A review of fMRI studies during visual emotive processing in major depressive disorder.

Objectives. This review synthesized literature on brain activity, indexed by functional magnetic resonance imaging (fMRI), during visual affective information processing in major depressive disorder (MDD). Activation was examined in regions consistently implicated in emotive processing, including the anterior cingulate cortex (ACC), prefrontal cortex (PFC), amygdala, thalamus/basal ganglia and hippocampus. We also reviewed the effects of antidepressant interventions on brain activity during emotive processing. Methods. Sixty-four fMRI studies investigating neural activity during visual emotive information processing in MDD were included. Results. Evidence indicates increased ventro-rostral ACC activity to emotive stimuli and perhaps decreased dorsal ACC activity in MDD. Findings are inconsistent for the PFC, though medial PFC hyperactivity tends to emerge to emotive information processing in the disorder. Depressed patients display increased amygdala activation to negative and arousing stimuli. MDD may also be associated with increased activity to negative, and decreased activity to positive, stimuli in basal ganglia/thalamic structures. Finally, there may be increased hippocampus activation during negative information processing. Typically, antidepressant interventions normalize these activation patterns. Conclusion. In general, depressed patients have increased activation to emotive, especially negative, visual stimuli in regions involved in affective processing, with the exception of certain PFC regions; this pattern tends to normalize with treatment.


Introduction
Major depressive disorder (MDD) can broadly be conceptualized as a disorder stemming from disturbances in the complex interplay between neural networks implicated in affective and cognitive processing as well as autonomic system activity, resulting in a heterogeneous array of emotive, cognitive and behavioural abnormalities. Although abnormalities in emotive circuits typically accompany depression, the nature of these disturbances varies.
The ability to identify and process the meaning of emotive visual information, such as facial expressions, is critical in communicating emotions and regulating social interactions (Gosselin et al. 1995). One dominant cognitive theory posits that depressed individuals are more likely to appraise their environment in accordance with negative schemas and exhibit cognitive biases toward negative stimuli (Kovacs and Beck 1978;Giesler et al. 1996;Gotlib and Joormann 2010). Indeed, depressed individuals, and even those at-risk for MDD, display enhanced memory and attention for sad faces and interpret neutral ones more negatively than non-depressed individuals Lepp ä nen et al. 2004;Surguladze et al. 2005;Le Masurier et al. 2007;Gollan et al. 2008). Others have extended this bias to all negative expressions, such as fear, anger and disgust (Bouhuys et al. 1999;Leyman et al. 2007). Similar fi ndings exist to emotive scenes or words, with depressed individuals exhibiting preferential attention to negative stimuli (Kellough et al. 2008;Leung et al. 2009;Koster et al. 2010). A handful of behavioural studies also indicate that blunted emotive information processing, in general, may exist in MDD (Gur et al. 1992;Rubinow and Post 1992;Koschack et al. 2003;Lepp ä nen et al. 2004). Other cognitive theories propose that depressed individuals display reduced perceptual sensitivity to positive Studies examining only remitted patients; 2. Studies that included patients with psychiatric co-morbidity other than anxiety.
A small number of studies were identifi ed that consisted of an adolescent population ( N ϭ 3 studies) and were included in this review; the majority of the studies examined adult populations. In total, 64 studies were included. Tables I -V present sample characteristics, methodology and relevant fMRI fi ndings across studies.

Defi ning regions of interest
We attempted to report cortical fi ndings in terms of Brodmann areas (BA). If BAs were not specifi ed, they were extracted from Talairach or MNI coordinates; occasionally, BAs were inferred from activation maps (consensus approach between NJ and X-RY). For the purposes of this review, BAs comprising the ventro-rostral ACC ( " affective " ACC aspect) included BA24, BA32 and BA25 (subgenual ACC/sgACC). The dorsal ACC ( " cognitive " ACC aspect) consisted of BA24 and BA32. When possible, we indicated whether activation in the dorsal or ventro-rostral ACC aspects was reported. The PFC tends to be subdivided into the medial and lateral aspects. In this review, the medial PFC (mPFC) included the frontopolar (BA10) and orbitofrontal cortices (OFC; BA11); together, they comprise the ventromedial PFC (vmPFC; BA10/11). The mPFC can also include dorsomedial PFC aspects (BA6/8/9). Lateral PFC areas include the dorsolateral PFC (DLPFC; BA9/46). The DLPFC may encompass the superior (BA9/8/6) and middle (BA9/46) frontal gyri. The lateral inferior frontal gyrus constitutes ~ BA44/45, while the most inferior lateral PFC aspect includes BA47.

Anterior cingulate cortex
The ACC is involved in modulating emotion-driven behaviours (Drevets et al. 2008), including confl ict monitoring, error detection and in evaluating the emotional signifi cance of stimuli (George et al. 1995;Pizzagalli et al. 2006). The dorsal ACC, referred to as the " cognitive " ACC aspect, is intimately connected with the DLPFC, a region highly implicated in cognitive control (Davidson et al. 2002). As such, dorsal ACC activation may represent a call for further processing and cognitive control of emotive stimuli. The ventro-rostral ACC, known as the " affective " aspect, has extensive connections with the amygdala and hippocampus. The sgACC projects to various nuclei of the basal ganglia and thalamus and stimuli, but engage in more in-depth and sustained processing of negative ones (Gollan et al. 2008). However, few studies have demonstrated this double-dissociation in MDD (Goeleven et al. 2006). Despite the utility of behavioural data in informing cognitive theories of depression, assessments of brain activity can provide insight into the neural modulations associated with emotive information processing in MDD that may not be captured behaviourally.
Converging data have identifi ed circuits involved in processing emotive stimuli and in emotive regulation (Davidson et al. 2002). Specifi cally, the prefrontal cortex (PFC), anterior cingulate cortex (ACC), amygdala, hippocampus and several subcortical regions, including the thalamus and basal ganglia, are critically involved in these functions. As such, most functional magnetic resonance (fMRI)indexed brain activity studies during emotive processing in MDD has focused on the circuits comprising these regions. This review focused on studies of brain activity to visual emotive information processing, indexed via blood oxygen level dependent (BOLD)-fMRI signals in individuals with MDD.
Higher rostral ACC activity during the processing of emotive, particularly negative, stimuli has been associated with an enhanced antidepressant response. Depressed individuals with greater baseline ACC activity to negative images showed a robust response to a serotonin norepinephrine reuptake inhibitor (SNRI; Davidson et al. 2003). Greater right sgACC (BA25) responses to sad facial expressions in the early stages of treatment were also associated with good clinical outcome to various antidepressants in melancholic depressed individuals (Keedwell et al. 2009(Keedwell et al. , 2010; increased baseline activity to happy expressions was associated with a smaller response (Keedwell et al. 2010). Faster improvement following chronic treatment with a selective serotonin reuptake inhibitor (SSRI) was also predicted by higher baseline ACC activity (BA25/32) to sad faces (Chen et al. 2007); a similar fi nding predicted remission with cognitive behavioural therapy (CBT; Costafreda et al. 2009). Although exceptions exist (Siegle et al. 2006), this enhanced rostral ACC activation may represent greater emotive monitoring, which may be a precursor for a positive response. Successful treatment may normalize ACC activation to emotive information processing (Fu et al. 2004(Fu et al. , 2008Robertson et al. 2007), but further work is required to confi rm this (Table I).

Prefrontal cortex
The PFC is important for the planning and execution of or restraining from actions, especially those guided by emotions (Fuster 2001;Davidson et al. 2002). The DLPFC receives input from sensory cortices and is densely connected with premotor areas, frontal eye fi elds and lateral parietal cortex The mPFC receives hypothalamic and brainstem projections that mediate visceral autonomic activity associated with emotion elicitation/processing, as well as from the ventral striatum, which signals reward and motivational value. The DLPFC has been primarily associated with cognitive functions, whereas the mPFC is largely ascribed affective roles. In healthy individuals, the DLPFC tends to be active during negative emotion regulation (Herwig et al. 2007;Stein 2008;Domes et al. 2010); mPFC activity is evoked when generating and regulating emotions (Jung et al. 2006;Kensinger and Schacter 2006;Gillihan et al. 2011). The vmPFC is implicated in representing reward and punishment (Davidson et al. 2002).
Several groups have reported that depressed individuals exhibit enhanced right mid/superior DLPFC receives dense dopaminergic innervations; it is likely implicated in regulating reward contingencies (Gaspar et al. 1989;Wise and Rompre 1989). The ventro-rostral ACC also plays a role in modulating automatic aspects of emotional processing and in forming emotive associations (Drevets et al. 2008).
Depressed individuals tend to exhibit ACC hyperactivity within the ventro-rostral region, specifi cally the sgACC, to negative stimuli (Gotlib et al. 2005;Fales et al. 2008Fales et al. , 2009Tao et al. 2012), although increased activation has also been found to positive stimuli Baeken et al. 2010). Depression severity has also been found to correlate with sg/ventro-rostral ACC activity to sad expressions (Keedwell et al. 2005b(Keedwell et al. , 2010. Hyperactivity to emotive and negative stimuli extending beyond the sgACC in MDD has also been observed (BA24/32; Beauregard et al. 1998;Elliott et al. 2002;Fu et al. 2004;Anand et al. 2005;Mitterschiffthaler et al. 2008;Dichter et al. 2009;Fales et al. 2009;Sheline et al. 2009;Yoshimura et al. 2010). Some investigators have found increased rostral ACC activity to processing happy and emotive facial expressions and pictures (Gotlib et al. 2005;Grimm et al. 2009a;Yang et al. 2010) or to cues signalling upcoming presentations of emotional pictures (BA32; Bermpohl et al. 2009) in MDD. In contrast, depressed individuals exhibited decreased left ACC activation (BA32) to targets presented after sad image distracters. This may refl ect preferential capture by negative images, which may interfere with ACC-mediated attention switching (Wang et al. 2008b).
A few studies have reported decreased ACC activity to negative stimuli (Davidson et al. 2003;Kumari et al. 2003), implicit sad face processing (Fu et al. 2008) and positive pictures (Lee et al. 2007). This decreased ACC activity to emotive processing may be related to volumetric ACC decreases in MDD (Drevets et al. 1997;Wagner et al. 2008) or refl ect ACC activity normalization following antidepressant treatment.
Some work suggests that observations of decreased ACC activity during emotive processing in MDD may be mainly attributed to decreased dorsal ACC activity. For instance, MDD patients exhibited decreased activity in the dorsal ACC (BA24/32) extending to the superior frontal gyrus (BA8) to processing and judging negative stimuli (Fu et al. 2008;Grimm et al. 2009b). Additionally, during the processing of neutral and fearful faces, MDD patients exhibited decreased dorsal and superior-rostral ACC (BA24/32; Fales et al. 2009). Symptomatic patients also exhibited decreased dorsal ACC (BA24) activity to processing threatening expressions compared with controls and those in remission (van Wingen et al.  dynamic range for activation (i.e., affective load response) to processing sad facial expressions of varying intensities in DLPFC regions (BA9/44; Fu et al. 2004). Chronic treatment with the non-SSRI antidepressant bupropion, however, reduced activation to sad pictures in mPFC regions ( ~ BA10/11, ~ BA8/9) pre-to post-treatment, though increased activation in the left inferior frontal cortex persisted ( ~ BA47; Robertson et al. 2007). G reater bilateral DLPFC (BA9/47) activity to emotive faces at baseline may predict a poor antidepressant response (Keedwell et al. 2009(Keedwell et al. , 2010 although another group noted that increased baseline activity to sad faces in the dorsomedial PFC ( ~ BA6/9/32) and superior frontal gyrus (BA6/8/9) was associated with a greater response to antidepressant pharmacotherapy (Samson et al. 2011; Table II).

Amygdala
The amygdala is implicated in recruiting and coordinating cortical resources to novel or ambiguous stimuli or those that elicit action tendencies (e.g., fear-eliciting cues; Davis and Whalen 2001). It is also implicated in the rapid processing of emotive stimuli (Zald 2003), emotional learning and memory formation (Barbas 2000;Phelps and LeDoux 2005). Amygdala projections are widespreadreciprocal connections exist with the forebrain, midbrain and brainstem regions; amygdalar nuclei receive information from all sensory modalities, though visual projections are most plentiful. The basolateral amygdalar nucleus has strong reciprocal connections with the PFC and hippocampus (Sah et al. 2003). The centromedial amygdalar nucleus projects primarily to the hypothalamus where it can infl uence hormonal release. Its tracts also terminate in other brainstem structures where they can modulate visceral autonomic functions in relation to the emotional signifi cance of stimuli (Knapska et al. 2003). Amygdalar hyperactivity to negative stimuli is typically evident in MDD (Sheline et al. 2001;Siegle et al. 2002Siegle et al. , 2007Fu et al. 2004;Anand et al. 2005;Surguladze et al. 2005;Wang et al. 2008a;Beesdo et al. 2009;Peluso et al. 2009;Suslow et al. 2010;van Wingen et al. 2011;Tao et al. 2012). Depression severity and lifetime psychiatric hospitalizations have been found to correlate with amygdala activity to negative stimuli (Lee et al. 2007;Dannlowski et al. 2008;Peluso et al. 2009;Mingtian et al. 2012), while inverse correlations exist with activity to positive stimuli (Keedwell et al. 2005a,b;Suslow et al. 2010). Finally, increased amygdala activation to stimuli of varied valences, including positive, has also been noted in patients with MDD (Matthews et al. 2008; (BA6/9/46) activity relative to controls while processing positive stimuli (Elliott et al. 2002;Grimm et al. 2008;Bermpohl et al. 2009;Demenescu et al. 2011;Liao et al. 2012). In general, however, mid/ superior DLPFC hyperactivity occurs in anticipation of and during the processing of negative stimuli (Canli et al. 2004;Keedwell et al. 2005a,b;Anand et al. 2005;Frodl et al. 2009;Herwig et al. 2010;Scheuerecker et al. 2010;Ritchey et al. 2011;Samson et al. 2011;Mingtian et al. 2012). A similar fi nding was noted in the inferior DLPFC (BA45; van Wingen et al. 2011).
Other studies, however, have reported decreased DLPFC and inferior lateral PFC activity in MDD during emotive processing (BA9/46/44; Davidson et al. 2003;Mitterschiffthaler et al. 2003;Canli et al. 2004;Lawrence et al. 2004;Schaefer et al. 2006;Siegle et al. 2007;Fales et al. 2008Fales et al. , 2009Lee et al. 2008;Wang et al. 2008b;Ritchey et al. 2011). Fu et al. (2004 also noted a decreased dynamic range for activation (i.e., affective load response) to processing varying intensities of sadness in MDD in the inferior lateral PFC and DLPFC (BA9/44). In an earlier meta-analysis by Fitzgerald et al. (2006), no consistent pattern of results could be extracted from studies examining DLPFC activity to presentations of positive and negative stimuli.
Results of studies examining mPFC activity during emotive information are also inconsistent, with some (Beauregard et al. 1998;Anand et al. 2005;Sheline et al. 2009;Herwig et al. 2010;Yoshimura et al. 2010;Mingtian et al. 2012) studies reporting increased activity to negative stimuli and others (Fu et al. 2008;Bermpohl et al. 2009;Grimm et al. 2009b) reporting decreased activity. Activity in the vmPFC (BA10/11/47), encompassing the OFC and frontopolar cortices, appears to be increased to negative stimuli (Elliott et al. 2002;Gotlib et al. 2005;Dichter et al. 2009;Sheline et al. 2009;Surguladze et al. 2010;Tao et al. 2012), and may even correlated with depression severity (Lee et al. 2007), though not everyone has found this (Lawrence et al. 2004;Lee et al. 2008;Friedel et al. 2009;Hsu et al. 2010). vmPFC activation has also been reported in response to positive images in MDD Mitterschiffthaler et al. 2003) and for cues signaling upcoming presentations of emotional (vs. neutral) pictures ). Finally, vmPFC hyperactivity to emotive stimuli correlated with depression severity and feelings of hopelessness (Grimm et al. 2009a).
A few studies have examined PFC activity to emotive information processing following antidepressant treatment. SSRIs enhanced initially blunted DLPFC (BA9) activity to unattended fearful stimuli in MDD (Fales et al. 2009) and increased the   Yang et al. 2010;Etkin and Schatzberg 2011;Liao et al. 2012;van Tol et al. 2012), although there are notable exceptions (Davidson et al. 2003;Gotlib et al. 2005;Frodl et al. 2009;Hsu et al. 2010;Ritchey et al. 2011). Several studies have examined the effects of antidepressant treatment on amygdala activity during emotive processing. Chronic SSRI treatment diminishes amygdala hyperactivity to negative/emotive stimuli (Sheline et al. 2001;Fu et al. 2004;Canli et al. 2005;Tao et al. 2012), normalizes pretreatment amygdala hypoactivity to fearful faces (Fales et al. 2009) and normalizes decreased activity to happy and increased activity to sad faces (Victor et al. 2012). Comparable normalization of hyperactivity to negative expressions was found with chronic bupropion treatment (Robertson et al. 2007), although, SNRI treatment did not alter amygdala activity to emotive processing in an earlier study (Davidson et al. 2003).
Psychotherapy may also infl uence amygdala dysregulation in MDD. In one study, initially diminished discriminant amygdala activity to emotive (vs. neutral) pictures increased with CBT (Ritchey et al. 2011). Successful CBT in MDD also attenuated initially elevated amygdala activity to sad expressions (Fu et al. 2008). Stronger recovery with CBT was also associated with increased pretreatment amygdala activity to negative words (Siegle et al. 2006). Work is required to elucidate the predictive nature of amygdala activity to antidepressant outcome -particularly to specifi c types of antidepressant interventions (Supplementary Table I

Basal ganglia and thalamus
The basal ganglia refer to a collection of subcortical nuclei surrounding the thalamus. Core basal ganglia structures typically include the caudate, putamen, globus pallidus (GP) and nucleus accumbens (NAc). The basal ganglia subserve aspects of cognition and emotive processing, including guiding actions towards motivationally salient stimuli, evaluating stimuli and regulating motoric expressions associated with emotive states (e.g., facial expressions; Ring and Serra-Mestres 2002;Camara et al. 2008). The basal ganglia and thalamus are richly interconnected with the DLPFC, OFC and ACC; the caudate also receives extensive amygdalar projections. Similarly, the dorsomedial nucleus of the thalamus is densely connected with medial cortical and limbic regions (Taber et al. 2004) and may thus be important in affective processing and regulation (Drevets et al. 2008).
The relatively sparse literature on basal ganglia activity during emotive visual information processing suggests increased activation to negative stimuli and decreased activation to positive stimuli in MDD. For instance, depressed individuals exhibited increased left putamen activity to sad faces while controls showed increased right putamen activity to happy ones (Surguladze et al. 2005). Patients also had greater putamen, caudate and thalamus activity when processing negative pictures and expressions (Scheuerecker et al. 2010;Mingtian et al. 2012). Similarly, when expecting unknown pictures, depressed patients showed medial thalamus activity similar to when expecting negative ones (Herwig et al. 2010). Conversely, they demonstrated decreased ventral striatum and overall basal ganglia activity to positive stimuli (Epstein et al. 2006;Schaefer et al. 2006), which correlated negatively with depression and anhedonia scores (Keedwell et al. 2005a(Keedwell et al. ,b, 2009. Though exceptions exist Mitterschiffthaler et al. 2003;Lee et al. 2008), extant evidence suggests increased activation to negative and decreased activation to positive visual stimuli in the basal ganglia/ thalamus in MDD. Fu et al. (2004) found that depressed subjects treated with SSRIs had reduced ventral striatum activation to sad expressions; symptom improvement was also associated with a reduced dynamic range of activation to varying intensities of sadness. The same group also reported that greatest clinical improvement with CBT was associated with low activity in the left putamen and GP during implicit sad expression processing (Fu et al. 2008). Similarly, right caudate activity decreased to emotional distracters in patients treated with bupropion (Robertson et al. 2007). Furthermore, depressed patients with initially decreased basal ganglia activity to positive images exhibited increased activity following SNRI treatment (Schaefer et al. 2006). Finally, patients who had diminished discriminant caudate activity to neutral versus emotive pictures prior to treatment showed increased discriminant activity following CBT (Ritchey et al. 2011). Preliminary data therefore suggests that antidepressant therapies may decrease basal ganglia activity during emotive, particularly negative, information processing (Supplementary Table II

Hippocampus
The hippocampus is a primary structure comprising the limbic system; it has a central role in episodic memory formation and spatial navigation (Duvernoy 1988). It is intimately connected with the amygdala and its efferents project to the NAc and ventral striatum as well as prefrontal and temporal cortices (Campbell and MacQueen 2004). Anterior hippocampal regions are strongly connected with the mPFC, whereas posterior regions are primarily connected with the DLPFC (Witter et al. 2000).
Studies examining hippocampus activation during visual emotive information processing in MDD are sparse, but suggest hippocampal hyperactivity to processing negative stimuli (Lee et al. 2007;Fu et al. 2008) and reduced activation to positive stimuli (Schaefer et al. 2006;van Tol et al. 2012; but see also Lee et al. 2008). Symptom severity may correlate with hippocampal activity to sad expressions (Keedwell et al. 2005b(Keedwell et al. , 2009.
Data on fMRI-indexed hippocampal activity during emotive processing following antidepressant therapy are also limited. One group reported that low baseline activity in the hippocampus and other limbic regions normalized with chronic SNRI treatment (Schaefer et al. 2006). Increased activation to emotive pictures was also noted following CBT in MDD (Ritchey et al. 2011). In a study examining the potential antidepressant effects of erythropoietin (vs. placebo), patients exhibited reduced hippocampal activity to fearful versus happy faces after treatment (Miskowiak et al. 2009;Supplementary

Discussion
This review aimed to synthesize research on neural activity during visual emotive information processing in MDD. Overall, evidence suggests increased ventro-rostral/sgACC activation to emotive, particularly negative, stimuli and perhaps decreased dorsal ACC activity during emotive processing. Findings are inconsistent for the PFC. There appears to be increased amygdalar activation in MDD to negative visual stimuli, but also to emotive/arousing stimuli in general. The degree of amygdalar activation may correlate with depression severity. In the basal ganglia and thalamus, patients seem to have increased activation to negative stimuli and decreased activation to positive stimuli compared to healthy controls. Finally, some evidence suggests that depressed individuals also demonstrate increased hippocampal activation during negative emotion processing.
There are few, if any, regions that have not had divergent reports with respect to activation in response to emotive stimuli. Such inconsistency may be partially accounted for by variability in regional labels (i.e., reporting BAs, Talairach or MNI coordinates) and in the tasks utilized (e.g. passive or active). Other sources of variability may be related to heterogeneity of the clinical populations studied, including factors such as depressive symptom severity, co-morbidity, age and methodological issues such as not controlling for total brain volume or volume loss. Gender and medication status are other aspects that may infl uence brain activity to emotive information processing (Stevens and Hamann 2012). Both were investigated in this review, though the scarcity of research directly assessing the contributions of these factors (especially gender) on brain activity prevents us from drawing conclusions. Generally, studies examining antidepressant pharmacotherapy or psychotherapy have reported that treatment normalizes brain activity abnormalities during emotive information processing.
Certain limitations must be acknowledged with respect to our review. First, unlike a meta-analytic approach, our review cannot comment on the degree of activation reliability in specifi c spatial coordinates implicated in emotive processing. Second, neuroimaging meta-analyses and reviews, including ours, are affected by publication biases, which favor the increased probability of published studies reporting signifi cant (versus null) results (Wager et al. 2007). Third, we did not include non-English studies in the current review and may have thus missed pertinent research in the fi eld.
Networks consisting of a number of brain areas comprise the brain's emotion circuitry. Evidence points toward dysregulation of these circuits in patients with MDD. Future studies may focus more on exploring the connectivity/dysconnectivity in these emotive networks rather than activation of isolated regions. Ideally, such studies will control for clinical sources of variation (or whole-brain assessments to increase statistical rigour) and use rigorous defi nitions of regions of interest, such that variation in results across studies can be better understood.
Ultimately, greater understanding into the dominant brain activity anomalies during emotive processing in MDD may facilitate in characterizing individuals at-risk for developing the disorder. This, in turn, may help with the development and employment of targeted prevention/intervention strategies. Further, greater insight into the functional correlates of depression may aid in predicting treatment response in different MDD subtypes. Finally, better understanding of the putative mediating effects of gender on brain function in depression may help us to better integrate gender-focused treatment strategies in the future.
Visual emotive processing in MDD 467