Advertisement

Reversed Frontotemporal Connectivity During Emotional Face Processing in Remitted Depression

Open AccessPublished:June 08, 2012DOI:https://doi.org/10.1016/j.biopsych.2012.04.031

      Background

      Vulnerability to relapse persists after remission of an acute episode of major depressive disorder. This has been attributed to abnormal biases in the processing of emotional stimuli in limbic circuits. However, neuroimaging studies have not so far revealed consistent evidence of abnormal responses to emotional stimuli in limbic structures, such as the amygdala, in remitted depression. This suggests the problem might lie in the integrated functioning of emotion processing circuits.

      Methods

      We recruited 22 unmedicated patients in remission from major depressive disorder (rMDD) and 21 age-matched healthy control subjects. Functional magnetic resonance imaging was performed during a face emotion processing task. Dynamic causal modeling was used with Bayesian model selection to determine the most likely brain networks and valence-specific modulation of connectivity in healthy control subjects and rMDD.

      Results

      In healthy volunteers, sad faces modulated bi-directional connections between amygdala and orbitofrontal cortex and between fusiform gyrus and orbitofrontal cortex. Happy faces modulated unidirectional connections from fusiform gyrus to orbitofrontal cortex. In rMDD, the opposite pattern was observed, with evidence of happy faces modulating bidirectional frontotemporal connections and sad faces modulating unidirectional fusiform–orbitofrontal connections.

      Conclusions

      Participants with rMDD have abnormal modulation of frontotemporal effective connectivity in response to happy and sad face emotions, despite normal activations within each region. Specifically, processing of mood incongruent happy information was associated with a more richly modulated frontotemporal brain network, whereas mood congruent sad information was associated with less network modulation. This supports a hypothesis of dysfunction within cortico–limbic connections in individuals vulnerable to depression.

      Key Words

      Major depressive disorder (MDD) is characterized by impaired cognitive and emotional processing (
      • Chamberlain S.R.
      • Sahakian B.J.
      The neuropsychology of mood disorders.
      ,
      • Phillips M.L.
      • Drevets W.C.
      • Rauch S.L.
      • Lane R.
      Neurobiology of emotion perception II: Implications for major psychiatric disorders.
      ,
      • Browning M.
      • Holmes E.A.
      • Harmer C.J.
      The modification of attentional bias to emotional information: A review of the techniques, mechanisms, and relevance to emotional disorders.
      ,
      • Elliott R.
      • Zahn R.
      • Deakin J.F.
      • Anderson I.M.
      Affective cognition and its disruption in mood disorders.
      ), including recognition and categorization of face emotion (
      • Gilboa-Schechtman E.
      • Erhard-Weiss D.
      • Jeczemien P.
      Interpersonal deficits meet cognitive biases: Memory for facial expressions in depressed and anxious men and women.
      ,
      • Harmer C.J.
      • O'Sullivan U.
      • Favaron E.
      • Massey-Chase R.
      • Ayres R.
      • Reinecke A.
      • et al.
      Effect of acute antidepressant administration on negative affective bias in depressed patients.
      ,
      • Joormann J.
      • Gotlib I.H.
      Selective attention to emotional faces following recovery from depression.
      ,
      • Anderson I.M.
      • Shippen C.
      • Juhasz G.
      • Chase D.
      • Thomas E.
      • Downey D.
      • et al.
      State-dependent alteration in face emotion recognition in depression.
      ). Emotion categorization studies have usually but not always found reduced discrimination of face emotions in MDD (
      • Anderson I.M.
      • Shippen C.
      • Juhasz G.
      • Chase D.
      • Thomas E.
      • Downey D.
      • et al.
      State-dependent alteration in face emotion recognition in depression.
      ). Some studies have additionally reported biases toward sad and/or away from happy faces. There is also evidence for altered recognition after remission from MDD (
      • Joormann J.
      • Gotlib I.H.
      Selective attention to emotional faces following recovery from depression.
      ,
      • Anderson I.M.
      • Shippen C.
      • Juhasz G.
      • Chase D.
      • Thomas E.
      • Downey D.
      • et al.
      State-dependent alteration in face emotion recognition in depression.
      ,
      • Bhagwagar Z.
      • Cowen P.J.
      • Goodwin G.M.
      • Harmer C.J.
      Normalization of enhanced fear recognition by acute SSRI treatment in subjects with a previous history of depression.
      ), although studies are inconsistent (
      • Elliott R.
      • Zahn R.
      • Deakin J.F.
      • Anderson I.M.
      Affective cognition and its disruption in mood disorders.
      ,
      • Leppanen J.M.
      Emotional information processing in mood disorders: A review of behavioral and neuroimaging findings.
      ,
      • Thomas E.J.
      • Elliott R.
      • McKie S.
      • Arnone D.
      • Downey D.
      • Juhasz G.
      • et al.
      Interaction between a history of depression and rumination on neural response to emotional faces.
      ). Functional magnetic resonance imaging (fMRI) has identified neuronal substrates of face processing abnormalities in MDD, particularly enhanced amygdala responses to sadness (
      • Fu C.H.
      • Williams S.C.
      • Cleare A.J.
      • Brammer M.J.
      • Walsh N.D.
      • Kim J.
      • et al.
      Attenuation of the neural response to sad faces in major depression by antidepressant treatment: A prospective, event-related functional magnetic resonance imaging study.
      ,
      • Surguladze S.
      • Brammer M.J.
      • Keedwell P.
      • Giampietro V.
      • Young A.W.
      • Travis M.J.
      • et al.
      A differential pattern of neural response toward sad versus happy facial expressions in major depressive disorder.
      ,
      • Victor T.A.
      • Furey M.L.
      • Fromm S.J.
      • Ohman A.
      • Drevets W.C.
      Relationship between amygdala responses to masked faces and mood state and treatment in major depressive disorder.
      ). Some evidence suggests enhanced responses to sad faces persisting into remission (
      • Victor T.A.
      • Furey M.L.
      • Fromm S.J.
      • Ohman A.
      • Drevets W.C.
      Relationship between amygdala responses to masked faces and mood state and treatment in major depressive disorder.
      ), although most studies have shown that abnormalities are state-dependent (
      • Harmer C.J.
      • O'Sullivan U.
      • Favaron E.
      • Massey-Chase R.
      • Ayres R.
      • Reinecke A.
      • et al.
      Effect of acute antidepressant administration on negative affective bias in depressed patients.
      ,
      • Thomas E.J.
      • Elliott R.
      • McKie S.
      • Arnone D.
      • Downey D.
      • Juhasz G.
      • et al.
      Interaction between a history of depression and rumination on neural response to emotional faces.
      ,
      • Fu C.H.
      • Williams S.C.
      • Cleare A.J.
      • Brammer M.J.
      • Walsh N.D.
      • Kim J.
      • et al.
      Attenuation of the neural response to sad faces in major depression by antidepressant treatment: A prospective, event-related functional magnetic resonance imaging study.
      ,
      • Sheline Y.I.
      • Barch D.M.
      • Donnelly J.M.
      • Ollinger J.M.
      • Snyder A.Z.
      • Mintun M.A.
      Increased amygdala response to masked emotional faces in depressed subjects resolves with antidepressant treatment: An fMRI study.
      ).
      Although there is an established literature on abnormal function of specific brain regions during face emotion processing in MDD, there has been relatively little direct exploration of how the disorder affects connectivity between brain regions. Many neurobiological models of depression focus on network dysfunction, particularly dysfunction of a cortical–limbic “mood-regulating circuit” (
      • Mayberg H.S.
      Limbic-cortical dysregulation: A proposed model of depression.
      ,
      • Seminowicz D.A.
      • Mayberg H.S.
      • McIntosh A.R.
      • Goldapple K.
      • Kennedy S.
      • Segal Z.
      • Rafi-Tari S.
      Limbic-frontal circuitry in major depression: A path modeling metanalysis.
      ,
      • Anand A.
      • Li Y.
      • Wang Y.
      • Wu J.
      • Gao S.
      • Bukhari L.
      • et al.
      Activity and connectivity of brain mood regulating circuit in depression: A functional magnetic resonance study.
      ). In a specific model of emotion perception in MDD, Phillips et al. (
      • Phillips M.L.
      • Drevets W.C.
      • Rauch S.L.
      • Lane R.
      Neurobiology of emotion perception II: Implications for major psychiatric disorders.
      ) argued that limbic overactivity during initial evaluation, combined with failure of cortical control, causes a bias toward processing negative information.
      Recent research has investigated resting state functional connectivity in depression (
      • Zhang J.
      • Wang J.
      • Wu Q.
      • Kuang W.
      • Huang X.
      • He Y.
      • Gong Q.
      Disrupted brain connectivity networks in drug-naive, first-episode major depressive disorder.
      ,
      • Zhu X.
      • Wang X.
      • Xiao J.
      • Liao J.
      • Zhong M.
      • Wang W.
      • Yao S.
      Evidence of a dissociation pattern in resting-state default mode network connectivity in first-episode, treatment-naive major depression patients.
      ,
      • Greicius M.D.
      • Flores B.H.
      • Menon V.
      • Glover G.H.
      • Solvason H.B.
      • Kenna H.
      • et al.
      Resting-state functional connectivity in major depression: Abnormally increased contributions from subgenual cingulate cortex and thalamus.
      ) and reported abnormal connectivity in regions of a mood-regulating circuit, which might relate to abnormal limbic–prefrontal white matter connectivity observed with diffusion tensor imaging (
      • Versace A.
      • Thompson W.K.
      • Zhou D.
      • Almeida J.R.
      • Hassel S.
      • Klein C.R.
      • et al.
      Abnormal left and right amygdala-orbitofrontal cortical functional connectivity to emotional faces: State versus trait vulnerability markers of depression in bipolar disorder.
      ). However, resting state studies do not optimally test network dysfunction models (
      • Phillips M.L.
      • Drevets W.C.
      • Rauch S.L.
      • Lane R.
      Neurobiology of emotion perception II: Implications for major psychiatric disorders.
      ), which propose specific abnormalities during processing of emotional stimuli rather than at rest. Connectivity analyses of fMRI data can provide explicit tests of network interactions in response to emotional challenges. Functional connectivity, assessed by psychophysiological interactions (PPIs) and related methods, explores context-dependent correlations between brain regions. This technique has implicated abnormal connectivity during emotional processing in depression (
      • Ruhe H.G.
      • Booij J.
      • Veltman D.J.
      • Michel M.C.
      • Schene A.H.
      Successful pharmacologic treatment of major depressive disorder attenuates amygdala activation to negative facial expressions: A functional magnetic resonance imaging study.
      ,
      • Hamilton J.P.
      • Gotlib I.H.
      Neural substrates of increased memory sensitivity for negative stimuli in major depression.
      ,
      • Frodl T.
      • Bokde A.L.
      • Scheuerecker J.
      • Lisiecka D.
      • Schoepf V.
      • Hampel H.
      • et al.
      Functional connectivity bias of the orbitofrontal cortex in drug-free patients with major depression.
      ,
      • Erk S.
      • Mikschl A.
      • Stier S.
      • Ciaramidaro A.
      • Gapp V.
      • Weber B.
      • Walter H.
      Acute and sustained effects of cognitive emotion regulation in major depression.
      ,
      • Smith R.
      • Fadok R.A.
      • Purcell M.
      • Liu S.
      • Stonnington C.
      • Spetzler R.F.
      • Baxter L.C.
      Localizing sadness activation within the subgenual cingulate in individuals: A novel functional MRI paradigm for detecting individual differences in the neural circuitry underlying depression.
      ) and bipolar disorder (
      • Versace A.
      • Thompson W.K.
      • Zhou D.
      • Almeida J.R.
      • Hassel S.
      • Klein C.R.
      • et al.
      Abnormal left and right amygdala-orbitofrontal cortical functional connectivity to emotional faces: State versus trait vulnerability markers of depression in bipolar disorder.
      ), including abnormalities of amygdala–prefrontal connectivity in remitted patients. Cremers et al. (
      • Cremers H.R.
      • Demenescu L.R.
      • Aleman A.
      • Renken R.
      • van Tol M.J.
      • van der Wee N.J.
      • et al.
      Neuroticism modulates amygdala-prefrontal connectivity in response to negative emotional facial expressions.
      ) suggest that amygdala–prefrontal connectivity during face processing is influenced by neuroticism, a trait associated with depression vulnerability.
      The PPI approaches are data-driven and give no information about direction or causality. To determine causal influences among brain regions and more directly assess theoretical anatomical models requires testing of “effective connectivity,” with models that embody causal connections such as structural equation modeling (
      • Penny W.D.
      • Stephan K.E.
      • Mechelli A.
      • Friston K.J.
      Modelling functional integration: A comparison of structural equation and dynamic causal models.
      ) or generative models such as dynamic causal modeling (DCM) (
      • Stephan K.E.
      • Penny W.D.
      • Moran R.J.
      • den Ouden H.E.
      • Daunizeau J.
      • Friston K.J.
      Ten simple rules for dynamic causal modeling.
      ,
      • Friston K.J.
      • Harrison L.
      • Penny W.
      Dynamic causal modelling.
      ). Studies using structural equation modeling have reported abnormalities in limbic–prefrontal networks in MDD during emotional face processing (
      • Carballedo A.
      • Scheuerecker J.
      • Meisenzahl E.
      • Schoepf V.
      • Bokde A.
      • Moller H.J.
      • et al.
      Functional connectivity of emotional processing in depression.
      ). Structural equation modeling allows a prespecified model to be tested; however, a major advantage of DCM over other connectivity approaches is the evidence-based model comparison procedure. This supports inferences of directionality and allows comparison of models to infer changes in network organization, over and above changes in connection strength within a standard network. Dynamic causal modeling employs an optimized neurovascular forward model enabling inferences to be made at the neuronal level of intrinsic and modulatory connections among multiple brain regions. A Bayesian model selection procedure allows direct comparison of different brain network models, determining which model of connectivity is most likely, given the data.
      Dynamic causal modeling analyses have been carried out previously for face-processing tasks in healthy volunteers (
      • de Marco G.
      • de Bonis M.
      • Vrignaud P.
      • Henry-Feugeas M.C.
      • Peretti I.
      Changes in effective connectivity during incidental and intentional perception of fearful faces.
      ,
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ,
      • Ishai A.
      Let's face it: It's a cortical network.
      ,
      • Dima D.
      • Stephan K.E.
      • Roiser J.P.
      • Friston K.J.
      • Frangou S.
      Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
      ), showing emotional modulation of effective connectivity between fusiform gyrus (FG), amygdala, and orbitofrontal cortex (OFC). There have been relatively few studies using DCM in MDD, although recent studies with very simple models (typically modeling intrinsic connectivity between two nodes) have reported abnormal effective connectivity to faces in bipolar disorder and distinct abnormalities to happy faces in MDD (
      • Almeida J.R.
      • Versace A.
      • Hassel S.
      • Kupfer D.J.
      • Phillips M.L.
      Elevated amygdala activity to sad facial expressions: A state marker of bipolar but not unipolar depression.
      ). To our knowledge there have been no attempts to assess altered effective connectivity to emotional stimuli, in remitted major depressive disorder (rMDD).
      The goal of the present study was to use DCM to explore abnormalities of connectivity during processing of happy and sad faces in rMDD. Rowe et al. (
      • Rowe J.B.
      • Hughes L.E.
      • Barker R.A.
      • Owen A.M.
      Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
      ) recently demonstrated that DCM is sufficiently robust and sensitive to study clinical populations. Indeed, a connectivity approach can be more sensitive to neuropsychiatric disorders than classical imaging analysis of regional activations (
      • Rowe J.B.
      Connectivity analysis is essential to understand neurological disorders.
      ). We adopted the Bayesian model selection approach of Rowe et al. (
      • Rowe J.B.
      • Hughes L.E.
      • Barker R.A.
      • Owen A.M.
      Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
      ) and predicted that happy and sad emotion would differentially modulate effective connectivity associated with face processing in rMDD compared with healthy control subjects.

      Methods and Materials

      Participants

      All participants were right-handed with normal or corrected-to-normal vision and no contraindications to fMRI. Volunteers with current or past history of neurological disorder, substance dependence, or Axis 1 psychiatric disorder other than MDD or anxiety disorders were excluded, as were people taking current medications.

      Healthy Control Subjects

      Twenty-nine healthy control volunteers were recruited, of whom 21 were included in DCM analysis (see Results). On the Mini-International Neuropsychiatric Interview (M.I.N.I.)-screen for Structured Clinical Interview for DSM interview (
      • Sheehan D.V.
      • Lecrubier Y.
      • Sheehan K.H.
      • Amorim P.
      • Janavs J.
      • Weiller E.
      • et al.
      The Mini-International Neuropsychiatric Interview (M.I.N.I.): The development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10.
      ), these volunteers had no personal or family history of psychiatric illness (Table 1).
      Table 1Participant Characteristics
      Healthy Control (n = 21)rMDD (n = 22)
      Male:Female7:146:16
      Mean Age (SEM)31.1 (9.97)33.73 (10.69)
      MADRS (SEM).92 (1.44)2.31 (3.24)
      Past Episodes3.13 (2.6)
      MADRS, Montgomery-Åsberg Depression Rating Scale; rMDD, remitted major depressive disorder.

      rMDD Subjects

      Thirty remitted depressed volunteers were also recruited, of whom 22 were included in DCM analysis (see Results). These subjects met criteria for major depression in full remission on Structured Clinical Interview for DSM interview (
      • First M.B.
      • Gibbon M.
      • Williams J.B.W.
      Structured Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, Patient Edition.
      ). For inclusion, they were required to have been remitted for at least 3 months, with current scores of <12 on the Montgomery-Åsberg Depression Rating Scale (MADRS) and to be medication-free (Table 1). A t test revealed that, although within the normal range, rMDD volunteers had significantly higher MADRS scores than healthy control subjects (t = −2.15, p = .04). A history of anxiety disorder was not an exclusion criterion; however, only 4 of 22 subjects in the DCM analysis had history of clinically significant anxiety, and none had current symptoms.

      Emotional Face Processing Task

      A series of faces was presented, each displayed for 3 sec with a 500-msec gap. There were four different emotion conditions—neutral (N), happy (H), sad (S), and fear (F)—and a rest condition (R), which was fixation on a central cross. The emotions of interest in this study were happy and sad, because abnormalities in processing these mood-relevant emotions have been most consistently associated with MDD. Each face block lasted 21 sec and consisted of six faces (three male; three female). Standard face stimuli of Ekman and Friesen were used, in conjunction with a morphing procedure (
      • Young A.W.
      • Rowland D.
      • Calder A.J.
      • Etcoff N.L.
      • Seth A.
      • Perrett D.I.
      Facial expression megamix: Tests of dimensional and category accounts of emotion recognition.
      ) such that faces displayed emotions at 80% intensity. There were 22 blocks presented in an NHNSNFNRSNHNFNRFNSNHNR order; total duration was 7 min 42 sec. Volunteers responded with a button press to each face indicating whether it was male or female. We present results for sad and happy emotions, in line with fMRI literature emphasizing the particular significance of these mood-related emotions to depressive disorders.

      Imaging

      Images were acquired on a 1.5T Philips Intera scanner (Eindhoven, the Netherlands) with a single-shot echo-planar sequence. Two hundred eighteen volumes were acquired, each comprising 29 contiguous axial slices (repetition time/echo time = 2100/40 msec, 3.5 × 3.5 mm in-plane resolution, slice thickness 4.5 mm, .5 mm slice gap). A T1-weighted structural scan (256 × 256 matrix, repetition time = 8.99 msec, echo time = 4.2 msec, 160 axial slices, voxel size .875 × .875 × 1 mm) was acquired for each participant to be used in spatial normalization.

      Analysis of Region-Specific Responses

      Pre-processing was carried out with SPM8 (http://www.fil.ion.ucl.ac.uk/spm). All images were realigned to the first to correct for movement, spatially normalized to standard space (Montreal Neurological Institute) and smoothed with a Gaussian kernel of full width at half maximum 7 × 7 × 10 mm. A high-pass filter of twice the maximum stimulus repetition time was applied.
      To determine coordinates of regional maxima for DCM analysis, random effects analysis was carried out to identify regional changes in blood oxygen level–dependent response to all faces compared with rest across the whole subject group. We performed a region of interest (ROI) analysis in four regions comprising a network of face-processing regions where effective connectivity is modulated by emotional valence (
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ,
      • Dima D.
      • Stephan K.E.
      • Roiser J.P.
      • Friston K.J.
      • Frangou S.
      Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
      ): primary visual cortex (V1), FG, amygdala, and OFC. Maxima were identified in each ROI in the random effects analysis. For each individual, local maxima within 14 mm of the maximum group activation were found, and data were extracted from a 6-mm radius sphere centered at these local maxima. Regional responses at these maxima are reported in Table 2.
      Table 2Coordinates of Activations Found for DCM Analysis
      RegionBAxyzCluster Size (Voxels)t Score
      Left V117−14−98086512.95
      Right V11718−95086511.45
      Left Fusiform−35−77−208657.64
      Right Fusiform42−60−158658.02
      Left Amygdala−18−7−151424.03
      Right Amygdala25−4−151134.67
      Left OFC47−4221−15331.97
      Right OFC472832−15462.06
      Threshold set at p < .05 uncorrected.
      BA, Brodmann area; DCM, dynamic causal modeling; OFC, orbitofrontal cortex; V1, visual cortex.

      DCM of Effective Connectivity

      Dynamic causal modeling was used to examine the effect of rMDD on effective connectivity with a set of models specified in SPM8, following methods recommended by Stephan et al. (
      • Stephan K.E.
      • Penny W.D.
      • Moran R.J.
      • den Ouden H.E.
      • Daunizeau J.
      • Friston K.J.
      Ten simple rules for dynamic causal modeling.
      ,
      • Stephan K.E.
      • Penny W.D.
      • Daunizeau J.
      • Moran R.J.
      • Friston K.J.
      Bayesian model selection for group studies.
      ), which have reliably detected between-group differences in connectivity (
      • Rowe J.B.
      • Hughes L.E.
      • Barker R.A.
      • Owen A.M.
      Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
      ). The analysis schema is given in Supplement 1.
      The anatomical network model proposed by Fairhall and Ishai (
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ) includes V1, superior temporal sulcus, FG, amygdala, inferior frontal gyrus, and OFC. However, connections between V1, FG, amygdala, and OFC were particularly sensitive to modulation by emotion, and therefore we specifically include only these areas; see also Dima et al. (
      • Dima D.
      • Stephan K.E.
      • Roiser J.P.
      • Friston K.J.
      • Frangou S.
      Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
      ). Left hemisphere responses were less extensive and less consistent across subjects, as has previously been observed (
      • Ishai A.
      • Schmidt C.F.
      • Boesiger P.
      Face perception is mediated by a distributed cortical network.
      ,
      • Ishai A.
      • Ungerleider L.G.
      • Martin A.
      • Haxby J.V.
      The representation of objects in the human occipital and temporal cortex.
      ). Reliable DCM analysis requires exclusion of subjects who do not show response in all network foci, and we therefore performed DCM only on the right hemisphere network. We have previously demonstrated with the model used here that adequate power can be achieved to detect subtle group differences (
      • Goulden N.
      • McKie S.
      • Suckling J.
      • Williams S.R.
      • Anderson I.M.
      • Deakin J.F.
      • Elliott R.
      A comparison of permutation and parametric testing for between group effective connectivity differences using DCM.
      ).
      A three-stage approach to model selection was used. First, connections between the ROIs were specified as intrinsic connections with all faces compared with rest modeled as input to V1. We tested seven structurally distinct models (Supplement 1). Bayesian Model Selection within each group clearly identified the fully interconnected model as most likely, with feedforward and feedback connections (Figure 1A).
      Figure thumbnail gr1
      Figure 1The models tested with Bayesian model selection. (A) Fully connected model of intrinsic connectivity with feedforward and feedback connections for each part of the network; (B) the patterns of modulatory influences of emotion on connectivity among fusiform gyrus (FG), amygdala (Amyg), and lateral orbitofrontal cortex (LOFC). These 21 models belong to one of seven model “families” (see Methods). V1, visual cortex.
      In the second step, the fully connected anatomical model, with intrinsic connectivity as in the preceding text, was used to make 21 models. Each of these models was subject to distinct profiles of modulation by emotion (Figure 1B). The 21 models were fitted with data from each subject, generating the model log-evidences and posterior probabilities—parameters that provide an index of the accuracy of the model, adjusting for model complexity and dependencies among parameters.
      The models were then partitioned into different families, according to shared model features. This facilitates optimum model selection and interpretation. Details are given in Supplement 1. Having identified the optimum model family, we then compared individual models within those families, again with Bayesian model selection (
      • Stephan K.E.
      • Penny W.D.
      • Moran R.J.
      • den Ouden H.E.
      • Daunizeau J.
      • Friston K.J.
      Ten simple rules for dynamic causal modeling.
      ,
      • Rowe J.B.
      • Hughes L.E.
      • Barker R.A.
      • Owen A.M.
      Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
      ,
      • Stephan K.E.
      • Penny W.D.
      • Daunizeau J.
      • Moran R.J.
      • Friston K.J.
      Bayesian model selection for group studies.
      ,
      • Penny W.D.
      • Stephan K.E.
      • Daunizeau J.
      • Rosa M.J.
      • Friston K.J.
      • Schofield T.M.
      • Leff A.P.
      Comparing families of dynamic causal models.
      ).

      Functional Connectivity Analysis with PPI

      The DCM with Bayesian model selection approach allows us to test our hypotheses that the best-fitting models for happy and sad faces differ between patients and control subjects but rests on several assumptions and limitations in model design and fitting. Following the approach of Passamonti et al. (
      • Passamonti L.
      • Crockett M.J.
      • Apergis-Schoute A.M.
      • Clark L.
      • Rowe J.B.
      • Calder A.J.
      • Robbins T.W.
      Effects of acute tryptophan depletion on prefrontal-amygdala connectivity while viewing facial signals of aggression.
      ,
      • Passamonti L.
      • Rowe J.B.
      • Ewbank M.
      • Hampshire A.
      • Keane J.
      • Calder A.J.
      Connectivity from the ventral anterior cingulate to the amygdala is modulated by appetitive motivation in response to facial signals of aggression.
      ), DCM analysis was therefore supplemented by an analysis of PPIs with general linear models, with fewer anatomical constraints and no inverse modeling of neurovascular coupling but without model comparison or inferences of directionality. Separate PPIs were carried out for happy versus neutral and sad versus neutral with the right amygdala as a seed region. Data from the amygdala were extracted from an 8-mm sphere, constructed around the focus used in the DCM analysis (25, −4, −15). A time series was calculated for each participant with the first eigenvariate from the time series of all voxels within the sphere. The PPI regressor was calculated as the product of the right amygdala neuronal time series and a vector coding for the task comparison (happy–neutral or sad–neutral). Subject-specific PPI contrast images were entered into second-level analysis, identifying brain areas where the change in emotion-related connectivity with the amygdala differed between groups. We also examined whether in the rMDD group, there were correlations between connectivity and MADRS depression ratings. This was a post hoc ROI analysis to supplement the DCM findings, and we therefore used a composite ROI comprising FG and OFC, defined anatomically, thresholded at p < .05 corrected.

      Results

      fMRI Activations

      When responses to emotional faces were contrasted with rest in all subjects, right-sided maxima were observed in the pre-specified ROIs: V1 (18, 95, 0), FG (42, 60, 15), amygdala (25, 4, 15), and OFC (28, 32, 15) (Table 2) (note: left hemisphere maxima also reported for completeness). No other regions of significant activation were observed, and there were no differences between the groups at p < .001 uncorrected.
      For DCM analysis, we selected subjects who exhibited activity (p < .05 uncorrected) within 14 mm of all regional maxima in the t contrast of emotional faces compared with rest. For the right hemisphere, significant response within 14 mm of all maxima was observed in 21 of 29 healthy control subjects and 22 of 30 rMDD volunteers, and only these participants were included in DCM. Results of the standard subtraction analysis in the full sample are reported and discussed elsewhere (
      • Thomas E.J.
      • Elliott R.
      • McKie S.
      • Arnone D.
      • Downey D.
      • Juhasz G.
      • et al.
      Interaction between a history of depression and rumination on neural response to emotional faces.
      ).

      DCM Analysis

      Table 3 and Figure 2 show the results of Bayesian model selection for each group, indicating very significant differences in model evidences (
      • Kass R.E.
      • Raftery A.E.
      Bayes Factors.
      ), with opposite directions of effect in the two groups.
      Table 3Bayesian Model Selection within Best-Fitting Families of Models
      Healthy ControlRemitted Depressed
      Most Likely FamilyMost Likely ModelPosterior ProbabilityMost Likely FamilyMost Likely ModelPosterior Probability
      Happy15.86721.98
      Sad721.81121
      Figure thumbnail gr2
      Figure 2Results of Bayesian Model Selection for identification of the most likely family (F1–F7) and, within the winning family, the most likely network models (Models 1–3 in Family F7, Models 1–6 in Family F1). Results are shown separately for: (A) Healthy Control Group Happy Condition; families and models; (B) Healthy Control Group Sad Condition; families and models; (C) Remitted Depressed Group Happy Condition; families and models; and (D) Remitted Depressed Group Sad Condition; families and models. For each of A–D, the differences in log-evidences between the first- and second-place models within Families 1 and 7 are highly significant (very strong evidence for a difference), with standard Bayesian thresholds (
      • Kass R.E.
      • Raftery A.E.
      Bayes Factors.
      ), as can most easily be seen from the differences in the posterior probabilities of each family/model.
      For healthy control subjects viewing happy faces, the most likely family was Family 1, and within that family, the most likely model was Model 5 (posterior probability .86). This network is defined by modulation of the single backward connection from OFC to FG. For healthy control subjects viewing sad faces, the most likely family was Family 7, and within that family, the most likely model was Model 21 (posterior probability .81). This network is defined by modulation of bidirectional connections between OFC and amygdala and between OFC and FG.
      For rMDD subjects viewing happy faces, the most likely family was Family 7, and within that family, the most likely model was Model 21 (posterior probability .98). Thus the most likely model for rMDD participants viewing happy faces was the same model that best fitted control subjects viewing sad faces. For rMDD subjects viewing sad faces, the most likely family was Family 1, and within that family, the most likely model was Model 2 (posterior probability 1.0), with modulation of the single forward connection from the OFC to FG. Thus the best-fitting model for rMDD participants viewing sad faces was similar to the model that best fitted control subjects viewing happy faces, with only the unidirectional OFC–FG connection modulated but in a different direction.

      PPI Results

      For happy faces, right amygdala connectivity to FG (31, −56, 5) and OFC (25, 32, −10) was significantly reduced (p < .05, family-wise error corrected) in healthy control subjects compared with rMDD. For sad faces, right amygdala connectivity to FG (28, −63, −15) and OFC (11, 40, −5) was significantly reduced (p < .05, family-wise error corrected) in rMDD patients compared with healthy control subjects (Figure 3). There were no correlations between amygdala connectivity and either Clinical Anxiety Scale or MADRS scores in any brain regions even at a liberal threshold of p < .01 uncorrected.
      Figure thumbnail gr3
      Figure 3Results of psychophysiological interaction analysis. (A) Relative strengths of connectivity from the amygdala for processing sad versus neutral faces in remitted major depressive disorder (rMDD) and healthy control subjects (HC). (B) Relative amygdala connection strengths for processing happy versus neutral faces. OFC, orbitofrontal cortex.

      Discussion

      In this study we identified how emotional valence modulated connectivity within a network of regions involved in face processing in healthy control subjects and rMDD. Happy and sad emotions modulated connectivity differently in the two groups, even though there were no significant differences in either region-specific responses or intrinsic connectivity associated with face processing.
      Inferences about the directionality of connections (forward, backward, bidirectional, or absent) are important in the context of emotion networks and made possible by DCM model comparison procedures. Because we were primarily interested in the relationship between rMDD and the emotional modulation of network connectivity, we used a first stage of DCM modeling to identify the most likely network. Within this fully connected network, we then characterized the pattern of emotional modulation of connections in each group. Strikingly, the family-based Bayesian model selection suggested a reversal of the association between emotion, connectivity, and group. Although DCM does not directly compare model fits from different datasets (e.g., different groups), the reversal of the Bayesian model comparisons between groups implies highly significant differences in underlying neural architectures for emotional processing in the two groups (
      • Passamonti L.
      • Crockett M.J.
      • Apergis-Schoute A.M.
      • Clark L.
      • Rowe J.B.
      • Calder A.J.
      • Robbins T.W.
      Effects of acute tryptophan depletion on prefrontal-amygdala connectivity while viewing facial signals of aggression.
      ). We went on to corroborate our findings with PPIs, a complementary technique that does not support inferences of causality or directionality but uses simpler general linear models to infer differences in functional connectivity.
      Dynamic causal modeling indicated that the pattern of modulation by sad faces in control subjects was the same as for happy faces in patients, involving bi-directional connections between OFC and both amygdala and FG. By contrast, happy faces in control subjects and sad faces in rMDD participants were associated with modulation of unidirectional connection between the OFC and FG (the backward connection in control subjects and the forward connection in rMDD). This was corroborated by a direct comparison of functional connectivity between the two groups with PPI. This analysis showed relatively reduced amygdala connectivity in rMDD during processing of sad faces but relatively enhanced amygdala connectivity during processing of happy faces.
      Our finding that emotional valence specifically modulates effective connectivity in a face processing network in healthy control subjects is broadly consistent with previous reports (
      • de Marco G.
      • de Bonis M.
      • Vrignaud P.
      • Henry-Feugeas M.C.
      • Peretti I.
      Changes in effective connectivity during incidental and intentional perception of fearful faces.
      ,
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ,
      • Ishai A.
      Let's face it: It's a cortical network.
      ). However, there are some discrepancies. Fairhall and Ishai (
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ) reported enhanced forward connectivity between inferior occipital gyrus, FG, and amygdala, whereas we observed enhanced OFC–FG feedback connectivity for happy faces and enhanced bidirectional connectivity to OFC to both FG and amygdala for sad faces. One reason for the difference is that Fairhall and Ishai (
      • Fairhall S.L.
      • Ishai A.
      Effective connectivity within the distributed cortical network for face perception.
      ) combined positive and negative emotions, simply assessing effects of “emotional” faces. Here we show that happy and sad emotions modulate effective connectivity in different ways, suggesting that—as for standard fMRI analysis—combining different emotions might obscure significant effects. Dima et al. (
      • Dima D.
      • Stephan K.E.
      • Roiser J.P.
      • Friston K.J.
      • Frangou S.
      Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
      ) recently showed that different negative emotions modulate connectivity in specific ways, with sad faces particularly modulating the FG–OFC connection. A second critical difference is in the nature of the task demands. The Fairhall and Ishai paradigm was passive viewing. Here, as in Dima et al. (
      • Dima D.
      • Stephan K.E.
      • Roiser J.P.
      • Friston K.J.
      • Frangou S.
      Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
      ), we required subjects to make active decisions about faces. Evidence suggests that precise cognitive and attentional demands of a task might be critical determinants of neuronal response to emotional faces, both in region-specific fMRI analysis (
      • Ishai A.
      Let's face it: It's a cortical network.
      ) and effective connectivity modeling (
      • Phillips M.L.
      • Williams L.M.
      • Heining M.
      • Herba C.M.
      • Russell T.
      • Andrew C.
      • et al.
      Differential neural responses to overt and covert presentations of facial expressions of fear and disgust.
      ).
      Our main finding is that emotion-specific modulation of effective connectivity is abnormal in rMDD. Neither behavioral performance nor region-specific neuronal responses to emotional faces have conclusively distinguished rMDD from control subjects ([4] for review). A number of studies, including our own, suggest that rMDD subjects show increased recognition of emotions, particularly negative emotions (
      • Anderson I.M.
      • Shippen C.
      • Juhasz G.
      • Chase D.
      • Thomas E.
      • Downey D.
      • et al.
      State-dependent alteration in face emotion recognition in depression.
      ,
      • Bhagwagar Z.
      • Cowen P.J.
      • Goodwin G.M.
      • Harmer C.J.
      Normalization of enhanced fear recognition by acute SSRI treatment in subjects with a previous history of depression.
      ,
      • Hayward G.
      • Goodwin G.M.
      • Cowen P.J.
      • Harmer C.J.
      Low-dose tryptophan depletion in recovered depressed patients induces changes in cognitive processing without depressive symptoms.
      ), although the effects are subtle. Studies of region-specific neural abnormalities have reported mixed results, with both enhanced (
      • Victor T.A.
      • Furey M.L.
      • Fromm S.J.
      • Ohman A.
      • Drevets W.C.
      Relationship between amygdala responses to masked faces and mood state and treatment in major depressive disorder.
      ) and normal (
      • Anderson I.M.
      • Shippen C.
      • Juhasz G.
      • Chase D.
      • Thomas E.
      • Downey D.
      • et al.
      State-dependent alteration in face emotion recognition in depression.
      ) amygdala responses observed in rMDD. Probably the most consistent finding is normalization of abnormalities found in current MDD with short-term successful treatment (
      • Harmer C.J.
      • O'Sullivan U.
      • Favaron E.
      • Massey-Chase R.
      • Ayres R.
      • Reinecke A.
      • et al.
      Effect of acute antidepressant administration on negative affective bias in depressed patients.
      ,
      • Fu C.H.
      • Williams S.C.
      • Cleare A.J.
      • Brammer M.J.
      • Walsh N.D.
      • Kim J.
      • et al.
      Attenuation of the neural response to sad faces in major depression by antidepressant treatment: A prospective, event-related functional magnetic resonance imaging study.
      ,
      • Surguladze S.
      • Brammer M.J.
      • Keedwell P.
      • Giampietro V.
      • Young A.W.
      • Travis M.J.
      • et al.
      A differential pattern of neural response toward sad versus happy facial expressions in major depressive disorder.
      ). In the cohort described here, there were no significant group differences in region-specific responses to happy or sad faces.
      However, in the absence of region-specific differences, the best fitting model families for the two groups are reversed for the two emotions, providing evidence for group differences in connectivity. In support of the DCM results, a PPI analysis reveals a reversal of emotion-specific face processing-related amygdala connectivity between the two groups on direct statistical comparison. Behaviorally, healthy control subjects are biased toward processing happy information in various contexts, whereas rMDD might be associated with bias toward negative, specifically sad, information (
      • Elliott R.
      • Zahn R.
      • Deakin J.F.
      • Anderson I.M.
      Affective cognition and its disruption in mood disorders.
      ). If we consider—as a shorthand—happy and sad faces as “congruent” with the biases seen in control subjects and rMDD subjects, respectively, and the other emotion “incongruent,” our results can be interpreted as indicating that the processing of congruent emotion in both control subjects and rMDD subjects is associated with modulation of FG–OFC connectivity only. Conversely, incongruent face emotion is associated with the same connectivity modulation in both groups: increased modulation of bidirectional connectivity of the OFC to both FG and amygdala. Thus, for control subjects and rMDD, only incongruent emotion modulates the amygdala–OFC connection. Enhanced processing of mood-congruent information might depend on decreased inhibitory cortico–limbic connectivity in both patients and control subjects. Almeida et al. (
      • Almeida J.R.
      • Versace A.
      • Mechelli A.
      • Hassel S.
      • Quevedo K.
      • Kupfer D.J.
      • Phillips M.L.
      Abnormal amygdala-prefrontal effective connectivity to happy faces differentiates bipolar from major depression.
      ) have previously reported abnormal top-down OFC–amygdala connectivity during overt processing of happy faces in MDD, which they interpret as reflecting increased inhibition of positive emotion. This is consistent with our suggestion that mood-incongruent information might be associated with greater inhibitory connectivity, whereas mood-congruent information is associated with decreased inhibitory connectivity. It cannot be concluded from the present results whether different patterns of connectivity between patients and control subjects represent a general mechanism for mediating subtle attentional biases in rMDD or whether the findings are specific to face emotion-processing. However, the results indicate clearly that connectivity approaches (both DCM and PPI) might importantly detect abnormalities that are not observed with “classic” region-specific fMRI.
      Abnormal connectivity between networks connecting OFC, FG, and amygdala are consistent with previous studies using various techniques. Anatomical connectivity can be assessed with tractography techniques such as diffusion tensor imaging. Gschwind et al. (
      • Gschwind M.
      • Pourtois G.
      • Schwartz S.
      • Van De Ville D.
      • Vuilleumier P.
      White-matter connectivity between face-responsive regions in the human brain [published online ahead of print February 23].
      ) demonstrated white matter (structural) connectivity between visual processing areas and amygdala, although FG–amygdala connectivity was weak (perhaps consistent with the absence of modulation of this connection in the present study). The OFC–amygdala connection is well-established (
      • Kringelbach M.L.
      • Rolls E.T.
      The functional neuroanatomy of the human orbitofrontal cortex: Evidence from neuroimaging and neuropsychology.
      ) and has been demonstrated in a recent tractography study (
      • Bach D.R.
      • Behrens T.E.
      • Garrido L.
      • Weiskopf N.
      • Dolan R.J.
      Deep and superficial amygdala nuclei projections revealed in vivo by probabilistic tractography.
      ). Diffusion tensor imaging studies in depression suggest abnormalities in structural connectivity of OFC–amygdala, both in currently depressed patients and those at high risk (
      • Versace A.
      • Thompson W.K.
      • Zhou D.
      • Almeida J.R.
      • Hassel S.
      • Klein C.R.
      • et al.
      Abnormal left and right amygdala-orbitofrontal cortical functional connectivity to emotional faces: State versus trait vulnerability markers of depression in bipolar disorder.
      ,
      • Frodl T.
      • Carballedo A.
      • Fagan A.J.
      • Lisiecka D.
      • Ferguson Y.
      • Meaney J.F.
      Effects of early-life adversity on white matter diffusivity changes in patients at risk for major depression.
      ). Direct connections between FG and OFC are less well-established; however, both regions connect via major fiber tracts to temporal pole (
      • Martino J.
      • De Witt Hamer P.C.
      • Vergani F.
      • Brogna C.
      • de Lucas E.M.
      • Vazquez-Barquero A.
      • et al.
      Cortex-sparing fiber dissection: An improved method for the study of white matter anatomy in the human brain.
      ), and therefore indirect anatomical connection is plausible.
      Studies of functional and effective connectivity also support a hypothesis of abnormal connectivity within the network suggested here. Resting state connectivity studies (
      • Zhang J.
      • Wang J.
      • Wu Q.
      • Kuang W.
      • Huang X.
      • He Y.
      • Gong Q.
      Disrupted brain connectivity networks in drug-naive, first-episode major depressive disorder.
      ,
      • Zhu X.
      • Wang X.
      • Xiao J.
      • Liao J.
      • Zhong M.
      • Wang W.
      • Yao S.
      Evidence of a dissociation pattern in resting-state default mode network connectivity in first-episode, treatment-naive major depression patients.
      ,
      • Greicius M.D.
      • Flores B.H.
      • Menon V.
      • Glover G.H.
      • Solvason H.B.
      • Kenna H.
      • et al.
      Resting-state functional connectivity in major depression: Abnormally increased contributions from subgenual cingulate cortex and thalamus.
      ) describe aberrant connectivity associated with depression. More relevant studies of functional and effective connectivity in response to emotional challenges also suggest impaired fronto–limbic connectivity (
      • Ruhe H.G.
      • Booij J.
      • Veltman D.J.
      • Michel M.C.
      • Schene A.H.
      Successful pharmacologic treatment of major depressive disorder attenuates amygdala activation to negative facial expressions: A functional magnetic resonance imaging study.
      ,
      • Hamilton J.P.
      • Gotlib I.H.
      Neural substrates of increased memory sensitivity for negative stimuli in major depression.
      ,
      • Frodl T.
      • Bokde A.L.
      • Scheuerecker J.
      • Lisiecka D.
      • Schoepf V.
      • Hampel H.
      • et al.
      Functional connectivity bias of the orbitofrontal cortex in drug-free patients with major depression.
      ,
      • Erk S.
      • Mikschl A.
      • Stier S.
      • Ciaramidaro A.
      • Gapp V.
      • Weber B.
      • Walter H.
      Acute and sustained effects of cognitive emotion regulation in major depression.
      ,
      • Smith R.
      • Fadok R.A.
      • Purcell M.
      • Liu S.
      • Stonnington C.
      • Spetzler R.F.
      • Baxter L.C.
      Localizing sadness activation within the subgenual cingulate in individuals: A novel functional MRI paradigm for detecting individual differences in the neural circuitry underlying depression.
      ,
      • Carballedo A.
      • Scheuerecker J.
      • Meisenzahl E.
      • Schoepf V.
      • Bokde A.
      • Moller H.J.
      • et al.
      Functional connectivity of emotional processing in depression.
      ). An important advantage of our study is that we did not rely on a between-groups comparison of parameter estimates of connection strengths. Early applications of DCM for group effects have used the connectivity parameter as a dependent variable, with a single connection (
      • Almeida J.R.
      • Versace A.
      • Mechelli A.
      • Hassel S.
      • Quevedo K.
      • Kupfer D.J.
      • Phillips M.L.
      Abnormal amygdala-prefrontal effective connectivity to happy faces differentiates bipolar from major depression.
      ) or the same model for patients and control subjects (
      • Desseilles M.
      • Schwartz S.
      • Dang-Vu T.T.
      • Sterpenich V.
      • Ansseau M.
      • Maquet P.
      • Phillips C.
      Depression alters “top-down” visual attention: A dynamic causal modeling comparison between depressed and healthy subjects.
      ). However, realistic models can be more complex, with multiple connections and modulatory influences. In these circumstances, reliability of parameter estimates can be poor (
      • Rowe J.B.
      • Hughes L.E.
      • Barker R.A.
      • Owen A.M.
      Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
      ), and model-level inference, such as our staged Bayesian model selection procedure, is more appropriate (
      • Stephan K.E.
      • Penny W.D.
      • Moran R.J.
      • den Ouden H.E.
      • Daunizeau J.
      • Friston K.J.
      Ten simple rules for dynamic causal modeling.
      ,
      • Stephan K.E.
      • Penny W.D.
      • Daunizeau J.
      • Moran R.J.
      • Friston K.J.
      Bayesian model selection for group studies.
      ).
      A recent study, with a similar approach to ours, showed that modulation of normal serotonin (5-HT) function by acute tryptophan depletion altered the connectivity of amygdala and ventrolateral frontal cortex during processing of emotional faces (
      • Passamonti L.
      • Crockett M.J.
      • Apergis-Schoute A.M.
      • Clark L.
      • Rowe J.B.
      • Calder A.J.
      • Robbins T.W.
      Effects of acute tryptophan depletion on prefrontal-amygdala connectivity while viewing facial signals of aggression.
      ). Regions of the network tested here are rich in 5-HT receptors (
      • Varnas K.
      • Halldin C.
      • Hall H.
      Autoradiographic distribution of serotonin transporters and receptor subtypes in human brain.
      ), and studies (
      • Fusar-Poli P.
      • Allen P.
      • Lee F.
      • Surguladze S.
      • Tunstall N.
      • Fu C.H.
      • et al.
      Modulation of neural response to happy and sad faces by acute tryptophan depletion.
      ,
      • Anderson I.M.
      • Del-Ben C.M.
      • McKie S.
      • Richardson P.
      • Williams S.R.
      • Elliott R.
      • Deakin J.F.
      Citalopram modulation of neuronal responses to aversive face emotions: A functional MRI study.
      ,
      • Del-Ben C.M.
      • Deakin J.F.
      • McKie S.
      • Delvai N.A.
      • Williams S.R.
      • Elliott R.
      • et al.
      The effect of citalopram pretreatment on neuronal responses to neuropsychological tasks in normal volunteers: An FMRI study.
      ,
      • Murphy S.E.
      • Norbury R.
      • O'Sullivan U.
      • Cowen P.J.
      • Harmer C.J.
      Effect of a single dose of citalopram on amygdala response to emotional faces.
      ,
      • Norbury R.
      • Taylor M.J.
      • Selvaraj S.
      • Murphy S.E.
      • Harmer C.J.
      • Cowen P.J.
      Short-term antidepressant treatment modulates amygdala response to happy faces.
      ) show that manipulation of 5-HT in healthy volunteers modulates responses to emotional faces in these regions. Fisher et al. (
      • Fisher P.M.
      • Meltzer C.C.
      • Price J.C.
      • Coleman R.L.
      • Ziolko S.K.
      • Becker C.
      • et al.
      Medial prefrontal cortex 5-HT(2A) density is correlated with amygdala reactivity, response habituation, and functional coupling.
      ) used multimodal imaging to show that 5-HT receptors are specifically implicated in emotion-modulated coupling of amygdala and OFC. It therefore seems plausible that the differences in connectivity we observe depend on 5-HT mechanisms. There is abundant evidence for impaired 5-HT neurotransmission in depression (
      • Elliott R.
      • Zahn R.
      • Deakin J.F.
      • Anderson I.M.
      Affective cognition and its disruption in mood disorders.
      ), although evidence for ongoing abnormalities of 5-HT function in rMDD is more mixed. Our present results might suggest ongoing subtle abnormalities of 5-HT mediated connectivity in response to emotional stimuli, although this hypothesis requires explicit testing.
      One limitation of this study is that not all participants activated all regions in response to faces at a standard threshold. The requirements of plausible DCM analysis necessitate exclusion of those who do not activate all nodes, but this raises questions about whether included participants are a distinct subset of rMDD patients. It should be noted that: 1) similar proportions of rMDD and healthy control subjects were excluded, and 2) included participants did not differ from the whole group on any demographic measures. A further limitation is that, although scoring within the normal range, rMDD patients had significantly higher MADRS scores than control subjects. In a traditional fMRI analysis, it would be possible to examine the contribution of residual symptomatology with correlations, but this cannot yet be applied within DCM model selection. However, in our PPI analysis, covarying for symptom scores did not influence the principal results.
      In conclusion, we have shown that rMDD is associated with a reversal of the normal pattern of valence-specific modulation of face-processing connectivity. The rMDD participants show greater bidirectional OFC–amygdala connectivity for happy faces than control subjects and less OFC–amygdala connectivity for sad faces. This pattern might reflect increased inhibitory control of mood-incongruent information and decreased inhibitory control of mood-congruent information. Our findings further support the use of DCM to explore aberrant connectivity in depression. Future studies are required to determine whether abnormal connectivity predicts relapse to MDD or severity of depressive episodes.
      This study was funded by the NEWMOOD (New Molecules for Mood Disorders) EU Integrated Programme LSHM-CT-2004-503474 . Author ET was supported by a Medical Research Council Studentship. Author JBR was supported by the Wellcome Trust ( 088324 ). We thank, for all their assistance, the staff of the Wellcome Trust Clinical Research Facility, Manchester, where this research was carried out. We also thank Dr. Diana Chase, Kathryn Lloyd-Williams, Dr. Ivan Koychev, and Christine Holliday for their hard work and assistance with recruitment and data acquisition.
      JFWD has carried out consultancy and speaking engagements for Bristol Myers Squibb, AstraZeneca, Eli Lilly, Schering Plough, Janssen-Cilag, and Servier. All fees are paid to the University of Manchester to reimburse for the time taken. He has share options in P1vital. IMA has received grant support from AstraZeneca , consultancy fees/honoraria for speaking/support to attend conferences from Wyeth, Servier, Eli Lilly, Lundbeck, Cephalon, and Bristol Myers Squibb. SRW has received grant support from AstraZeneca . RE has carried out consultancy for Cambridge Cognition. All other authors report no biomedical financial interests or potential conflicts of interest.

      References

        • Chamberlain S.R.
        • Sahakian B.J.
        The neuropsychology of mood disorders.
        Curr Psychiatry Rep. 2006; 8: 458-463
        • Phillips M.L.
        • Drevets W.C.
        • Rauch S.L.
        • Lane R.
        Neurobiology of emotion perception II: Implications for major psychiatric disorders.
        Biol Psychiatry. 2003; 54: 515-528
        • Browning M.
        • Holmes E.A.
        • Harmer C.J.
        The modification of attentional bias to emotional information: A review of the techniques, mechanisms, and relevance to emotional disorders.
        Cogn Affect Behav Neurosci. 2010; 10: 8-20
        • Elliott R.
        • Zahn R.
        • Deakin J.F.
        • Anderson I.M.
        Affective cognition and its disruption in mood disorders.
        Neuropsychopharmacology. 2011; 36: 153-182
        • Gilboa-Schechtman E.
        • Erhard-Weiss D.
        • Jeczemien P.
        Interpersonal deficits meet cognitive biases: Memory for facial expressions in depressed and anxious men and women.
        Psychiatry Res. 2002; 113: 279-293
        • Harmer C.J.
        • O'Sullivan U.
        • Favaron E.
        • Massey-Chase R.
        • Ayres R.
        • Reinecke A.
        • et al.
        Effect of acute antidepressant administration on negative affective bias in depressed patients.
        Am J Psychiatry. 2009; 166: 1178-1184
        • Joormann J.
        • Gotlib I.H.
        Selective attention to emotional faces following recovery from depression.
        J Abnorm Psychol. 2007; 116: 80-85
        • Anderson I.M.
        • Shippen C.
        • Juhasz G.
        • Chase D.
        • Thomas E.
        • Downey D.
        • et al.
        State-dependent alteration in face emotion recognition in depression.
        Br J Psychiatry. 2011; 198: 302-308
        • Bhagwagar Z.
        • Cowen P.J.
        • Goodwin G.M.
        • Harmer C.J.
        Normalization of enhanced fear recognition by acute SSRI treatment in subjects with a previous history of depression.
        Am J Psychiatry. 2004; 161: 166-168
        • Leppanen J.M.
        Emotional information processing in mood disorders: A review of behavioral and neuroimaging findings.
        Curr Opin Psychiatry. 2006; 19: 34-39
        • Thomas E.J.
        • Elliott R.
        • McKie S.
        • Arnone D.
        • Downey D.
        • Juhasz G.
        • et al.
        Interaction between a history of depression and rumination on neural response to emotional faces.
        Psychol Med. 2011; 41: 1845-1855
        • Fu C.H.
        • Williams S.C.
        • Cleare A.J.
        • Brammer M.J.
        • Walsh N.D.
        • Kim J.
        • et al.
        Attenuation of the neural response to sad faces in major depression by antidepressant treatment: A prospective, event-related functional magnetic resonance imaging study.
        Arch Gen Psychiatry. 2004; 61: 877-889
        • Surguladze S.
        • Brammer M.J.
        • Keedwell P.
        • Giampietro V.
        • Young A.W.
        • Travis M.J.
        • et al.
        A differential pattern of neural response toward sad versus happy facial expressions in major depressive disorder.
        Biol Psychiatry. 2005; 57: 201-209
        • Victor T.A.
        • Furey M.L.
        • Fromm S.J.
        • Ohman A.
        • Drevets W.C.
        Relationship between amygdala responses to masked faces and mood state and treatment in major depressive disorder.
        Arch Gen Psychiatry. 2011; 67: 1128-1138
        • Sheline Y.I.
        • Barch D.M.
        • Donnelly J.M.
        • Ollinger J.M.
        • Snyder A.Z.
        • Mintun M.A.
        Increased amygdala response to masked emotional faces in depressed subjects resolves with antidepressant treatment: An fMRI study.
        Biol Psychiatry. 2001; 50: 651-658
        • Mayberg H.S.
        Limbic-cortical dysregulation: A proposed model of depression.
        J Neuropsychiatry Clin Neurosci. 1997; 9: 471-481
        • Seminowicz D.A.
        • Mayberg H.S.
        • McIntosh A.R.
        • Goldapple K.
        • Kennedy S.
        • Segal Z.
        • Rafi-Tari S.
        Limbic-frontal circuitry in major depression: A path modeling metanalysis.
        Neuroimage. 2004; 22: 409-418
        • Anand A.
        • Li Y.
        • Wang Y.
        • Wu J.
        • Gao S.
        • Bukhari L.
        • et al.
        Activity and connectivity of brain mood regulating circuit in depression: A functional magnetic resonance study.
        Biol Psychiatry. 2005; 57: 1079-1088
        • Zhang J.
        • Wang J.
        • Wu Q.
        • Kuang W.
        • Huang X.
        • He Y.
        • Gong Q.
        Disrupted brain connectivity networks in drug-naive, first-episode major depressive disorder.
        Biol Psychiatry. 2011; 70: 334-342
        • Zhu X.
        • Wang X.
        • Xiao J.
        • Liao J.
        • Zhong M.
        • Wang W.
        • Yao S.
        Evidence of a dissociation pattern in resting-state default mode network connectivity in first-episode, treatment-naive major depression patients.
        Biol Psychiatry. 2011; 71: 611-617
        • Greicius M.D.
        • Flores B.H.
        • Menon V.
        • Glover G.H.
        • Solvason H.B.
        • Kenna H.
        • et al.
        Resting-state functional connectivity in major depression: Abnormally increased contributions from subgenual cingulate cortex and thalamus.
        Biol Psychiatry. 2007; 62: 429-437
        • Versace A.
        • Thompson W.K.
        • Zhou D.
        • Almeida J.R.
        • Hassel S.
        • Klein C.R.
        • et al.
        Abnormal left and right amygdala-orbitofrontal cortical functional connectivity to emotional faces: State versus trait vulnerability markers of depression in bipolar disorder.
        Biol Psychiatry. 2010; 67: 422-431
        • Ruhe H.G.
        • Booij J.
        • Veltman D.J.
        • Michel M.C.
        • Schene A.H.
        Successful pharmacologic treatment of major depressive disorder attenuates amygdala activation to negative facial expressions: A functional magnetic resonance imaging study.
        J Clin Psychiatry. 2011; 73: 451-459
        • Hamilton J.P.
        • Gotlib I.H.
        Neural substrates of increased memory sensitivity for negative stimuli in major depression.
        Biol Psychiatry. 2008; 63: 1155-1162
        • Frodl T.
        • Bokde A.L.
        • Scheuerecker J.
        • Lisiecka D.
        • Schoepf V.
        • Hampel H.
        • et al.
        Functional connectivity bias of the orbitofrontal cortex in drug-free patients with major depression.
        Biol Psychiatry. 2011; 67: 161-167
        • Erk S.
        • Mikschl A.
        • Stier S.
        • Ciaramidaro A.
        • Gapp V.
        • Weber B.
        • Walter H.
        Acute and sustained effects of cognitive emotion regulation in major depression.
        J Neurosci. 2011; 30: 15726-15734
        • Smith R.
        • Fadok R.A.
        • Purcell M.
        • Liu S.
        • Stonnington C.
        • Spetzler R.F.
        • Baxter L.C.
        Localizing sadness activation within the subgenual cingulate in individuals: A novel functional MRI paradigm for detecting individual differences in the neural circuitry underlying depression.
        Brain Imaging Behav. 2011; 5: 229-239
        • Cremers H.R.
        • Demenescu L.R.
        • Aleman A.
        • Renken R.
        • van Tol M.J.
        • van der Wee N.J.
        • et al.
        Neuroticism modulates amygdala-prefrontal connectivity in response to negative emotional facial expressions.
        Neuroimage. 2010; 49: 963-970
        • Penny W.D.
        • Stephan K.E.
        • Mechelli A.
        • Friston K.J.
        Modelling functional integration: A comparison of structural equation and dynamic causal models.
        Neuroimage. 2004; 23: S264-S274
        • Stephan K.E.
        • Penny W.D.
        • Moran R.J.
        • den Ouden H.E.
        • Daunizeau J.
        • Friston K.J.
        Ten simple rules for dynamic causal modeling.
        Neuroimage. 2010; 49: 3099-3109
        • Friston K.J.
        • Harrison L.
        • Penny W.
        Dynamic causal modelling.
        Neuroimage. 2003; 19: 1273-1302
        • Carballedo A.
        • Scheuerecker J.
        • Meisenzahl E.
        • Schoepf V.
        • Bokde A.
        • Moller H.J.
        • et al.
        Functional connectivity of emotional processing in depression.
        J Affect Disord. 2011; 134: 272-279
        • de Marco G.
        • de Bonis M.
        • Vrignaud P.
        • Henry-Feugeas M.C.
        • Peretti I.
        Changes in effective connectivity during incidental and intentional perception of fearful faces.
        Neuroimage. 2006; 30: 1030-1037
        • Fairhall S.L.
        • Ishai A.
        Effective connectivity within the distributed cortical network for face perception.
        Cereb Cortex. 2007; 17: 2400-2406
        • Ishai A.
        Let's face it: It's a cortical network.
        Neuroimage. 2008; 40: 415-419
        • Dima D.
        • Stephan K.E.
        • Roiser J.P.
        • Friston K.J.
        • Frangou S.
        Effective connectivity during processing of facial affect: Evidence for multiple parallel pathways.
        J Neurosci. 2011; 31: 14378-14385
        • Almeida J.R.
        • Versace A.
        • Hassel S.
        • Kupfer D.J.
        • Phillips M.L.
        Elevated amygdala activity to sad facial expressions: A state marker of bipolar but not unipolar depression.
        Biol Psychiatry. 2010; 67: 414-421
        • Rowe J.B.
        • Hughes L.E.
        • Barker R.A.
        • Owen A.M.
        Dynamic causal modelling of effective connectivity from fMRI: Are results reproducible and sensitive to Parkinson's disease and its treatment?.
        Neuroimage. 2010; 52: 1015-1026
        • Rowe J.B.
        Connectivity analysis is essential to understand neurological disorders.
        Front Syst Neurosci. 2010; 4: 144
        • Sheehan D.V.
        • Lecrubier Y.
        • Sheehan K.H.
        • Amorim P.
        • Janavs J.
        • Weiller E.
        • et al.
        The Mini-International Neuropsychiatric Interview (M.I.N.I.): The development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10.
        J Clin Psychiatry. 1998; 59: 22-33
        • First M.B.
        • Gibbon M.
        • Williams J.B.W.
        Structured Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, Patient Edition.
        (SR) Oxford University Press, New York2002
        • Young A.W.
        • Rowland D.
        • Calder A.J.
        • Etcoff N.L.
        • Seth A.
        • Perrett D.I.
        Facial expression megamix: Tests of dimensional and category accounts of emotion recognition.
        Cognition. 1997; 63: 271-313
        • Stephan K.E.
        • Penny W.D.
        • Daunizeau J.
        • Moran R.J.
        • Friston K.J.
        Bayesian model selection for group studies.
        Neuroimage. 2009; 46: 1004-1017
        • Ishai A.
        • Schmidt C.F.
        • Boesiger P.
        Face perception is mediated by a distributed cortical network.
        Brain Res Bull. 2005; 67: 87-93
        • Ishai A.
        • Ungerleider L.G.
        • Martin A.
        • Haxby J.V.
        The representation of objects in the human occipital and temporal cortex.
        J Cogn Neurosci. 2000; 12: 35-51
        • Goulden N.
        • McKie S.
        • Suckling J.
        • Williams S.R.
        • Anderson I.M.
        • Deakin J.F.
        • Elliott R.
        A comparison of permutation and parametric testing for between group effective connectivity differences using DCM.
        Neuroimage. 2010; 50: 509-515
        • Penny W.D.
        • Stephan K.E.
        • Daunizeau J.
        • Rosa M.J.
        • Friston K.J.
        • Schofield T.M.
        • Leff A.P.
        Comparing families of dynamic causal models.
        PLoS Comput Biol. 2010; 6: e1000709
        • Passamonti L.
        • Crockett M.J.
        • Apergis-Schoute A.M.
        • Clark L.
        • Rowe J.B.
        • Calder A.J.
        • Robbins T.W.
        Effects of acute tryptophan depletion on prefrontal-amygdala connectivity while viewing facial signals of aggression.
        Biol Psychiatry. 2012; 71: 36-43
        • Passamonti L.
        • Rowe J.B.
        • Ewbank M.
        • Hampshire A.
        • Keane J.
        • Calder A.J.
        Connectivity from the ventral anterior cingulate to the amygdala is modulated by appetitive motivation in response to facial signals of aggression.
        Neuroimage. 2008; 43: 562-570
        • Kass R.E.
        • Raftery A.E.
        Bayes Factors.
        J Amer Statist Assoc. 2005; 90: 773-795
        • Phillips M.L.
        • Williams L.M.
        • Heining M.
        • Herba C.M.
        • Russell T.
        • Andrew C.
        • et al.
        Differential neural responses to overt and covert presentations of facial expressions of fear and disgust.
        Neuroimage. 2004; 21: 1484-1496
        • Hayward G.
        • Goodwin G.M.
        • Cowen P.J.
        • Harmer C.J.
        Low-dose tryptophan depletion in recovered depressed patients induces changes in cognitive processing without depressive symptoms.
        Biol Psychiatry. 2005; 57: 517-524
        • Almeida J.R.
        • Versace A.
        • Mechelli A.
        • Hassel S.
        • Quevedo K.
        • Kupfer D.J.
        • Phillips M.L.
        Abnormal amygdala-prefrontal effective connectivity to happy faces differentiates bipolar from major depression.
        Biol Psychiatry. 2009; 66: 451-459
        • Gschwind M.
        • Pourtois G.
        • Schwartz S.
        • Van De Ville D.
        • Vuilleumier P.
        White-matter connectivity between face-responsive regions in the human brain [published online ahead of print February 23].
        Cereb Cortex. 2012;
        • Kringelbach M.L.
        • Rolls E.T.
        The functional neuroanatomy of the human orbitofrontal cortex: Evidence from neuroimaging and neuropsychology.
        Prog Neurobiol. 2004; 72: 341-372
        • Bach D.R.
        • Behrens T.E.
        • Garrido L.
        • Weiskopf N.
        • Dolan R.J.
        Deep and superficial amygdala nuclei projections revealed in vivo by probabilistic tractography.
        J Neurosci. 2011; 31: 618-623
        • Frodl T.
        • Carballedo A.
        • Fagan A.J.
        • Lisiecka D.
        • Ferguson Y.
        • Meaney J.F.
        Effects of early-life adversity on white matter diffusivity changes in patients at risk for major depression.
        J Psychiatry Neurosci. 2012; 37: 37-45
        • Martino J.
        • De Witt Hamer P.C.
        • Vergani F.
        • Brogna C.
        • de Lucas E.M.
        • Vazquez-Barquero A.
        • et al.
        Cortex-sparing fiber dissection: An improved method for the study of white matter anatomy in the human brain.
        J Anat. 2011; 219: 531-541
        • Desseilles M.
        • Schwartz S.
        • Dang-Vu T.T.
        • Sterpenich V.
        • Ansseau M.
        • Maquet P.
        • Phillips C.
        Depression alters “top-down” visual attention: A dynamic causal modeling comparison between depressed and healthy subjects.
        Neuroimage. 2011; 54: 1662-1668
        • Varnas K.
        • Halldin C.
        • Hall H.
        Autoradiographic distribution of serotonin transporters and receptor subtypes in human brain.
        Hum Brain Mapp. 2004; 22: 246-260
        • Fusar-Poli P.
        • Allen P.
        • Lee F.
        • Surguladze S.
        • Tunstall N.
        • Fu C.H.
        • et al.
        Modulation of neural response to happy and sad faces by acute tryptophan depletion.
        Psychopharmacology (Berl). 2007; 193: 31-44
        • Anderson I.M.
        • Del-Ben C.M.
        • McKie S.
        • Richardson P.
        • Williams S.R.
        • Elliott R.
        • Deakin J.F.
        Citalopram modulation of neuronal responses to aversive face emotions: A functional MRI study.
        Neuroreport. 2007; 18: 1351-1355
        • Del-Ben C.M.
        • Deakin J.F.
        • McKie S.
        • Delvai N.A.
        • Williams S.R.
        • Elliott R.
        • et al.
        The effect of citalopram pretreatment on neuronal responses to neuropsychological tasks in normal volunteers: An FMRI study.
        Neuropsychopharmacology. 2005; 30: 1724-1734
        • Murphy S.E.
        • Norbury R.
        • O'Sullivan U.
        • Cowen P.J.
        • Harmer C.J.
        Effect of a single dose of citalopram on amygdala response to emotional faces.
        Br J Psychiatry. 2009; 194: 535-540
        • Norbury R.
        • Taylor M.J.
        • Selvaraj S.
        • Murphy S.E.
        • Harmer C.J.
        • Cowen P.J.
        Short-term antidepressant treatment modulates amygdala response to happy faces.
        Psychopharmacology (Berl). 2009; 206: 197-204
        • Fisher P.M.
        • Meltzer C.C.
        • Price J.C.
        • Coleman R.L.
        • Ziolko S.K.
        • Becker C.
        • et al.
        Medial prefrontal cortex 5-HT(2A) density is correlated with amygdala reactivity, response habituation, and functional coupling.
        Cereb Cortex. 2009; 19: 2499-2507