Mayberg et al., 1997, 2005). This evi-dence is so compelling that a recently introduced interventionfor treatment-resistant depression involves direct stimulation ofthis area (Mayberg et al., 2005). This phenomenon is of particularinterest given the common comorbidity of PTSD with other Axis-1disorders, particularly depression, and that Axis-1 comorbidity isa known inhibitor of the recovery process (Blanchard et al., 1996;Zlotnick et al., 2004). However, it should be noted that the inclusionof changes in comorbid depressive symptoms (BDI scores) did nothave any effect on the relationship between the change in sgACCactivity and the reduction in PTSD symptoms. This suggests that thechange in sgACC activity we observed may be indicative of a neuralmechanism of recovery that is independent from that known to bein effect for depression.

4.3. Limitations

There was substantial attrition between Time 1 and Time 2,which was expected due to the relatively long time betweensessions. Nonetheless, our final sample size was similar to mostneuroimaging studies of PTSD (average sample size of 12.3across 52 published studies; unpublished observation). Moreover,because of the within-subject design of this experiment, statisticalpower was substantially larger than that of a typical between-subjects design study employing the same number of patients.Importantly, no significant differences in any of the measuredvariables (when available) existed between those who failed tocomplete both scanning sessions and those for whom full data wasavailable (see Supplemental Table S1).

As previously stated, we were not involved in these patients’treatment and did not monitor the nature of psychotherapy thatwas available. Therefore the results of the present study are notinterpretable in terms of treatment response to any particularregimen, but instead as an ecologically valid investigation of therecovery process.

4.4. Conclusions

Our findings underline the importance of longitudinal investi-gations for the understanding of neural correlates of recovery fromPTSD. We observed that amygdala and vmPFC activity was a markerof current symptom severity, whereas changes in activity in the hip-pocampus and sgACC were correlated with changes in symptomseverity. The pattern of neural activity we identified supports cur-rent theories of extinction learning in PTSD pathology and recovery(Milad, Rauch, Pitman, & Quirk, 2006).

Conflict of interest

The Authors declare no conflicts of interest.

Acknowledgments

This project was funded by grants from Fonds de la recherche ensanté du Québec and the Canadian Psychiatric Research Foundationto J.L.A. and A.B. E.W.D. was supported by a scholarship from theNatural Sciences and Engineering Research Council of Canada. Weare grateful to Dr. L. Gaston and the staff at Traumatys Clinic, aswell as the nurses, especially France Lauriault, at Charles-LeMoyneHospital for their invaluable help in recruiting subjects. We alsothank the technical staff at the MNI for their help during scanningand A.-C. David for help with clinical data entry and verification.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.neuropsychologia.2011.02.055.

References

Armony, J. L., Corbo, V., Clement, M. H., & Brunet, A. (2005). Amygdala response inpatients with acute PTSD to masked and unmasked emotional facial expressions.American Journal of Psychiatry, 162(10), 1961–1963.

Armony, J. L., & LeDoux, J. E. (1997). How the brain processes emotional information.Annals of the New York Academy of Sciences, 821, 259–270.

Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventoryfor measuring depression. Archives of General Psychiatry, 4, 561–571.

Birmes, P., Brunet, A., Benoit, M., Defer, S., Hatton, L., Sztulman, H., et al. (2005). Val-idation of the Peritraumatic Dissociative Experiences Questionnaire self-reportversion in two samples of French-speaking individuals exposed to trauma. Euro-pean Psychiatry, 20(2), 145–151.

Blake, D. D., Weathers, F. W., Nagy, L. M., Kaloupek, D. G., Gusman, F. D., Charney, D.S., et al. (1995). The development of a clinician-administered Ptsd scale. Journalof Traumatic Stress, 8(1), 75–90.

Blanchard, E. B., Hickling, E. J., Barton, K. A., Taylor, A. E., Loos, W. R., &Jones-Alexander, J. (1996). One-year prospective follow-up of motor

1778 E.W. Dickie et al. / Neuropsychologia 49 (2011) 1771–1778

vehicle accident victims. Behaviour Research and Therapy, 34(10),775–786.

Bremner, J. D., Innis, R. B., Ng, C. K., Staib, L. H., Salomon, R. M., Bronen, R. A.,et al. (1997). Positron emission tomography measurement of cerebral metaboliccorrelates of yohimbine administration in combat-related posttraumatic stressdisorder. Archives of General Psychiatry, 54(3), 246–254.

Bremner, J. D., Staib, L. H., Kaloupek, D., Southwick, S. M., Soufer, R., & Charney,D. S. (1999). Neural correlates of exposure to traumatic pictures and sound inVietnam combat veterans with and without posttraumatic stress disorder: Apositron emission tomography study. Biological Psychiatry, 45(7), 806–816.

Bremner, J. D., Vythilingam, M., Vermetten, E., Southwick, S. M., McGlashan, T.,Nazeer, A., et al. (2003). MRI and PET study of deficits in hippocampal struc-ture and function in women with childhood sexual abuse and posttraumaticstress disorder. American Journal of Psychiatry, 160(5), 924–932.

Bremner, J. D., Vythilingam, M., Vermetten, E., Southwick, S. M., McGlashan, T., Staib,L. H., et al. (2003). Neural correlates of declarative memory for emotionallyvalenced words in women with posttraumatic stress disorder related to earlychildhood sexual abuse. Biological Psychiatry, 53(10), 879–889.

Breslau, N. (2001). Outcomes of posttraumatic stress disorder. Journal of ClinicalPsychiatry, 62(Suppl 17), 55–59.

Breslau, N., Kessler, R. C., Chilcoat, H. D., Schultz, L. R., Davis, G. C., & Andreski, P.(1998). Trauma and posttraumatic stress disorder in the community – The 1996Detroit Area Survey of Trauma. Archives of General Psychiatry, 55(7), 626–632.

Brewin, C. R., Kleiner, J. S., Vasterling, J. J., & Field, A. P. (2007). Memory for emo-tionally neutral information in posttraumatic stress disorder: A meta-analyticinvestigation. Journal of Abnormal Psychology, 116(3), 448–463.

Britton, J. C., Phan, K. L., Taylor, S. F., Fig, L. M., & Liberzon, I. (2005). Corticolim-bic blood flow in posttraumatic stress disorder during script-driven imagery.Biological Psychiatry, 57(8), 832–840.

Brunet, A., Weiss, D. S., Metzler, T. J., Best, S. R., Neylan, T. C., Rogers, C., et al. (2001).The peritraumatic distress inventory: A proposed measure of PTSD criterion A2.American Journal of Psychiatry, 158(9), 1480–1485.

Bryant, R. A., Felmingham, K., Kemp, A., Das, P., Hughes, G., Peduto, A., et al. (2008).Amygdala and ventral anterior cingulate activation predicts treatment responseto cognitive behaviour therapy for post-traumatic stress disorder. PsychologicalMedicine, 38(4), 555–561.

Bryant, R. A., Felmingham, K., Whitford, T. J., Kemp, A., Hughes, G., Peduto, A.,et al. (2008). Rostral anterior cingulate volume predicts treatment responseto cognitive-behavioural therapy for posttraumatic stress disorder. Journal ofPsychiatry and Neuroscience, 33(2), 142–146.

Bryant, R. A., Kemp, A. H., Felmingham, K. L., Liddell, B., Olivieri, G., Peduto, A., et al.(2008). Enhanced amygdala and medial prefrontal activation during noncon-scious processing of fear in posttraumatic stress disorder: An fMRI study. HumanBrain Mapping, 29(5), 517–523.

Chen, C. H., Ridler, K., Suckling, J., Williams, S., Fu, C. H., Merlo-Pich, E., et al. (2007).Brain imaging correlates of depressive symptom severity and predictors ofsymptom improvement after antidepressant treatment. Biological Psychiatry,62(5), 407–414.

Corcoran, K. A., & Quirk, G. J. (2007). Activity in prelimbic cortex is necessary forthe expression of learned, but not innate, fears. Journal of Neuroscience, 27(4),840–844, 1

Davidson, R. J., Irwin, W., Anderle, M. J., & Kalin, N. H. (2003). The neural substratesof affective processing in depressed patients treated with venlafaxine. AmericanJournal of Psychiatry, 160(1), 64–75.

(2000). Diagnostic and statistical manual of mental disorders, text revision (DSM-IV-TR)(4th ed.). Washington, DC: American Psychiatric Association.

Dickie, E. W., Brunet, A., Akerib, V., & Armony, J. L. (2008). An fMRI investigation ofmemory encoding in PTSD: Influence of symptom severity. Neuropsychologia,46(5), 1522–1531.

Dolcos, F., LaBar, K. S., & Cabeza, R. (2004). Interaction between the amygdala andthe medial temporal lobe memory system predicts better memory for emotionalevents. Neuron, 42(5), 855–863.

Ehlers, A., Hackmann, A., & Michael, T. (2004). Intrusive re-experiencing in post-traumatic stress disorder: Phenomenology, theory, and therapy. Memory, 12(4),403–415.

Evans, A., Kamber, M., Collins, D., & MacDonald, D. (1994). In S. Shorvon, D. Fish, F.Andermann, & G. Bydder (Eds.), Magnetic resonance scanning and epilepsy (pp.263–274). New York: Plenum.

Felmingham, K., Kemp, A., Williams, L., Das, P., Hughes, G., Peduto, A., et al. (2007).Changes in anterior cingulate and amygdala after cognitive behavior therapy ofposttraumatic stress disorder. Psychological Science, 18(2), 127–129.

Francati, V., Vermetten, E., & Bremner, J. D. (2007). Functional neuroimaging stud-ies in posttraumatic stress disorder: Review of current methods and findings.Depression and Anxiety, 24(3), 202–218. doi:10.1002/da.20208

Genovese, C. R., Lazar, N. A., & Nichols, T. (2002). Thresholding of statistical mapsin functional neuroimaging using the false discovery rate. Neuroimage, 15(4),870–878.

Geuze, E., Vermetten, E., Ruf, M., de Kloet, C. S., & Westenberg, H. G. (2008). Neuralcorrelates of associative learning and memory in veterans with posttraumaticstress disorder. Journal of Psychiatric Research, 42(8), 659–669.

Goldapple, K., Segal, Z., Garson, C., Lau, M., Bieling, P., Kennedy, S., et al. (2004). Mod-ulation of cortical-limbic pathways in major depression: Treatment-specificeffects of cognitive behavior therapy. Archives of General Psychiatry, 61(1), 34–41.

Isaac, C. L., Cushway, D., & Jones, G. V. (2006). Is posttraumatic stress disorder associ-ated with specific deficits in episodic memory? Clinical Psychology Review, 26(8),939–955.

Josephs, O., Turner, R., & Friston, K. (1997). Event-related f MRI. Human Brain Map-ping, 5(4), 243–248.

Karl, A., Schaefer, M., Malta, L. S., Dorfel, D., Rohleder, N., & Werner, A. (2006). Ameta-analysis of structural brain abnormalities in PTSD. Neuroscience and Biobe-havioral Reviews, 30(7), 1004–1031.

Koenigs, M., Huey, E. D., Raymont, V., Cheon, B., Solomon, J., Wassermann, E. M.,et al. (2008). Focal brain damage protects against post-traumatic stress disorderin combat veterans. Nature Neuroscience, 11(2), 232–237.

Maldjian, J. A., Laurienti, P. J., Kraft, R. A., & Burdette, J. H. (2003). An automatedmethod for neuroanatomic and cytoarchitectonic atlas-based interrogation offMRI data sets. Neuroimage, 19(3), 1233–1239.

Mayberg, H. S., Brannan, S. K., Mahurin, R. K., Jerabek, P. A., Brickman, J. S., Tekell, J. L.,et al. (1997). Cingulate function in depression: A potential predictor of treatmentresponse. Neuroreport, 8(4), 1057–1061.

Mayberg, H. S., Lozano, A. M., Voon, V., McNeely, H. E., Seminowicz, D., Hamani, C.,et al. (2005). Deep brain stimulation for treatment-resistant depression. Neuron,45(5), 651–660.

Milad, M. R., Rauch, S. L., Pitman, R. K., & Quirk, G. J. (2006). Fear extinction in rats:Implications for human brain imaging and anxiety disorders. Biological Psychol-ogy, 73(1), 61–71.

North, C. S., Smith, E. M., & Spitznagel, E. L. (1997). One-year follow-up of survivorsof a mass shooting. American Journal of Psychiatry, 154(12), 1696–1702.

Osuch, E. A., Benson, B., Geraci, M., Podell, D., Herscovitch, P., McCann, U. D.,et al. (2001). Regional cerebral blood flow correlated with flashback inten-sity in patients with posttraumatic stress disorder. Biological Psychiatry, 50(4),246–253.

Piefke, M., Pestinger, M., Arin, T., Kohl, B., Kastrau, F., Schnitker, R., et al. (2007).The neurofunctional mechanisms of traumatic and non-traumatic memoryin patients with acute PTSD following accident trauma. Neurocase, 13(5–6),342–357.

Quirk, G. J., Russo, G. K., Barron, J. L., & Lebron, K. (2000). The role of ventromedialprefrontal cortex in the recovery of extinguished fear. Journal of Neuroscience,20(16), 6225–6231.

Rauch, S. L., Shin, L. M., & Phelps, E. A. (2006). Neurocircuitry models of posttraumaticstress disorder and extinction: Human neuroimaging research – past, present,and future. Biological Psychiatry, 60(4), 376–382.

Rauch, S. L., Whalen, P. J., Shin, L. M., McInerney, S. C., Macklin, M. L., Lasko, N. B.,et al. (2000). Exaggerated amygdala response to masked facial stimuli in post-traumatic stress disorder: A functional MRI study. Biological Psychiatry, 47(9),769–776.

Sergerie, K., Lepage, M., & Armony, J. (2005). The amygdala and emotional memory:An FMRI investigation of its role in successful encoding and retrieval. Journal ofCognitive Neuroscience, 13–113.

Sheehan, D. V., Lecrubier, Y., Sheehan, K. H., Amorim, P., Janavs, J., Weiller, E.,et al. (1998). The Mini-International Neuropsychiatric Interview (M. I. N. I.):The development and validation of a structured diagnostic psychiatric inter-view for DSM-IV and ICD-10. Journal of Clinical Psychiatry, 59(Suppl. 20), 22–33,quiz 34–57.

Shin, L. M., Orr, S. P., Carson, M. A., Rauch, S. L., Macklin, M. L., Lasko, N. B., et al.(2004). Regional cerebral blood flow in the amygdala and medial prefrontal cor-tex during traumatic imagery in male and female Vietnam veterans with PTSD.Archives of General Psychiatry, 61(2), 168–176.

Shin, L. M., Shin, P. S., Heckers, S., Krangel, T. S., Macklin, M. L., Orr, S. P., et al.(2004). Hippocampal function in posttraumatic stress disorder. Hippocampus,14(3), 292–300.

Shin, L. M., Wright, C. I., Cannistraro, P. A., Wedig, M. M., McMullin, K., Martis, B., et al.(2005). A functional magnetic resonance imaging study of amygdala and medialprefrontal cortex responses to overtly presented fearful faces in posttraumaticstress disorder. Archives of General Psychiatry, 62(3), 273–281.

Thomaes, K., Dorrepaal, E., Draijer, N. P., de Ruiter, M. B., Elzinga, B. M., van Balkom,A. J., et al. (2009). Increased activation of the left hippocampus region in com-plex PTSD during encoding and recognition of emotional words: A pilot study.Psychiatry Research, 171(1), 44–53.

Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix,N., et al. (2002). Automated anatomical labeling of activations in SPM using amacroscopic anatomical parcellation of the MNI MRI single-subject brain. Neu-roimage, 15(1), 273–289.

Wall, P. M., Blanchard, R. J., Yang, M., & Blanchard, D. C. (2004). Differential effectsof infralimbic vs. ventromedial orbital PFC lidocaine infusions in CD-1 mice ondefensive responding in the mouse defense test battery and rat exposure test.Brain Research, 1020(1–2), 73–85.

Werner, N. S., Meindl, T., Engel, R. R., Rosner, R., Riedel, M., Reiser, M.,et al. (2009). Hippocampal function during associative learning in patientswith posttraumatic stress disorder. Journal of Psychiatric Research, 43(3),309–318.

Wild, J., & Gur, R. C. (2008). Verbal memory and treatment response in post-traumaticstress disorder. British Journal of Psychiatry, 193(3), 254–255.

Williams, L. M., Kemp, A. H., Felmingham, K., Barton, M., Olivieri, G., Peduto, A.,et al. (2006). Trauma modulates amygdala and medial prefrontal responses toconsciously attended fear. Neuroimage, 29(2), 347–357.

Yehuda, R., & LeDoux, J. (2007). Response variation following trauma: A translationalneuroscience approach to understanding PTSD. Neuron, 56(1), 19–32.

Zlotnick, C., Rodriguez, B. F., Weisberg, R. B., Bruce, S. E., Spencer, M. A., Culpepper,L., et al. (2004). Chronicity in posttraumatic stress disorder and predictors of thecourse of posttraumatic stress disorder among primary care patients. Journal ofNervous and Mental Disease, 192(2), 153–159.