Introduction
Acute major depression is associated with pervasive negative biases in the processing of emotional information. Developments in functional brain imaging have allowed the neural circuitry supporting emotional processing to be increasingly well characterized. For example, a ventral system that includes the amygdala, ventral striatum and orbitofrontal cortex has been identified (Phillips et al. Reference Phillips, Ladouceur and Drevets2008), which is involved in detecting the emotional significance of environmental stimuli and representing ensuing affective states. A complementary dorsal system, including the hippocampus, dorsal anterior cingulate and dorsolateral prefrontal cortex (DLPFC), appears to have a regulatory role in emotional processing and integrates emotional input with cognitive processes. In particular, the DLPFC is activated by explicit tasks involving emotional content as compared with more implicit tasks (Hariri et al. Reference Hariri, Drabant, Munoz, Kolachana, Mattay, Egan and Weinberger2005).
Neuroimaging studies of acutely depressed patients have consistently reported amygdala hyperactivity (Sheline et al. Reference Sheline, Barch, Donnelly, Ollinger, Snyder and Mintun2001; Fu et al. Reference Fu, Williams, Cleare, Brammer, Walsh, Kim, Andrew, Pich, Williams, Reed, Mitterschiffthaler, Suckling and Bullmore2004; Siegle et al. Reference Siegle, Thompson, Carter, Steinhauer and Thase2007) and DLPFC hypoactivity (Johnstone et al. Reference Johnstone, van Reekum, Urry, Kalin and Davidson2007; Siegle et al. Reference Siegle, Thompson, Carter, Steinhauer and Thase2007; Fales et al. Reference Fales, Barch, Rundle, Mintun, Snyder, Cohen, Mathews and Sheline2008, Reference Fales, Barch, Rundle, Mintun, Mathews, Snyder and Sheline2009) in response to negative emotional stimuli. Taken together, these findings provide evidence for a disrupted cortico-limbic circuitry in major depressive disorder implicating both bottom-up and top-down dysfunction in emotional processing. Recovered unmedicated depressed patients continue to show some evidence of negative emotional biases in behavioural studies (Bhagwagar et al. Reference Bhagwagar, Cowen, Goodwin and Harmer2004; Bhagwagar & Cowen, Reference Bhagwagar and Cowen2008); however, the neural correlates of these enduring changes in emotional processing are unclear; it is not known, for example, whether the pattern of amygdala hyperactivity and DLPFC hypoactivity persists. Treatment of depressed patients with antidepressant medication does apparently normalize amygdala and DLPFC activity but whether such changes persist once medication is withdrawn is not known (Fu et al. Reference Fu, Williams, Cleare, Brammer, Walsh, Kim, Andrew, Pich, Williams, Reed, Mitterschiffthaler, Suckling and Bullmore2004; Fales et al. Reference Fales, Barch, Rundle, Mintun, Mathews, Snyder and Sheline2009).
In the current study we used functional magnetic resonance imaging (fMRI) to assay neural responses in recovered depressed (RD) patients and healthy controls (HC) during an emotional face-matching (fearful and happy faces) paradigm designed to activate both the amygdala and regulatory areas involving the DLPFC (Hariri et al. Reference Hariri, Drabant, Munoz, Kolachana, Mattay, Egan and Weinberger2005; Drabant et al. Reference Drabant, McRae, Manuck, Hariri and Gross2009). We hypothesized that relative to controls, the recovered patients would show increased amygdala activation to fearful versus happy faces but normal DLPFC activity, reflecting regulatory control of limbic activity during clinical recovery.
Method
Participants
We recruited 16 patients who had suffered at least two previous episodes of depression (nine females and seven males, age range 18–57 years) (RD) and 21 never-depressed healthy volunteers (10 females and 11 males, age range 19–63 years) (HC). RD participants had been euthymic and medication free for at least 8 months; three patients were drug naive. Nine patients had a family history of mood disorder. Three RD patients had co-morbid illness; one had social phobia and two subjects had a history of alcohol misuse but not dependence. HC subjects and those with previous depression were similar in terms of their age [mean 32.3 (s.d.=12.9) and 36.2 (s.d.=13.9) years, respectively], gender (χ2=0.007, df=1, p=0.993), Beck Depression Inventory score [2.3 (s.d.=2.4) and 3.5 (s.d.=3.7)] and State-Trait Anxiety Inventory scores [state 31.9 (s.d.=6.6) and 30.75 (s.d.=7.9), trait 31.9 (s.d.=7.3) and 36.7 (s.d.=9.5)] as assessed immediately prior to scanning. All subjects provided written informed consent prior to entry into the study, which was approved by the Oxford Research Ethics Committee, and received an honorarium for their participation. Subjects were briefed on scanner safety and had no contraindications for fMRI examination.
fMRI task design
During fMRI scanning, subjects completed a simple perceptual task involving the matching of fearful and happy facial expressions known to robustly engage the amygdala (Hariri et al. Reference Hariri, Drabant, Munoz, Kolachana, Mattay, Egan and Weinberger2005; Paulus et al. Reference Paulus, Feinstein, Castillo, Simmons and Stein2005; Stein et al. Reference Stein, Simmons, Feinstein and Paulus2007). In this task, nine 30 s blocks of a sensorimotor control task (condition A) were interleaved with eight 30 s blocks of the emotional task [four blocks of fear (condition B) and four blocks of happy (condition C)]. To reduce potential carry-over effects, cycles of alternation between conditions were counterbalanced across subjects. Thus, during the course of the 8.5 min experiment half of the subjects completed the following order: ABACABACABACABACA, the remaining subjects ACABACABACABACABA. During the emotional matching task subjects viewed a trio of faces, all derived from a standard set of pictures of facial affect (Matsumoto & Eckman, Reference Matsumoto and Eckman1988). Faces were presented in a triangular configuration and subjects were asked to select the one of two bottom faces (probes) that expressed the same emotion as the top (target) face. Each emotional block consisted of six trials, presented sequentially for 5 s. During the sensorimotor control task, subjects viewed a trio of geometric shapes (rectangles) in a triangular configuration and selected the one of two bottom shapes that matched the orientation (either vertical or horizontal) of the target (top shape). Each sensorimotor control block consisted of six trials, presented sequentially for 5 s. Stimuli were presented on a personal computer using e-prime (version 1.0; Psychology Software Tools Inc., USA) and projected onto an opaque screen at the foot of the scanner bore, which subjects viewed using angled mirrors. Subject responses were made via an MRI-compatible keypad. Both emotion matching accuracy and reaction times were recorded by e-prime.
fMRI data acquisition
All imaging data were collected using a Varian 3T scanner located at the Oxford Centre for Functional MRI of the Brain (FMRIB; University of Oxford, UK). Functional imaging consisted of 28 T 2*-weighted echo-planar image (EPI) coronal oblique slices [3 mm slice thickness; 64×64 in plane resolution; field of view 192; 258 volumes; repetition time (TR)/echo time (TE)/flip angle=2000 ms/30 ms/60°]. Slices were prescribed perpendicular to the anterior commissure–posterior commissure (AC–PC) line then tilted about 30° in the rostral direction to provide coverage of the amygdala and prefrontal cortex including the anterior cingulate. These parameters were selected to optimize the amygdala signal and have been used extensively in earlier studies assessing amygdala function (Johnstone et al. Reference Johnstone, Somerville, Alexander, Oakes, Davidson, Kalin and Whalen2005; Whalen et al. Reference Whalen, Johnstone, Somerville, Nitschke, Polis, Alexander, Davidson and Kalin2008). The first three EPI volumes in each session were discarded to avoid T 1 equilibration effects.
fMRI data analyses
fMRI data were pre-processed and analysed using freely available software tools within FMRIB's Software Library (FSL, version 4.1; Smith et al. Reference Smith, Jenkinson, Woolrich, Beckman, Behrens, Johansen-Berg, Bannister, De Luca, Drobnjak, Flitney, Niazy, Saunders, Vickers, Zhang, De Stefano, Brady and Mathews2004). Pre-processing included within-subject image realignment (Jenkinson et al. Reference Jenkinson, Bannister, Brady and Smith2002), non-brain removal (Smith, Reference Smith2002), spatial normalization to a study-specific custom template using an affine procedure (Jenkinson & Smith, Reference Jenkinson and Smith2001) and spatial smoothing using a Gaussian kernel (5 mm full-width-half-maximum). The time series was high pass-filtered (to a maximum of 0.008 Hz).
Analyses of data from individual subjects were computed using the general linear model with local autocorrelation correction (Woolrich et al. Reference Woolrich, Ripley, Brady and Smith2001). Two explanatory variables were modelled: ‘fearful faces’ and ‘happy faces’. These explanatory variables were modelled by convolving each emotion block with a haemodynamic response function, using a variant of a γ function (i.e. a normalization of the probability density function of the γ function) with a standard deviation of 3 s and a mean lag of 6 s). In addition, temporal derivatives were included in the model as covariates of no interest to increase statistical sensitivity.
Individual subjects' data were combined at the group level using a full mixed-effects analysis (Woolrich et al. Reference Woolrich, Behrens, Beckmann, Jenkinson and Smith2004). This mixed-effects approach enables generalization of the results beyond the sample of subjects tested. Characterising between-group differences on task-specific brain activity may be confounded by the possibility that changes in activation are actually produced by a shifting baseline, rather than by a change in the brain response to the task itself. With this in mind, all comparisons we report directly contrast fearful with happy facial expressions (i.e. group×emotion interactions) rather than to an under-specified, low level baseline or resting condition. Significant activations were identified using cluster-based thresholding of statistical images with a height threshold of Z=2.3 and a (corrected) spatial extent threshold of p<0.05. Cluster-based thresholding using Gaussian random field theory [as implemented here (Friston et al. Reference Friston, Worsley, Frackowiak, Mazziotta and Evans1994)] is an alternative to voxel-based correction and may be more sensitive to activation under conditions where one can assume multiple contiguous activated voxels (Forman et al. Reference Forman, Cohen, Fitzgerald, Eddy, Mintun and Noll1995). Group statistical analyses also included age as a covariate, which was not found to affect any of the results reported here. For those regions where a significant group×emotion interaction was identified, percentage blood oxygenation level-dependent (BOLD) signal change for each contrast was calculated in order to identify the profile of group effect. For our a priori amygdala regions of interest (ROI), masks were generated using a standard space anatomical atlas (Maldjian et al. Reference Maldjian, Laurienti, Kraft and Burdette2003). Finally, we used simple linear regression to assess the relationship between BOLD response in the amygdala and significant clusters arising from the whole-brain analysis.
Behavioural data analysis
Behavioural data were analysed using a repeated-measures analysis of variance model with group as the between-subject factor and valence as the within-subjects factor implemented in SPSS version 15.0 (SPSS Inc., USA).
Results
Behavioural results
Response accuracy and latency
Due to technical difficulties, response accuracy and latency for one HC subject were not acquired. Subsequent analysis of these measures, therefore, included 20 HC subjects and 16 RD patients. Subjects were highly accurate in their behavioural performance (>80% correct matching identifications) which was not affected by previous depression [F(1, 34)=2.078, p=0.16]. Similarly, latency was not affected by previous depression [F(1, 34)=3.6, p=0.07]. There was a main effect of face type [F(2, 68)=97.48, p<0.001], indicating increased latency when matching fearful faces as opposed to happy faces and shapes.
fMRI results
Amygdala ROI
Consistent with earlier neuroimaging studies, HC subjects showed a significant bilateral amygdala response to fearful faces [independent-samples t test: left amygdala, t(20)=4.098, p=0.001; right amygdala, t(20)=2.561, p=0.019]. Contrary to our original hypothesis, however, we did not observe a significant group×emotion interaction in the left or right amygdala [left, F(1, 35) <1; right, F(1, 35)=1.52, n.s.].
Correlations between amygdalae and DLPFC
There were no significant correlations between the BOLD response (as percentage signal change) in the amygdala and the DLPFC clusters identified in the whole-brain analysis.
Whole-brain analysis
Does previous depression modulate neural processing of fearful facial expressions?
In our exploratory whole-brain analysis we observed significant group×emotion interactions in left middle frontal gyrus [Brodmann's area (BA) 9, x=−36, y=14, z=36, F(1, 35)=27.28, p<0.001], right middle frontal/superior frontal gyrus [BA 8/9, x=26, y=16, z=52, F(1, 35)=10.36, p<0.003] and right caudate [x=16, y=8, z=8, F(1, 35)=10.31, p<0.003]. For details, see Table 1 and Figs. 1–3. Post-hoc analysis showed that in the left DLPFC the BOLD response (as percentage signal change) to fear was significantly greater in the RD compared with the HC participants [t(35)=−3.54, p=0.001]. By contrast, in the same brain region, HC subjects showed a significantly greater BOLD response to happy faces than the RD participants [t(35)=2.58, p=0.01]. In the right DLPFC and right caudate, the significant interaction was driven by greater activation to fear in the RD participants [t(35)=−2.41, p=0.02; t(35)=−3.58, p=0.001, respectively], with no group differences in response to happy faces.
(a) Coronal, sagittal and axial images depicting significantly greater activation in the left dorsolateral prefrontal cortex in recovered depressed patients versus healthy controls for the orthogonal contrast fearful versus happy faces. Images were thresholded at Z>2.3, p<0.05, corrected. Numerals refer to coordinates in Montreal Neurological Institute space. (b) Group×emotion interaction for the above significant cluster. (□), Healthy controls; (▪), recovered depressed patients. Values are means, with standard errors represented by vertical bars.
(a) Coronal, sagittal and axial images depicting significantly greater activation in the right dorsolateral prefrontal cortex in recovered depressed patients versus healthy controls for the orthogonal contrast fearful versus happy faces. Images were thresholded at Z>2.3, p<0.05, corrected. Numerals refer to coordinates in Montreal Neurological Institute space. (b) Group×emotion interaction for the above significant cluster. (□), Healthy controls; (▪), recovered depressed patients. Values are means, with standard errors represented by vertical bars.
(a) Coronal and axial images depicting significantly greater activation in the right caudate in recovered depressed patients versus healthy controls for the orthogonal contrast fearful versus happy faces. Images were thresholded at Z>2.3, p<0.05, corrected. Numerals refer to coordinates in Montreal Neurological Institute space. (b) Group×emotion interaction for the above significant cluster. (□), Healthy controls; (▪), recovered depressed patients. Values are means, with standard errors represented by vertical bars.
Group×emotion interactions in regions of increased activation to fear in recovered depressed patients for the orthogonal contrast fearful versus happy facial expressions
MNI, Montreal Neurological Institute; BA, Brodmann's area.
a Coordinates refer to the position (x, y, z) for the peak voxel in each cluster according to the MNI template.
Discussion
The main finding of the current study is that compared with controls, individuals with a previous history of depression show increased processing of fearful versus happy expressions in the bilateral DLPFC. This effect occurred in the absence of any measurable differences in overt behavioural performance. The DLFPC has been implicated in emotion regulation and visuospatial attention (Drabant et al. Reference Drabant, McRae, Manuck, Hariri and Gross2009). Our data, therefore, suggest recovery from depression may be associated with more effortful regulation of negative stimuli.
There are a number of limitations associated with this study, and these should be taken into consideration when interpreting the results. For example, the inclusion criterion for our study was a personal history of major depression, which has been shown to be a significant risk factor for future episodes (for a review, see Bhagwagar & Cowen, Reference Bhagwagar and Cowen2008). We did not specify that participants should also have an affected first-degree relative, which may have resulted in greater sample homogeneity. Future studies sufficiently powered to examine previous depression with and without family history of depression are required. Another limitation is that the majority of our patients had been previously treated with antidepressant medication. Although these participants had been drug free for long periods of time (see Method) it is possible that long-lasting neuroplastic effects of antidepressant medication could have influenced our imaging findings (Pittenger & Duman, Reference Pittenger and Duman2008).
Overactivity of the amygdala in response to negative emotional stimuli is a common finding in functional imaging studies of patients with acute major depression. In addition, earlier studies suggest that increased amygdala activity to negative stimuli may persist in RD patients. For example, Neumeister et al. (Reference Neumeister, Drevets, Belfer, Luckenbaugh, Henry, Bonne, Herscovitch, Goldman and Charney2006) reported increased amygdala activation to sad (though not fearful) faces in medication-free remitted depressed patients performing an implicit gender discrimination task. We found a similar effect in a different group of RD patients when we employed a subliminal task with masked fearful faces (C. Harmer, unpublished observations). The current study, however, did not reveal an abnormality of amygdala reactivity in RD patients. A possible reason for this is that, relative to the use of implicit or masked emotional faces, the task in the present study was more explicit and therefore more likely to engage the amygdala-regulating effects of the DLPFC.
As noted in the Introduction, patients with acute major depression manifest lowered activity in the DLPFC in response to negative emotional stimuli (Johnstone et al. Reference Johnstone, van Reekum, Urry, Kalin and Davidson2007; Siegle et al. Reference Siegle, Thompson, Carter, Steinhauer and Thase2007; Fales et al. Reference Fales, Barch, Rundle, Mintun, Snyder, Cohen, Mathews and Sheline2008, Reference Fales, Barch, Rundle, Mintun, Mathews, Snyder and Sheline2009). This has been taken as evidence of impaired cognitive control of emotion, particularly when the negative emotional stimulus needs to be suppressed in order to carry out a particular cognitive task. Interestingly, in our study, the recovered patients showed increased activity in the DLPFC relative to controls in response to the fearful stimulus. This could suggest that recovered patients are treating the fearful face as more salient than controls and this is reflected as increased recruitment of the DLFPC to suppress the emotional impact of the fearful stimulus.
Previous studies have shown increased DLPFC activation after antidepressant treatment in depressed patients relative to the depressed state. For example, Fales et al. (Reference Fales, Barch, Rundle, Mintun, Mathews, Snyder and Sheline2009) reported that selective serotonin reuptake inhibitor treatment of depressed patients normalized activation of the DLPFC in response to unattended fear stimuli. Our finding, while in the same direction, indicates that clinical recovery in unmedicated RD patients may be associated with exaggerated DLPFC activation compared with HC subjects in the presence of an aversive stimulus. Interestingly, the regions of increased BOLD activation observed here closely resemble those found during explicit emotion reappraisal tasks (Johnstone et al. Reference Johnstone, van Reekum, Urry, Kalin and Davidson2007).
Also, Drabant et al. (Reference Drabant, McRae, Manuck, Hariri and Gross2009), using the same emotion matching task implemented in the current study, reported a significant positive association between reappraisal [as measured by the reappraisal scale of the Emotion Regulation Questionnaire (Gross & John, Reference Gross and John2003)] and DLPFC activation in healthy volunteers. Taken together (Johnstone et al. Reference Johnstone, van Reekum, Urry, Kalin and Davidson2007; Drabant et al. Reference Drabant, McRae, Manuck, Hariri and Gross2009), these data suggest that during a simple emotional reactivity task, as used here, spontaneous emotion reappraisal ensues, and this is reflected through DLPFC activation. One interpretation of the current finding, namely increased DLPFC activation to fearful faces in RD participants, is that reappraisal is more effortful and greater cortical resources are required to achieve control over amygdala activation. This implies a compensatory change during recovery, mediated in part by the DLFPC, from one of an inability to exert cognitive control over emotion (acutely depressed), to successful, although less efficient than HC subjects, emotion regulation.
We did not observe any significant overall correlations between DLFPC and amygdala activity. However, such effects would be likely to be indirectly mediated (Ghashghaei & Barbas, Reference Ghashghaei and Barbas2002), via ventromedial regions (such as rostral and subgenual cingulate gyri and orbitofrontal cortex) (DeRubeis et al. Reference DeRubeis, Siegle and Hollon2008). Future studies employing rapid event-related designs and task-dependent measures of functional connectivity (e.g. dynamic causal modelling or psychophysiological and physiophysiological interactions) are therefore required to fully explore the role of the DLPFC during explicit emotional processing in RD patients.
We also found that fearful faces were associated with increased activity in the right caudate nucleus in RD patients relative to controls. The caudate, one of the five major nuclei of the basal ganglia, receives inputs from cortical association areas and projects, via the thalamus, to prefrontal regions. The caudate, therefore, is anatomically well placed to have an impact on both cognitive and emotional processes. In acutely depressed patients, structural and resting metabolism studies have produced conflicting results. For example, morphometric studies of the basal ganglia have reported reduced (Krishnan et al. Reference Krishnan, McDonald, Escalona, Doraiswamy, Na, Husain, Figiel, Boyko, Ellinwood and Nemeroff1992; Hickie et al. Reference Hickie, Naismith, Ward, Scott, Mitchell, Schofield, Scimone, Wilhelm and Parker2007) and unchanged (Bremner et al. Reference Bremner, Narayan, Anderson, Staib, Miller and Charney2000; Hannestad et al. Reference Hannestad, Taylor, McQuoid, Payne, Krishnan, Steffens and Macfall2006) caudate volumes. Resting caudate metabolism, as measured by 18F-fluoro-deoxy-glucose, has been shown to be increased (Brody et al. Reference Brody, Saxena, Stoessel, Gillies, Fairbanks, Alborzian, Phelps, Huang, Wu, Ho, Ho, Au, Maidment and Baxter2001), reduced (Mayberg et al. Reference Mayberg, Lozano, Voon, McNeely, Seminowicz, Hamani, Schwalb and Kennedy2005) and unchanged (Baxter et al. Reference Baxter, Phelps, Mazziotta, Guze, Schwartz and Selin1987).
By contrast, functional activation studies in depressed patients have largely reported increased caudate activation. For example, caudate response to sad faces has been shown to be elevated in depressed patients (Fu et al. Reference Fu, Williams, Cleare, Scott, Mitterschiffthaler, Walsh, Donaldson, Suckling, Andrew, Steiner and Murray2008). Chen et al. (Reference Chen, Ridler, Suckling, Williams, Fu, Merlo-Pich and Bullmore2007), using the same sad faces task, reported an inverse relationship between caudate activation and baseline depression severity. Others have demonstrated greater hippocampal–caudate coupling in depressed patients, as compared with HC participants, during successful encoding of negative images (Hamilton & Gotlib, Reference Hamilton and Gotlib2008). Thus, across a number of studies and activation paradigms, caudate activity appears to be elevated in acute depression. The current finding of increased activation to fearful faces in a population of RD individuals suggests that caudate reactivity may be a trait marker of depression vulnerability. Future studies that include populations at increased risk for depression are warranted.
Overall, our findings suggest that clinical recovery in unmedicated depressed patients is associated with an increased activity of structures in fronto-striatal networks believed to play a role in the inhibitory control of emotional responses. We suggest that this represents a compensatory mechanism presumably to attenuate an underlying susceptibility to emotional dysregulation. If this suggestion is correct, it is possible that RD individuals with greater increases in DLPFC activity to negative emotional stimuli would have a better long-term prognosis and may, like the patients we studied, be able to maintain euthymia for long periods in the absence of antidepressant medication. Follow-up studies will be needed to test this prediction.
Acknowledgements
The authors are grateful to all the individuals who took part in this study.
Declaration of Interest
P.J.C. has served as a paid member of advisory boards for Eli Lilly, Servier, Xytis and Wyeth and has received fees for speaking at events organized by Eli Lilly, GlaxoSmithKline, Lundbeck and Servier. C.J.H. has acted as a consultant for P1vital, Lundbeck, Servier and Merck, Sharpe and Dohme and is on the advisory board of P1vital.
