Psychiatry Research: Neuroimaging

Transcription

1 Psychiatry Research: Neuroimaging 172 (2009) Contents lists available at ScienceDirect Psychiatry Research: Neuroimaging journal homepage: Cortico-limbic response to personally challenging emotional stimuli after complete recovery from depression Jill M. Hooley a,, Staci A. Gruber b, Holly A. Parker a, Julien Guillaumot c, Jadwiga Rogowska b, Deborah A. Yurgelun-Todd b a Department of Psychology, Harvard University, Cambridge, MA, USA b Cognitive Neuroimaging and Neuropsychology Laboratory, McLean Hospital, Belmont, MA, USA c McLean Hospital, Belmont MA, USA article info abstract Article history: Received 24 June 2007 Received in revised form 5 April 2008 Accepted 6 April 2008 Keywords: Expressed emotion fmri Vulnerability markers Amygdala Anterior cingulate cortex Prefrontal cortex People vulnerable to depression are at increased risk of relapse if they live in highly critical family environments. To explore this link, we used neuroimaging methods to examine cortico-limbic responding to personal criticisms in healthy participants and participants with known vulnerability to major depression. Healthy controls and fully recovered participants with a past history of major depression were scanned while they heard praising, critical, and neutral comments from their own mothers. Prior to scanning, the formerly depressed and the control participants were indistinguishable with respect to self-reported positive, negative, or anxious mood. They also reported similar mood changes after being praised or criticized. However, formerly depressed participants responded to criticism with greater activation in the amygdala and less activation in the dorsolateral prefrontal cortex (DLPFC) and anterior cingulate cortex (ACC) than did controls. During praise and neutral commentary, amygdala activation was comparable in both groups, although lower levels of activation in the DLPFC and ACC still characterized formerly depressed participants. Vulnerability to depression may be associated with abnormalities in cortico-limbic activation that are independent of mood state and that remain even after full recovery. Criticism may be a risk factor for relapse because it activates the amygdala and perturbs the affective circuitry that underlies depression Elsevier Ireland Ltd. All rights reserved. 1. Introduction Recovery from an episode of major depression is often slow. Nonetheless, most patients eventually experience remission. In some cases, mild, sub-clinical symptoms may remain; in other cases, the recovery will be complete with no residual symptoms and a full return to the former level of functioning (Boland and Keller, 2002). However, even after clinical improvement, people who have experienced depression are at increased risk of relapse or recurrence (Burcusa and Iacono, 2007) especially if they live in highly critical family environments (Butzlaff and Hooley,1998). This suggests that trait vulnerability markers of depression may exist even after full clinical recovery. Here, we investigate cortico-limbic responses to critical commentary to determine whether altered neurobiological responding to criticism is associated with a past history of depression. Despite the high prevalence of mood disorders, their underlying neuropathology is still poorly understood. Current models of depression implicate widespread functional interactions between neocortical, limbic, and subcortical brain regions in the pathogenesis of affective illness DOIs of original article: /j.pscychresns , /j.pscychresns Corresponding author. Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, MA 02138, USA. Tel.: (Mayberg et al., 1997; Drevets, 2000; Davidson et al., 2002). Altered functioning in these systems is supported by imaging data that have highlighted neural networks underlying emotional processing deficits related to mood disorders. For example, Mayberg et al. (1999) have suggested a cortico-limbic model that includes ventral, dorsal and rostral compartments(mayberg et al., 1999). A number of frontal brain regions, including the dorsolateral and dorsomedial prefrontal cortex, the dorsal anterior cingulate (dacc), posterior cingulate and the ventral prefrontal cortex, insula, and subgenual anterior cingulate have been shown to be associated with the cognitive, somatic and affective disturbances associated with depression. Compared with healthy controls, currently depressed individuals show decreased activation in the dorsolateral prefrontal cortex (DLPFC; Buchsbaum et al.,1997; Drevets et al.,1997; Mayberg et al.,1997; Keedwell et al., 2005a; Siegle et al., 2007). Activation in the DLPFC has also been shown to be decreased when non-depressed individuals are exposed to a sad mood induction (Gemar et al., 1996; Baker et al., 1997; Liotti et al., 2000). Anterior cingulate activity is also dysfunctional in depression (Mayberg et al., 1999; Seminowicz et al., 2004; Pizzagalli et al., 2006). This region, which is functionally quite heterogenous, has been parcellated into a cognitive and an affective subdivision based on distinct afferent and efferent projection systems. The affective subdivision encompasses ventral (rostral, subcallosal, and subgenual) areas of the ACC and has /$ see front matter 2008 Elsevier Ireland Ltd. All rights reserved. doi: /j.pscychresns

2 84 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) extensive connections with limbic and paralimbic regions including the amygdala (Devinsky et al., 1995; Drevets et al., 1997). Emotional tasks are thought to activate the affective division (Whalen et al., 1998; Mayberg et al., 1999; Davidson et al., 2002). In contrast, cognitive tasks such as the classic Stroop test activate the cognitive subdivision of the ACC (Bush et al., 1998; Mayberg et al., 1999). The cognitive subdivision of the ACC involves more dorsal regions and has connections with brain regions such as the DLPFC, posterior cingulate, and parietal cortex (Devinsky et al., 1995; Mayberg et al., 1999; Davidson et al., 2002). Dorsal regions of the ACC have been shown to be hypoactive in depression (Davidson et al., 2002). In contrast, greater rostral cingulate activity has been associated with better response to pharmacological treatments and ECT (Mayberg et al., 1997) as well as to the antidepressant effects of sleep deprivation (Wu et al., 1999). Deep brain stimulation of the subgenual ACC has also been shown to improve clinical symptoms in treatment refractory-patients (Mayberg et al., 2005). Other key findings concern the amygdala. Drevets (1999) has suggested that this limbic region is hyperactive in people who are depressed. Although group differences between controls and depressed patients were not reported by Abercrombie and colleagues (Abercrombie et al., 1998), these investigators did find that right amygdalar glucose metabolism was positively correlated with the level of dispositional negative affect reported by the depressed participants. Siegel and colleagues (2002, 2007) have reported increased and sustained amygdalar responses to emotional stimuli (words) in medicated and unmedicated depressed patients compared with never depressed controls. Moreover, bilateral amygdala response to valenced words has been linked to increased recall of negative selfreferent information in remitted patients exposed to a sad mood challenge (Ramel et al., 2007). This suggests the amygdala may play a role in mood-congruent memory in depression. The disinhibition of the amygdala system may be linked to a failure of inhibition from integrative cortical structures such as the DLPFC (Davidson, 2000; Drevets, 2003). Increased limbic activity in response to emotional stimuli has been linked to decreased DLPFC activity during cognitive tasks; there also appears to be a decreased functional relationship between these brain structures in depressed patients (Siegle et al., 2007). For some patients, increased tonic amygdala activity could lead to decreased DLPFC function (Siegle et al., 2002). In other cases, there may be a failure to effectively recruit the DLPFC, leading to a failure to inhibit limbic activity. Although neuroimaging studies suggest that depression involves disruptions in limbic-cortical pathways, (Drevets, 1999; Mayberg et al., 1999; Siegle et al., 2006), the extent to which activity in these pathways normalizes after full recovery from depression is unexplored. Certainly, there is reason to believe that vulnerability factors may be evident in clinically remitted patients. Liotti, Mayberg, and colleagues (Liotti et al., 2002) exposed currently depressed, remitted and never depressed participants to a sad mood provocation. They then used PET to examine cerebral blood flow abnormalities in the three groups. When in a sad mood, the remitted depressed patients showed patterns of brain activity that were more like those found in currently depressed patients as opposed to never depressed healthy participants. Clinical remission may still involve the presence of sub-clinical symptoms, however. This raises the question of the extent to which vulnerability markers for depression may be found in people who have a past history of the disorder but who have no residual symptoms even minor ones. The current study used functional magnetic resonance imaging (fmri) techniques to compare neural activity in healthy controls and fully recovered, symptom-free participants who had experienced one or more past episodes of clinical depression. We sought to challenge the neural circuitry implicated in depression, not by inducing a sad mood, but by using a psychosocial stressor that has been empirically linked to the relapse process. Criticism from a close family member is a reliable predictor of relapse in depression (Butzlaff and Hooley, 1998; Hooley and Gotlib, 2000). Accordingly, we exposed participants to critical, praising, and neutral comments from their own mothers and used fmri to examine the relationship between exposure to personally relevant emotional stimuli and activity in the DLPFC, ACC, and amygdala. Using a larger and different sample, we sought to replicate and extend our preliminary finding of reduced activation in the DLPFC in people with a vulnerability to depression during exposure to criticism (Hooley et al., 2005). In addition we hypothesized that compared with healthy controls, fully well participants with a past history of depression would show increased activation in the amygdala when they were being criticized. Differences in both sub-regions of ACC activity during criticism were also anticipated because of the connections of these areas to the DLPFC and amygdala. A number of studies have reported decreased blood flow within regions of the anterior cingulate cortex in patients with depression (Bench et al., 1992; Drevets et al., 1997; George et al., 1997), although sub-regions of this area demonstrate different patterns of activation. Whereas the dorsal region of the ACC has been shown to be hypoactive in patients with depression, the rostral ACC appears to be overactive in patients who are eventual treatment responders (Mayberg et al., 1997; Pizzagalli et al., 2001). Moreover, remission of clinical symptoms has been shown to be related to greater activation and increased gray matter volume in the ACC (Chen et al., 2007), and also to increased dorsal ACC activity (Buchsbaum et al., 1997; Mayberg et al., 1999). Given that our participants included people who had recovered from major depressive disorder, we sought to explore changes in both the ventral and dorsal ACC in response to the affectively challenging stimuli. 2. Methods 2.1. Participants Twenty-three right-handed female participants aged between 20 and 30 years were recruited by advertisements in local media. We focused on women because rates of depression are higher in women than they are in men (Nolen-Hoeksema, 2002) and because we sought to minimize heterogeneity in our data due to gender effects. Potential participants who appeared likely to meet study entry criteria based on a brief telephone screening interview were invited for a further diagnostic assessment by a trained interviewer using the patient version of the Structured Clinical Interview for DSM-IV (SCID) (First et al., 1997). To be considered a healthy control (n=12), participants had to have no evidence of current or past psychopathology based on the SCID. Participants in the formerly depressed (FD) study group (n=11) were required to have a past history of one or more episodes of unipolar depression (mean=1.9; range 1 5) that met DSM-IV criteria (American Psychiatric Association, 2000) but to be free from other current or past Axis I disorders, including all anxiety disorders. They were also required to be fully recovered, report no symptoms of depression (even mild ones) and to have been completely well for a minimumof6 months (mean=20.45 months; range 7 62months). Inaddition, they hadto firmly believe that they were now back to their old self, or better. Participants with a history of head injury or neurological problems were excluded from participation. However, antidepressant medicationwas not an exclusion criterion. Five of the 11 FD participants were currently taking antidepressant medications. Three of the six FD participants who were not taking antidepressant medications had done so in the past. The remaining three participants had no history of pharmacological treatment. All participants werewell educated. Mostwerecollege graduatesandallhad at least somecollegeexperience (some were still students). Using a 1 4 rating scale for education (1 = completed high school, 4 = completed a Master's degree or a Ph.D.), there was no difference (P=1.0) in the level of education between the healthy controls (mean=3.0, S.D.=0.47) and the formerly depressed participants (mean=3.0, S.D.=0.77). Participants in the two groups also did not differ in mean age (24.33 years (S.D.=2.4) forcontrols and years (S.D.=3.0) for the formerly depressed, P=0.92) Procedure Subjects participated in two separate research sessions conducted approximately 1 month apart. During the first visit they were interviewed with the SCID and completed several self-report questionnaires. These included the Beck Depression Inventory (BDI) (Beck et al., 1961), and the trait form of the Positive and Negative Affect Scales (PANAS) (Watson et al., 1988). The PANAS contains 10 positive (e.g., interested, proud) and 10 negative (e.g., ashamed, irritable) mood descriptors rated on a 1- to 5-point scale (1=very slightly or notatall;5=extremely). The fmri scans were conducted at the second visit. Just before the scanning session, participants completed the BDI to ensure that they scored in the non-depressed range (b10;

3 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) see Beck et al., 1988) and reported no depressed mood on item 1 of the BDI. In addition, the anxious arousal and the anhedonia subscales of the Mood and Anxiety Symptom Questionnaire (MASQ) (Watson et al., 1995) were administered at this time. Participants were also questioned about their positive and negative mood states on five separate occasions during the affective challenge protocol using the PANAS. PANAS ratings were made after participants entered the scanner, after localizing scans had been completed, and also after hearing criticism, praise, and neutral comments from their mothers Affective challenge stimuli The affective stimuli used in this study were provided by participants' own mothers. First, we asked all of our eligible participants to confirm that their mothers were willing to becontactedby theresearchteam. Afterwehad receivedthis assurance, andagainwith the consent of all our participants, mothers ofeligible subjects were sent a letterdescribing the nature and purpose of the study. The letter invited mothers to assist in the research by providing the auditory stimuli that their daughters would hear when they were in the scanner. After they had received the letter, mothers were contacted by the senior author and invitedto participate. If they provided informedconsent, motherswerescheduled fora telephone appointment during which the auditory stimuli were recorded. These procedures led us to exclude two potential participants from our study. One mother told her daughter than she did not wish to be approached about participation. In a second case the mother gave permission to be contacted about the study but then said she was too busy to participate when she was contacted directly. The mothers who did participate expressed great interest in the research and were enthusiastic about their role in the study. Mothers were asked to make specific comments about their daughters that were critical or praising in their affective valence. Mothers were free to choose the content of each criticizing or praising statement. Each comment lasted 30 s and began with the mother addressing her daughter by name. Mothers were asked to begin all critical and praising remarks in a standard way (e.g., Stephanie, one thing that really bothers me about you is or Stephanie, one thing I really like about you is ). All comments were recorded over the telephone. If mothers made a mistake when they were making a specific comment or if the length of the comment was too long or too short, we allowed them to make as many attempts as they wanted to get the comment to sound natural and be of the right length. In addition to providing critical and praising comments, mothers were also asked to provide neutral comments that were not personal to their daughters. Rather, these comments described situations or events that mothers believed would not be of direct interest to their daughters such as the current weather, or a description of construction at the local high school. Table 1 provides examples of the kinds of praising, critical, and neutral comments that were made by mothers. We asked mothers to refrain from talking to their daughters about the content of their telephone conversations with the research team until after the scanning had been completed. All recorded telephone conversations with mothers were later edited to extract the stimulus comments that were to be used in the study from the general conversational content. The stimulus comments were then transferred to CD. During fmri, participants heard these comments over non-ferrous, gradient damping headphones, in the context of a blocked design. Half of the subjects heard the critical comments first, followed by the praise. The other participants heard praise first, followed by criticism. Neutral comments were always heard last. There were no significant effects (PN0.10) associated with order of presentation for any of the analyses to be reported Presentation of stimuli There were a total of two comments of each type (e.g., criticism, praise, neutral) within a given scanning epoch. Each epoch, which lasted for 2:31, began with a 30-s rest period, Table 1 Examples of praising, critical and neutral comments Praise: Joanna, one of the things I really like about you, is your kindness. You are one of the kindest people I know. You are always so accepting and accommodating, and willing to help whenever possible. You make my life so much easier. You always go the extra mile for me, and for daddy, for granddad, for Aunty Gail, whoever needs you. You are always there, and anytime we need you we can always call you. I want you to know how much I appreciate your kindness. Criticism: One thing that bothers me about you, Joanna, is your need to pierce and tattoo. Please, stop piercing and tattooing. I know it is something you like but it is something I think has no beneficial value to you as a person, and it takes away from your ability to move into other people's lives without unnecessary judgment. Grandma and I agree that it also takes away from your beautiful face, and the stretched ears are not that easy to fix, if you ever come to your senses. Neutral: Joanna, let me tell you a little bit about the weather today. It finally got nice today and the sun was shining and it was relatively warm. And, although last night it was very very cold, it got down into the low 40s, it warmed up this morning and has been really nice all day. And it is still warm enough to go outside with a jacket on tonight. followed by 30 s of commentary, another rest period, another 30 s of the same type of commentary, and then another rest period. Thus, in each epoch, participants heard two 30-s segments of each type of commentary. There was no commingling of comment type in the same scan epoch; subjects heard either two praising, two critical, or two neutral comments from their mothers. Each individual comment was heard only once and subjects did not hear any of the recorded maternal comments prior to the scanning Protection of participants Because this study involved exposing vulnerable participants to potentially distressing stimuli, we made everyeffort to protect the emotional well-beingof our subjects. We also did everything we could to ensure that their relationships with their mothers were not damaged by our procedures. The senior author, an experienced clinician, made the initial contact phone call to all of the mothers. She also recruited all of the recovered depressed participants, was present during their scanning sessions, and debriefed them after the scanning had been completed. Great care was taken to make sure that participants knew that their mothers had only made the critical remarks because this was required by the experimental protocol. We also asked mothers only to criticize behaviors that they had criticized before and that their daughters would already know they disliked or disapproved of. These protections worked well. No participant reported being emotionally distressed at the end of the study protocol and it was quite common for participants to tell us how interesting they thought the study was. Some said they had been expecting something much worse than the criticism they had heard; others reported feeling pleased about the praise that they received Imaging methods Functional images were acquired on a 1.5 Tesla GE LX MRI scanner equipped with a birdcage quadrature RF head coil (TR=3 s, TE=40 ms, flip angle=90). Images were collected over 20 coronal slices with a 20-cm field of view and a acquisition matrix (in-plane resolution= mm). For each 151 s scan, 50 echoplanar images were collected each consisting of five alternating 30-s control/task periods. Three dummy images were taken at the outset of each functional scan and discarded from analysis. Matched T1-weighted high-resolution images were also collected for every subject. Head motion was minimized with foam padding and a stabilization strap across the forehead. Preprocessing and statistical analysis were conducted in SPM99 (Friston et al., 1995a,b). Because of concerns about signal dropout near the air-filled sinus cavities, we manually examined each of the data sets. None of the scans that were used in analyses had evidence of significant signal loss in any of the regions of interest. The functional data sets were motion-corrected (intra-run alignment) within SPM99 using the first image as a reference. Any subject's data with motion exceeding 2 mm in any direction were eliminated. After motion correction, the image data were normalized to a standard template from the Montreal Neurological Institute (MNI) with an isotropic mm voxel size and smoothed using an isotropic Gaussian kernel (full width half maximum [FWHM]=10 mm). Within SPM99, the general linear model was used to generate a statistical parametric map for each subject (Friston et al., 1995a,b). We convolved the stimulus paradigm to the standard hemodynamic response function specified in SPM99. Individual contrast images were entered into a fixed effects multiple regression model. The criticism, praise, and neutral conditions were analyzed separately. Activation during the block of each commentary type was contrasted with the activation during the rest condition. This procedure yielded statistical parametric maps that isolated the activity unique to the commentary condition. Finally, using a random effects t-test analysis, we made a direct comparison between the data from the recovered depressed sample and the control sample. For the present study, our hypotheses were constrained to a few specific cortical and limbic regions believed to be sensitive to affective processing in depression (ACC, DLPFC, and amygdala). The region of interest (ROI) masks were created using the Wake Forest University Pickatlas utility (Tzourio-Mazoyer et al., 2002; Maldjian et al., 2003) and excluded activity in nonhypothesized regions, permitting adjustment of statistical thresholds using the small volume correction implemented in SPM99. We set the statistical height threshold at Pb0.05, corrected for multiple comparisons. Only active clusters with a minimum extent (k) threshold of 20 contiguous voxels were considered as significant. 3. Results 3.1. Clinical assessments There were no significant differences between the controls and the formerly depressed participants on the BDI at either of the two assessments. Upon entry into the study, the mean total BDI score was 0.58 (S.D.=1.4) for the controls and 1.81 (S.D.=2.1) for the FD participants. At the second assessment, the means were 0.58 (S.D. =1.1) and 2.27 (S.D.=2.8), respectively. The two groups also did not differ on the Anhedonic Depression and Anxious Arousal subscales of the MASQ. For the controls, the

4 86 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) Fig. 1. SPM maps displaying significantly greater activation in healthy comparison participants relative to FD participants during presentation of auditory affective stimuli. These results are based on whole brain analyses. mean Anhedonic Depression score was (S.D. =2.9); for the FD participants, the mean was (S.D. =3.0). On the Anxious Arousal subscale, the mean for the controls was (S.D. =1.7) versus (S.D. =1.8) for the FD participants. Controls and FD participants also reported similar levels of trait-like positive and negative mood on the PANAS. Controls had a mean of (S.D. =4.1) for trait positive mood; the corresponding mean for the FD participants was (S.D. =3.6). For trait negative mood, the means were (S.D.=5.1) for controls versus (S.D.=2.8) for participants with a past history of depression. Taken together, these findings support the full recovery status of our FD participants. With respect to their self-reported mood and anxiety ratings, the participants in the FD study group were indistinguishable from people who had always been emotionally healthy. Table 2 Brain regions displaying significantly greater activation in healthy comparison participants relative to formerly depressed participants during presentation of auditory affective stimuli Comment Type Voxels Coordinates Region of activation T-score Z-score Negative Left Sub-lobar, Lentiform Nucleus, Lateral Globus Pallidus, Putamen Positive Left Occipital Lobe, Middle Occipital Gyrus, Sub-Gyral, Inferior Occipital Gyrus, Brodmann area Right Frontal Lobe, Middle Frontal Gyrus, Brodmann area Left Occipital Lobe, Cuneus, Middle Occipital Gyrus, Brodmann area Left Cerebellum, Anterior Lobe, Culmen Left Cerebellum, Anterior Lobe, Culmen Right Occipital Lobe, Cuneus, Brodmann area Right Limbic Lobe, Parahippocampal Gyrus, Brodmann area Right Parietal Lobe, Supramarginal Gyrus, Brodmann area Left Sub-lobar, Insula, Brodmann area Right Parietal Lobe, Inferior Parietal Lobule, Brodmann area Right Occipital Lobe, Lingual Gyrus, Brodmann area Neutral Inter-Hemispheric, Left Frontal Lobe, Medial Frontal Gyrus, Right Frontal Lobe Inter-Hemispheric, Left Occipital Lobe

5 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) Fig. 2. SPM maps displaying significantly greater activation in FD participants relative to healthy comparison participants during presentation of neutral auditory stimuli. These results are based on whole brain analyses Mood changes in response to emotional stimuli Participants reported on their positive and negative mood using the PANAS. Planned contrasts showed that, across all participants, there was a significant increase in subjects' positive mood compared with baseline (mood after localizing scans) after they heard praise (means=33.52 v 36.00, t(22)=2.83, P=0.005, one tailed, r=0.59). In addition, after hearing criticism, there was a significant increase in subjects' negative mood from baseline (means=12.13 vs , t(22)=2.10, P=0.026, one-tailed, r=0.41). Exposure to the neutral comments did not result in a significant positive or negative mood change. Taken together, these findings confirm the validity of our experimental paradigm. They show that participants reported the predicted mood changes after exposure to the critical, praising, or neutral comments from their mothers. A group (controls vs. FD) mood (positive or negative) time (before or after criticism or praise) repeated measures analysis of variance (ANOVA) revealed a significant main effect for time, F(1,21)=7.75, P=0.011, partial η 2 =0.27, due to increased positive or negative mood after praise and criticism, respectively (see above). However, there was no significant group by time interaction, F(1,21)=0.18, ns, and no significant group mood time interaction, F(1,21)=0.38, ns. In other words, the controls and the FD participants responded to the affective challenges in similar ways. Follow-up analyses indicated that the magnitude of change in negative mood that participants reported after hearing critical remarks from their mothers was quite similar in both groups (means=2.0; S.D.=4.1 for FD participants and 0.9, S.D.=2.5 for controls). There was also no difference between the two groups with respect to how much change in their positive moods they experienced after hearing praise. Healthy control participants showed a 2.4 point increase (S.D.=4.6) in positive mood from baseline and FD participants showed a 2.5 (S.D.=3.9) point increase. Again, these data support the full recovery status of our FD participants and suggest that they responded to the critical and praising stimuli in a manner that was similar to the controls fmri data analysis Whole brain analyses The aim of this study was to examine group differences in the three regions of interest described earlier. However, for clarity, and to facilitate comparison of our data with those from other published studies, we also present significant results (Pb0.001) from a series of whole brain analyses. The scale is based on T values. Fig. 1 and Table 2 illustrate the brain areas that were more active in the healthy controls versus the FD participants during the critical, praising and neutral comments. There is a general pattern of increased activation in the controls relative to the FD participants; this is especially evident during praise. Fig. 2 and Table 3 show results of the comparison between FD participants and controls. No brain regions were significantly more activated in the FD participants compared with the controls when hearing criticism or praise. However, in the neutral condition, several brain areas showed significantly greater activation in the FD participants compared with the controls. Figs. A1 A3 and Tables A1 A3 (see supplementary materials available online) show the results of analyses comparing brain activation in control participants during criticism (versus resting baseline), praise (versus resting baseline) and neutral comments (versus resting baseline). As would be expected given that our task involved an auditory stimulus, the greatest activation in all conditions is seen in temporal areas. This is also true for the FD participants. Activation data (compared with resting baseline) for the FD participants during criticism, praise and neutral comments are presented in Figs. A4 A6 and in Tables A4 A6 (see supplementary online materials) ROI analyses A group (controls vs. FD) condition (positive, negative, or neutral comments) ROI repeated measures ANOVA was conducted to evaluate whether the number of voxels activated in DLPFC, ACC and amygdala differed in the healthy controls and FD groups during exposure to different types of comments from their mothers. This analysis revealed a significant main effect for condition, F(2,20)=4.78, P=.020, partial η 2 =.32, a significant main effect for brain region, F(2,20),=61.68, Pb.001, partial η 2 =.86,andasignificant condition by region interaction F(4,18)=4.47, P=.011, partial η 2 =.50. There was no significant group condition interaction, F(2,20) =1.77, and no significant group by region interaction, F(2,20)=1.24. The findings show that the number of voxels that were activated in the three regions of interest changed according to the experimental condition that participants experienced and that this was not associated with group. The results further indicate that the three brain regions we studied showed different overall patterns of activation and Table 3 Brain regions displaying significantly greater activation in formerly depressed participants relative to healthy comparison participants during presentation of neutral auditory stimuli Comment Type Voxels Coordinates Region of activation T-score Z-score Neutral Right Frontal Lobe, Inferior Frontal Gyrus Left Temporal Lobe, Superior Temporal Gyrus Left Sub-lobar, Insula, Brodmann area Right Frontal Lobe,Inferior Frontal Gyrus Right Temporal Lobe, Sub-Gyral Right Frontal Lobe, Inferior Frontal Gyrus No significant increase was evident for recovered depressed patients relative to healthy comparison subjects during presentation of negative and positive stimuli.

6 88 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) Fig. 3. Criticism. Illustrations of group subtraction BOLD activation results for (A) dorsolateral prefrontal cortex (BA 9 and 46), (B) anterior cingulate, and (C) amygdala. During criticism there is significantly greater activation in DLPFC and ACC in controls than in formerly depressed participants. Amgygala activation during criticism is significantly greater in formerly depressed participants than it is in controls. that this was also not associated with group. The significant condition by region interaction indicates that the number of voxels that were activated in the three different ROIs was influenced by the type of verbal material participants were hearing. Finally, there was a trend toward a significant ROI condition group interaction, F(4,18)=2.47, P=0.082, partial η 2 =0.35. This three-way interaction is important because it suggests that the patterns of brain activation during the experimental protocol are different in the healthy controls than in the FD subjects. To further explore these differences, the neuroimaging data from the FD sample were compared directly with the data from the healthy controls using random effects t-test analyses. Fig. 3 shows that when they heard criticism (compared with resting baseline), the FD sample showed significantly greater BOLD activation than the controls within the amygdala (k=39, max T=2.86; see Table 4). This was accompanied by significantly less BOLD activation than in the controls in both the DLPFC (k=106, max T=2.70) and anterior cingulate gyrus (k=339, max T=3.60). Levels of activation in the amygdala were similar in both groups when they heard praise. Again, however, during the praise condition (Fig. 4), the FD group showed significantly less activation in the DLPFC (k=450, max T=5.90; see Table 5)and ACC(k=452, max T=4.20) than did the controls. These differences were also paralleled in the neutral condition. Simply hearing their mothers talk, even when the topic was the weather, led to greateractivationinthedlpfc(k=147,maxt=4.35)andacc(k=416,max T=4.51) in the controls relative to the FD participants (see Fig. 5 and Table 6). Levels of activation in the amygdala were similar across both groups, however. The ACC region measured includes both cognitive and affective subdivisions of the ACC. The boundaries of this region are delineated by the mask created by the Pickatlas utility Activation in the DLPFC and amygdala during criticism A central finding of this study is that, when exposed to criticism, people who are completely symptom free but who have a past history of depression show less activation in the DLPFC and more activation in the amygdala than do healthy controls. As noted earlier, there is theoretical speculation about the role that cortical structures such as the DLPFC might play in the inhibition of the amygdala. Accordingly, we conducted two additional analyses correlating activation in the DLPFC with activation in the amygdala when our participants were hearing critical remarks from their mothers. For the controls, the number of voxels activated in the DLPFC was negatively correlated (r= 0.42) with the number of voxels activated in the amygdala. For the FD participants, however, there was no such association, r= Because of the small sample size, the difference between these two correlations was not significant, z= 0.838, P=0.20. Nonetheless, the overall pattern of the correlations is consistent with the idea that vulnerability to depression may be characterized by a decoupling of activity in the DLPFC and the amygdala and that this may not normalize even with full recovery Ratings of mothers' comments At the end of the scanning session, participants rated the comments that they had heard from their mothers using a 1 10 (not at all extremely) intensity scale. There were no differences between the control and and the FD participants with regard to how strongly positive they rated their mothers' praising comments as being (means=9.4 (0.67) vs. 8.1 (S.D.=2.90). The same was true for criticism, with control and FD participants rating the severity of their mothers' criticisms quite comparably (means=6.5 (S.D.=1.74) vs. 5.7 (S.D.=2.25). The stimulus comments were also rated by a panel of 10 independent judges (college students aged years). Again, no significant differences between the mothers' comments were found across the two study groups, either for criticism (mean=4.6 (S.D.=0.99) for mothers of controls vs. 4.8 (S.D.=0.69) for mothers of FD participants, or for praise (means=7.48 (S.D.=1.65) and 7.23 (S.D.=1.55), respectively. These results suggest that differences in the stimulus comments themselves do not explain the differences in the patterns of neural responding that we found in the FD participants relative to the controls. Table 4 Comparison of FD and control participants during critical auditory stimuli Region of Interest Voxels Coordinates Region of activation T-score Z-score BA 9 and Right Frontal Lobe, Medial Frontal Gyrus, Brodmann areas 8 and Right Frontal Lobe, Inferior Frontal Gyrus, Brodmann area Right Frontal Lobe, Middle Frontal Gyrus Right Frontal Lobe, Medial Frontal Gyrus, Brodmann areas 9 and ACC Left Limbic Lobe, Anterior Cingulate, Brodmann area Left Limbic Lobe, Anterior Cingulate, Brodmann areas 24 and Amygdala Right Limbic Lobe, Uncus, Amygdala, Parahippocampal Gyrus, Brodmann area Control participants show significantly greater activation in dorsolateral prefrontal cortex (BA 9 and 46) and ACC relative to FD participants. FD participants show significant greater activation in amygdala relative to controls.

7 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) Fig. 4. Praise. Illustrations of group subtraction BOLD activation results for (A) dorsolateral prefrontal cortex (BA 9 and 46), and (B) anterior cingulate during praise. Significantly greater activation is found in controls compared with formerly depressed participants. No group differences were found for the amygdala. It is important to note, however, that the praising comments and the critical comments were not matched in intensity. Paired t-tests revealed that the praising comments were rated (on the 1 10 intensity scale) as more strongly positive than the negative comments were rated as being strongly negative. This was true for the ratings of the participants themselves (praise=8.78 vs for criticism, t(23)=6.36 Pb0.001) and also for the ratings of the independent judges (praise=7.72 vs for criticism, t(18)=12.42, Pb0.001). This is not especially surprising, because, to protect our subjects, we asked mothers not to make remarks that were extremely negative or hostile. In contrast, no such limits were placed on the intensity of mothers' praising remarks. The difference in the intensity of the praising and critical comments is not an issue with respect to the interpretation of the between-group, within-condition comparisons. However, within-group comparisons across the various conditions should be interpreted keeping in mind that praise is a more intense stimulus than criticism. 4. Discussion The current study is the first to demonstrate abnormalities in cortico-limbic pathways in people who have experienced complete clinical recovery from depression when they are challenged by a psychosocial stressor implicated in the relapse process. Even when they have been entirely symptom free for more than 6 months, people with a past history of major depression do not respond to criticism in the same way that healthy controls do. When they are being criticized, people with a history of depression activate the amygdala significantly more and activate the DLPFC and the ACC significantly less than is characteristic of healthy controls. The observed activation differences between controls and formerly depressed participants cannot be explained by differences in their current mood states; the formerly depressed participants had similar scores to the controls on the BDI both at entry into the study and on the day of scanning. Both study groups also experienced similar levels of anxious arousal and anhedonia on the MASQ. Moreover, independent ratings of the mothers' critical comments provided no evidence that the comments made about the formerly depressed participants were different in intensity from the comments made about the healthy controls. There were also no group differences when the comments were rated for intensity by the participants themselves. These results suggest that even after full recovery from an episode of major depression neural responses to criticism do not normalize. The findings are particularly interesting in light of the reliable association between criticism and relapse in depression (Butzlaff and Hooley, 1998). Criticism may be linked to relapse because it is capable of perturbing nodes in the neural circuitry that underlies depression. Increased amygdala activation to criticism may be a vulnerability marker for depression that endures even in the face of full and complete symptom recovery. During praise, or when exposed to neutral comments, formerly depressed and control participants showed comparable levels of activation in the amygdala, suggesting similar arousal of this region with full recovery. However, under both criticism and praise conditions, formerly depressed participants continued to show reduced activation in the DLPFC and the ACC relative to the controls. Again, participants' ratings of their mothers' praising comments failed to reveal any differences in the intensity of the stimulus comments that could account for the betweengroup brain activation differences. This suggests that hypoactivation in the DLFPC and the ACC may not just characterize people who are actively depressed (Mayberg et al., 1999; Davidson et al., 2002; Pizzagalli et al., 2006) but may also be associated with trait vulnerability to depression. During the presentation of neutral comments, control participants demonstrated increased ACC activity relative to formerly depressed participants. This may indicate their greater monitoring of potentially salient stimuli, even when these have no explicit emotional valence. Healthy controls may be characterized by an increased tendency to attend to auditory input, checking it for relevance. In contrast, we might speculatethatpeoplewhoarevulnerabletodepressionhaveanaccthat is (relatively) turned off perhaps to reduce the risk of being challenged by emotional material. In contrast, control participants, who do not need to employ this protective strategy, may be more able to remain engaged and to more fully monitor the stimuli that they hear. This could render them better able to receive praise when it occurs, and also allow them to process negative affective challenges more efficiently. The difference between the FD participants and the controls in the neutral comment condition makes it unlikely that rumination provides an Table 5 Comparison of FD and control participants during positive auditory stimuli Region of Interest Voxels Coordinates Region of activation T-score Z-score BA 9 and Right Frontal Lobe, Middle Frontal Gyrus, Brodmann area Left Frontal Lobe, Precentral Gyrus, Sub-Gyral Left Frontal Lobe, Middle Frontal Gyrus Right Frontal Lobe, Medial Frontal Gyrus, Brodmann areas 8 and Left Frontal Lobe, Medial Frontal Gyrus, Limbic Lobe, Cingulate Gyrus Brodmann areas 6, 9 and Left Frontal Lobe, Superior Frontal Gyrus, Middle Frontal Gyrus, Brodmann areas 8 and ACC Right Limbic Lobe, Anterior Cingulate Control participants show significantly greater activation in dorsolateral prefrontal cortex (BA 9 and 46) and ACC relative to FD participants. No significant group differences were evident for the amygdala during positive auditory stimuli.

8 90 J.M. Hooley et al. / Psychiatry Research: Neuroimaging 172 (2009) Fig. 5. Neutral. Illustrationsof group subtractionbold activation resultsfor (A) dorsolateral prefrontal cortex (BA 9 and 46), and (B) anterior cingulate during neutral comments. Significantly greater activation is found in controls compared with formerly depressed participants. No group differences were found for the amygdala. adequate explanation of the group activation differences reported during criticism and praise. Certainly, it is plausible to suggest that FD participants mightbemoreinclinedtoruminateaboutthecriticismtheyhadjust heard during the following rest period. This would result in a relative decrease in brain activation in any criticism-rest period subtraction and an overall decrease in brain activation relative to (non-ruminating) controls in any between-groups comparison. However, it is much less plausible to suggest that FD participants also ruminate after they hear a comment about the weather. For this reason, differences in rumination do not appear to provide a satisfactory explanation of our findings. These results replicate and extend our preliminary finding of reduced DLPFC activation in fully recovered unipolar depressed participants during criticism (Hooley et al., 2005). Our findings are also consistent with earlier reports of resting blood flow abnormalities in subgenual ACC in recovered depressed patients (Liotti et al., 2002) and with decreased activation in prefrontal cortex in unmedicated remitted depressed patients exposed to transient sadness (Gemar et al., 2006). The DLPFC has been shown to play an important role in emotional processing, and neuroimaging studies using both resting state approaches and challenge techniques have found decreased activity in this region in depressed patients. Decreased activity of the DLPFC has also been associated with increased amygdala activation, although the specific interactions between these brain regions have not been fully characterized. As noted earlier, it has been hypothesized that the DLPFC exerts inhibitory control over the amygdala, and that the increased amygdalar activity seen in patients with depression may be the result of reduced control from the DLPFC. However, it has also been suggested that increased amygdalar activity may lead to decreased DLPFC function or that the DLPFC may not be effectively recruited, independent of emotional reactivity. Findings from the current investigation are consistent with changes in connectivity in these pathways, as they revealed a negative correlation between the number of voxels activated in the DLPFC and the amygdala in the control participants during exposure to criticism. This correlation was not evident in the formerly depressed subjects. Siegle and colleagues (2007) have reported that the functional connectivity between the DLPFC and the amygdala is weaker in depressed subjects than in control subjects. Connectivity between the rostral cingulate and the amygdala has also been reported to be reduced in patients with depression (Anand et al., 2005; Siegle et al., 2007). Although we did not conduct a formal connectivity analysis, our data hint that reduced connectivity between the DLPFC and the amygdala may remain even after full recovery from depression. We constrained our primary analyses to specified regions of interest that have previously been shown to mediate affective processing in depressed adults. Future research will need to expand the analyses to include all brain regions that may be implicated in the affective processing network. The generalizability of our findings is also limited by the focus on female subjects. Another potential concern is that we did not exclude participants who were taking medications. However, it should be noted that exploratory analyses revealed no significant differences between medicated and unmedicated FD participants for any of the regions of interest during any of the experimental conditions. Moreover, Drevets and colleagues (2002) have noted that treatment with antidepressants may suppress (rather than enhance) amygdalar reactivity and Gemar et al. (2006) found decreased activation in prefrontal cortex in drug-free remitteddepressedpatientswhoexperiencedasadmoodinduction. Maintenance antidepressant usage is therefore unlikely to account for our finding of different cortico-limbic responding to criticism in formerly depressed participants versus controls. Indeed, our data show that even if they take medications designed to normalize mood, fully recovered people with a history of depression still do not show a normalization of responding to affectively challenging social stimuli. A unique aspect of our study is that it involved daughters receiving criticism and praise from their mothers. In our future work we hope to explore how attachment and other measures of relationship quality impact the processing of emotional stimuli in this paradigm. It will also be important to explore the role of current depressed mood state in the neural processing of personally salient affective stimuli. Overall, the findings suggest that even when there are no selfreported differences between the formerly depressed and the never depressed, vulnerability to depression may be characterized by abnormalities in nodes along cortico-limbic pathways that can be revealed by certain types of affective challenges. Moreover, these abnormalities, which are independent of clinical state, are operating under the radar of self-report, highlighting the value of neuroimaging approaches. These abnormalities warrant further exploration as potential trait-markers for vulnerability to depression. Table 6 Comparison of FD and control participants during neutral auditory stimuli Region of Interest Voxels Coordinates Region of activation T-score Z-score BA 9 and Left Frontal Lobe, Middle Frontal Gyrus Right Frontal Lobe, Medial Frontal Gyrus, Brodmann area ACC Left Limbic Lobe, Anterior Cingulate, Brodmann area Right Frontal Lobe, Medial Frontal Gyrus, Limbic Lobe,Anterior Cingulate Inter-Hemispheric, Right Limbic Lobe, Anterior Cingulate, Brodmann area Control participants show significantly greater activation in dorsolateral prefrontal cortex (BA 9 and 46) and ACC relative to FD participants. No significant group differences were evident for the amygdala during neutral auditory stimuli.