Modulation of Cortical-Limbic Pathways in Major Depression: Treatment-Specific Effects of Cognitive Behavior Therapy

Background  Functional imaging studies of major depressive disorder demonstrate response-specific regional changes following various modes of antidepressant treatment.

Objective  To examine changes associated with cognitive behavior therapy (CBT).

Methods  Brain changes underlying response to CBT were examined using resting-state fluorine-18–labeled deoxyglucose positron emission tomography. Seventeen unmedicated, unipolar depressed outpatients (mean ± SD age, 41 ± 9 years; mean ± SD initial 17-item Hamilton Depression Rating Scale score, 20 ± 3) were scanned before and after a 15- to 20-session course of outpatient CBT. Whole-brain, voxel-based methods were used to assess response-specific CBT effects. A post hoc comparison to an independent group of 13 paroxetine-treated responders was also performed to interpret the specificity of identified CBT effects.

Results  A full course of CBT resulted in significant clinical improvement in the 14 study completers (mean ± SD posttreatment Hamilton Depression Rating Scale score of 6.7 ± 4). Treatment response was associated with significant metabolic changes: increases in hippocampus and dorsal cingulate (Brodmann area [BA] 24) and decreases in dorsal (BA 9/46), ventral (BA 47/11), and medial (BA 9/10/11) frontal cortex. This pattern is distinct from that seen with paroxetine-facilitated clinical recovery where prefrontal increases and hippocampal and subgenual cingulate decreases were seen.

Conclusions  Like other antidepressant treatments, CBT seems to affect clinical recovery by modulating the functioning of specific sites in limbic and cortical regions. Unique directional changes in frontal cortex, cingulate, and hippocampus with CBT relative to paroxetine may reflect modality-specific effects with implications for understanding mechanisms underlying different treatment strategies.

RANDOMIZED CLINICAL TRIALS in patients with both mild and severe majordepression consistently demonstrate similar rates of response to cognitivebehavior therapy (CBT) and antidepressant pharmacotherapy.^1,2 Althoughit is generally assumed that these disparate treatments have different primary targets of action, with cortical "top-down" vs subcorticalor "bottom-up" mechanisms theorized,^3-5 definitiveneural mechanisms that mediate antidepressant response arenot yet characterized for either treatment modality.

Preclinical studies^6-10 ofantidepressant medications emphasize a bottom-up chain of events, includingaminergic reuptake inhibition and associated presynaptic autoregulatory desensitization,up- and down-regulation of multiple postsynaptic receptor sites, and receptor-mediatedsecond messenger and neurotrophic intracellular signaling effects. Requisitebrain regions that mediate these events are unknown, although putative primarysites of action in the dorsal raphe, locus ceruleus, hippocampus, and hypothalamusare well described, with documented secondary changes in frontal cortex alsoreported.^11-16 Neuroimagingstudies¹⁷ of medication effects show a similartime course of differential acute and chronic subcortical and cortical changes.Across studies^17-21 ofantidepressant response, frontal cortex changes arethe most consistently reported, with normalization of frontal overactivityand underactivity described. Additionally, changes have been seen in limbicand subcortical regions, including the subgenual cingulate, hippocampus, posteriorcingulate, and insula, with decreased activity the most commonly observedeffect.^17,19-23

In contrast, little is known about brain mechanisms that mediate clinicalresponse to CBT for depression. The literature^24-26 characterizingbrain changes associated with CBT response is sparse and based largely onthe treatment of obsessive-compulsive and anxiety disorders. Theoretical modelsof CBT action in the treatment of depression generally implicate top-downmechanisms, because the intervention focuses on modifying attention and memoryfunctions involved in the mediation of depression-relevant cognitions, affectivebias, and maladaptive information processing.^27-32 Thetime course of symptom changes with CBT further supports an initial corticalsite of action, as improvement in hopelessness and views of self and moodgenerally precede changes in vegetative and motivational symptoms—atimeline not seen in patients treated with pharmacotherapy.^3,33 Braincorrelates of this chronology are, however, untested. Recent functional imagingstudies^34,35 examining brain changesfollowing interpersonal psychotherapy report a variety of regional effects,but there is no consistent pattern across the few published studies.

A critical question is whether disparate antidepressant treatments resultin common or modality-specific neural effects. As a first step in addressingthis issue, this study examined changes in regional glucose metabolism measuredwith positron emission tomography (PET) associated with depression remissionfollowing 15 to 20 sessions of CBT. Metabolic change patterns with CBT responsewere contrasted post hoc with those of a previous study²¹ ofparoxetine treatment to further test the hypothesis that modulation of distinctneural targets by different interventions within a putative limbic-corticaldepression "network" occurs with clinical remission, regardless of the specifictreatment modality.^36,37

Seventeen unmedicated, depressed patients (6 men, 11 women; mean ±SD age, 41 ± 9 years; mean ± SD 17-item Hamilton DepressionRating Scale [HDRS] score, 20 ± 3) with symptoms that required treatmentwere recruited to the Mood and Anxiety Disorders Program at the Centre forAddiction and Mental Health in Toronto, Canada, through newspaper advertisement.The clinical diagnosis of a major depressive episode, unipolar type, was confirmedusing the Structured Clinical Interview for DSM-III-R and DSM-IV criteria.^38,39 Byhistory, none of the enrolled patients were considered treatment refractory.Mean ± SD education was 16 ± 2 years, and 10 of 14 were unmarried.Exclusion criteria included history of neurological disease, head trauma,or other Axis I psychiatric diagnoses, as well as current psychotic symptoms,substance abuse, antidepressant treatment within the preceding month, andpregnancy. Six patients were completely drug naive, and none had been treatedwith CBT for depression in the past. One patient required antidepressant washoutfor 4 weeks. Written informed consent was obtained from all participants,and the study was conducted as approved by the Centre for Addiction and MentalHealth Ethics Committee.

All patients received 15 to 20 individualized outpatient sessions ofCBT. Treatment was conducted by 1 of 2 trained CBT therapists (M.L. and P.B.)with 10 and 8 years of experience, respectively, according to the treatmentmanual described by Beck et al.²⁷ All CBT sessionswere audiotaped to enable ratings of treatment fidelity, which were confirmedby the supervising psychologist (Z.S.). Patients undergoing CBT used a numberof therapeutic strategies intended to reduce automatic reactivity to negativethoughts or attitudes and to combat dysphoric mood. Behavioral activationwas used to address the disruption of routine often brought on by depressionand focused on increasing the frequency of pleasant and masterful events inpatients' lives, especially in those areas where marked avoidance and withdrawalwere noted. Cognitive monitoring taught patients how to dismantle seeminglycomplex chains of thinking and feeling into separate components that couldthen be evaluated for evidence of biased information processing. Between sessions,patients were asked to test their interpretations and beliefs through theuse of behavioral experiments and to record their thinking using thought records.During the sessions, the therapists used collaborative inquiry to guide thepatient to a more evidence-based and less reactive construal of their experience.

Clinical response was monitored weekly using the Beck Depression Inventory.⁴⁰ The HDRS scores (17-item)⁴¹ wereassessed at study onset, at study completion, and once midway through therapy(eighth session). Patients were classified as responders based on the criteriaof at least a 50% reduction in HDRS or nonresponders for those with a decreasein HDRS score of less than 20%.⁴²

Positron emission tomography measurements of regional cerebral glucosemetabolism were obtained at baseline and again at the end of treatment usingstandard imaging methods⁴³ and a previouslypublished protocol.^17,44 Bothscans were acquired within 1 week of the first and last therapeutic session.For each scan, a 5-mCi (185-Mbq) dose of fluorine-18–labeled deoxyglucose(FDG) was injected intravenously, with image acquisition beginning after 40minutes (PC 2048b; GEMS-Scanditronix, Uppsala, Sweden). All scans were acquiredwith patients supine, awake, and in the resting state, with eyes closed andears uncovered. Patients were asked to refrain from food, coffee, and alcoholintake for a minimum of 6 hours before each scan session. None of the participantswere smokers. Patients were taking no medications at the time of either scanwith the exception of 1 woman who was taking long-standing estrogen and thyroidtherapy. Patients were given no explicit cognitive instructions but were askedto avoid ruminating on any one topic during the FDG uptake period. Wakefulnesswas additionally monitored every 10 minutes by a study investigator. A debriefingsession took place following the uptake period to document compliance. Presenceof active random thoughts was not quantitatively assessed. Emission data wasacquired during a 35-minute period (approximately 1 million counts per slice;10-cm field of view). A customized, thermoplastic face mask was used to minimizehead movement for the initial scan and for accurate repositioning at the secondsession. Raw images (15 parallel slices; 6.5-mm center-to-center interslicedistance) were corrected for attentuation, reconstructed, and smoothed toa final in-plane resolution of 7.0 mm at full width at half maximum.

Statistical analyses were performed using SPM99 statistical software(Wellcome Department of Cognitive Neurology, London, England) implementedin Matlab (version 5.3; Mathworks Inc, Sherborn, Mass). The data were firstscreened for distributional properties, outliers, and missing values. Thisprocess rejected no scans. All scans were then normalized to the MontrealNeurological Institute's ICBM 152 stereotactic template within SPM99, whichreferences brain locations in 3-dimensional space relative to the anteriorcommissure.^45,46 The images werethen corrected for differences in the whole-brain global mean and smoothedusing a gaussian kernel to a final in-plane resolution of 10 mm at full widthat half maximum. Absolute glucose metabolic rates were not calculated.

Response-specific CBT effects were the primary focus of this study,reflected by the following series of statistical analyses. Significant regionalchanges before and after treatment were first assessed using SPM and a pairwiserandom-effects design.^47,48 Basedon previous results of antidepressant medication effects,¹⁷ peakvoxel value significance thresholds were set at P<.01(uncorrected) for 5 targeted regions (ventral subgenual cingulate Brodmannarea [BA] 25, dorsal anterior cingulate BA 24, dorsolateral prefrontal cortexBA 9/46, hippocampus, and posterior cingulate BA 23/31) and at P<.001 (uncorrected) for all other regions. Cluster significancethresholds were set at 50 contiguous voxels (voxel = 8 mm³) tofurther reduce type I errors introduced by potential noise. Resulting t values were converted to z scores,with brain locations reported as x, y, and z coordinates in Montreal NeurologicalInstitute space with approximate BAs identified by mathematical transformationof SPM99 coordinates into Talairach space⁴⁹ (additionalinformation available at http://www.mrc-cbu.cam.ac.uk/Imaging/)(Table 1).

To assist in interpreting any identified metabolic changes with CBT,several additional post hoc analyses were performed. Metabolic changes withresponse to CBT were statistically contrasted to those seen in a previouslypublished data set of comparably recruited depressed men (n = 13; mean ±SD age, 36 ± 10 years; mean ± SD education, 15 ± 2 years;7 unmarried; mean ± SD HDRS score, 22.4 ± 3.6) who had beensimilarly scanned following clinical response to 6 weeks of paroxetine treatment.²¹ A conjunctional analysis using statistical criteriaidentical to those described herein was performed to directly compare thechange pattern of CBT responders to that of paroxetine responders ([CBT scan2-1] − [paroxetine scan 2-1]). The specific paroxetine change patternwas also examined separately to determine if significant differences in theconjunctional analysis were due to differences in magnitude of the same changeor distinct treatment-specific effects of each intervention. Scans from theparoxetine treatment group were acquired with the same PET camera and an identicalscanning protocol to that used for the CBT study. Furthermore, the paroxetineraw data were reprocessed and reanalyzed in SPM99 to match all variables usedfor the primary CBT analyses. In the absence of a controlled randomized trialof CBT and medication, this set of post hoc analyses provided a critical perspectivetoward interpreting the main CBT response findings. Baseline scans for the2 groups were also compared.

Fourteen of the 17 patients completed the full treatment course (mean± SD number of sessions, 17.7 ± 2 for 26 ± 7 weeks).Three participants withdrew within the first 2 weeks due to worsening of symptoms(2 patients) or inability to comply with CBT instructions (1 patient); nosecond scan was acquired for these patients. For the 14 completers, the mean± SD HDRS scores were 20 ± 3 before treatment and 6.7 ±4 after treatment, with a decrease of 66% ± 22% (t = 9.66, P<.001). Of these 14 completers,9 patients met the 50% decrease criteria for full response (final mean ±SD HDRS score, 4.7 ± 3.5; decrease of 78 ± 17). The remaining5 patients had no less than a 35% decrease in their HDRS scores (final mean± SD HDRS score, 10.4 ± 0.7). Because of the small overall samplesize and lack of a pure CBT nonresponder group, all patients were includedin the pretreatment-to-posttreatment analysis. Patients in the paroxetine-treatedcomparison group had a similar severity of symptoms at baseline (mean ±SD HDRS score, 22.8 ± 3.6) and showed a comparable clinical response(posttreatment mean ± SD HDRS score, 6.0 ± 4.1; mean ±SD decrease of 75% ± 14%; t = 17.2, P<.001).

Treatment with CBT was associated with significant regional metabolicchanges (Table 1, left; Figure 1, top).Areas of increased metabolism before to after treatment included the hippocampusand dorsal midcingulate (BA 24b/c). In addition, widespread decreases wereobserved in dorsolateral prefrontal (BA 9/46), ventrolateral prefrontal (BA11/47), and superior and inferior medial frontal regions (BA 9/10/11), aswell as posterior cingulate (BA 31), inferior parietal (BA 40), and inferiortemporal cortex (BA 20). The same significant metabolic change pattern wasseen when the 5 patients who showed less than the 50% response rate were excludedfrom the analysis. The findings seem specific for clinical response ratherthan solely the passage of time, because covarying for the HDRS score nullifiedthe between-occasion effects.

The conjunctional analysis contrasting CBT response change to paroxetineresponse change identified significant differences between the 2 treatmentsin numerous cortical and limbic regions (Table 1): dorsolateral prefrontal (BA 9), ventromedial frontal (BA10/11), and inferior parietal (BA 40) cortices, as well as insula, hippocampus,ventral subgenual cingulate (BA 25), anterior and dorsal midcingulate (BA24), posterior cingulate (BA 31), insula, brainstem, and cerebellum. The separateanalyses of the 2 change patterns were in fact necessary to determine whichgroup drove these differences and in what direction (Table 1).

The dorsolateral prefrontal, inferior parietal, and hippocampal differencesidentified in this conjunctional analysis represented an inverse pattern forCBT and paroxetine. The between-treatment differences in dorsal midcingulate,ventromedial frontal, and posterior cingulate were related to unique changeswith CBT treatment and were not seen with paroxetine at any statistical threshold(Table 1). Differences involvingsubgenual cingulate (BA 25), insula, brainstem, and cerebellum likewise weredue to unique paroxetine treatment effects (Table 1). Similar for the 2 treatments were decreases in ventralprefrontal cortex (BA 47).

Direct comparison of baseline scans for the CBT and paroxetine groupsdemonstrated no significant differences. There were also no significant correlationsbetween metabolism and weeks of treatment across groups. Finally, covaryingthe pretreatment and posttreatment changes with the HDRS score nullified thechanges in both groups, providing additional evidence that the divergent changepatterns reflect treatment-specific response effects.

Reciprocal limbic increases (hippocampus, dorsal midcingulate) and corticaldecreases (dorsolateral, ventrolateral, and medial orbital frontal; inferiortemporal and parietal) were identified following successful treatment withCBT. These regional changes involve sites similar, and in some cases identical,to those seen previously with paroxetine and other pharmacotherapies,^21,37 but the changes were in the oppositedirection.

Interpreted in the context of an extensive PET and functional magneticresonance imaging behavior mapping literature, the metabolic change patternseen with resolution of depressive symptoms following CBT provides tentativeneural correlates of the long-theorized psychological or top-down mechanismsthat mediate CBT response^33,50 (Figure 2). Examples of such parallels includelocalization of tasks involving directed attention, reward-based decisionmaking, and monitoring of emotional salience to the anterior cingulate andorbital frontal cortex^51-58;memory encoding, retrieval, and consolidation to the hippocampus^59-61;and working memory, self-referential processing, and cognitive ruminationsto dorsolateral, medial frontal, and ventral prefrontal cortex, respectively.^62-65 Componentsof these behaviors have been implicated in the initiation and maintenanceof the depressive state^31,32,57,66 andseem to be specifically targeted in CBT.^27,31,32 Althoughspeculative, hippocampal and mid and anterior cingulate increases coupledwith decreases in medial frontal, dorsolateral, and ventrolateral prefrontalactivity with CBT treatment might be nonetheless interpreted as correlatesof CBT-conditioned increases in attention to personally relevant emotionaland environmental stimuli associated with a learned ability to reduce onlinecortical processes at the level of encoding and retrieval of maladaptive associativememories, as well as a reduction in both ruminations and the overprocessingof irrelevant information.

In further support of a critical role for medial frontal modulationwith CBT response compared with medication are the unique changes in anteriorand dorsal midcingulate (BA 24), medial frontal (BA 10), and orbital frontal(BA 11) with treatment response. Although both groups demonstrated hyperactivityin medial frontal regions before treatment, only CBT was associated with widespreadchanges. Activation of these regions has been previously associated with emotionalprocessing tasks in nondepressed control participants, including the activerethinking and reappraisal of emotional feelings.^58,63-65,67 Exaggeratedactivity in this region has been similarly reported in depressed patientsin response to sad words, supporting the previously recognized negative emotionalbias in this patient population.⁶⁶ These observationsare consistent with nonimaging studies that demonstrate increased relapserisk in those remitted depressed patients with persistent mood-linked reactivityto negative emotional stimuli³¹ and increasedsustained remission for patients in whom this reactivity is reduced.^32,68 Referable to the patients in thisstudy, selective changes with CBT in these regions may reflect a reduced biastoward the processing of negative information in the recovered state, withimplications for future relapse risk.

The frontal decreases seen with CBT response are strikingly similarto those reported in a recent FDG-PET study³⁴ ofinterpersonal psychotherapy for major depression. Regional changes with CBTtreatment for other disorders also describe areas of overlap with those reportedherein for depression. For instance, changes in the hippocampus are reportedwith CBT treatment for social phobia, although the changes are in the reversedirection.²⁶ A third distinct pattern of caudateand posterior orbital cortex decreases has been shown with CBT treatment forobsessive-compulsive disorder.^24,25 Together,these findings suggest that brain change pattern variations with CBT for variousdisorders likely reflect both fundamental differences in the underlying psychopathology(depression, obsessive-compulsive disorder, social phobia) and proceduraldifferences inherent in the cognitive method used to treat each condition.⁶⁹Method in this context notonly refers to the subtleties of the CBT procedures themselves but also includespatient expectation and conditioned learning facilitated by both the specificintervention and the individual physician-patient interaction.

These various elements, inherent in any specified therapy, likely alsoexplain the differences between the pattern of response to CBT and that reportedfor placebo medication,⁷⁰ a response consideredby some to be an uncontrolled psychological form of treatment. In the caseof fluoxetine treatment for depression, the change pattern for placebo fluoxetineoverlapped that seen with response to the active medication to which it wasexperimentally linked (frontal, parietal increases, subgenual cingulate decreases)rather than the psychological intervention patternseen here with CBT (frontal decreases, hippocampal increases). Brain changeswith placebo response, in fact, directly shadowed the true drug-response pattern,similar to that shown with both an acute dose of a dopamine-agonist in patientswith Parkinson disease (striatal dopamine changes)⁷¹ andan acute dose of an opiate analgesic (cingulate and brainstem blood flow changes),⁷² suggesting a complex interaction of the specifictreatment and expected behavioral effects. Obviously, a placebo-controlledCBT trial will be necessary to fully test the hypothesis that placebo-responsechanges mirror the specific intervention to which they are paired, meaningthat placebo CBT would be expected to overlap true CBT changes, not thoseseen with placebo medication. A wait-list control group will also be neededto address effects potentially attributable to spontaneous remission witheither treatment.

There are other potential explanations for reported change-pattern differencesacross various psychological treatment studies for depression, including thetype of cognitive intervention (CBT vs interpersonal psychotherapy), the imagingmodality (PET vs single-photon emission computed tomography; glucose metabolismvs blood flow), and the point of the second scan within the treatment course(6-8 vs >15 weeks). Although it is possible that brain changes with an incompletecourse of a nonpharmacologic treatment may be similar to those seen with fullclinical response, this is clearly not the case with antidepressant medication,where analyses of time course (1 vs 6 weeks) and response effects (respondervs nonresponder at 6 weeks) show significantly different metabolic changepatterns.¹⁷ This may also explain differencesin the nonpharmacotherapy treatment change patterns reported across otherpublished reports.^34,35 Explicitstudies of the time course of brain changes with various cognitive interventions,including a parallel assessment of both responders and nonresponders, areneeded to further test these hypotheses. Examination of the time course ofchange in HDRS scores in this CBT responder group (without corresponding PETscans) would suggest that metabolic change effects might be reasonably seenafter 8 sessions (eighth session mean ± SD HDRS score, 11.5 ±6), perhaps providing an early indication of who is most likely to respondto a full treatment course.⁵⁰

Despite the absence of a prospective, randomized study design and obviousdifferences in treatment duration, the post hoc paroxetine comparison performedherein provided several critical clues for further interpreting the identifiedCBT change effects. Most notably, the conjunctional analyses demonstrateda complex set of change pattern differences between CBT and paroxetine responders.Most significantly, in contrast to the CBT increases in hippocampus and decreasesin frontal cortex, the independent paroxetine analyses demonstrated the reversepattern—frontal increases and hippocampal decreases. The localizationand pattern of changes seen in the paroxetine group, including the uniquechanges in subgenual cingulate (BA 25), insula, and brainstem, replicate previoushuman and animal metabolic studies of various pharmacotherapies, includingother selective serotonin reuptake inhibitors and tricyclics.^16-19,21,37

This divergent pattern of frontal decreases and hippocampal increaseswith CBT relative to paroxetine is not explained by pretreatment metabolicabnormalities, because the 2 groups show no significant differences when directlycompared. The differential change patterns also appear not to be simply theresult of differences in the mean duration of treatment between the 2 groups,because there were no significant correlations between brain metabolism andweeks of treatment. Interestingly, in both groups, there is considerable overlapbetween the regions of metabolic change and areas of reported glial cell lossin postmortem studies, notably, dorsolateral, ventrolateral, and medial frontalcortices.⁷³ That said, neither group showedsignificant baseline hypometabolism in either frontal cortex or hippocampus,suggesting a more complex relationship among glial abnormalities, brain atrophy,and metabolic change patterns than previously suggested.^74-76 Quantitativemagnetic resonance imaging volumetric analyses, however, were not performed.

Taken together, the treatment-specific change patterns in CBT and paroxetineresponders support our initial hypothesis that each treatment targets differentprimary sites with differential top-down and bottom-up effects—medialfrontal and cingulate cortices with cognitive therapy (Figure 2, green) and limbic and subcortical regions with pharmacotherapy(brainstem, insula, subgenual cingulate; Figure 2, blue), both resulting in a net change in critical prefrontal-hippocampalpathways (Figure 2, red). The overallmodulation of this complex system rather than any one focal regional changemay be most critical for disease remission. As previously stated, definitiveconclusions regarding treatment-specific effects will require a randomizeddesign of depressed patients seeking either treatment.

It has been previously suggested that variations in scan patterns bothat baseline and following treatment reflect such clinical factors as illnessseverity, cognitive impairment, anxiety, psychomotor retardation, and depressivesubtypes.^77-81 Inthis study, there were no significant differences in illness severity, demographics,HDRS factor scores, or any other depression-related variable that might alternativelyexplain the differential metabolic change patterns across the 2 treatmentgroups. Detailed neuropsychological testing, however, was not performed. Althoughthe paroxetine comparison group was exclusively composed of men, no significantsex differences were seen in either the baseline scans or change patternsof the CBT group, although statistical power was inadequate to definitelyexclude sex effects.

Another potential confounder is the ongoing behavior of each patientat the time of each scan, particularly since patients were studied in a relativelyuncontrolled state (eyes-closed rest). Previous studies^82,83 duringa variety of cognitive tasks demonstrate that medial frontal regions showdecreases relative to rest, suggesting an ongoing activation of these regionsin the resting state. The medial frontal increases, seen at baseline in boththe CBT and paroxetine patients relative to healthy controls, although possiblyinterpretable as a pretreatment marker of increased attention to self, donot appreciably change with treatment, despite clinical improvement. Furthermore,the localization of these reported self-directed resting state markers isconsiderably more caudal to those demonstrated herein either at baseline orwith CBT response, suggesting that these baseline and change effects reflectdisease rather than a confounding of the short-term behavioral state.

Similarly, in test-retest studies^84-86 thatexamined effects of test environment, novelty, and levels of anxiety, publishedreports demonstrate a pattern of hyperactivity in lateral frontal corticesassociated with the first test condition. Again, neither group in this studyshowed this dorsolateral prefrontal pattern at rest, although both groupsshowed significant changes in these regions following clinical recovery. Althoughneither group was tested explicitly for state anxiety at the time of the scan,anxiety subscales of the HDRS performed just before each scan session showedno differences between the groups at baseline. In addition, comorbid anxietydisorders were among protocol exclusion criteria. It is possible that theabsence of prefrontal findings at baseline reflect a first-test effect inboth groups, in essence, counteracting the expected frontal hypometabolismtypical of many published studies of major depression.⁷⁹ This,however, would not explain the differential changes in frontal cortex seenfollowing treatment where again both groups showed comparable anxiety subscalescores. In the absence of more subtle behavioral measures, there is no evidenceto support the conclusion that the disparate changes in frontal activity forone group relative to the other are a function of state anxiety. The potentialcontributions of other uncontrolled individual variables, such as family history,specific gene polymorphisms, temperament, early life abuse, or previous depressiveepisodes, were not examined.^87-89

Finally, although the 2 groups were studied as independent cohorts,met identical inclusion criteria, and were recruited through the same mediaoutlets, the possibility of a selection bias still exists. A trial with randomassignment of patients to 1 of the 2 treatments of comparable duration isneeded to fully address this concern and is the focus of an ongoing study.That said, it is worth noting that the self-selection by patients of a specificantidepressant intervention may reflect their probabilistic calculation ofbenefit, taking past treatment into account. Anecdotally, many of those inthe CBT group who had previously been treated with medication expressed strongdisinterest in repeating pharmacotherapy. In fact, many demonstrated considerableinsight, believing that their negative thoughts and beliefs were causing andmaintaining their depressive state. In addition, those who had taken antidepressantmedications in the past tended to minimize their effectiveness due to associatedadverse effects. These subjective reports may provide important targets forfuture investigations of the predictive value of patient treatment preferencesand their neural correlates.⁹⁰

Corresponding author: Helen Mayberg, MD, Rotman Research Institute,Baycrest Centre, 3560 Bathurst St, Toronto, Ontario, Canada M6A 2E1 (e-mail: hmayberg@rotman-baycrest.on.ca).

Submitted for publication September 11, 2002; final revision receivedJune 17, 2003; accepted June 24, 2003.

This study was supported in part by the Sandra Rotman Chair in Neuropsychiatry,Rotman Research Institute (Toronto), the Canadian Institute for Health Research,and a University of Toronto Institute of Medical Science Open Fellowship Award(Ms Goldapple).

This study was presented in part at the 2002 Annual Meeting of the Societyof Biological Psychiatry; May 17, 2002; Philadelphia, Pa.

We thank Doug Hussey, BSc, RTNM, and Kevin Cheung, RTNM, for their experttechnical assistance with PET scan acquisition and data reconstruction.