Functional Neuroimaging of Major Depressive Disorder: A Meta-Analysis and New Integration of Baseline Activation and Neural Response Data
Abstract
Objective:
Functional neuroimaging investigations of major depressive disorder can advance both the neural theory and treatment of this debilitating illness. Inconsistency of neuroimaging findings and the use of region-of-interest approaches have hindered the development of a comprehensive, empirically informed neural model of major depression. In this context, the authors sought to identify reliable anomalies in baseline neural activity and neural response to affective stimuli in major depressive disorder.
Method:
The authors applied voxel-wise, whole-brain meta-analysis to neuroimaging investigations comparing depressed to healthy comparison groups with respect to baseline neural activity or neural response to positively and/or negatively valenced stimuli.
Results:
Relative to healthy subjects, those with major depression had reliably higher baseline activity, bilaterally, in the pulvinar nucleus. The analysis of neural response studies using negative stimuli showed greater response in the amygdala, insula, and dorsal anterior cingulate cortex and lower response in the dorsal striatum and dorsolateral prefrontal cortex in individuals with major depressive disorder than in healthy subjects.
Conclusions:
The meta-analytic results support an elegant and neuroanatomically viable model of the salience of negative information in major depressive disorder. In this proposed model, high baseline pulvinar activity in depression first potentiates responding of the brain's salience network to negative information; next, and owing potentially to low striatal dopamine levels in depression, this viscerally charged information fails to propagate up the cortical-striatal-pallidal-thalamic circuit to the dorsolateral prefrontal cortex for contextual processing and reappraisal.
Advancing neural theory of major depressive disorder is of great importance given both that this disorder is predicted to be the most burdensome disease in the world in the 21st century (1) and that it is often unresponsive to conventional pharmacologic intervention (2). Over the past two decades, investigations using functional neuroimaging have provided important insights about its neural substrates. Several of these investigations have supported a neural model of depressed mood in which, within a mutually inhibitory limbic and dorsal cortical architecture, limbic activation overcomes dorsal cortical activation, which reciprocally fails to inhibit limbic activation in major depressive disorder (3); researchers examining neural aspects of emotional reactivity in major depression have proposed similar frameworks (4). Investigators examining network-level connectivity in major depressive disorder have reported evidence of limbic inhibition of dorsal cortical activation, partially supporting corticolimbic inhibition models of major depression (5).
While current neural models of major depressive disorder have been important in both understanding and treating this disorder (6), they are equivocal with respect to the primary neural mechanisms of the illness and do not integrate contemporary knowledge of major neuroanatomical pathways and functional subdivisions of the brain. In formulating a more mechanistic and neuroanatomically viable model of major depression, it is important that we first identify and address issues concerning the data that inform neural theories of the disorder. In this context, it is noteworthy that even the neural findings that are considered most foundational in the neuroscience of depression, such as higher than normal baseline activity in the subgenual anterior cingulate cortex (6) and low activation in the dorsolateral prefrontal cortex (7), have not gone unchallenged (8–10). Further, region-of-interest analytic approaches, which are widely used in clinical neuroimaging (in reference 11, for instance) both to bolster statistical power and to test specific neural models of depression, may nonetheless bias neural models of major depressive disorder by ignoring anomalies in regions that are not of current interest in the study of depression. Thus, we propose that in order to advance neural theory of depression, we must identify, across the whole brain, reliable patterns of anomalous neural response and baseline brain activation in this disorder.
In this context, voxel-wise, whole-brain meta-analysis has the potential to identify robust functional neural findings from the corpus of functional neuroimaging research in major depression; indeed, such an approach has been applied successfully to identifying reliable and interpretable neural responses in anxiety (12) and in emotional responding more generally (13). Further, voxel-wise meta-analysis allows investigators to combine and examine patterns of activation obtained across studies using different types of experimental conditions. For example, findings from studies of neural reactivity to positive stimuli in major depressive disorder can be contrasted with results of investigations of reactivity to negative stimuli, yielding findings that are typically not possible to obtain in a single study. Finally, and particularly important for developing a comprehensive neural theory of major depressive disorder, meta-analysis of whole-brain functional neuroimaging data has the potential to identify anomalous patterns of activation that may not have been accorded adequate attention in previous neural conceptualizations of depression.
While meta-analyses reported by other researchers have produced intriguing results (14), only recently has the number of functional neural investigations of major depression—particularly those using functional MRI (fMRI) to measure neural response—reached a level sufficient to warrant meta-analysis. In the present study we used a whole-brain, voxel-wise meta-analytic technique to identify reliable anomalies in major depressive disorder from both resting-state regional cerebral blood flow (rCBF) and affective-challenge fMRI studies. For affective-challenge fMRI investigations, we examined the results of studies of neural responses to negative relative to positive stimuli in depression. In conducting these analyses, our objective was to develop an empirically informed and rigorous neural model of major depressive disorder.
Method
Overview
We used multilevel kernel density analysis (12, 13) in conducting the present meta-analysis. In this analysis, we first identified functional neuroimaging studies that reported neural functional aberrations among participants diagnosed with major depressive disorder in 1) rCBF as measured with positron emission tomography (PET) or single photon emission computed tomography (SPECT) or 2) neural reactivity (task-based fMRI) to negative and/or positive stimuli. Next we converted results from each comparison of interest in each study into summary binary indicator maps (12). Finally, we merged and conducted voxel-wise comparisons of these maps separately for rCBF and fMRI studies.
Study Selection
We used the ISI Web of Science database to identify functional neuroimaging studies of depression conducted up to Jan. 15, 2011. To identify fMRI studies of major depression, we entered the terms “depress* OR mood” and “functional MRI OR fMRI” in the topic-based search field. To identify PET and SPECT studies of depression, we entered “depress* OR mood” and “positron emission tomography OR PET OR single photon emission OR SPECT OR cerebral-blood flow” in the topic-based search field. Among the studies identified—1,448 fMRI studies and 5,147 rCBF studies—we kept for subsequent meta-analysis only those that were conducted with adults between the ages of 18 and 60 who were diagnosed with a current episode of primary unipolar depression as assessed with the Structured Clinical Interview for DSM (SCID) (15) or with the International Statistical Classification of Diseases and Related Health Problems (ICD) (16). In addition, we required that studies took a whole-brain analysis approach in which the investigators reported coordinates of all suprathreshold clusters in comparing neural activity in subjects with major depressive disorder to that in a comparison group of demographically matched, SCID- or ICD-assessed nondepressed individuals. We also searched the reference lists of studies that met these criteria to ensure that our original web-based search was comprehensive. For PET and SPECT studies to be included, we required that they used a tracer amenable to estimating brain activity: [18F]FDG or [15O]H2O for PET and [99mTc]ECD for SPECT. For fMRI studies to be included, we added the criterion that there was at least one condition in the study in which activation during processing of affectively valenced information was either contrasted to a neutral stimulus condition or parametrically varied. Finally, we required that typical regression analyses, i.e., those using regression to estimate the extent or “height” of the blood-oxygen-level-dependent response, were conducted on fMRI data. Fourteen rCBF studies (6, 17–29) and 24 fMRI studies (7, 8, 30–51) met our inclusion criteria; 5,133 rCBF and 1,424 fMRI studies identified in our original search were excluded for not meeting one or more of these criteria. While the fMRI studies included in this meta-analysis collectively used a broad range of negatively and positively valenced affective challenges and stimuli, we have combined these studies given that valence is a fundamental component of emotions and affective stimuli in affective circumplex models of emotion (see, for instance, reference 52). Thus, examining differential neural response in major depression to stimulus valence, regardless of its specific form, can inform us from a theoretical perspective about emotional functioning in major depressive disorder.
Computing the Meta-Analytic Statistic
As in other studies employing multilevel kernel density analysis (for instance, 12), the data for the present meta-analysis were coordinates that were reported in three-dimensional standard brain space of peak activation differences between groups. For each study that met the inclusion criteria, we constructed at 1-mm isotropic voxel resolution an indicator map in Talairach space (53) as represented in the AFNI statistical package (54). We converted coordinates reported in the standard space of the Montreal Neurological Institute to Talairach space by using the MATLAB script mni2tal.m (http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach). Each binary indicator map, I, was composed of spheres (value=1) of 10-mm radius—a size chosen to account for limitations in spatial resolution in functional neuroimaging data while maintaining adequate sensitivity to detect neural anomalies in smaller structures—centered at reported locations of peak between-groups difference; in instances of overlap between spheres, we valued the region of intersection at 1.
We merged indicator maps into six meta-analytic statistical maps: four from fMRI studies and two from PET/SPECT studies. The fMRI meta-analytic maps showed regions where activation in depressed subjects was greater than in comparison subjects in response to negative stimuli (20 contrasts, 92 coordinates) and positive stimuli (10 contrasts, 40 coordinates) and where activation was greater in comparison subjects than in depressed subjects in response to negative stimuli (19 contrasts, 90 coordinates) and positive stimuli (14 contrasts, 76 coordinates). The PET/SPECT meta-analytic maps indicated where activation was greater in depressed subjects than in comparison subjects (10 contrasts, 37 coordinates) and where it was greater in comparison subjects than in depressed subjects (11 contrasts, 37 coordinates). Indicator maps from N studies per category were merged into meta-analytic statistical maps, where the meta-analytic statistic (^Pv) at each voxel, v, was calculated as
[See PDF for formula]
where wn is the square root of the number of subjects in the full study group (depressed plus comparison subjects) of the nth of N studies. Weighting each map in this fashion meant that studies with larger groups and presumably more reliable results were accorded more weight in the meta-analytic statistic. This approach has several advantages relative to other meta-analytic methods used to examine patterns of brain activation in depression, such as those that use activation likelihood estimation techniques (for instance, reference 14). First, because we treated comparison maps within studies as random effects, no single map could contribute disproportionately to our meta-analytic results. Second, weighting by group size increases accuracy without introducing the complexity of z-score-based weighting for different types of analyses. Finally, relative to methods that incorporate z-score weighting, such as activation likelihood estimation, the approach taken in the present meta-analysis is better suited to examining the replicability of neural responding across studies.
Comparing Meta-Analytic Maps
We subtracted meta-analytic maps showing regions where activation was greater in comparison subjects than in depressed subjects from those showing regions in which activation was greater in depressed subjects for both rCBF studies (21 contrasts, 74 coordinates) and negative- and positive-valence fMRI studies (39 contrasts, 182 coordinates; and 24 contrasts, 106 coordinates, respectively). Further, for the fMRI studies, to test whether significant differences from these comparisons were valence specific, we also subtracted positive from negative meta-analytic maps for comparisons showing where activation was greater in depressed subjects and those showing where activation was greater in comparison subjects (63 contrasts, 288 coordinates). Null-hypothesis distributions for each comparison were calculated from 10,000 Monte Carlo simulations performed in MATLAB (www.mathworks.com) by using relevant variables from the included studies (group size, number of regions reported) but with the assumption of a random distribution of peaks in a standardized brain (AFNI's TT_N27+tlrc). We used the AFNI program AlphaSim to set the significance threshold for each comparison at α=0.05, family-wise-error-rate (FWE) controlled for comparison of multiple voxels across the whole brain. AlphaSim takes as input the parameters used in neuroimaging analysis (e.g., voxel-wise statistical threshold, size of mask over which statistical calculations are performed) and determines, using Monte Carlo simulation, a cluster threshold necessary to assure FWE control. We used a voxel-wise threshold of α=0.001 (resulting in a cluster threshold, k, of 177 voxels) for our fMRI meta-analysis and α=0.005 (k=384 voxels) for our meta-analysis of PET data. We used a more liberal voxel-wise threshold—and, therefore, a more stringent cluster size threshold—for the PET meta-analysis given the lower spatial resolution of PET data and the consequent lower incidence of intersection of PET-derived statistical maps. In addition, to address potential outlier effects and to ensure that significant meta-analytic findings are representative of results from a range of studies, we established that at least three studies had to contribute to a given meta-analytic finding for it to be reported here. Finally, we note that our null-hypothesis distributions were created under the assumption of independence across studies. Several studies that met our inclusion criteria were conducted in the same laboratory (fMRI: Cynthia Fu [7, 35, 36, 43], Thomas Frodl [34, 46], Ian Gotlib [32, 33, 37, 39], Mary Phillips [8, 48]; rCBF: Lewis Baxter [19, 27], Omer Bonne [24, 25]) or included more than one comparison per valence. Therefore, we checked significant meta-analytic findings to ensure that no study was represented more than once. We took the following exclusionary precautions to ensure independence across analyses from the same study: 1) if results from both parametric and neutral stimulus contrasts were reported, only results from the presumably more reliable parametric analyses were included; 2) if a study reported results from contrasting two levels of an emotion variable to a neutral stimulus, then results from only the comparison of the highest-level emotion and the neutral stimulus were included; and 3) if a study reported results from contrasting an experimental condition to more than one control condition, we included results from only one comparison, i.e., that which included the most coordinates. In cases where different studies from the same laboratory contributed to a significant finding, we contacted investigators to ensure that the study groups were independent.
Results
Characteristics of Included Studies
Table 1 and Table 2 present, for the included rCBF and fMRI studies, respectively, the number of subjects, gender composition, and proportion of medicated subjects in each group of depressed participants. Further, for fMRI studies we report the task that was used and the contrasts that were included in the meta-analysis.
Number of Subjects in Group | Characteristics of Depressed Subjects | Imaging Technique | ||||
---|---|---|---|---|---|---|
Study | Major Depressive Disorder | Comparison | Medicated (%) | Female (%) | Method | Tracer |
Aihara et al. (17) | 24 | 23 | 0 | 63 | PET | [18F]FDG |
Bench et al. (18) | 33 | 23 | 58 | 36 | PET | [15O]H2O |
Brody et al. (19) | 24 | 16 | 0 | 54 | PET | [18F]FDG |
Drevets et al. (20) | 13 | 33 | 0 | 54 | PET | [15O]H2O |
Germain et al. (21) | 12 | 13 | 0 | 83 | PET | [18F]FDG |
Kennedy et al. (22) | 13 | 24 | 0 | 0 | PET | [18F]FDG |
Kimbrell et al. (23) | 38 | 37 | 0 | 66 | PET | [18F]FDG |
Kohn et al. (24) | 33 | 25 | 0 | 58 | SPECT | [99mTc]ECD |
Krausz et al. (25) | 10 | 10 | 0 | 90 | SPECT | [99mTc]ECD |
Mayberg et al. (6) | 6 | 5 | 100 | 50 | PET | [15O]H2O |
Périco et al. (26) | 15 | 15 | 0 | 80 | SPECT | [99mTc]ECD |
Saxena et al. (27) | 27 | 17 | 0 | 52 | PET | [18F]FDG |
Skaf et al. (28) | 9 | 12 | 0 | 44 | SPECT | [99mTc]ECD |
Videbech et al. (29) | 42 | 47 | 95 | 71 | PET | [15O]H2O |
Studies Measuring Differences in Baseline Regional Cerebral Blood Flow Between Persons With Major Depressive Disorder and Healthy Comparison Subjects
Number of Subjects in Group | Characteristics of Depressed Subjects | Included Contrasts | |||||
---|---|---|---|---|---|---|---|
Study | Major Depressive Disorder | Comparison | Medicated (%) | Female (%) | Affective Challenge | Negative Valence | Positive Valence |
Abler et al. (30) | 12 | 12 | 100 | 100 | Anticipating and viewing negative, neutral, and positive pictures | Anticipating negative versus neutral; viewing negative versus neutral | Viewing positive versus neutral |
Bär et al. (31) | 13 | 13 | 0 | 100 | Experiencing variable-intensity thermal pain | Parametric thermal pain response | None |
Canli et al. (32) | 15 | 15 | 47 | 80 | Performing lexical decision task for neutral, happy, sad, and threat-related words | Sad versus neutral; physical threat versus neutral; social threat versus neutral | Happy versus neutral |
Cooney et al. (33) | 14 | 14 | 43 | 57 | Thinking about personal traits, abstract ideas, and physical objects | Personal traits versus abstract ideas | None |
Frodl et al. (34) | 12 | 12 | 67 | 58 | Emotion (explicit) or gender (implicit) matching of negative faces to target; shape-matching control | Explicit versus control; implicit versus control | None |
Fu et al., 2004 (36) | 19 | 19 | 0 | 68 | Gender identification of variable-intensity sad faces | Parametric sad face response | None |
Fu et al., 2007 (35) | 19 | 19 | 0 | 68 | Gender identification of variable-intensity happy faces | None | Parametric happy face response |
Fu et al., 2008 (7) | 16 | 16 | 0 | 81 | Gender identification of variable-intensity sad faces | Parametric sad face response | None |
Gotlib et al. (37) | 18 | 18 | 50 | 72 | Viewing happy, sad, and neutral faces | Sad versus neutral | Happy versus neutral |
Herwig et al. (38) | 14 | 14 | 100 | 57 | Viewing negative and neutral pictures | Negative versus neutral | None |
Keedwell et al. (8) | 12 | 12 | 92 | 67 | Remembering happy, sad, and neutral experiences | Sad versus neutral | Happy versus neutral |
Knutson et al. (39) | 14 | 12 | 0 | 64 | Anticipating and experiencing monetary gains, losses, and neutral outcomes | None | Anticipating gain versus neutral outcome; experiencing gain versus neutral outcome |
Kumari et al. (40) | 6 | 6 | 100 | 100 | Viewing positive, negative, and neutral picture-caption pairs | Negative versus neutral | Positive versus neutral |
Lawrence et al. (41) | 9 | 11 | 100 | 44 | Viewing high- and low-intensity happy, sad, and fearful faces; neutral-face control | High-intensity fearful versus neutral; high-intensity sad versus neutral | High-intensity happy versus neutral |
Mitterschiffthaler et al., 2003 (42) | 7 | 7 | 100 | 100 | Viewing positive and neutral images | None | Positive versus neutral |
Mitterschiffthaler et al., 2008 (43) | 17 | 17 | 0 | 82 | Identifying colors of sad and neutral words | Sad versus neutral | None |
Osuch et al. (44) | 16 | 17 | 6 | 69 | Listening to endorsed favorite and neutral music | None | Favorite versus neutral |
Pizzagalli et al. (45) | 30 | 31 | 0 | 50 | Anticipating and experiencing monetary gains, losses, and neutral outcomes | Anticipating loss versus neutral outcome; experiencing loss versus neutral outcome | Anticipating gain versus neutral outcome; experiencing gain versus neutral outcome |
Scheuerecker et al. (46) | 13 | 15 | 0 | 77 | Emotion (explicit) or gender (implicit) matching of negative faces to target; shape-matching control | Explicit plus implicit versus control | None |
Strigo et al. (47) | 15 | 15 | 0 | 80 | Anticipating and experiencing painful and nonpainful thermal stimuli | Anticipating pain versus nonpain; experiencing pain versus nonpain | None |
Surguladze et al. (48) | 16 | 14 | 100 | 38 | Viewing variable-intensity happy and sad faces | Parametric sad face response | Parametric happy face response |
Townsend et al. (49) | 15 | 15 | 0 | 40 | Matching negative faces to facial emotion or word target; shape-matching control | Face matching versus control | None |
Wang et al. (50) | 19 | 20 | 58 | 63 | Oddball target detection with sad or neutral picture as background | Sad versus neutral | None |
Yang (51) | 10 | 10 | 0 | 100 | Viewing erotic and neutral film clips | None | Erotic versus neutral |
Studies Using Functional Magnetic Resonance Imaging to Estimate Neural Response to Negative and/or Positive Stimuli in Persons With Major Depressive Disorder and Healthy Comparison Subjects
rCBF Meta-Analysis
Our meta-analysis of resting-state PET and SPECT studies showed reliably greater activity in the depressed participants, relative to the comparison subjects, in both the right and left pulvinar nuclei of the thalamus (FWE-controlled p<0.05); see Figure 1. The data supplement accompanying the online version of this article contains otherwise significant regional cerebral blood flow anomalies in major depression supported by only two studies.
fMRI Meta-Analysis
Negative stimuli.
Our analysis showed that individuals diagnosed with major depressive disorder had reliably greater responses to negative stimuli than did the comparison subjects in the amygdala, dorsal anterior cingulate cortex, and insula/superior temporal gyrus; we also found greater activation in response to negative stimuli in depressed individuals in precentral and middle temporal gyri. It is important to note that these effects were also evident in depressed participants when the responses to negative stimuli were contrasted with responses to positive stimuli, indicating that these aberrant responses in depression were specific to negative stimuli. We also found reliably lower activation in response to negative stimuli in depressed participants in the dorsolateral prefrontal cortex bilaterally and in the caudate body (dorsal striatum); however, only the lower response in the right dorsolateral prefrontal cortex in the depressed subjects was found to be specific to negative stimuli. All findings were significant at FWE-controlled p<0.05; see Figure 2 for a graphical rendering of these results.
Positive stimuli.
Given that aberrant responses to positive stimuli in major depression do not figure prominently into the framework of this article, we present meta-analytic results relating to positive stimuli in the online supplement to this article.
Discussion
In the present study, we applied whole-brain, voxel-wise meta-analysis to a large corpus of functional neuroimaging studies of baseline activation and neural response in major depressive disorder. Our analysis of rCBF studies showed higher than normal baseline activity, bilaterally, in the pulvinar nucleus in depression. Further, our examination of fMRI studies showed that depressed individuals had greater responses than healthy comparison subjects in the amygdala, dorsal anterior cingulate cortex, and insula and lower responses to negative stimuli in the dorsolateral prefrontal cortex and dorsal striatum.
High Baseline Activity of Pulvinar Nucleus in Depression
In this meta-analysis we found high baseline activation in major depressive disorder in the pulvinar, a large (2,200 mm3) nucleus of the thalamus. In light of evidence of monosynaptic projections from the pulvinar to the amygdala (55), early formulations of pulvinar function proposed that this structure is part of a fast, unconscious processing stream for priming behavioral response in the presence of threat (56). A more recent conceptualization of pulvinar function posits, instead, that this structure plays a key role in emotional attention and awareness (57). This position is supported by the finding that the primary neural drivers of pulvinar activity are cortical, and not subcortical, structures; indeed, both the insula and dorsal anterior cingulate cortex have strong bidirectional connectivity with the pulvinar nucleus (57, 58). Also consistent with the formulation of a central role for the pulvinar in emotional attention is evidence that this structure is necessary for feature binding (59), a key function of attention in integrating the contributions of distinct cell ensembles that code for different perceptual features (60). This model has been confirmed by a recent fMRI study examining pulvinar responding to stimuli that were either affective or neutral and were either detected or undetected. Consistent with the model just described, this study found pulvinar response only to detected affective stimuli (61). Given this role of the pulvinar nucleus in emotional attention and awareness and its connectivity with the amygdala, insula, and dorsal anterior cingulate cortex, we propose that high baseline pulvinar activation in major depressive disorder acts to potentiate response in neural structures that subserve orienting to, awareness of, and responding to affective information.
Overreactive Salience Network and Low Response of Dorsolateral Prefrontal Cortex and Dorsal Striatum to Negative Information in Depression
We found in this meta-analysis that the amygdala, dorsal anterior cingulate cortex, and insula are selectively overreactive to negative stimuli in major depressive disorder; these structures are all prominent nodes in the salience network of the brain (62). In this network, the amygdala is considered essential for sensory tuning to enable adaptive responding (63), whereas the dorsal anterior cingulate cortex is crucial for engaging motor and visceral efferents that constitute the sympathetic division of the autonomic nervous system (64). The insula, which shares extensive bidirectional connectivity with both the amygdala (65) and dorsal anterior cingulate cortex (66), has been postulated to subserve conscious awareness of interoceptive states (67). We also found in this meta-analysis that depressed persons have less response to negative stimuli in the dorsolateral prefrontal cortex than do healthy comparison subjects. The dorsolateral prefrontal cortex is a key component in the brain's executive network (68) and has been implicated in a variety of cognitive control processes, including allocation of attention (69) and interference resolution (70). Complementing its role in subserving cognitive processes is its role in the representation of information (71). Often in tandem with the dorsal striatum—found here to exhibit a lower than normal response to negative information in depression—the dorsolateral prefrontal cortex undergirds representation and maintenance of information (72, 73). The representational capacities of the dorsolateral prefrontal cortex and dorsal striatum have also been observed as components of more complex processes, such as the contextualization and reappraisal of information (71, 74). In the following, we present a model of how low responses of the dorsolateral prefrontal cortex and dorsal striatum to negative stimuli, as well as overreactivity of the salience network to negative information, in major depression might contribute to the pathophysiology of this disorder.
Patterns of Abnormal Neural Responding to Negative Stimuli in Depression: The Ascending Spiral Model of the Cortical-Striatal-Pallidal-Thalamic Circuit
The amygdala and dorsal anterior cingulate cortex, found in this meta-analysis to be overly responsive to negative stimuli in major depressive disorder, are primary input structures of the limbic subdivision of the cortical-striatal-pallidal-thalamic circuit. The basic architecture of this circuit has been identified from a large body of anterograde and retrograde neuronal tracing data from animals, and it provides a general architecture for the flow of information in the brain. In this architecture, information from cortical and allocortical structures travels to the striatum, then to the globus pallidus, then on to the thalamus, and finally, back to the cortex (75). Moreover, investigators have identified segregation within the circuit, in which functionally distinct inputs from limbic and cognitive subdivisions largely maintain segregation through the striatum, globus pallidus, and thalamus (76). Haber and colleagues (77) further refined this model by integrating a large body of anterograde and retrograde tracing data; these researchers posit an ascending spiral configuration of the cortical-striatal-pallidal-thalamic circuit in which information directed to limbic structures follows an upward spiraling trajectory toward dorsal cortical structures, such as the dorsolateral prefrontal cortex. It is important to note here that this upward spiraling trajectory is mediated through dopamine-dependent interactions between the striatum and substantia nigra (77). This ascending trajectory suggests that emotional significance is mapped first to sensory stimuli and that only later are higher-level attributes, such as interpretations and reappraisals, ascribed to these stimuli.
Integrative Neural Model of Heightened Salience of Negative Stimuli in Depression
For over three decades, theorists have postulated that negative cognitive biases play a critical role in the onset and maintenance of major depressive disorder (78, 79). The findings of this meta-analysis concerning baseline activity and neural response to affective stimuli cohere into a remarkably clear and neuroanatomically viable model of this greater salience of negative stimuli in depression. Given the conceptualizations we have presented of the pulvinar nucleus, salience network, and dorsal cortical function and connectivity, as well as the dopamine-mediated ascending spiral trajectory of information through the cortical-striatal-pallidal-thalamic circuit, the present meta-analytic findings support a two-step model of enhanced processing of negative information in depression. We propose, first, that through monosynaptic projections to the amygdala, dorsal anterior cingulate cortex, and insula, abnormally high baseline activation in the pulvinar nucleus in depression potentiates responding of these components of the brain's salience network to negative stimuli; this processing may be sustained both by mutually excitatory connections between the insula and amygdala and between the insula and dorsal anterior cingulate cortex and by connections from the insula and dorsal anterior cingulate cortex back to the pulvinar nucleus (Figure 3A). Next, we propose that sensory data fail to propagate to the dorsal caudate and dorsolateral prefrontal cortex (Figure 3B), leading depressed individuals to fail to ascribe emotionally attenuating appraisals and contextual information to viscerally overcharged data from limbic structures. Given that nigrostriatal dopamine projections are instrumental in conducting information up the cortical-striatal-pallidal-thalamic circuit to the dorsolateral prefrontal cortex, both in vivo (80, 81) and postmortem (82) findings of low striatal dopamine activity in major depressive disorder suggest a potential mechanism by which data transfer through the dorsal caudate to the dorsolateral prefrontal cortex is attenuated in major depressive disorder. Recent meta-analytic results from Delaveau and colleagues (83) are important to consider in the present context, as they showed that key aspects of this model—overreaction of the amygdala and insula and underreaction of the dorsolateral prefrontal cortex in depression—resolve with antidepressant treatment, indicating that the present model accounts for state- as opposed to trait-related neural anomalies in major depressive disorder.
This proposed model is distinct in important ways from current cortical-limbic inhibition models of major depressive disorder. First, this model posits a unidirectional, as opposed to a cyclical, pattern of neural influence between limbic and dorsal cortical structures in depression. Consistent with this formulation are the findings in our recent study of depressed participants, providing strong evidence of greater than normal limbic-to-dorsal cortical inhibition but no evidence of lower dorsal cortical-to-limbic inhibition (84) in depression. Second, our model posits a primary neural mechanism, exaggerated tonic pulvinar activation, as critical to initiating a bias toward the processing of negative information in major depressive disorder. Finally, this model accounts for patterns of limbic overresponsiveness and dorsal striatal and dorsal cortical underresponsiveness in depression within a well-validated and integrative neural processing architecture: the upward spiral model of the cortical-striatal-pallidal-thalamic circuit (77). This aspect of the model also accounts well for the lack of dorsal cortical responding—as opposed to dorsal cortical inhibition—that is often observed in major depression (85).
While this proposed model of negative information processing in major depressive disorder offers an integrative account of findings from a large body of neuroimaging research, it nevertheless leaves an important question unaddressed. The finding from our meta-analysis of excess baseline pulvinar activation in depression is primary in our proposed model. However, the reason for this difference is not clear. In this context, it is important to note that genetics may contribute to the anomalies in pulvinar activation reported in major depressive disorder. Indeed, a postmortem brain volumetry study (86) showed higher pulvinar volume in individuals who were homozygous for the short allele in the promoter region of the serotonin transporter gene (5-HTTLPR), a polymorphism that has been associated with greater risk for major depression (87, 88).
Limitations
We should note five limitations of the present study. First, because only a few neuroimaging studies that met the criteria for our meta-analysis involved participants with remitted depression, it was not possible in this meta-analysis to determine whether the anomalies we identified in major depressive disorder are related to the state of depression or are trait-related characteristics of individuals at risk for this disorder. However, the recent meta-analytic finding that amygdala and insula overactivation and dorsolateral prefrontal cortex underreaction normalize after pharmacotherapy for depression (83) indicates that the present meta-analytic results represent state- as opposed to trait-related aspects of neural functioning in depression. Second, while the present meta-analysis was adequately powered to detect reliable functional neural anomalies in major depression, it was not adequately powered to determine the effects of extraexperimental variables, such as psychotropic medication use by the depressed participants. While the meta-analysis of Delaveau and colleagues (83) indicates that key neural anomalies found in the present analysis actually normalize with use of psychotropic medication, our findings cannot speak to the issue of medication effects. Third, we did not include in this meta-analysis data from groups experiencing forms of psychopathology other than major depression. Thus, while we were able to provide a comprehensive neural characterization of major depressive disorder, the issue of the specificity of these findings to major depression remains to be addressed. Fourth, our meta-analysis of resting-state activation studies used data from three different functional neuroimaging modalities; noise introduced by including heterogeneous modalities may have decreased the sensitivity of our analysis. Similarly, noise introduced by the variety of affective challenge techniques used in the fMRI studies that were included in our meta-analysis may have affected the power of this analysis. Finally, because in this meta-analysis we set criteria for finding the strongest and most reliable findings in the literature concerning functional neuroimaging of depression, it is possible that we failed to detect weaker findings that are nevertheless important in understanding neural functioning in major depressive disorder. Indeed, results from previous meta-analytic investigations of major depression (14) indicated a broader array of neural anomalies in depression—both similar to those reported in the present analysis (e.g., greater amygdalar response to negative stimuli in depressed subjects than in healthy subjects) and different from the current findings (e.g., greater bilateral inferior frontal gyrus response to negative stimuli in depression). Certainly, these results should be accorded serious consideration in neural conceptualizations of major depressive disorder.
Summary and Future Directions
From the present meta-analytic synthesis, we formulated what we believe is an elegant and viable model of increased salience of negative information in major depressive disorder. We posit that through monosynaptic projections to the amygdala, dorsal anterior cingulate cortex, and insula, heightened baseline activity in the pulvinar nucleus in depression potentiates responding of these input nodes to the affective subdivision of the cortical-striatal-pallidal-thalamic circuit. We propose further that, potentially due to deficient striatal dopamine activity in major depression, there is a failure at nigrostriatal relays to propagate information to the dorsal striatum and dorsolateral prefrontal cortex; thus, attributions and appraisals that are typically initiated by the dorsolateral prefrontal cortex to reduce the impact of negative stimuli are not applied to viscerally charged sensory data in major depressive disorder.
Theorists have contended that biased processing of negative information plays a critical role in the etiology and maintenance of major depression (89). Moreover, the model supported by the current meta-analytic findings specifies crucial and independent neural anomalies underlying the enhanced processing of negative information in depression—specifically, high baseline pulvinar activation and low striatal dopamine levels. Thus, this model suggests that novel neural-level interventions, including neuroregulatory therapies for down-modulating pulvinar activity and pharmacotherapies for bolstering striatal dopamine availability, may be effective in the treatment of major depressive disorder.
1.
2.
3. : Limbic-cortical dysregulation: a proposed model of depression. J Neuropsychiatry Clin Neurosci 1997; 9:471–481Crossref, Medline, Google Scholar
4. : Increased amygdala and decreased dorsolateral prefrontal bold responses in unipolar depression: related and independent features. Biol Psychiatry 2007; 61:198–209Crossref, Medline, Google Scholar
5. : Investigating neural primacy in major depressive disorder: multivariate Granger causality analysis of resting-state fMRI time-series data. Mol Psychiatry 2011; 16:763–772Crossref, Medline, Google Scholar
6. : Deep brain stimulation for treatment-resistant depression. Neuron 2005; 45:651–660Crossref, Medline, Google Scholar
7. : Neural responses to sad facial expressions in major depression following cognitive behavioral therapy. Biol Psychiatry 2008; 64:505–512Crossref, Medline, Google Scholar
8. : The neural correlates of anhedonia in major depressive disorder. Biol Psychiatry 2005; 58:843–853Crossref, Medline, Google Scholar
9. : Functional but not structural subgenual prefrontal cortex abnormalities in melancholia. Mol Psychiatry 2004; 9:325, 393–405Crossref, Google Scholar
10. : An fMRI study of prefrontal brain activation during multiple tasks in patients with major depressive disorder. Hum Brain Mapp 2008; 29:490–501Crossref, Medline, Google Scholar
11. : Neural substrates of increased memory sensitivity for negative stimuli in major depression. Biol Psychiatry 2008; 63:1155–1162Crossref, Medline, Google Scholar
12. : Functional neuroimaging of anxiety: a meta-analysis of emotional processing in PTSD, social anxiety disorder, and specific phobia. Am J Psychiatry 2007; 164:1476–1488Link, Google Scholar
13. : Valence, gender, and lateralization of functional brain anatomy in emotion: a meta-analysis of findings from neuroimaging. Neuroimage 2003; 19:513–531Crossref, Medline, Google Scholar
14. : A meta-analytic study of changes in brain activation in depression. Hum Brain Mapp 2008; 29:683–695Crossref, Medline, Google Scholar
15. : Structured Clinical Interview for DSM-III-R (SCID). New York, New York State Psychiatric Institute, Biometrics Research, 1985Google Scholar
16. : Progress toward achieving a common language in psychiatry: results from the field trial of the clinical guidelines accompanying the WHO classification of mental and behavioral disorders in ICD-10. Arch Gen Psychiatry 1993; 50:115–124Crossref, Medline, Google Scholar
17. : HPA axis dysfunction in unmedicated major depressive disorder and its normalization by pharmacotherapy correlates with alteration of neural activity in prefrontal cortex and limbic/paralimbic regions. Psychiatry Res 2007; 155:245–256Crossref, Medline, Google Scholar
18. : The anatomy of melancholia—focal abnormalities of cerebral blood-flow in major depression. Psychol Med 1992; 22:607–615Crossref, Medline, Google Scholar
19. : Brain metabolic changes associated with symptom factor improvement in major depressive disorder. Biol Psychiatry 2001; 50:171–178Crossref, Medline, Google Scholar
20. : A functional anatomical study of unipolar depression. J Neurosci 1992; 12:3628–3641Crossref, Medline, Google Scholar
21. : Diurnal variation in regional brain glucose metabolism in depression. Biol Psychiatry 2007; 62:438–445Crossref, Medline, Google Scholar
22. : Changes in regional brain glucose metabolism measured with positron emission tomography after paroxetine treatment of major depression. Am J Psychiatry 2001; 158:899–905Link, Google Scholar
23. : Regional cerebral glucose utilization in patients with a range of severities of unipolar depression. Biol Psychiatry 2002; 51:237–252Crossref, Medline, Google Scholar
24. : Tc-99m-HMPAO SPECT study of cerebral perfusion after treatment with medication and electroconvulsive therapy in major depression. J Nuclear Med 2007; 48:1273–1278Crossref, Medline, Google Scholar
25. : Brain SPECT study of common ground between hypothyroidism and depression. Int J Neuropsychopharmacology 2007; 10:99–106Crossref, Medline, Google Scholar
26. : Relationship between regional cerebral blood flow and separate symptom clusters of major depression: a single photon emission computed tomography study using statistical parametric mapping. Neurosci Lett 2005; 384:265–270Crossref, Medline, Google Scholar
27. : Cerebral metabolism in major depression and obsessive-compulsive disorder occurring separately and concurrently. Biol Psychiatry 2001; 50:159–170Crossref, Medline, Google Scholar
28. : Psychotic symptoms in major depressive disorder are associated with reduced regional cerebral blood flow in the subgenual anterior cingulate cortex: a voxel-based single photon emission computed tomography (SPECT) study. J Affect Disord 2002; 68:295–305Crossref, Medline, Google Scholar
29. : The Danish PET/depression project: PET findings in patients with major depression. Psychol Med 2001; 31:1147–1158Crossref, Medline, Google Scholar
30. : Anticipation of aversive stimuli activates extended amygdala in unipolar depression. J Psychiatr Res 2007; 41:511–522Crossref, Medline, Google Scholar
31. : Increased prefrontal activation during pain perception in major depression. Biol Psychiatry 2007; 62:1281–1287Crossref, Medline, Google Scholar
32. : Brain activation to emotional words in depressed vs healthy subjects. Neuroreport 2004; 15:2585–2588Crossref, Medline, Google Scholar
33. : Neural correlates of rumination in depression. Cogn Affect Behav Neurosci 2010; 10:470–478Crossref, Medline, Google Scholar
34. : Neuronal correlates of emotional processing in patients with major depression. World J Biol Psychiatry 2009; 10:202–208Crossref, Medline, Google Scholar
35. : Neural responses to happy facial expressions in major depression following antidepressant treatment. Am J Psychiatry 2007; 164:599–607Link, Google Scholar
36. : 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; 61:877–889Crossref, Medline, Google Scholar
37. : Subgenual anterior cingulate activation to valenced emotional stimuli in major depression. Neuroreport 2005; 16:1731–1734Crossref, Medline, Google Scholar
38. : Neural correlates of 'pessimistic' attitude in depression. Psychol Med 2010; 40:789–800Crossref, Medline, Google Scholar
39. : Neural responses to monetary incentives in major depression. Biol Psychiatry 2008; 63:686–692Crossref, Medline, Google Scholar
40. : Neural abnormalities during cognitive generation of affect in treatment-resistant depression. Biol Psychiatry 2003; 54:777–791Crossref, Medline, Google Scholar
41. : Subcortical and ventral prefrontal cortical neural responses to facial expressions distinguish patients with bipolar disorder and major depression. Biol Psychiatry 2004; 55:578–587Crossref, Medline, Google Scholar
42. : Neural response to pleasant stimuli in anhedonia: an fMRI study. Neuroreport 2003; 14:177–182Crossref, Medline, Google Scholar
43. : Neural basis of the emotional Stroop interference effect in major depression. Psychol Med 2008; 38:247–256Crossref, Medline, Google Scholar
44. : Brain activation to favorite music in healthy controls and depressed patients. Neuroreport 2009; 20:1204–1208Crossref, Medline, Google Scholar
45. : Reduced caudate and nucleus accumbens response to rewards in unmedicated individuals with major depressive disorder. Am J Psychiatry 2009; 166:702–710Link, Google Scholar
46. : Orbitofrontal volume reductions during emotion recognition in patients with major depression. J Psychiatry Neurosci 2010; 35:311–320Crossref, Medline, Google Scholar
47. : Association of major depressive disorder with altered functional brain response during anticipation and processing of heat pain. Arch Gen Psychiatry 2008; 65:1275–1284Crossref, Medline, Google Scholar
48. : A differential pattern of neural response toward sad versus happy facial expressions in major depressive disorder. Biol Psychiatry 2005; 57:201–209Crossref, Medline, Google Scholar
49. : fMRI activation in the amygdala and the orbitofrontal cortex in unmedicated subjects with major depressive disorder. Psychiatry Res 2010; 183:209–217Crossref, Medline, Google Scholar
50. : Prefrontal mechanisms for executive control over emotional distraction are altered in major depression. Psychiatry Res 2008; 163:143–155Crossref, Medline, Google Scholar
51. : Functional neuroanatomy in depressed patients with sexual dysfunction: blood oxygenation level dependent functional MR imaging. Korean J Radiol 2004; 5:87–95Crossref, Medline, Google Scholar
52. : In what sense are positive and negative affect independent? a reply to Tellegen, Watson, and Clark. Psychol Science 1999; 10:304–306Crossref, Google Scholar
53. : Co-Planar Stereotaxic Atlas of the Human Brain. Stuttgart, Germany, Thieme, 1988Google Scholar
54. : AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996; 29:162–173Crossref, Medline, Google Scholar
55. : Projection from medial pulvinar to amygdala in primates. Brain Res 1976; 104:142–147Crossref, Medline, Google Scholar
56. : The Emotional Brain. New York, Simon & Schuster, 1996Google Scholar
57. : Emotion processing and the amygdala: from a 'low road' to 'many roads' of evaluating biological significance. Nat Rev Neurosci 2010; 11:773–782Crossref, Medline, Google Scholar
58. : Thalamic connections of the insula in the rhesus-monkey and comments on the paralimbic connectivity of the medial pulvinar nucleus. J Comp Neurol 1984; 227:109–120Crossref, Medline, Google Scholar
59. : Deficits in spatial coding and feature binding following damage to spatiotopic maps in the human pulvinar. Nat Neurosci 2002; 5:99–100Crossref, Medline, Google Scholar
60. : Solutions to the binding problem: progress through controversy and convergence. Neuron 1999; 24:105–110Crossref, Medline, Google Scholar
61. : Pulvinar and affective significance: responses track moment-to-moment stimulus visibility. Front Hum Neurosci 2010; 4:64Crossref, Medline, Google Scholar
62. : Dissociable intrinsic connectivity networks for salience processing and executive control. J Neurosci 2007; 27:2349–2356Crossref, Medline, Google Scholar
63. : The amygdala: vigilance and emotion. Mol Psychiatry 2001; 6:13–34Crossref, Medline, Google Scholar
64. : Human cingulate cortex and autonomic control: converging neuroimaging and clinical evidence. Brain 2003; 126:2139–2152Crossref, Medline, Google Scholar
65. : Insular interconnections with the amygdala in the rhesus-monkey. Neuroscience 1981; 6:1231–1248Crossref, Medline, Google Scholar
66. : The von Economo neurons in frontoinsular and anterior cingulate cortex in great apes and humans. Brain Struct Funct 2010; 214:495–517Crossref, Medline, Google Scholar
67. : How do you feel—now? the anterior insula and human awareness. Nat Rev Neurosci 2009; 10:59–70Crossref, Medline, Google Scholar
68. : An information theoretical approach to prefrontal executive function. Trends Cogn Sci 2007; 11:229–235Crossref, Medline, Google Scholar
69. : Cooperation of the anterior cingulate cortex and dorsolateral prefrontal cortex for attention shifting. Neuroimage 2004; 23:670–679Crossref, Medline, Google Scholar
70. : Interference resolution: insights from a meta-analysis of neuroimaging tasks. Cogn Affect Behav Neurosci 2007; 7:1–17Crossref, Medline, Google Scholar
71. : Human prefrontal cortex: processing and representational perspectives. Nat Rev Neurosci 2003; 4:139–147Crossref, Medline, Google Scholar
72. : A parametric study of prefrontal cortex involvement in human working memory. Neuroimage 1997; 5:49–62Crossref, Medline, Google Scholar
73. : Striatal contributions to working memory: a functional magnetic resonance imaging study in humans. Eur J Neurosci 2004; 19:755–760Crossref, Medline, Google Scholar
74. : Rethinking feelings: an fMRI study of the cognitive regulation of emotion. J Cogn Neurosci 2002; 14:1215–1229Crossref, Medline, Google Scholar
75. : Connexions of striatum and globus pallidus—synthesis and speculation. Philos Trans R Soc Lond B Biol Sci 1971; 262:441–457Crossref, Medline, Google Scholar
76. : Parallel organization of functionally segregated circuits linking basal ganglia and cortex. Annu Rev Neurosci 1986; 9:357–381Crossref, Medline, Google Scholar
77. : Striatonigrostriatal pathways in primates form an ascending spiral from the shell to the dorsolateral striatum. J Neurosci 2000; 20:2369–2382Crossref, Medline, Google Scholar
78. : Cognitive Therapy and the Emotional Disorders. New York, International Universities Press, 1976Google Scholar
79. : Cognition and depression: current status and future directions. Annu Rev Clin Psychol 2010; 6:285–312Crossref, Medline, Google Scholar
80. : Lower dopamine transporter binding potential in striatum during depression. Neuroreport 2001; 12:4121–4125Crossref, Medline, Google Scholar
81. : Elevated putamen D2 receptor binding potential in major depression with motor retardation: an [11C]raclopride positron emission tomography study. Am J Psychiatry 2006; 163:1594–1602Link, Google Scholar
82. : Reduced dopamine turnover in the basal ganglia of depressed suicides. Brain Res 1997; 769:135–140Crossref, Medline, Google Scholar
83. : Brain effects of antidepressants in major depression: a meta-analysis of emotional processing studies. J Affect Disord 2011; 130:66–74Crossref, Medline, Google Scholar
84. : Investigating neural primacy in major depressive disorder: multivariate Granger causality analysis of resting-state fMRI time-series data. Mol Psychiatry 2011; 16:763–772Crossref, Medline, Google Scholar
85. : Activation in dorsolateral prefrontal cortex in response to maternal criticism and praise in recovered depressed and healthy control participants. Biol Psychiatry 2005; 57:809–812Crossref, Medline, Google Scholar
86. : 5HTTLPR polymorphism and enlargement of the pulvinar: unlocking the backdoor to the limbic system. Biol Psychiatry 2007; 61:813–818Crossref, Medline, Google Scholar
87. : Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science 2003; 301:386–389Crossref, Medline, Google Scholar
88. : HPA axis reactivity: a mechanism underlying the associations among 5-HTTLPR, stress, and depression. Biol Psychiatry 2008; 63:847–851Crossref, Medline, Google Scholar
89. : Cognitive Therapy of Depression. New York, Guilford, 1979Google Scholar