The Extended Functional Neuroanatomy of Emotional Processing Biases for Masked Faces in Major Depressive Disorder

Background Major depressive disorder (MDD) is associated with a mood-congruent processing bias in the amygdala toward face stimuli portraying sad expressions that is evident even when such stimuli are presented below the level of conscious awareness. The extended functional anatomical network that maintains this response bias has not been established, however. Aims To identify neural network differences in the hemodynamic response to implicitly presented facial expressions between depressed and healthy control participants. Method Unmedicated-depressed participants with MDD (n = 22) and healthy controls (HC; n = 25) underwent functional MRI as they viewed face stimuli showing sad, happy or neutral face expressions, presented using a backward masking design. The blood-oxygen-level dependent (BOLD) signal was measured to identify regions where the hemodynamic response to the emotionally valenced stimuli differed between groups. Results The MDD subjects showed greater BOLD responses than the controls to masked-sad versus masked-happy faces in the hippocampus, amygdala and anterior inferotemporal cortex. While viewing both masked-sad and masked-happy faces relative to masked-neutral faces, the depressed subjects showed greater hemodynamic responses than the controls in a network that included the medial and orbital prefrontal cortices and anterior temporal cortex. Conclusions Depressed and healthy participants showed distinct hemodynamic responses to masked-sad and masked-happy faces in neural circuits known to support the processing of emotionally valenced stimuli and to integrate the sensory and visceromotor aspects of emotional behavior. Altered function within these networks in MDD may establish and maintain illness-associated differences in the salience of sensory/social stimuli, such that attention is biased toward negative and away from positive stimuli.

In a recent functional MRI (fMRI) study of implicit emotional processing biases in depressed and healthy participants [8] we reported that the negative bias in MDD is evinced by the hemodynamic responses of the amygdala to emotionally expressive face stimuli presented below conscious awareness using a backward masking technique. Specifically, the amygdala response was greater to masked-sad faces versus masked-happy or maskedneutral faces in unmedicated depressed subjects, but was greater toward masked-happy faces versus masked-sad or masked-neutral faces in healthy controls (see also [16,17]).
The amygdala plays a pivotal role in evaluating the behavioral salience of sensory stimuli, partly through subcortical networks that rapidly respond to stimulus features and objects processed at both conscious and non-conscious levels [18][19][20]. In this role the amygdala forms part of extended anatomical networks that involve other mesiotemporal lobe structures, the sensory thalamus, the primary and associative sensory cortices, and the medial and orbital prefrontal cortex. Together, these regions function to modulate the neural and behavioral responses to emotional stimuli on the basis of learning, context and changing reinforcement contingencies [21][22][23][24]. The current study aimed to elucidate the extended functional anatomical network that participates with the amygdala in maintaining emotional processing biases in MDD, by identifying other brain regions where the hemodynamic response to masked-sad versus masked-happy face stimuli differed between depressed and healthy control participants.

Ethics Statement
The research was conducted at the National Institute of Mental Health Intramural Research Program under Protocol 04-M-0002: ''The Functional Neuroanatomy of Emotion Regulation in Major Depressive Disorder (MDD)'', which received initial NIMH IRB approval on August 19, 2003 with subsequent approval at each annual continuing review. The most recent Continuing Renewal application for this protocol was approved by the NIMH Combined Neuroscience (CNS) Institutional Review Board (IRB) on August 4, 2009.

Participants
The study samples were the same as those characterized in our initial report of data primarily limited to the amygdala [8], and were composed of 22 unmedicated adults who met the Diagnostic and Statistical Manual of Mental Disorders-IV-TR [25] criteria for having MDD in a current major depressive episode (MDD) and 25 healthy controls (HC). Briefly, right-handed individuals [26], ages 18 to 50 years were recruited from the Washington, D.C. metropolitan area. Prior to enrolling in the study, each person underwent a screening evaluation including a medical and psychiatric history, laboratory testing that included a urine drug screen, and neuromorphological MRI. The psychiatric diagnosis was established by a semi-structured interview with a psychiatrist and the Structured Clinical Interview for DSM-IV-TR (SCID) [27]. A family history of psychiatric disorders was evaluated by The Family Interview for Genetic Studies [28].
Subjects were excluded from participation if they manifested any of the following conditions: 1) serious suicidal ideation or behavior, 2) major medical or neurological disorders, 3) exposure within three weeks (eight weeks for fluoxetine) to psychotropic drugs or to other medications likely to affect cerebral blood flow or cognitive function, 4) history of drug or alcohol abuse within the past one-year or a lifetime history of drug or alcohol dependence [27], excepting nicotine, 5) current pregnancy or breastfeeding, or 6) general MRI exclusions. The healthy volunteers were excluded if they had a history of any major psychiatric disorder or a firstdegree relative with a mood or anxiety disorder. Each participant received a thorough explanation of the study, provided written informed consent as approved by the National Institute of Health Combined Neuroscience Institutional Review Board, and received financial compensation for their involvement in the study.

fMRI Backward Masking Task
Participants were scanned while performing a backward masking task [8]. A complete description of the backward masking task design was included in our initial report [8]. To summarize, participants were shown two neutral target faces before the start of each of four ten-minute runs. During the task they responded to the presence of a target face by pressing the top button on a twobutton response box or a non-target face by pressing the bottom button on a response box. The stimuli displayed sad, happy or neutral faces, however, the participants were asked to respond to the faces as target or non-target based on their identity, irrespective of the emotional expression. Each task-trial consisted of an emotional stimulus pair of faces including either sad-neutral (SN), happy-neutral (HN), neutral-sad (NS), neutral-happy (NH) or neutral-neutral (NN) faces. The first face was displayed for 26 ms, followed by a 2 nd ''masking'' face for 107 ms to inhibit conscious perception of the first face. The current paper describes the data limited to trials in which the emotional face stimuli were presented in the masked position. The SN, HN, NS, and NH stimulus types each were shown eight times and the NN type was presented 16 times during each of four runs in a pseudorandomized, mixed-trial design. Within a single trial, the identity of the masked face was never the same as the identity of the masking face, but the gender of the two face stimuli was always the same. A 10 to 13 second inter-stimulus interval followed each stimulus pair to allow the hemodynamic response to return to baseline prior to the next presentation.
Stimuli were presented to subjects in the scanner gantry using E-Prime software (Psychological Software Tools, Pittsburgh, PA) and a cloned display projection on a Monarch Hornet PC computer with a cathode ray tube monitor at 75 Hz to enable the short presentation times. The accuracy of the face stimulus timings was verified using a photodiode and oscilloscope. Faces were selected from the NimStim Set of Facial Expressions [35].

fMRI Data Acquisition
Functional and structural MRI scans were acquired on a General Electric 3.0 Tesla scanner (GE Signa, Milwaukee, WI) with an 8-channel phased-array head coil. The fMRI was performed using a gradient-echo echoplanar imaging (EPI) pulse sequence (39 continuous slices, TE = 20 ms, TR = 2000 ms, flip angle = 90u, 64664 matrix, field of view = 22 cm, voxel dimensions = 3.463.463.0 mm 3 ). A total of 290 fMRI images were obtained during each of four 10-minute runs while participants performed the task. Four images were discarded at the beginning of each run to allow for steady-state tissue magnetization. A highresolution anatomical scan was obtained during the same scan session using a fast spoiled gradient echo (FSPGR) sequence (TR = 780 ms, TE = 2.7 ms, flip angle = 12 degrees, FOV = 22 cm, matrix = 2246224, number of axial slices = 128, slice thickness = 1.2 mm, in-plane resolution = 0.98 mm 2 ) to provide an anatomical framework for the functional imaging analyses.

Analysis of Behavioral Performance
Behavioral performance data were analyzed using SPSS 14.0 (SPSS Inc., Chicago, IL). The accuracy of the subject's response to each stimulus presentation and reaction times for identification of face stimuli as a ''target'' face were recorded using E-Prime software. The validity and effectiveness of the backward masking task design were assessed and established as described in our initial report [8].

Analysis of Functional Neuroimaging Data
Image analyses were performed using the general linear model within SPM5 (Wellcome Trust Center for Neuroimaging, London, England, http://www.fil.ion.ucl.ac.uk/spm). Whole brain fMRI volumes were realigned to the first volume, co-registered to each subject's anatomical MRI scan, normalized to the Montreal Neurological Institute (MNI) standard brain template and smoothed using an 8 mm full-width at half-maximum Gaussian kernel. A high-pass with a cutoff period of 128 sec was used to correct low-frequency artifacts, serial correlations were corrected by choosing an autoregressive model of the order 1 [AR(1)] model, and the non-specific effects of global fluctuations in blood-oxygenlevel dependent (BOLD) signal were removed using global normalization. The realignment parameters were modeled into the analysis as regressors of no interest to correct the image data for motion artifacts. Data for a task run were removed from the analyses if the subject showed movement of more than one-half voxel (1.5 mm) translation or 1.25u rotation. The fMRI data from three or four runs were included for all participants.
Single-subject t-contrast maps within the groups (HC, MDD) were generated by computing the difference maps between each emotional condition (SN, HN, NN) versus baseline (crosshair image). For the group level analysis, a 263 (group6emotion) ANOVA was performed using a random-effects analysis of the eigenvariate values generated from the single-subject analyses to assess the hemodynamic differences across conditions. Post hoc ttests were conducted to further evaluate differences between the MDD and HC groups by comparing specific contrasts involving each of the emotional face conditions (SN, HN, NN) versus baseline. A comparison of masked-sad versus masked-happy faces (SN-HN) was defined as (SN MDD -HN MDD ) versus (SN HC -HN HC ), a comparison of masked-sad versus masked-neutral faces (SN-NN) was defined as (SN MDD -NN MDD ) versus (SN HC -NN HC ), and finally a comparison of masked-happy versus maskedneutral faces (HN-NN) was defined as (HN MDD -NN MDD ) versus (HN HC -NN HC ). The significance threshold for a reported group difference in the mean regional BOLD signal was set at a height threshold for a cluster's peak voxel at p#0.001 and a minimum cluster size of k = 23 voxels. The minimum cluster size was determined using Monte Carlo simulations (10,000 iterations) implemented in AFNI's AlphaSim program [36] to calculate the spatial extent threshold that would be equivalent to correcting for multiple comparison testing while maintaining a cluster-level falsepositive detection rate of p,0.05. Regions that remained significant after applying this correction for multiple comparisons were indicated in the tables using an asterisk. In addition, results were reported in the tables that remained significant after using the same minimum cluster size applied in our earlier amygdala region-of-interest analysis, in order to afford continuity with our previously published results (i.e. 10 voxels- [8]). The coordinates for the voxel with the peak p-value within each cluster were transformed from MNI coordinates to the stereotaxic array of Talairach and Tournoux [37]. Anatomical localization was performed with reference to the Talairach and Tournoux [37] and Mai et al. [38] atlases.

Results
The demographic and clinical characteristics of the study samples appear in Table 1. Subject groups were not significantly different in gender composition, mean age or mean intelligence quotients. Thirteen of the unmedicated MDD subjects were naïve to psychotropic drug treatment. For those subjects who previously had received antidepressant drug treatment the mean (SD) drugfree period was 21 (23) months (range 1 to 72 months). The mean (SD) age at illness onset was 16.7 (6.0) years and the mean (SD) number of major depressive episodes were 3.2 (2.0).

Behavioral Performance on the fMRI task
Upon debriefing after the scanning session, participants reported no awareness of the backward masking technique (i.e., they denied perception of the masked face stimuli). Participants performed at chance level for the detection of target faces that had been presented in the masked face position [8].
The behavioral manifestation of the negative affective bias was demonstrated using a group6masked-condition interaction in the reaction time data, previously described in [8]. The reaction time for target faces was faster to masked-happy faces versus maskedsad faces in controls (p,0.001) and was faster to masked-sad faces versus masked-happy faces in currently-depressed subjects versus controls (p,0.05).  Tables 2 and 3 list results of the t-tests performed to characterize the specific contrasts that accounted for the significant interactions. Within the group6 emotion model these tests evaluated the significance of differences between masked-sad versus masked-happy faces (SN-HN), masked-sad versus masked-neutral faces (SN-NN) and maskedhappy versus masked-neutral faces (HN-NN). The eigenvariate values were extracted from each significant cluster's peak and reported in Figures 1, 2, and Tables S1, S2, S3.
For the SN-HN contrast, the between conditions difference in BOLD activity was greater for the depressed subjects versus the controls in the left hippocampus (t 90 = 3.74, p,0.001; Figure 1a For the SN-NN contrast, the between conditions difference in BOLD activity was greater for the depressed subjects versus the controls in left and right rSTG (t 90 = 3.84, p,0.001 and t 90 = 3.19, p = 0.001, respectively; Figure 2a) and right anterior orbitofrontal cortex (OFC; t 90 = 3.28, p = 0.001; Figure 2b). Conversely, the mean BOLD activity was greater for the controls than the depressed subjects in the left inferior parietal cortex (t 90 = 3.52, p,0.001) and right frontal polar cortex (t 90 = 3.43, p,0.001).
Finally, for the HN-NN contrast, the between conditions difference in BOLD activity was greater in MDD versus HC in the left anterior OFC (t 90 = 3.49, p,0.001; Figure 2c) left anterior      Exploratory analyses were performed post hoc to characterize potential relationships between the neuroimaging findings and the clinical assessments of illness course or severity in the MDD subjects. Spearman rank order correlation coefficients were computed to explore associations between the eigenvariate values and the total number of major depressive episodes, the age at illness onset, and the HAM-D, IDS-SR, ATQ, TCQ and STAI scores. The HAM-D scores correlated inversely with the hemodynamic response to HN-NN in the right rSTG and left anterior insula, such that the response to masked-happy faces decreased as depression severity increased (r = 20.43, p,0.05 and r = 20.51, p,0.05, respectively; Figures S1a-S1b). In contrast, the IDS-SR, ATQ and STAI-T scale scores positively correlated with the hemodynamic response to SN-NN in the right rSTG, such that the response to masked-sad faces increased as the selfreported depression severity (IDS-SR), frequency of automatic negative thoughts (ATQ) and trait levels of anxiety (STAI-T) increased (r = 0.66, p,0.001; r = 0.55, p,0.01 and r = 0.49, p,0.05, respectively; Figures S1c-S1e). In addition, for the HN-NN condition, the age at onset correlated inversely with the hemodynamic response to masked-happy faces in the left anterior insula (r = 20.45, p,0.05) and the STAI-State scores correlated positively with the hemodynamic response in the left anterior orbitofrontal cortex (r = 0.49, p,0.05; Figures S1f-S1g).

Discussion
These data identify neuroanatomical regions that, in addition to the amygdala, show changes in hemodynamic activity associated with automatic negative emotional processing biases in unmedicated, currently-depressed subjects with MDD. During exposure to masked-sad versus masked-happy faces, the depressed group showed greater BOLD activity than the control group in the left hippocampus and right anterior inferotemporal cortex, in addition to the left amygdala (Table 2). During exposure to masked-sad versus masked-neutral faces, the depressed group showed greater BOLD activity than the control group in the bilateral rSTG and right anterior OFC (Table 3). In the hippocampus, amygdala, rSTG and right anterior OFC, the group difference was attributable both to the depressed group showing a higher BOLD response to masked-sad versus masked-happy or neutral faces, and to the control group showing a lower BOLD response to maskedsad versus masked-happy or neutral faces (Figures 1 and 2, Tables S1, S2). Finally, the BOLD response was higher in the depressed subjects but lower in the controls to masked-happy versus maskedneutral faces in the bilateral rSTG, left anterior OFC, left anterior insula, left pregenual ACC, right ventral thalamus and right postcentral gyrus (Figure 2, Table S3). These data thus revealed several structures in which the hemodynamic response patterns appeared opposite in direction between groups for masked-sad or masked-happy faces relative to masked-neutral faces, implying that a shared neural circuitry may be processing stimuli distinctly between depressives and controls depending upon the emotional valence of the stimuli.
Several of the regions implicated in this study reflect structures that share substantial monosynaptic projections with the amygdala. The networks formed by these projections include the anterior inferotemporal cortex that participates in visual processing [39][40][41][42][43], and the networks formed between the medial and orbital prefrontal cortex, amygdala, hippocampus, and rSTG [21,22], [43,44] that play roles in integrating sensory information and modulating the behavioral, emotional and visceral responses to salient stimuli [21,45,46]. The most striking hemodynamic changes with respect to their effect size and the direction of the Table 3. Regional differences in the BOLD response to masked-sad faces versus masked-neutral faces (SN-NN) and masked-happy versus masked-neutral faces (HN-NN) in patients with major depressive disorder and healthy controls.  changes paralleling those within the amygdala included the hippocampus, right anterior OFC, and bilateral rSTG. Each of these structures manifested an exaggerated BOLD response to masked-sad faces relative to masked-happy or masked-neutral faces in the depressed subjects. The projections from these regions putatively convey information to the amygdala regarding context (hippocampus) and sensory integration (anterior OFC, parts of rSTG) [21,[47][48][49]. In contrast, the regions that showed hemodynamic responses that were opposite in direction to those seen in the amygdala, namely the left pregenual anterior cingulate cortex, left anterior OFC (BA 10o) and bilateral areas of the rSTG form part of the extended medial prefrontal, or ''visceromotor'', network [21,49]. The monosynaptic projections from several regions within the visceromotor network have been characterized for their role in influencing the outflow of amygdala activity related to emotional expression and to tuning cortical neuronal responses to sensory stimuli [50].
Most previous neuroimaging studies investigating the functional anatomical correlates of emotional processing biases have focused on the amygdala. This structure contains cells that are selectively tuned to stimulus characteristics that allow for the early detection of salient information, including faces [18,51]. Moreover, the amygdala projections to the primary sensory cortices both directly and indirectly through the cholinergic inputs originating in the basal forebrain tune the firing parameters of neurons in the sensory cortices so they become differentially sensitive to salient sensory signals [52,53]. These interactions between the sensory cortices and the lateral nucleus of the amygdala also are modulated by input from hippocampal and orbitofrontal cortical projections during the associative learning processes that underlie appetitive and aversive conditioning [54,55]. The extended network formed by projections between these structures thus integrates limbic and sensory input as part of assessing stimulus valence and incentive value within a particular context, based upon the tendency for a stimulus to be associated with rewarding or punishing reinforcement [21,22,47,55]. The differential hemodynamic responses of the hippocampus and OFC to implicitly presented sad and happy face stimuli observed between depressives and controls thus may reflect and/or mediate the differences in stimulus salience that maintain a positive processing bias in healthy subjects, but a negative bias in depressed subjects. Although the causal mechanism underlying the distinct directions in the emotional bias between groups remains unknown, the circuits involving these structures appear to be responding automatically to the salience of the happy face in healthy individuals, but responding automatically to the salience of the sad face in depression.
In the current study, the group differences that were greatest with respect to statistical significance localized to the hippocampus and anterior insula (Figures 1a, 2e). Lesions of the hippocampus interrupt afferent neurotransmission to the amygdala that conveys information regarding environmental contexts that hold emotional salience [47,56]. Thus it is conceivable that the hippocampus plays a role in MDD in setting the context for a negative emotional processing bias. In other words the hippocampal input may specifically drive the amygdala to respond differently to sad versus happy stimuli within the context of the depressive episode.
Notably, the anterior insula changed activity most robustly in response to masked-happy faces (HN versus NN), a condition under which the BOLD signal increased in the depressives but decreased in the controls. This pattern of group differences thus was opposite to that observed in the amygdala. In healthy humans the anterior insula consistently activates in response to stimuli that are aversive or disgusting, or to face stimuli that show expressions of disgust [57,58]. These data, taken together with our finding that depressed subjects show increased BOLD activity in this region in response to happy faces, suggest the hypothesis that pathological activation of the anterior insula plays a role in diverting attention away from positive stimuli in MDD.
In healthy volunteers previous research had shown greater BOLD responses in the amygdala [16,17] and the ACC [16] to masked-happy faces relative to either fixation point or to masked-sad or neutral face control conditions. In our original report comparing HC to MDD participants [8], we replicated these findings in the amygdala in the healthy subjects, who showed a greater BOLD response to masked-happy faces versus masked-sad faces. Here we report a pregenual ACC area (located.one cm from the ACC areas described in Killgore et al. [17] that showed a greater response to masked-happy versus neutral faces in depressed subjects versus healthy controls (Figure 2e).
A previous study by Suslow et al. [9] in antidepressantmedicated but persistently depressed MDD subjects also showed hemodynamic responses in the right amygdala that were exaggerated to masked-sad faces and blunted to masked-happy faces, compatible with our findings in unmedicated depressed subjects in this region. Suslow et al. [9] additionally reported the results of a whole brain analysis in which they observed significant group-byemotion interaction effects in the left anterior insula and left STG, structures where we also observed significant group-by-emotion interactions. Although Suslow et al. [9] did not further characterize these interactions, we found here that the interaction in left anterior insula was attributable to a greater BOLD response to masked-happy versus masked-neutral faces in depressives compared to controls, while the interaction in left STG was accounted for by a greater response to masked-sad versus neutral faces in depressives compared to controls.
The manifestation of MDD may result partly as a consequence of a pathological change in the top-down control from the medial prefrontal cortical network over structures in the limbic system, such as the amygdala and hippocampus (see [24]). Impairment in top-down modulatory control exerted by the medial network over amygdala neuronal activity conceivably may result in exaggerated processing of aversive or potentially threatening emotional stimuli, together with attenuated processing of potentially positive or rewarding stimuli [49,59].
The cross-sectional design of this study did not allow us to address directly whether the findings presented herein are mood state-dependent, or instead reflect trait-like abnormalities that precede and/or persist across mood episodes. Nevertheless, our work has shown an increased hemodynamic response to masked-sad faces vs. masked-happy faces in the amygdala [8] and hippocampus (unpublished data) in unmedicated-remitted subjects with MDD, suggesting that the emotional processing bias manifest in these structures may constitute a persistent bias toward the processing of negative stimuli. Further evidence of a trait-like bias in the processing of negative information has been reported for subjects with genetic risk carrying risk alleles in the 5-HTTLPR polymorphism (see [60]), including for nonconsciously presented stimuli [61], and in healthy subjects with environment risk suffering from childhood maltreatment [62]. Additionally, however, there is strong evidence for the hypothesis [63] that antidepressant drugs exert their effects by normalizing the bias toward negative information processing (e.g. [8,12,15]). While this normalization may appear to provide support for a processing bias associated with the current mood state, additional work is needed to assess whether or not this bias represents a biomarker underlying the vulnerability to relapse in MDD that can be modulated by antidepressant drug treatment. A longitudinal study of individuals at high familial risk for MDD may assist in determining the developmental time course of these emotional processing biases associated with MDD.
Other limitations of our study design also merit comment. First, there is no explicitly modeled baseline in the slow eventrelated design of the task. The stimulus contrasts in the analyses compared one emotional stimulus type directly to another stimulus type that differed only in their emotional valence. The slow event-related design was selected to allow the hemodynamic response to return to baseline and to reduce overlap of the neural response to each stimulus condition. However, the result is additional noise in the baseline measure estimates, thereby making reductions relative to baseline more difficult to interpret (Supplementary Tables S1, S2, S3). Nevertheless, the results highlight the robustness of the difference in emotional conditions compared to baseline. A rapid event-related design conceivably may allow for better modeling of a specific baseline measure within the design construct because more trials can be completed within the scanning session. Second, the study design did not allow determination of a causal relationship between the abnormal BOLD responses identified herein and the pathogenesis of MDD. Finally, this study did not address the generalizability of the results to other categories of mood disorders and did not assess potential sex differences in the processing of emotional information in MDD.
In conclusion, our findings provide support for automatic neurophysiological biases toward negative and away from positive emotional stimuli in MDD, and show that these processing biases are supported by an extended network of structures known to share substantial projections with the amygdala and to participate in the evaluation of stimulus salience. Future studies are needed to explore whether the abnormal hemodynamic response pattern to negative versus positive stimuli in MDD can sensitively and specifically discriminate individuals with mood disorders from healthy controls, and to characterize the potential relationship of this pattern with the onset, maintenance and recurrence of depressive episodes. Figure S1 Relationships between the neuroimaging findings and the clinical assessments of illness course or severity in MDD participants: (a,b) Hamilton Depression Rating Scale, (c) Inventory of Depressive Symptomatology, (d) Automatic Thoughts Questionnaire, (e) State-Trait Anxiety Inventory-Trait Score, (f) age of onset and (g) State-Trait Anxiety Inventory-State Score. (TIF) Table S1 Mean (SD) eigenvariates extracted from the peak voxel regions where hemodynamic activity was significantly different between healthy controls and participants with major depressive disorder for masked-sad faces (SN) versus maskedhappy faces (HN). The coordinates correspond to regions reported in Table 2 of the manuscript. (DOCX) Table S2 Mean (SD) eigenvariates extracted from the peak voxel regions where hemodynamic activity was significantly different between healthy controls and participants with major depressive disorder for masked-sad faces (SN) versus maskedneutral faces (NN). The coordinates correspond to regions reported in Table 3 of the manuscript. (DOCX)

Supporting Information
Table S3 Mean (SD) eigenvariates extracted from the peak voxel regions where hemodynamic activity was significantly different between healthy controls and participants with major depressive disorder for masked-happy faces (HN) versus maskedneutral faces (NN). The coordinates correspond to regions reported in Table 3 of the manuscript. (DOCX)