Attenuation of the Neural Response to Sad Faces in Major Depressionby Antidepressant Treatment: A Prospective, Event-Related Functional Magnetic Resonance ImagingStudy

Background  Depression is associated with interpersonal difficulties related to abnormalities in affective facial processing.

Objectives  To map brain systems activated by sad facial affect processing in patients with depression and to identify brain functional correlates of antidepressant treatment and symptomatic response.

Design  Two groups underwent scanning twice using functional magnetic resonance imaging (fMRI) during an 8-week period. The event-related fMRI paradigm entailed incidental affect recognition of facial stimuli morphed to express discriminable intensities of sadness.

Setting  Participants were recruited by advertisement from the local population; depressed subjects were treated as outpatients.

Patients and Other Participants  We matched 19 medication-free, acutely symptomatic patients satisfying DSM-IVcriteria for unipolar major depressive disorder by age, sex, and IQ with 19 healthy volunteers.

Intervention  After the baseline assessment, patients received fluoxetine hydrochloride, 20 mg/d, for 8 weeks.

Main Outcome Measures  Average activation (capacity) and differential response to variable affective intensity (dynamic range) were estimated in each fMRI time series. We used analysis of variance to identify brain regions that demonstrated a main effect of group (depressed vs healthy subjects) and a group × time interaction (attributable to antidepressant treatment). Change in brain activation associated with reduction of depressive symptoms in the patient group was identified by means of regression analysis. Permutation tests were used for inference.

Results  Over time, depressed subjects showed reduced capacity for activation in the left amygdala, ventral striatum, and frontoparietal cortex and a negatively correlated increase of dynamic range in the prefrontal cortex. Symptomatic improvement was associated with reduction of dynamic range in the pregenual cingulate cortex, ventral striatum, and cerebellum.

Conclusions  Antidepressant treatment reduces left limbic, subcortical, and neocortical capacity for activation in depressed subjects and increases the dynamic range of the left prefrontal cortex. Changes in anterior cingulate function associated with symptomatic improvement indicate that fMRI may be a useful surrogate marker of antidepressant treatment response.

The first functional neuroimaging studies of depression measured restingcerebral blood flow using xenon Xe 133 inhalation during a resting state.¹ Since then, researchers have used single-photon emissiontomography and positron emission tomography to correlate abnormalities inresting state activity with clinical symptoms^2-4 andto examine trait abnormalities and state changes after a variety of treatments.^5-10 Morerecent studies have examined specific cognitive^11-13 andaffective^13-16 processesand, using functional magnetic resonance imaging (fMRI), have combined cognitiveactivation paradigms with antidepressant treatments.^17,18

A fundamental neuropsychological impairment in depression is a mood-congruentprocessing bias such that ambiguous or positive events tend to be perceivedas negative.^19-23 Inparticular, depressed patients show a diminished ability to discern affective,eg, happy and sad, facial expressions.^24-27 Thisimpairment contributes significantly to psychosocial and interpersonal difficultiescommonly observed during an acute depressive episode,^24,25 andits persistence during remission of clinical symptoms is associated with avulnerability for future episodes.^27,28

The neurocognitive systems involved in identifying affective facialexpressions have been well studied in healthy individuals.^29-31 Keynodes in this network³² include the fusiformface area in the ventral occipitotemporal cortex, which shows a selectiveresponse for faces^29-31;the superior temporal sulcus, which is responsive to mouth and eye movementsinvolved in facial expressions³³; and the amygdala(usually the left amygdala), which shows a selective response to emotionalfacial expressions³⁴ such as fear,^35-37 sadness,^38,39 anger,⁴⁰ anddisgust.^41,42

In depression, neural correlates of negative affective visual processingwere investigated using fMRI in 2 recent studies. Davidson et al¹⁸ observedactivation by negative visual stimuli, greater in the left fusiform gyrusin acutely depressed patients compared with healthy volunteers, and in theamygdala bilaterally in both groups. With masked emotional faces, Shelineet al¹⁷ found increased left amygdala activityin acutely depressed patients. These findings are consistent with studiesof healthy individuals that found greater activation in the fusiform gyrusduring explicit processing of sad relative to neutral facial expressions⁴³ and in the amygdala with sad faces.^38,39 Shelineet al¹⁷ also found that increased activationof the left amygdala resolved after 8 weeks of antidepressant treatment witha selective serotonin reuptake inhibitor (sertraline hydrochloride, 100 mg/d).

This effect of antidepressant treatment on amygdala activation is complementedby the findings of other studies, which indicate reverse of brain functionalabnormalities by antidepressant treatments.^5-8 Inparticular, the subgenual component of the anterior cingulate cortex showsincreased activity with provocation of sadness in normal volunteers⁴⁴ and attenuation of initially enhanced basal metabolicactivity in patients with depression after effective antidepressant treatment.¹⁵ Higher resting metabolism at baseline in the anteriorcingulate cortex has been replicably associated with a better symptomaticresponse to antidepressant treatment^6,18,45-48 (reviewedby Fu et al⁹).

In this study, we used fMRI to study brain systems activated specificallyby incidental processing of sad facial affect in the following 2 parallelgroups of participants who underwent imaging twice in 8 weeks: 19 depressivepatients treated with fluoxetine hydrochloride after baseline assessment and19 healthy volunteers. On the basis of 2 key previous fMRI studies,^17,18 we hypothesized that an abnormallyenhanced amygdala response to sad facial affect processing in the patientsat baseline would be attenuated by 8 weeks of antidepressant treatment. Onthe basis of previous data implicating the anterior cingulate cortex in recoveryfrom depression,^6,18,45-48 wealso predicted an association between symptomatic response and changes inactivation of the pregenual anterior cingulate cortex.

There are methodological distinctions between this study and its immediatepredecessors^17,18 in terms ofdesign and analysis. We used an event-related fMRI paradigm and presentedsad faces at 3 discriminable degrees of affective intensity. This allowedus to map the dynamic range of brain response to affective stimuli, as wellas the overall capacity or mean difference between affective and baselinetrials. Also, we used whole-brain analysis by nonparametric statistical methodsto address the following 2 key questions: (1) do patients with depressiondiffer from comparison subjects in activation of negative affect–processingsystems? and (2) do patients with depression show antidepressant treatment–relatedchanges in activation of such systems? These questions were tested by meansof mixed-effects modeling of a balanced factorial design consisting of a largernumber of participants (n = 19 in each group) than previously described incomparable studies.

Twenty-one participants (15 women; age range, 29-58 years) meeting DSM-IV criteria for major depressive disorder⁴⁹ accordingto the Structured Clinical Interview for DSM-IV AxisI Disorders⁵⁰ were recruited through localnewspaper advertisements. Inclusion criteria were an acute episode of majordepressive disorder of the unipolar subtype⁴⁹ anda score of at least 18 on the 17-item Hamilton Rating Scale for Depression(HRSD).⁵¹ Exclusion criteria were a historyof neurological trauma resulting in loss of consciousness; current neurologicaldisorder; current comorbid Axis I disorder, including bipolar disorder andanxiety disorder; or a history of substance abuse within 2 months of studyparticipation. All patients were free of psychotropic medication for a minimumof 4 weeks at recruitment. Functional MRI data from 2 patients were subsequentlyexcluded because of a neuroradiological abnormality (age-inappropriate ventriculomegalyand periventricular leukoaraiosis) in one case and failure to attend follow-upappointments in the other. Therefore, 19 patients constitute the sample reportedherein (Table 1).

Nineteen healthy comparison subjects (11 women) with HRSD scores ofless than 8 and no history of any psychiatric disorder, neurological disorder,or head injury resulting in a loss of consciousness were recruited by advertisementfrom the local community and matched to the patients in terms of age, sex,and IQ.

All participants provided written, informed consent. The project wasapproved by the Ethics Research Committee, Institute of Psychiatry, London,England.

We adopted a parallel-group, repeated-measures design in which all subjectsunderwent imaging in 4 separate sessions. The first session served to acquirea structural MRI data set for neuroradiological examination and to familiarizesubjects with the MRI unit and imaging environment. Then, 3 separate 60- to90-minute sessions were scheduled to acquire fMRI data sets at baseline orweek 0, 2 weeks after baseline, and 8 weeks after baseline. During each session,subjects participated in a number of activation paradigms, but (for the sakeof brevity and clarity) only data acquired at weeks 0 and 8 for the sad facialaffect recognition task and the visual stimulation task will be presentedherein. The sad facial affect recognition task was always the final cognitivetask presented in each imaging session to prevent any possible residual effectof an induced negative mood state on performance of subsequent tasks.⁵²

Patients received antidepressant treatment with oral fluoxetine hydrochloride,a selective serotonin reuptake inhibitor, 20 mg/d in a single dosage, startingas soon as possible (typically <1 day) after the baseline fMRI sessionand continuing until their completion of the study protocol 8 weeks later.For the duration of their participation in the study, the patients underwenta clinical assessment every 2 weeks with a psychiatrist (C.H.Y.F.), and depressivesymptoms were serially rated using the HRSD.⁵¹ Somepatients reported minor adverse effects (nausea/vomiting or headache) soonafter the initiation of treatment, but all patients recruited into the studywere able to complete the protocol satisfactorily. Subjects were reimbursedfor their travel expenses for each clinical session and received £20(approximately US $35) for the initial MRI session and £30 (US $45)for each fMRI session.

Ten faces (5 male) from a standardized series of facial expressionsof sadness⁵³ were morphed to represent low,medium, and high intensities of sadness (Figure 1). Further detail on prior behavioral validation of thesestimuli is available as supplemental material at http://www-bmu.psychiatry.cam.ac.uk/DATA (accessed July 5, 2004). For the event-related fMRI paradigm, facialstimuli and baseline trials (crosshair fixation) were presented in randomorder. Each facial stimulus was presented twice at each intensity of sadness(60 faces in total), along with 12 baseline trials (crosshair visual fixationpoint), for a total of 72 trials. Each trial was presented for 3 seconds,and the intertrial interval was randomly varied according to a Poisson distributionwith mean intertrial interval of 5 seconds. Total duration of the experimentwas therefore 360 seconds. The same stimulus set was used at baseline andat 8 weeks.

For each facial trial, subjects were asked to indicate the sex of theface (male or female) by lateral movement of a joystick; no hand movementwas required in response to a baseline trial. Latency (or reaction time) andaccuracy of the sex decision during imaging were recorded for each trial.We used this strategy to elicit incidental or implicit affective processingbecause of previous data suggesting that the affective evaluation of facialexpressions initially occurs at an implicit level,^54-56 beforeexplicit judgments of the type or intensity of affect, which involve furthercognitive processing.^57,58 Thelevel of processing also appears to have a significant effect on neural activation,as a greater amygdalar signal has been observed with implicit compared withexplicit affective facial expression tasks in most^40,54,56 butnot all^55,59 studies.

Alternating checkerboard stimuli were visually presented in a gradedblock design with 32-second blocks of stimulation alternating periodicallywith 32-second blocks of darkness. Frequency of checkerboard alternation wasvaried between blocks as 2, 4, or 8 Hz. A total of 9 cycles of visual stimulation/darknesswas presented in the course of an experiment lasting 9 minutes 36 secondsin total. The participants were instructed to lie quietly in the imaging unitwith their eyes open throughout the experiment. Three patients were unableto tolerate the extra imaging time required for acquisition of these dataat both time points; subsequent analysis of this experiment, therefore, usesonly data from 16 patients and 16 matched comparison subjects.

Gradient-echo single-shot echoplanar imaging was used to acquire T2-weightedimage volumes on a neuro-optimized 1.5-T IGE LX System (General Electric,Milwaukee, Wis) at the Maudsley Hospital, South London, and Maudsley NHS Trust,London. We acquired 180 volumes for the sad facial affect task and 144 volumesfor the visual auditory stimulation control task. For each volume, 16 noncontiguousaxial planes parallel to the intercommissural plane were collected with thefollowing parameters: repetition time, 2000 milliseconds; echo time, 40 milliseconds;section thickness, 7 mm; section skip, 0.7 mm; and in-plane resolution, 3× 3 mm.To facilitate later coregistration of the fMRI data in standardspace, we also acquired in the first fMRI session a 43-section, high-resolutioninversion recovery echo planar image of the whole brain in the intercommissuralplane with the following parameters: repetition time, 16000 milliseconds;echo time, 73 milliseconds; inversion time,180 milliseconds; and section thickness,3 mm.

Following correction of section-timing differences and head movement–relatedeffects in the fMRI time series,⁶⁰ linear regressionwas used to estimate experimentally induced signal changes. Regression analysismodeled the following 2 mutually orthogonal aspects of brain activation ateach voxel: (1) average facial-processing capacity, ie, the response elicitedby the difference on average between baseline trials and all facial trialstaken together; and (2) facial-processing dynamic range or load response,ie, the response elicited by the difference between facial trials presentedat low, medium, and high intensities of sadness or levels of negative affectiveload. Before model fitting, each column of the regression matrix was convolvedwith a pair of Poisson kernels (λ = 4 or 8 seconds) to model locallyvariable hemodynamic response functions. Statistic maps or t maps representing each of these 2 standardized effects for each individualat each imaging session were registered in the standard space of Talairachand Tournoux⁶¹ by means of an affine transformationto a template image.⁶²

Factorial effects of interest were identified in a second stage of analysis;methodological detail and validation have been provided by Suckling and Bullmore.⁶³ A 2 × 2 analysis of variance (ANOVA) modelwas specified, including a main effect of time (weeks 0 vs 8), a main effectof group (depressed patients vs healthy comparison subjects), and a group× time interaction. This ANOVA model was fitted at all intracerebralvoxels in standard space (n = 76 at each voxel), and a set of 3 F maps wasestimated, 1 map for each factorial F statistic.

To identify brain regions associated with symptomatic response, thechange in HRSD symptom score for each patient during 8 weeks of treatmentwas regressed on the change in the affective load response estimated duringthe same period by subtracting the t map for polynomialload response at week 0 from the corresponding map at week 8.

The statistical significance of these (ANOVA and regression) effectswas decided by means of a cluster-level permutation test that involved applyinga preliminary probability threshold (P<.05) tothe corresponding voxel statistic maps and setting all subthreshold voxelsto 0, thus creating a set of suprathreshold voxel clusters that were spatiallycontiguous in 3 dimensions. The sum of the suprathreshold voxel statistics,or cluster mass M, was tested by means of a permutation test⁶⁴ withclusterwise probability of a type I error of P<.005.At this size of test, and over the search volume of clusters tested (typicallyin the range 100-1000), we expect less than 1 false-positive cluster per map.

Further methodological details and software for nonparametric analysisof factorial designs can be accessed at http://www-bmu.psychiatry.cam.ac.uk/BAMM (accessed July 5, 2004).

Anatomical aspects of the fMRI results are summarized below and in Figure 2, Figure 3, and Figure 4.Further details, including tables of Talairach coordinates and whole-brainANOVA maps, are available as supplemental material at http://www-bmu.psychiatry.cam.ac.uk/DATA.

Do patients with depression differ from comparison subjects in activationof negative affect–processing systems?

There was increased capacity in the patients compared with the healthycomparison subjects in the following regions of the left brain: hippocampusextending to amygdala and parahippocampal gyrus, hypothalamus, ventral striatum(putamen/globus pallidus), insula, caudate nucleus, thalamus, dorsal cingulategyrus, and inferior parietal cortex (Figure2; Table 2).

There was increased dynamic range in the patients compared with thehealthy comparison subjects in the bilateral cerebellum and anterior cingulategyrus extending bilaterally to rostral prefrontal cortex (Table 3).

Do patients with depression show antidepressant treatment–relatedchanges in activation of negative affect–processing systems?

For capacity of activation, there was a significant interaction betweengroup and time in the following regions of the left brain: the amygdala, ventralstriatum (putamen/globus pallidus), insula, caudate nucleus, thalamus, anterior,dorsal and posterior cingulate cortex, precentral gyrus (approximate Brodmannarea [BA] 4) extending to the lateral premotor cortex, postcentral gyrus,and inferior parietal lobule. There was also a significant interaction inthe right ventral striatum and thalamus and right inferior parietal lobule(Figure 2 and Figure 3; Table 4).

Post hoc analysis showed that amplitude of response to sad faces wassignificantly increased in patients compared with healthy volunteers at baseline(independent samples t₃₆ = −4.32; P<.001) and reduced significantly in patients duringthe course of 8 weeks of treatment (repeated-samples t₁₈ = 4.75; P<.001).

For dynamic range of activation, there was a significant interactionin the following left brain regions: the inferior and middle frontal gyri(BAs 44 and 9), postcentral gyrus, and putamen/globus pallidus (Table 5).

Post hoc analysis showed that the dynamic range or affective load responsein these regions was reduced in patients compared with healthy comparisonsubjects at baseline (independent-samples t₃₆ = 3.17; P = .003) and increased significantlyin patients during the course of 8 weeks of antidepressant treatment (repeated-samples t₁₈ = −3.16; P =.005).

The effect of group on facial-processing capacity in limbic, subcortical,and neocortical regions was positively correlated with group effects on thedynamic range of response in anterior cingulate and prefrontal cortex (r = 0.26; n = 76; P = .02). Thus,depression tended to enhance both measures of activity in these systems.

There was also a strong negative correlation between the group ×time effect on facial-processing capacity in limbic and subcortical regionsand the group × time effect on dynamic range of response in frontaland striatal regions (r = −0.36; n = 76; P = .001) (Figure 3A).In other words, as limbic and subcortical activation by sad faces was reducedon average by antidepressant treatment, so the dynamic range of neocortical(frontal) regions was increased. This coupling between complementary treatmenteffects occurred because the prefrontal cortical load-response curve at baselinewas flat at the high level of activation (response ceiling). As sad facialaffect–processing capacity was generally reduced by antidepressant treatmentexposure, prefrontal activation by low-intensity faces was selectively reduced,and the gradient of the load-response curve became correspondingly steeper.

There was a significant association between the changes in HRSD scoresand the dynamic range of activation in the following brain regions: the bilateral(right>left) pregenual anterior cingulate cortex (BAs 24 and 32), right ventralstriatum, and bilateral (left>right) cerebellum. This association was suchthat patients who showed the greatest reduction in depressive symptoms aftertreatment also tended to show the greatest reduction in dynamic range in thecingulate and cerebellar regions (Figure 4).

There was a reduction of capacity over time in the following regionsof the left brain: the hippocampus extending to the amygdala and parahippocampalgyrus, ventral striatum extending dorsally to the head of the caudate nucleus,and ventral occipital cortex (fusiform and lingual gyri) extending anteriorlyto the inferior temporal cortex (BA 37) and superiorly to the inferior parietalcortex (Figure 2; Table 6). There was a reduction of dynamic range over time in thecerebellum, ventral occipital cortex (fusiform and lingual gyri), posteriorcingulate gyrus, thalamus, and left inferior parietal cortex (Table 7). There was a significant positive correlation between theeffect of time on facial-processing capacity in the medial temporal and subcorticalregions and the effect of time on affective load response in the cerebellum,thalamus, and ventral occipital and posterior cingulate cortices (r = 0.34; n = 76; P = .003).

There was a main effect of group on latency (ANOVA; F_1,34 =5.1; P = .03); patients with depression were slowerto respond on average over all trials in both sessions. There were no othersignificant effects of group or group × time on latency or accuracyof explicit sex recognition. There was a significant main effect of affectiveintensity or load (repeated-measures ANOVA; F_2,33 = 3.9; P = .03); participants in both groups were slower to respondto facial stimuli depicting more intense degrees of sadness (Table 8).

A region of occipital (calcarine) cortex activated by photic stimulationon average across all checkerboard frequencies in the comparison subjectsat baseline was defined as a region of interest. There were no significanteffects of group (F_1,30 = 0.11; P = .74)or group × time (F_1,30 = 1.05; P =.31) on capacity or dynamic range of response to photic stimulation in thisregion of occipital cortex.

Response to sad faces in the patients was abnormally exaggerated atbaseline but significantly reduced during the course of treatment. The amygdalahas been identified as a fundamental component of neuropsychological modelsof depression.^65,66 Of more directrelevance to our results, Sheline et al¹⁷ reportedgreater activation of the left amygdala by fearful faces in acutely depressedpatients that was reduced after 8 weeks of treatment with sertraline hydrochloride,100 mg/d. Similarly, Davidson et al¹⁸ reportedgreater bilateral amygdalar activation by aversive visual stimuli in patientswith depression at baseline that was significantly reduced after 8 weeks oftreatment with venlafaxine hydrochloride, 225 mg/d. The Talairach coordinatesfor left amygdalar treatment effects in their report,¹⁸ namelyx = −18 mm, y = −6 mm, z = −10 mm, are comparable to thefollowing coordinates for a left amygdalar effect reported herein (Table 4): −12, −5, and −8mm and −11, −7, and −12 mm. Thus, we conclude that thereis now convergent evidence from 3 independent fMRI studies that antidepressantdrugs generically act to reduce abnormal amygdalar responsivity to negativelyaffective faces in patients with depression.

We also found evidence in patients with depression for reduction ofan initially exaggerated response to sad faces in the ventral striatum andthalamus. Decreased activity in the basal ganglia and thalamus of depressedpatients after treatment has been noted in other studies with the serotoninreuptake inhibitors sertraline,⁴⁵ fluoxetine,⁶ and paroxetine,^7,48 butsome earlier studies did not report a treatment effect.^5,67-69 Thismay be a diagnostically nonspecific effect of serotonin reuptake inhibitors,as similar reductions in caudate activity have been observed in obsessive-compulsivedisorder after treatment with paroxetine and may be mediated by posttreatmentreductions in serotonin transporter density in the striatum.⁷⁰

These treatment effects on sad facial affect–processing capacityin limbic and subcortical regions were associated with comparable changesin frontal and parietal neocortical regions that have also been repeatedlyimplicated in the functional neuroanatomy of depression.⁶ Mayberget al⁴⁶ have proposed that experimental orpathological changes in mood state are associated with reciprocal changesin activity of limbic-subcortical systems and a frontoparietal attentionalsystem,⁷¹ ie, limbic-subcortical regions aremore activated and frontoparietal circuits are less activated by sadness orlowering mood. Some of our results are consistent with this formulation. Asoverall capacity for activation by sad faces was decreased by treatment inlimbic-subcortical systems, there was a reciprocal increase in differentialactivation of prefrontal cortex by the highest levels of affective load (mostintensely sad faces). This was a significant (negative) correlation that canbe explained in terms of antidepressant treatment exposure reducing prefrontalactivation selectively at the lowest levels of affective load and thereforeincreasing the dynamic range available for differential activation by thehighest levels of affective load.

When dynamic range was considered, the most interesting result was foundin the subgenual anterior cingulate cortex. Increased dynamic range in acutelydepressed, medication-free patients at baseline was associated with a greatersymptom reduction at 8 weeks. Increased activity in the rostral pregenualanterior cingulate cortex has been observed in depressed patients who subsequentlyrespond to treatments, including antidepressant drugs,^{6,7,18,45,46,48} themood-stabilizing agent carbamazepine,⁴⁷ orsleep deprivation.⁷² Activation of the rostralanterior cingulate cortex has been elicited by a number of affectively challengingparadigms, including induction of anxiety,^73-75 sadness,^15,72,76 anger,³⁸ fear,³⁶ and affective "unpleasantness" associated with pain.⁷⁷ Mayberg et al⁶ havesuggested that the rostral anterior cingulate facilitates interactions betweendorsal cortical and ventral paralimbic systems and has a significant rolein the regulation of mood and cognitive and somatic functions. Our findingfurther implicates pregenual anterior cingulate function as a surrogate markerfor symptomatic response in depression and adds to the growing body of supportfor baseline neural activity in this region as an important predictor of antidepressanttreatment response.⁷⁸

Two additional regions showed a relationship between change in the HRSDscores and dynamic range changes, ie, the ventral striatum and the cerebellum.The ventral striatum also showed increased activation capacity at baselinein depressed patients that decreased after treatment. The striatum is typicallyassociated with reward⁷⁹ and has been elicitedby most neuroimaging mood-induction studies of happiness, as reviewed by Phanet al.⁸⁰ However, the striatum has also beenrecruited in studies of disgust⁸⁰ and in responseto aversive stimuli.^81,82 It hasbeen proposed that engagement of the striatum occurs with the initiation ofaction in response to a relevant stimulus, rather than to reward itself.⁷⁹ The overall reduction in activation of the striatumin patients likely reflects an effect of fluoxetine treatment.⁶

The cerebellum is classically associated with motor control. However,its activation is frequently reported in neuroimaging studies of mood induction,^44,80 and degenerative cerebellar diseasesare associated with mood disorders and personality changes.⁸³ Thesedata are compatible with our observation and suggest the cerebellum may bea major component of dysfunctional circuits in mood disorders.⁸⁴

Depressed patients tended to have greater capacity and greater dynamicrange of response in sad facial affect–processing systems. In futurestudies, it will be interesting to test the hypothesis that recognition biasfor negative stimuli is directly related to enhanced activation of limbicand subcortical brain regions in depressed subjects.

Attenuation of medial temporal and ventral occipital fMRI signals inresponse to repeated presentation of affectively valent stimuli has been previouslyreported.³⁵ Our results confirm that regionsof the left medial temporal lobe, consisting of the hippocampus and parahippocampalgyrus, and of the left ventral occipital cortex are sensitive to repeatedpresentation of negative affect in faces for a longer period than previouslyinvestigated. The implicit nature of our affect-processing task precludedinvestigation of the subjective correlates of these broadly habituating effects,but it is interesting to speculate that longer-term physiological attenuationof response to affectively valent stimulation may be correlated with subjectiveblunting of response to repeated presentation of emotive material.

We have used a comparison group of untreated, healthy volunteers tocontrol for the major effects of task repetition over time. Consequently,we have interpreted the group × time interaction as indicative of antidepressanttreatment. Although we regard this interpretation as tenable, such an interactioncould also occur because of nontherapeutic, trait differences between thegroups. The ideal control would have been a group of untreated (or placebo-treated)patients with depression. We considered this possibility initially, but ruledit out on ethical grounds because it would entail withholding an effectivetreatment from symptomatic patients for the duration of the study.

We chose to use an implicit or incidental facial affect–processingtask in which participants were not explicitly instructed to pay attentionto the sadness of stimuli. However, the behavioral data recorded during scanningindicates that the degree of sadness modified the participants' responsesto the explicit sex discrimination task, suggesting that awareness of affectmay have been conscious or at least able to interfere with simultaneous performanceof a conscious task.

The biophysical basis for drug-related changes in fMRI signals has notbeen entirely elucidated yet. It is possible that some drug-related effectson blood oxygenation level–dependent (BOLD) signal may be a neuronallynonspecific consequence of altered cerebral hemodynamics.⁸⁵ Arguingagainst this interpretation of our data, we note first that we have describedgraded antidepressant treatment–related BOLD signal changes in anteriorcingulate cortex that are associated with graded symptomatic recovery fromdepression in subjects receiving the same drug dose. Second, there was nosignificant antidepressant drug effect on the BOLD signal in primary visualcortex (as noted also in healthy volunteers⁸⁶),indicating a degree of region and task specificity compatible with a neuronalmechanism for treatment-related signal change.

Correspondence: Edward T. Bullmore, MRCPsych, PhD, University ofCambridge, Brain Mapping Unit, Department of Psychiatry, Addenbrooke's Hospital,Cambridge CB2 2QQ, England (etb23@cam.ac.uk) (URL:http://www-bmu.psychiatry.cam.ac.uk).

Submitted for publication December 2, 2003; final revision receivedMarch 11, 2004; accepted March 16, 2004.

This study was supported by an experimental medicine research grantfrom GlaxoSmithKline, Cambridge, England; by the Wellcome Trust, London, England(Drs Fu and Bullmore); and by a Human Brain Project grant from the NationalInstitute of Biomedical Imaging and Bioengineering and the National Instituteof Mental Health, Rockville, Md, for software development.

We thank the volunteers who participated in this study; Lidia Yaguez,PhD, Michael Dilley, MD, and Sri Kalidindi, MD, for their clinical assistance;the staff of the MRI Unit, Maudsley Hospital, London, for their technicalassistance; and Liqun Wang, PhD, and Alan Bye, PhD, GlaxoSmithKline, for theircontributions to the project.