|Home | About | Journals | Submit | Contact Us | Français|
Both major depressive disorder and bipolar disorder are the subject of a voluminous imaging and genetics literature. Here, we attempt a comprehensive review of MRI and metabolic PET studies conducted to date on these two disorders, and interpret our findings from the perspective of developmental and degenerative models of illness. Elevated activity and volume loss of the hippocampus, orbital and ventral prefrontal cortex are recurrent themes in the literature. In contrast, dorsal aspects of the PFC tend to display hypometabolism. Ventriculomegaly and white matter hyperintensities are intimately associated with depression in elderly populations and likely have a vascular origin. Important confounding influences are medication, phenotypic and genetic heterogeneity, and technological limitations. We suggest that environmental stress and genetic risk variants interact with each other in a complex manner to alter neural circuitry and precipitate illness. Imaging genetic approaches hold out promise for advancing our understanding of affective illness.
Depression is arguably the epidemic of our time. The lifetime prevalence of major depressive disorder (MDD) ranges from 10–30% (Kessler et al 2003) and depression arising within the context of bipolar disorder (BD) has equally serious implications for morbidity and mortality (Pini et al 2005).
Understanding the etiological and pathophysiological basis of affective illness is clearly an international imperative. Current nosological systems are based on symptomatology rather than etiology, and robust biological correlates of depression have not been identified. The identification of biomarkers of depression is crucial not only in the ascent towards etiological understanding, but also in evaluating the efficacy of treatment interventions.
While the limited resolution of previous generation neuroimaging paradigms blunted the sensitivity and specificity of preceding investigations; with the advent of new technology, glints of understanding are beginning to emerge. Neuroimaging data implicate a key emotion-regulating circuit, the visceromotor network, encompassing the medial prefrontal cortex (mPFC) and its reciprocal connections to the amygdala, hippocampus, ventral striatum, hypothalamus and brain-stem in the pathophysiology of both MDD and BD (Ongur and Price 2000a).
More specifically, regions of orbitofrontal cortex (OFC) and the mPFC have shown MRI and post-mortem-derived evidence of tissue loss. Parallel metabolic and volumetric changes to the limbic components of the visceromotor network such as the amygdala, hippocampus and ventral striatum have been widely recorded, although the data are often conflicting. Unfortunately, diminutive structures such as the habenula and periaqueductal gray (PAG), which play a prominent role in emotional behavior, remain largely unstudied because of technological limitations.
Here we review the biological correlates of MDD and BD as evinced by neuroimaging paradigms, and interpret these data from the perspective of neurodevelopmental and neurodegenerative pathoetiology. We attempt to integrate the conclusions drawn from the literature into a heuristic framework which characterizes affective illness as the consequence of a loss of top-down control (especially mPFC) over limbic structures such as the amygdala; or alternatively, the consequence of disinhibited limbic drive which overrides cortical regulation (Figure 1).
A literature search of the PUBMED database up until September 2007 was carried out using the following key words: depression, bipolar disorder, neuroimaging, MRI, PET, fMRI, amygdala, hippocampus, basal ganglia, caudate, prefrontal cortex, orbital frontal cortex, dorsolateral prefrontal cortex, anterior cingulate, subgenual prefrontal cortex, white matter, and ventricle. Furthermore, review articles were searched, and other publications cross-referenced for additional published articles. Our inclusion criteria were heavily biased towards analyses of resting state metabolism rather than responses to cognitive, emotional or biochemical challenges. Thus, we although we highlight the result of certain fMRI studies (primarily in the discussion section), we make no attempt to cover the substantial, but as yet inchoate, functional magnetic resonance imaging (fMRI) and magnetic resonance spectroscopy (MRS) literature comprehensively. Positron emission tomography (PET) neuroreceptor studies are also beyond the scope of this review.
Further, the following exclusion criteria obtained: Papers that were not written in English, book chapters, conference abstracts, and case studies were not reviewed. Computerised tomography (CT) and Single Photon Emission Tomography (SPECT) studies were omitted. CT has a significantly lower tissue contrast and spatial resolution than MRI, and is also subject to bony artifacts in brain structures situated near the skull. SPECT has reduced sensitivity for detecting areas of increased perfusion due to its reliance on radioligands that are not freely diffusible across the blood brain barrier. Analyses of gross neuroanatomical structures such the entire frontal or temporal cortices were generally not discussed.
Extensive inter-connecting neural networks are involved in the generation and regulation of affect. These networks can be at least partly subsumed under the iterative activity of three cortical-striatal-limbic circuits encompassing a dorsolateral/dorsomedial prefrontal circuit, an orbital prefrontal circuit, and a ventromedial prefrontal circuit, including the anterior cingulate cortex (ACC) (Tekin and Cummings 2002) (Drevets 2001) (Price 1999). These circuits operate in parallel with prefrontal cortex-originating bidirectional projections to different nuclei of the perirhinal and entorhinal cortices, striatum, pallidum, thalamus, amygdala, hippocampus, hypothalamus, habenula, and periaqueductal grey (Tekin and Cummings 2002) (Price 1999).
Nevertheless, the limitations of current neuroimaging modalities make it unfeasible to reliably discriminate between individual nuclei within these structures. The amygdala, for example, is a heterogeneous structure of at least 14 different nuclei (Bachevalier and Loveland 2006). We will therefore discuss imaging studies of the amygdala, hippocampus, and striatum, separately, rather then placing them under the rubric of the dorsal prefrontal, orbital frontal, and ventromedial cortical-striatal-limbic circuits described above.
Despite its apparent heterogeneity of function, a degree of consensus that the amygdala plays a pivotal role in evaluating the emotional significance of perceptual data has been reached (Phillips et al 2003). In coloring perceptual stimuli with emotion, however, the amygdala appears to emphasise the hues of fear, anger and sadness (Gloor et al 1982) (Davidson 2002), providing prima facie evidence for its involvement in depression.
Neuroimaging studies of the amygdala in patients with BD (Tables 1 and and2)2) are characterized by an interesting age-related dichotomy of findings. In adults, the predominant pattern is one of increased amygdala volume while in children and adolescents the reverse applies (Table 1). The findings in adults seem to hold even in samples with a long history of illness (Altshuler et al 2000) (Brambilla et al 2003a) (Frangou 2005).
Resting state functional analyses have been largely limited to the adult population (Table 2) and are indicative of increased baseline amygdalar activity, (Ketter et al 2001) (Sheline et al 2001) (Drevets et al 2002b) (Bauer et al 2005) (Mah et al 2007) which correlates positively with severity of depression (Drevets et al 1992) (Ketter et al 2001).
The literature regarding the structural and functional changes of this region in MDD is in disagreement (Tables 3 and and4).4). Several studies have reported gray matter (GM) volume loss in euthymic (Sheline et al 1998) and depressed (von Gunten et al 2000) (Caetano et al 2004) (Hickie et al 2006), or psychotic (Keller et al 2008) patients, but many negative results have also been published. Resting state functional data are largely suggestive of hypermetabolism (Drevets et al 1992) (Drevets et al 2002b) (Anand et al 2005).
The hippocampus, a key structure for the encoding of emotionally relevant data into memory, interacts with the amygdala to provide input regarding the context in which stimuli occur (LaBar and Cabeza 2006). This process is influenced by the hypothalamic-pituitary-adrenal (HPA) axis through the modulation of arousal (LaBar and Cabeza 2006) (Roozendaal et al 2006). In rodents the hippocampus in turn plays an inhibitory role in the regulation of the amygdala, and HPA axis activity (Jacobson and Sapolsky 1991).
Hippocampal volume reduction has been widely reported (Table 5). The authors of an earlier meta-analysis of the literature also came to the conclusion that hippocampal GM loss is characteristic of depression (Campbell et al 2004); an effect that (Stockmeier et al 2004) attribute to a decrease in neuropil - although see (Lucassen et al 2001).
A significant number of studies have, however, failed to find evidence of hippocampal atrophy in depressed patients (Table 5) and based on these data we suggest that the following caveat obtains: The majority of studies reporting evidence of hippocampal atrophy have made use of elderly, middle-aged or chronically ill populations (see discussion).
Regarding BD, although a few studies have indeed reported volumetric decrements, the majority of studies report preservation of hippocampal tissue (Table 7). Once again, pediatric and adult samples appear to produce different results with more evidence of volume loss in the former; although see (Ladouceur et al 2008) who found increased GM volume in the left hippocampus of the healthy offspring of parents with BD.
The basal ganglia (BG), made up of the caudate, putamen, globus pallidus (GP), subthalamic nucleus (STN), and substantia nigra (SN) were traditionally conceptualized as a center of motoric integration (Pollack 2001) (DeLong and Wichmann 2007). This view evolved over time as the neuropsychiatric symptoms of Parkinson’s (Lieberman 2006) and Huntington’s disease (PD and HD) (Slaughter et al 2001) patients became clear, and neuropsychological case studies of affectively disturbed patients with BG lesions began to surface in the literature (Lauterbach et al 1997). More recently, deep brain stimulation of the STN, nucleus accumbens, ventromedial caudate, and GP has provided some relief to patients with obsessive compulsive disorder and PD-related depression (Kopell and Greenberg 2008). The current understanding is that motor, sensory and emotional data travel in parallel but segregated pathways between cortical and sub-cortical structures such as the BG.
For example, the dorsal and orbito-frontal aspects of the prefrontal cortex send major efferent projections to the dorsal, and anterior caudate, respectively, while the anterior cingulate cortex projects to the ventral surface of the striatum, respectively (Ferry et al 2000). In addition, the ventral or limbic striatum (ventral caudate, accumbens and olfactory tubercle), receives dopaminergic input from the SN and ventral tegmental area (VTA) (Utter and Basso 2007), glutamatergic input from the amygdala and thalamus, and serotonergic input from the dorsal raphe nucleus (Pollack 2001) (Bonelli et al 2006).
Several early studies raised the possibility of BG volume reductions in MDD, although with the exception of a recent pediatric study (Matsuo et al 2008), these data have generally not been replicated in subsequent analyses (Table 9). As in the case of BD there has been some suggestion that volume loss is associated with late age-of-onset (Greenwald et al 1997) and chronicity or severity of illness (Pillay et al 1998) (Lacerda et al 2003) - although see (Sheline et al 1999). Most recently, (Hickie et al 2006) failed to detect a MDD-associated decrease in striatal volumes. However, when the sample was stratified by serotonin transporter promoter polymorphism genotype (5-HTTLPR), striatal volume loss was observed in short allele carriers; indicating that genetic factors may contribute to the heterogeneity characteristic of the literature.
At least five studies (Table 10) have reported decreased activity of the striatum in MDD, and a tryptophan depletion study reported that severity of depression was associated with diminished activity of the caudate (Smith et al 1999).
Compared to healthy controls, subjects with BD generally have not shown morphometric differences in the caudate or putamen (Table 11). These data are congruent with the reported absence of N-acetyl-aspartate (NAA) abnormalities in the BG of BD populations (Kato et al 1996) (Hamakawa et al 1998) (Ohara et al 1998). Nevertheless, a post-mortem study of a combined MDD and BD sample has reported volumetric reductions of the left nucleus accumbens, the bilateral pallidum, and the right putamen (Baumann et al 1999).
There has however, been some suggestion of BD-associated striatal enlargement in adult (Aylward et al 1994); (Noga et al 2001); (Strakowski et al 2002) and pediatric (DelBello et al 2004); (Wilke et al 2004) samples, a point we will return to in the discussion.
To complicate matters further, reduced striatal volumes have been associated with length of bipolar illness (Brambilla et al 2001b) and an older age of onset (Beyer et al 2004a) suggesting a potential role for chronicity or cerebrovascular disease. This notion is consistent with reports of white matter lesions of the BG in elderly patients with MDD (Murphy et al 1992); (Greenwald et al 1996) (Iidaka et al 1996) although striatal volume reductions have also been reported to be a marker of disease-diathesis in the relatives of BD probands (McDonald et al 2004a); (McIntosh et al 2004).
Finally, metabolic activity or blood flow in the BG has been reported to be both increased and decreased in BD samples (Table 12). In manic or hypomanic samples striatal activity appears more generally increased relative to controls during tasks that normally activate the striatum (Blumberg et al 2000) (Caligiuri et al 2003) (Caligiuri et al 2006). For example, a recent fMRI study using a monetary incentive task noted that the expected differences in functional activity of the nucleus accumbens and ventral tegmentum in response to trials where a reward is received, compared to trials where a reward is not received, are attenuated in patients with mania (Abler et al 2008). In other words, subjects with mania show an inappropriate, generalized activation of reward circuitry. In bipolar depressed samples activity also is increased in the ventral striatum (Ketter et al 2001) (Bauer et al 2005) (Dunn et al 2002) (Mah et al 2007), but the direction and existence of physiological abnormalities in the remainder of the striatum has been more variable across studies (Table 12).
Substantial tissue loss in the medial-temporal lobe, lateral PFC or BG in MDD or BD may be reflected by enlargement of the adjacent ventricular system. The evidence for ventricular enlargement (mostly of the third or lateral ventricles) in MDD and BD is mixed. Almost all reports of ventricular enlargement have been obtained in elderly or chonically depressed samples with late-onset illness (Coffey et al 1989) (Rabins et al 1991) (Salloway 1996) (Dahabra et al 1998) (Simpson et al 2001) and given the evidence that familial MDD usually manifests early in life (Kovacs et al 1997) (Kendler et al 2005) (Nierenberg et al 2007), this suggests that periventricular tissue loss does not have a purely genetic etiology. One possibility is cerebrovascular disease, as discussed in the following section.
A higher than normal incidence of deep frontal white matter hyperintensities (WMH), especially WMH of the deep frontal cortex and BG, appears characteristic of MDD and BD samples who manifest with late age-of-illness onset (Krishnan et al 1991); (Figiel et al 1991); (Hickie et al 1995) (Steffens et al 1999) (Hannestad et al 2006).
WMH appear as bright high intensity signals seen on T2-weighted MRI scans that are caused by circumscribed increases in water content (Ovbiagele and Saver 2006). As discussed by (Ovbiagele and Saver 2006) they are most likely indicative of leukoaraiosis: a decrease in the density of white matter due to demyelination, atrophy of the neuropil, and ischemia-associated microangiopathy, among others. The phenomenon is non-specific, being prevalent in elderly populations, generally. According to (Kertesz et al 1988) almost all individuals will display WMH by the age of 85.
Diffusion tensor imaging (DTI) is another method of assessing WM integrity. Theoretically, damage to cellular tissue causes parallel changes in the rate of diffusion of water across the affected cellular membranes, and this so-called proton diffusibility can be measured with DTI (Bammer 2003).
While the onset of MDD peaks in adolescence and young adulthood, an increased incidence is also seen in elderly individuals (Paykel et al 2005), contributing to the idea that vascular pathology plays a role. The term “vascular depression” was initially proposed to describe depressive symptomatology associated with multiple subcortical infarcts of an ischemic origin (Alexopoulos et al 1997) (Krishnan et al 1997).
With few exceptions (Dupont et al 1995a); (Greenwald et al 1996); (Sassi et al 2003); (Rainer et al 2006), extant evidence supports the existence of vascular depression. Both epidemiological studies of elderly community-based samples and cross-sectional analyses of matched control and MDD groups detail the intimate relationship between late-onset depression and WM lesions (Table 13).
In BD most studies suggestive of WM pathology involve DTI. Changes in diffusion coefficients, which are believed to reflect the integrity of WM fibres, have been reported in the WM bundles connecting the PFC (particularly BA 9 and 10) to subcortical regions (Adler et al 2004), OFC (Beyer et al 2005), internal capsule adjacent to the striatum and thalamus (Haznedar et al 2005), genu of the corpus callosum (Yurgelun-Todd et al 2007), and prefrontal and temporal cortex (Bruno et al 2008). Further, a DTI tractography (allows for the measurement of entire WM pathways) study reported a higher incidence of reconstructed WM fibres linking the left sgACC and the amygdalo-hippocampal complex in remitted adults with BD (Houenou et al 2007).
Most MRI-based reports of BD-associated WMH are, however, derived from pediatric samples. A significant minority of young BD patients with a relatively typical age-of-onset show WM abnormalities on MRI (Table 14).
(Adler 2004) has hypothesized that the pattern of deep frontal WM pathology commonly seen in affective illness, results in a disruption of the pathways linking subcortical regions such as the striatum to functionally homologous regions of the prefrontal cortex – a version of the so-called “disconnection syndrome” (Geschwind 1965). In the following three sections we discuss key areas of the PFC involved in the regulation of emotional behavior.
The orbital frontal cortex (OFC) on the ventral surface of the frontal lobe receives inputs from the ventrolateral amygdala and other limbic structures such as the entorhinal and perirhinal cortices, and the hippocampal subiculum (Price 1999). The OFC also projects directly to the amygdala, hypothalamus, and brain stem, modulating limbic-driven behavior (Ongur and Price 2000b). More specifically, it acts to integrate limbic data with sensory input thus providing a first-pass analysis of the reward or aversive value of stimuli (Price 1999) (Kringelbach 2005). These data then feed into higher level processing circuits such as that of the mPFC (vide infra) which act together with the OFC to guide behaviour in terms of these expected contingencies (Kringelbach and Rolls 2004) (Amodio and Frith 2006).
Functional imaging analyses of the OFC in MDD (Table 15) are generally suggestive of increased metabolism or blood flow in the eyes closed or resting state in young to middle-aged samples (Drevets et al 1992) (Biver et al 1994) (Drevets et al 2002a) – particularly acutely depressed subjects (Drevets 2007). In contrast, while performing a probabilistic reversal learning task, MDD patients demonstrated attenuated hemodynamic activity in the lateral OFC/ventrolateral PFC and dorsomedial PFC relative to both healthy controls and BD subjects on trials in which misleading negative feedback triggered a behavioral response reversal, compared to trials in which misleading negative feedback did not precipitate reversal (Taylor Tavares et al 2008). This physiological difference was associated with an increased likelihood for the MDD subjects to switch their behavioral responses as a result of the misleading negative feedback, although basic object reversal learning is intact in depression. Another study showed an increased hemodynamic response in the posterior orbital cortex in depressed versus non-depressed BD subjects imaged while performing a color-word Stroop task (Blumberg et al 2003b).
Concerning the structural neuroimaging data, a sample of elderly MDD patients were first reported by (Lai et al 2000) to show bilateral volume reduction of the OFC. This finding has been replicated a number of times in elderly populations with later-onset depression (Table 16) and a negative association between the integrity of the white matter tracts of the OFC and severity of depression in a late-onset MDD group has also been recorded (Nobuhara et al 2006). (Lacerda et al 2004) reported a reduction of right medial and left lateral OFC volume in a younger sample (mean age 39) although the effect was more pronounced in depressed compared with euthymic patients. Similar to the (Nobuhara et al 2006) report, (Chen et al 2007a) found that severity of depression correlated negatively with right OFC volume.
GM volume reductions of the OFC have also been reported in adult (Frangou 2005) (Haznedar et al 2005) (Lyoo et al 2006) (Nugent et al 2006) and pediatric BD samples (Wilke et al 2004) (Najt et al 2007); although see (Lopez-Larson et al 2002) and (Dickstein et al 2005). Thus unlike MDD, BD-associated volume reduction of the OFC appears to occur in individuals with a more typical age-of-onset (Table 18)
Concerning functional imaging studies of the OFC in BD (Table 17), (Blumberg et al 1999) showed that manic patients have reduced rCBF in this region while induction of a sad mood through psychological means resulted in decreased rCBF to the medial OFC in euthymic but not depressed patients compared with controls (Kruger et al 2003). A study using an affective Go/NoGo paradigm to examine neural activity during voluntary attentional control of emotion found that despite intact task performance, euthymic BD adults demonstrated abnormally increased activity in bilateral OFC and left dorsal ACC, together with increased activity in some subcortical limbic regions, when inhibiting responses to emotional versus neutral distractors (Wessa et al 2007).
Other studies showed reduced activity in euthymic BD relative to healthy adults during automatic attentional control paradigms, such as the non-emotional Stroop color-word or counting task, in the left OFC/VLPFC and mPFC (Blumberg et al 2003b) (Kronhaus et al 2006), and the right mPFC (Strakowski et al 2005) and dorsal ACG, but also increased activity in right dorsolateral PFC (Gruber et al 2004). Subjects imaged during the depressed or manic phases of BD while performing similar paradigms also showed reduced left OFC activity relative to healthy controls (Blumberg et al 2003b), although these findings were not replicated by others (Marchand et al 2007). Most of these studies showed intact task performance in BD adults, although in other cases performance was impaired relative to healthy controls (Strakowski et al 2005).
Neuroimaging data suggest that the peri-callosal tissue of the ventro-medial prefrontal cortex known as the anterior cingulate cortex (ACC) (BA 24, 25, and 32) 1 (Figure 2) plays a pivotal role in translating OFC-derived valenced data into actions and behavior (Bush et al 2000) (Devinsky et al 1995) (Drevets 2000a).
Drevets and colleagues (Drevets et al 1997) carried out the first well-controlled imaging study of the medial prefrontal cortex in affective illness. Both depressed BD and MDD patients with a family history of affective illness showed left hemisphere gray matter loss in a region immediately ventral to the genu of the corpus callosum – the subgenual anterior cingulate cortex (sgACC). Further, the medication free patients displayed reduced metabolism of the left sgACC as demonstrated by PET.
In the interim, at least seven other studies have provided further evidence for MDD or BD-associated GM loss in the sgACC (Hirayasu et al 1999b) (Botteron et al 2002) (Sharma et al 2003) (Hastings et al 2004) (Boes et al 2007) (Koo et al 2008) (Yucel et al 2008) (Tables 20 and and22).22). The recent meta-analysis of (Hajek et al 2008) confirms that on average both left and right sgACC volumes are decreased in mood disorders although interestingly, the effect (especially in the left hemisphere) appears to be driven by cases with a family history of illness.
Similarly, decreased activity of in particular, the left sgACC has been recorded in other studies of MDD and BD samples (Tables 19 and and21).21). These data apply equally to males (Hastings et al 2004) and females (Botteron et al 2002) as well as psychotic (Hirayasu et al 1999a) (Coryell et al 2005) (Adler et al 2006b) and bipolar-spectrum illness (Haznedar et al 2005), respectively. There are also data to suggest that metabolic changes predate the onset of clinical symptoms. (Kumano et al 2006) followed a cohort of cancer patients and found that those individuals who went on to develop depression had lower baseline metabolic rates of the subgenual PFC compared to their counterparts who did not become depressed during the course of the study.
Other researchers have however, produced evidence of increased MDD and BD-associated metabolic activity of the ventral emotion circuit (Tables 19 and and21).21). A highly consistent finding across the literature has been that activity is elevated in the sgACC in the depressed phase relative to the remitted phase of MDD (Drevets et al 2002a) (Mayberg et al 2005) (Clark et al 2006) (Neumeister et al 2006b) (Hasler et al 2008). Consistent with these data, an fMRI study reported decreased response to sad facial stimuli in the right sgACC after AD treatment (Keedwell et al 2008).
An anatomical region more rostral to the sgACC (24/25), BA 32 (Figure 2), is also an integral part of the ventral “emotion” circuit, and has been implicated in affective illness. A voxel-based morphometry (VBM) study reported a 7.3% decrease of the GM density of the left ventral aspect of this region in a sample of patients with BD I (Lyoo et al 2004a). The partially medicated BD I, BD II, and cyclothymic sample of (Haznedar et al 2005) yielded similar results with decreased white and GM volume of the BA 32 area, noted. In line with these data, (Kegeles et al 2003) noted hypometabolism of BA 32 in depression. In contrast, (Holthoff et al 2004) reported a remission-associated decrease in the activity of this region and (Nobler et al 2001) obtained analogous results after ECT administration.
Individuals with MDD display a predominantly left sided hypometabolism of the ventro-medial circuit (Table 19) (Drevets et al 1997) (Kumano et al 2006). Compatible with these findings, (Smith et al 2002) found that decreased metabolism of the right sgACC at baseline was predictive of an improved response to antidepressant therapy. Similarly, higher baseline activity of the left sgACC at baseline was positively correlated with response to sleep deprivation therapy; although after treatment, responders showed a drop in perfusion of the left ventral cingulate (Clark et al 2006). Parallel findings of volume reductions of the ventral aspects of the left ACC are observed in morphometric analyses (Table 20).
The volumetric reductions seen in the left SGACC in some mood disordered subgroups may contribute to the complex relationship between clinical state and sgACC metabolism. Partial volume effects may have contributed to the original suggestions of hypometabolism (see (Mah et al 2007) and discussion below).
A similar pattern obtains in BD with left-sided hypometabolism or right-sided hypermetabolism (Drevets et al 1997) (Bauer et al 2005) (Mah et al 2007), and left-sided tissue loss (Drevets et al 1997) (Hirayasu et al 1999a) (Sassi et al 2004) (Wilke et al 2004) (Kaur et al 2005) of the broader ventral ACC region (Tables 21 and and22).22). One might hypothesise the reverse pattern to apply to the manic state (Drevets et al 1997), a phenomenon observed in the sample of (Blumberg et al 2000) which was characterized by elevated rCBF in the left sgACC.
During functional brain mapping studies of BD a study employing an Affective Go/No Go task showed greater activity to happy distracters in left sgACC, medial PFC and bilateral DLPFC, and greater activity in response to sad relative to neutral distracters in right ventrolateral PFC and DLPFC, in manic BD versus healthy adults (Elliott et al 2004).
(Davidson and Irwin 1999) illustrate how the line between cognition and emotion is becoming increasingly blurred in their description of the DLPFC as mediator of “affective working memory” (Davidson and Irwin 1999) (p 11). The DLPFC allows for the initially “raw” limbic-derived emotional stimuli which is refined through rounds of ever more comprehensive processing in the orbital and ventromedial circuits to be represented “online” so that it can drive goal directed behaviour (Davidson and Irwin 1999). In this sense, the dorsal aspects of the PFC discharge traditionally characterized executive processes such as response selection, inhibition, and error detection - the monitoring of actions to insure that they match intentions.
As reviewed by (Davidson 2002), in depression, a deficit in the top-down inhibitory control of the DLPFC over the amygdala and sundry limbic tissue may result in chronic limbic over-activity and negative emotions. Implicit in this deliberately over-simplified description of the above model is the notion that positive affect is primarily a consequence of the suppression of negative emotions.
Hypometabolism of the dorsal PFC is one of the most robust findings in both MDD (Hurwitz et al 1990) (Cohen et al 1989) (Biver et al 1994) (Bench et al 1995) (Dunn et al 2002) (Davidson et al 2003) (Chen et al 2007a) and BD (Buchsbaum et al 1986) (Baxter et al 1989) (Drevets et al 2002a) (Kruger et al 2003), and in some studies appears to normalize with successful treatment (Mayberg et al 1999) (Mayberg et al 2000) (Kennedy et al 2001) (Mayberg et al 2005) (Fales et al 2008a), lending support to the veridicality of the above-mentioned model of depression.
Consistent with these data, (Chen et al 2007a) found that severity of depression correlated negatively with GM volume of DLPFC and (Davis et al 2004) reported global decrements of both GM and WM volume, including areas of the DLPFC. We note that there are conflicting reports in the literature – see (Brody et al 1999) (Brody et al 2001) (Drevets et al 2002a) (Table 24).
Nevertheless, the vast area of cortex that is subsumed under the term dorsal or DLPFC make the functional data difficult to interpret. The dorsal PFC consists of distinct functional units, some of which play the traditionally-described cognitive role, while others have closer ties to the more medial and ventral aspects of the PFC. For example, part of BA 9 in the medio-dorsal PFC is probably an important constituent of the viscero-motor network with projections to the lateral column of the PAG in macaques (An et al 1998).
Decreases in the volume of the dorsal cortex, although not always the lateral convexity, have been reported in BD (Table 23). More recently, (Frangou 2005) and (Haznedar et al 2005) described GM volume reductions of the DLPFC (BA 8, 9, 45, 46) in medicated and remitted BD I patients, and partially medicated, “stable” bipolar-spectrum individuals, respectively. More circumscribed volume reductions of BA 9 have also reported in a medicated, euthymic pediatric sample (Dickstein et al 2005). An optimized VBM analysis of a mixed BD I and BD II sample (Lochhead et al 2004) reported reduced GM volume of the ACC immediately dorsal to the corpus callosum (CC).
There is some preliminary evidence for decreased volume of the corpus callosum (CC) in BD (Coffman et al 1990) (Brambilla et al 2003b) (Atmaca et al 2007) and MDD (Lyoo et al 2002a) (Table 25). A meta-analysis of the BD studies supported the hypothesis that the collosal region is reduced in volume in BD (Arnone et al 2008). Congruent with these data, (Yurgelun-Todd et al 2007) reported BD-associated WM changes in the genu of the CC as measured by DTI.
It is unclear from the above studies, however, whether the genu or splenium is preferentially affected. (Ballmaier et al 2007) have proposed that the effect is age of onset-specific: they reported genu-specific thinning in early-onset depressed patients but evidence of thinning of both the genu and splenium in late-onset MDD.
The finding of increased amygdala reactivity to negative stimuli is one of the most consistent findings in the literature (Thomas et al 2001) (Fu et al 2004) (Surguladze et al 2005) (Neumeister et al 2006a). In addition, (Siegle et al 2002) found that amygdalar responses to negative words were no longer visible after 10 seconds in healthy controls but persisted in depressed patients for at least 25 seconds, on average. Similarly, patients with MDD reportedly remember negative words better than positive words (Watkins et al 1992), a finding that correlates with increased BOLD activity of the right amygdala (Hamilton and Gotlib 2008).
The phenomenon may be reversed by antidepressant treatment. (Sheline et al 2001) showed that the elevated BOLD response seen in the left amygdala of depressed patients in response to masked fearful faces was reduced by sertraline. The apparent efficacy of antidepressant medications in attenuating this hyper-reactivity has received support from more recent studies (Fu et al 2004) (Chen et al 2007b). There is some suggestion that rather than being a stable trait, baseline hypermetabolic activity seen in MDD samples may be reflective of acute depression (Abercrombie et al 1998) (Drevets et al 2002b) (Surguladze et al 2005).
The inconsistency in the volumetric MRI literature in MDD may reflect the intrinsic problem that all analyses of the amygdala conducted to date were acquired using 1.5T scanners, which do not have the resolution to allow for the adequate definition of the boundaries of this diminutive region because of its proximity to other grey matter structures such as the claustrum, entorhinal cortex, hippocampus, and basal forebrain.
Increased metabolism of the amygdala – both at base-line, and in response to emotional challenges such as negatively valenced faces (Yurgelun-Todd et al 2000) (Lawrence et al 2004) (Rich et al 2006) (Pavuluri et al 2007) - is commonly reported in BD samples, and may be attenuated by lamotrigine (Chang et al 2008). This increase in amygdala activity is often associated with greater MRI volume.
The amygdala may be an exception to this pattern with neuronal and dendritic hypertrophy reported in parallel to hypermetabolism (Vyas et al 2002) Elevated activity may be an adaptive mechanism that enhances sensitivity to aversive stimuli, thereby facilitating fear conditioning and anxiety-related phenomenology (McEwen and Chattarji 2004) (Radley and Morrison 2005).
Nevertheless, this hypothetical phenomenon cannot explain the decreased neuronal somal size (suggestive of reduced axo-dendritic connections) in the lateral amygdaloid and accessory basal parvocellular nuclei (Bezchlibnyk et al 2007), and the decreased number and density of lateral amygdaloid nucleus neurons (Berretta et al 2007) reported in postmortem BD subjects. The postulated hypertrophy of the amygdala is also inconsistent with the putative volume decrements seen in childhood BD, and a minority of adult samples (Pearlson et al 1997) (Blumberg et al 2003a).
One possibility is that studies reporting enlarged amygdala volumes in BD may be at least partly confounded by medication effects. (Chang et al 2005b) has suggested that hypertrophy of the amygdala may be related to exposure to mood stabilizers. This topic is addressed in detail below.
The amygdala plays a central role in the modulation of monoamine and corticosteroid release in response to aversive or novel stimuli. Disruption of the amygdala’s afferent and efferent connections may thus have negative implications for the regulation of mood and affect (Figure 1).
Hippocampal volume loss appears to be a sequala of depression, particularly in elderly or chronically ill samples. Further, hippocampal volume has been reported to be negatively related to risk of depressive relapse in patients followed over 2 years (Kronmuller et al 2008). Recurrent depressive illness appears to exert an insidious pathological effect on neural, and in particular, hippocampal tissue. Stress-induced dendritic atrophy has been shown to be mediated by an interaction of glucocorticoid receptor stimulation and excitatory amino acid transmission (specifically NMDA receptor stimulation) (McEwen and Magarinos 2001).
Notably, in rodents, the CA3 region of the dorsal hippocampus is the hippocampal area that shows the greatest vulnerability to excitotoxic processes (Sloviter 1996) (Ben-Ari and Cossart 2000) (Dudek and Sutula 2007). In humans with MDD and PTSD, the posterior hippocampus (the putative homologue of the rodent dorsal hippocampus) shows the most prominent reduction in volume relative to healthy controls (Neumeister et al 2005) (Bonne et al 2008).
At sufficiently high concentrations, glutamate acts as an excitotoxin and neuronal death may follow from intra-cellular calcium-driven cytoskeletal degeneration and free radical production (Lee et al 2002). At lower glutamate concentrations, hippocampal neurogenesis is inhibited and dendritic atrophy occurs in association with stress-induced glucocorticoid hormone secretion (McEwen and Magarinos 2001).
The latter effect may be mediated at least in part by down-regulated gene expression of brain-derived neurotrophic factor (BDNF) and other neurotrophins (Duman and Monteggia 2006). Rats exposed to corticosteroids or stressors show decreased expression of BDNF in the hippocampus (Gronli et al 2006) (Jacobsen and Mork 2006) (Tsankova et al 2006) (Xu et al 2006) although the effect may be limited to particular splice-variants of the protein (Nair et al 2006) (Tsankova et al 2006). Similarly humans with mood disorders have shown decreased serum (Karege et al 2002) and hippocampal BDNF levels at post-mortem (Dwivedi et al 2003) (Karege et al 2005).
Stress-induced plasticity of the hippocampus appears to be crucial for effective learning and memory, maintenance of goal directed behavior, and modulation of emotional behavior by contextual information (LeDoux 2000) (Frank et al 2006) (Morris 2006) (Hasler et al 2007). Prolonged prenatal and adult stress in primates, rodents and tree shrews leads to selective hippocampal damage including apoptosis, depressed long term potentiation (LTP) and neurogenesis, as well as, apical dendritic atrophy (Uno et al 1989) (Sapolsky et al 1990) (Watanabe et al 1992) (Magarinos et al 1996) (Czeh et al 2001). Similar effects have been noted in stress-related psychiatric disorders such as post-traumatic stress disorder (PTSD) (Bremner et al 1995) (Vythilingam et al 2005).
If the excitotoxic hypothesis holds true in humans then one would expect to see a negative correlation between the duration or severity of illness and hippocampal volume. There is mixed evidence for this potential effect (Table 5).
In BD the data are less clear, perhaps because of the widespread use of mood stabilizers which appear to increase gray matter volume. Animal studies have demonstrated that lithium promotes hippocampal neurogenesis (Kim et al 2004) and long term potentiation (LTP) (Son et al 2003). A sample of BD patients treated for 4 weeks with lithium showed a 3% (24 cm3) increase in whole brain gray matter volumes from baseline (Moore et al 2000), an effect that appears to result from the neurotrophic effect of the drug (Manji et al 2000). Four more recent studies (Beyer et al 2004b) (Sassi et al 2004) (Bearden et al 2007) (Yucel et al 2007) comparing lithium-treated and non-lithium treated groups demonstrated similar effects in large cortical areas, including the hippocampus. The phenomenon may not be restricted to lithium with comparable effects noted with other classes of mood stabilizers, especially valproate (Mark et al 1995) (Hao et al 2004). In contrast, with the exception of tianeptine (Watanabe et al 1992) (McEwen et al 2002) (McEwen and Olie 2005), the neurotrophic properties of antidepressants are less persuasive - although see (Stewart and Reid 2000) and (Duman and Monteggia 2006).
An alternative explanation for the differential impact of MDD and BD on hippocampal structure is that the two disorders differ in etiology. This hypothesis, however, stands in contradistinction to evidence that MDD tends to aggregate in the relatives of individuals with BD (McGuffin and Katz 1989).
A disruption of hippocampal function may contribute to the deficits in executive performance, learning and emotion-mediated memory formation observed in mood disorders (Figure 1).
Both hypometabolism or blood flow and reduced grey matter volume have commonly been reported in the MDD literature (Tables 9 and and10).10). The former result may thus reflect a partial volume averaging effect rather than a genuine diminution in metabolic activity. Interpretation of imaging studies is, however, rendered difficult by the mass of contradictory results yielded through diverse scanning methodologies and population samples. We suggest that medication and age of onset are important confounding factors. In contrast, increased resting metabolic activity of the anteroventral striatum is apparent during depression, and may be linked particularly to anhedonic symptoms (Hasler et al 2008), consistent with evidence that depressives show blunted caudate activation during exposure to positively valenced sensory stimuli (Epstein et al 2006).
No clear conclusions can yet be drawn from studies of the BG. There is some suggestion of increased striatal metabolism and volumetric GM increases in BD. As in the case of the hippocampus, it is plausible that these analyses are confounded by treatment effects. Enlargement of the BG nuclei is a well known effect of anti-psychotic drugs (Jernigan et al 1991); (Swayze et al 1992); (Chakos et al 1995); (Frazier et al 1996) and a perusal of Table 11 indicates that three of the studies reporting BG enlargement made use of partially manic samples treated with anti-psychotic medication (Strakowski et al 2002); (DelBello et al 2004); (Wilke et al 2004).
Metabolic activity or blood flow was consistently elevated in manic or hypomanic samples (Blumberg et al 2000) (Caligiuri et al 2003) (Caligiuri et al 2006) (Table 10). Nevertheless, hypermetabolism of the BG has also been reported in depressed samples, specifically in the ventral striatum (Ketter et al 2001) (Bauer et al 2005) (Mah et al 2007).
Impaired striatal function may explain the anhedonia and reduction in goal-seeking behavior that is observed in some patients with MDD and depressed BD cases. Conversely, the elevation in psychomotor activity is congruent with reports of increased metabolism or blood flow in more dorsal regions of the striatum in the manic phase of BD.
Ventricular enlargement, usually of the third or lateral ventricles, is characteristic of older and chronically ill patients with depression, or patients with a late age of depression-onset. The white matter hyperintensities seen concomitantly in these patient groups raise the possibility of a vascular etiology for depression (Krishnan 2002) (Knopman et al 2005) (Sneed et al 2008).
Clearly, the hypothesized excitotoxic processes which may be operating in medial temporal and striatal tissue could theoretically cause ventricular enlargement. Nevertheless, the degree of subcortical atrophy needed to manifest as ventricular enlargement remains unclear. If long-term pathological processes are a prerequisite for ventriculomegaly then this may explain why the phenomenon appears to predominate in elderly populations. Factors such as chronic alcohol abuse (Anstey et al 2006) and incipient neurological disorders with prodromal depression may also contribute to enlargement of ventricles.
These data are congruent with the neuropathological studies of (Thomas et al 2002) and (Thomas et al 2003) which describe WM disease of the DLPFC in a sample of elderly depressed patients, a result replicated in a middle-aged sample of depressed individuals who showed reduced levels of myelin staining in the deep WM tracts of the DLPFC (Regenold et al 2007). One possible explanation for this potential demyelination is change in oligodendrocyte function. At least three post-mortem studies have reported a down-regulation of oligodendrocyte-related gene expression in MDD and BD (Hakak et al 2001) (Tkachev et al 2003) (Aston et al 2005), while decreased oligodendrocyte density has also been noted in several studies (Cotter et al 2002) (Hamidi et al 2004) (Uranova et al 2004) (Vostrikov et al 2007).
An unresolved issue is whether the WM pathology-depression relationship is one of cause or effect. Certainly, there appear to be cases that are precipitated by subcortical infarcts. For example, depression is a relatively common sequala of the genetic disorder, cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) (Chabriat et al 1995) (Desmond et al 1999). Similarly, given the increased rate of risk factors for cerebrovascular disease observed in MDD samples (Steffens et al 1999); (Lyness et al 2000) (Wassertheil-Smoller et al 2004), it is likely that subcortical ischemic events play a role in the development of late-onset MDD.
Nevertheless, some studies have made attempts to match patients and controls (or statistically control) for the presence of cerebrovascular risk factors and still find elevated rates of WMH in their MDD samples (Table 13). A possible explanation is that depression may predispose to the development of small WM lesions via some as yet occult pathophysiological mechanism.
Based on extant evidence in the literature (Biegon et al 1990) (Musselman et al 1996); (Lenze et al 1999) and (Nemeroff and Musselman 2000) have speculated that excess depression-associated secretion of serotonin by blood platelet cells facilitates cellular aggregation and therefore predisposes to atherosclerosis, thrombosis and vasoconstriction. Another possibility is impaired regulation of vascular tone. Cerebrovascular reactivity, which describes the compensatory dilatory capacity of arterioles to dilatory stimuli, has been reported to be reduced in acutely depressed patients without any neurological, cardiac or vascular risk factors (de Castro et al 2006).
Ventriculomegaly is more commonly observed in BD across different age groups (Swayze et al 1990); (Figiel et al 1991); (Strakowski et al 1993); (Botteron et al 1995); (Zipursky et al 1997); (Lim et al 1999); (Strakowski et al 1999); (Hauser et al 2000); (Brambilla et al 2001a); (Strakowski et al 2002); (Davis et al 2004); (Strasser et al 2005); (Soares et al 2005).
It is plausible that as in the case of MDD, WM abnormalities are a long-term consequence rather than a cause of bipolar illness. One possibility is that people with BD have an excess of atherosclerotic risk factors that lead to microvascular pathology at an even earlier age than MDD. As reviewed by (Kilbourne et al 2004) and (Newcomer 2006), BD is associated with a significantly increased prevalence of cardiovascular disease risk factors such as smoking, obesity, diabetes mellitus (DM), hypertension and dyslipidemia. Hypertension (Dufouil et al 2001) (Gunstad et al 2005), obesity (Gustafson et al 2004), smoking (Dager and Friedman 2000) and DM (Novak et al 2006) have in turn been directly associated with the development of WMH. Further, drug abuse is prevalent in BD populations and stimulant drug-induced vasoconstriction may lead to WMH (Dupont et al 1990) (Lyoo et al 2004b). Marijuana may also interact in an additive fashion with WMH to predispose to depressive symptomatology (Medina et al 2007).
Although most published studies make an attempt to exclude patients with potentially confounding conditions, the whole gamut of risk conditions is very rarely controlled for, raising the possibility that WMH in BD are an artifact of medical co-morbidity ischemic risk factors.
Nevertheless, this hypothesis fails to account for the WM pathology noted in pediatric BD samples (Botteron et al 1995); (Lyoo et al 2002b); (Pillai et al 2002) as well as the high concentration of WMH in both BD subjects and their unaffected relatives reported by (Ahearn et al 1998). We suggest that although a proportion of BD individuals with significant WM pathology will present with cardiovascular risk factors, WMH may less commonly also co-occur with some kind of developmental insult. In other words, the WMH seen in pediatric or young adult BD samples are likely to have a different origin to those seen in older populations.
The etiology of these precocious lesions is unclear. Obstetric complications are well known to be associated with schizophrenia (see meta-analysis of (Cannon et al 2002)), but with a few exceptions (Kinney et al 1993); (Kinney et al 1998), appear less salient in BD. Nevertheless, it is possible that perinatal hypoxic events precipitate bipolar illness in a vulnerable minority (Pavuluri et al 2006).
The high incidence of familial WMH seen in the (Ahearn et al 1998) sample suggests that genetic factors may also be at play. A number of post-mortem analyses provided evidence for the altered gene expression of genes impacting myelin or oligodendrocyte function in both MDD and BD (Tkachev et al 2003) (Aston et al 2005) (Sequeira et al 2006). As reviewed by (Carter 2007a), (Carter 2007b), and (Sokolov 2007) variants of some of these genes such as oligodendrocyte lineage transcription factor 2 (OLIG2) [NCBI accession number 10215] Neuregulin 1 (NRG1) , and v-erb-a erythroblastic leukemia viral oncogene homolog 4 (ERBB4)  have been directly associated with affective illness and may determine how resilient these cells are to environmental stressors2. This hypothesis will be discussed in greater detail below.
Lesions to the deep WM tracts of the frontal cortex and BG disrupt communication between key components of the viscero-motor network, resulting in a so-called “disconnection syndrome” with sequalae that bear some resemblance to the symptomatology induced by GM lesions to the individual components of the network.
The evidence for increased metabolism of the OFC in depressed samples receives support from serotonin and catecholamine depletion studies which have elicited a similar effect in remitted patients as they develop depressive relapse (Neumeister et al 2004) (Hasler et al 2008). Congruent with these data, treatment with antidepressant medication (Brody et al 1999), psychotherapy (Goldapple et al 2004), and deep-brain stimulation (Mayberg et al 2005) has been associated with decreased metabolism of the OFC. The combination of elevated OFC activity and tissue loss in MDD samples is consistent with the operation of an excitotoxic process.
The structural MRI-derived evidence of OFC volume loss (Table 16) is supported by the post-mortem analysis of (Rajkowska et al 1999) which uncovered a 12% decrease in the thickness of the rostral (but not the caudal) OFC, and a 15% decrease in the middle OFC in depressed patients compared to healthy controls; a finding attributed to neuronal shrinkage with attendant reduction in volume of the neuropil. Further, a 15% decrease in glial cell density was observed in the caudal OFC.
Nevertheless, we note that a significant number of the patients in the samples demonstrating volume reductions became ill after the age of 40, raising the issue of etiological heterogeneity.
BD samples also appear to be characterized by tissue loss in the OFC as measured by MRI. These data are supported by the post-mortem analysis of (Cotter et al 2005) who found evidence of reduced glial cell density and neuronal size in the caudal OFC of patients with BD. Further, the structural integrity of WM tracts in the OFC, as measured by diffusion tensor imaging (DTI), may also be compromised in adult (Beyer et al 2005) (Haznedar et al 2005) and pediatric (Frazier et al 2007) patients. GM loss secondary to excitatory metabolic disturbances is supported by the MRS study of (Cecil et al 2002) who recorded reduction of NAA and choline concentrations in the orbitofrontal GM of a BD sample.
Too few resting state studies have been conducted to reach a conclusion about BD-associated OFC metabolism. Nevertheless, fMRI studies using cognitive probes of behavioral inhibition and decision-making, have reported decreased OFC BOLD response in patients with mania (Elliott et al 2004) (Altshuler et al 2005b). Similarly, the passive viewing of emotional material has been associated with a relative decrease in the BOLD response of the left OFC in a patient with post-partum psychosis compared with her monozygotic cotwin (Fahim et al 2007). It is unclear how these fMRI finding relate to the putative OFC volume loss observed in BD.
The OFC plays major roles in the regulation of emotional and by implication, social behaviour. Lesions to the OFC have been associated with depression, mood instability and anxiety in other clinical case-studies (Grafman et al 1996) (MacFall et al 2001), providing further evidence that this part of the brain is critical for the self-regulation of emotion and behaviour.
Theoretically, therefore, volumetric and perhaps metabolic changes to the OFC may account for the emotional and cognitive disinhibition often characteristic of affective illness. Reduced orbitofrontal activity in BD patients during performance of an emotional go/no-go task (Elliott et al 2002) (Elliott et al 2004) as well as the Stroop task, a measure of cognitive control, (Kronhaus et al 2006) has previously been noted. These data are relevant because impaired performance on the Stroop and go/no-go tasks have been separately demonstrated in MDD and BD samples (Erickson et al 2005) (Savitz et al 2007b).
The modulation of visceral responses to affective stimuli seems to be crucial for translating OFC-derived valenced data into actions and behaviour (Ongur et al 2003), an idea popularized by Damasio and colleagues (Damasio 1996) (Bechara et al 1997) as the “somatic marker” (SM) hypothesis.
According to the SM conjecture, a somatic marker, an unpleasant or positive viscerally-generated stimulus such as anxiety or reward, is temporally-paired with a predicted future outcome. This juxtaposition of cognitive and emotional inputs ensures competent decision-making in complex situations because it allows individuals to anticipate the future positive and negative consequences of their actions. When this process goes awry, as in the case of individuals with lesions to the ventro-medial cortex, impaired judgment results (Damasio 1996).
Recent studies are indicative of increased metabolism of the sgACC in the depressed phase relative to the remitted phase of MDD patients (Drevets et al 2008). Congruent with these data, a tryptophan depletion study of healthy subjects showed that serotonin depletion resulted in increased rCBF in the sgACC (BA 24 and 25) (Talbot and Cooper 2006). (Neumeister et al 2004) had previously reported the same phenomenon using glucose metabolic measures in remitted, unmedicated patients with a history of MDD. Similarly, the depressive relapse associated with alpha-methyl-para-tyrosine induced catecholamine depletion in remitted MDD subjects was also associated with increased metabolism in the sgACC (Hasler et al 2008).
Changes in the metabolic activity of the ventro-medial circuit have also been reported after treatment and improvement in symptomatology. (Mayberg et al 2000) found that people who responded to a trial of fluoxetine showed decreased bilateral activity of the sgACC, while treatment with sertraline resulted in significant decreases in the activity of the left sgACC (Drevets et al 2002a). Two sleep deprivation studies showed a drop in elevated baseline levels of sgACC activity in depressed patients who responded to treatment (Wu et al 1999) (Clark et al 2006). In another indication that these changes may be independent of treatment modality, a deep-brain stimulation study (Mayberg et al 2005) showed reduced perfusion of both the anterior and posterior portions of the sgACC (BA 25 and part of 24, respectively; Figure 2) in treatment responders..
The ventral ACC has also been demonstrated to mediate fluctuations within the normal range of emotion. For example, (Mayberg et al 1999) asked healthy and depressed subjects to recall sad personal experiences and found that negative mood induction was associated with increased activity of the sgACC. On the other hand, women grieving the break-up of a romantic relationship have been reported to display reduced activity in the left pregenual and sgACC (Najib et al 2004).
Anatomical MRI studies have provided evidence for decreased sgACC volume in both MDD and BD (Drevets et al 2008). Moreover, (Ongur et al 1998) found a reduction in the number of glial cells together with an increase in neuronal density in the sgACC of patients with familial BD and MDD, suggesting that the reduction in neuropil is accounted for by the reduction in sgACC volumes reported in the literature. Using the same set of brain tissue, these findings were partially replicated by (Cotter et al 2001) who detected reduced glial density and neuronal size (but not density) in the region of pregenual ACC in patients with major depression. An analogous effect in the sgACC in both familial and non-familial depression with glial cell density reductions of 24% and 41%, respectively (Ongur et al 1998) (Drevets 2000b). These data receive support from animal models of depression.
Rats subjected to 3 weeks of repeated restraint stress show a 16–20% reduction of apical dendritic spine density in the anterior cingulate region of the medial PFC (Liston et al 2006) (Radley et al 2006). A corresponding effect has also been noted in the pups of mothers exposed to prenatal stress (Murmu et al 2006). This dendritic remodeling process has been shown to depend on interactions between NMDA receptor stimulation and glucocorticoid release during chronic or repeated stress, suggesting that the dendritic atrophy is a type stress-induced, slow excitotoxicity (Reagan and McEwen 1997).
These anatomical changes may correlate with behavior: Restraint stress-associated retraction of apical dendrites in the medial PFC of rodents has been associated with retarded extinction learning (Miracle et al 2006). The same effect may hold in humans. Healthy individuals with thinner ventromedial PFC tissue show a greater galvanic skin response to conditioned stimuli during extinction learning (Milad et al 2005). Chronic psychosocial stress may also impact glial cell function: rats exposed to 5 weeks of daily social defeat have been reported to show a suppression of gliogenesis in the medial PFC; an effect reversed by fluoxetine (Czeh et al 2007).
Interestingly, ibotenic acid-induced lesions of the rat mPFC (infralimbic/prelimbic and ACC cortices) have been shown to produce evidence for a lateralized effect: right but not left-sided lesions were anxiolytic (Sullivan and Gratton 2002). In contrast, left-sided lesions are anxiogenic, and result in increased sympathetic activity and corticosterone response to stress. This result is reminiscent of an earlier study in which dopamine depletion of the left but not right medial PFC rendered animals vulnerable to stress-associated ulceration (Sullivan and Szechtman 1995). Further, (Sullivan and Gratton 1999) found that left-sided lesions potentiate autonomic stress responses to restraint stress while right-sided lesions diminish these endocrine reactions. The putatively differential effect of the two hemispheres appears to follow a phylogenetic continuum.
Babinski (Babinski 1914) first discussed the anosognosic and manic sequelae of right-sided lesions while Goldstein (Goldstein 1939) coined the term, “catastrophic reaction” to describe the emotional distress and depressive symptomatology associated with left-hemisphere lesions. More modern structural imaging studies have lent support to the original clinical reports (Robinson et al 1984) (Morris et al 1996) and studies of idiopathic anxiety, depression, and BD have produced analogous findings (Silberman and Weingartner 1986) (Davidson 1992) (Pascual-Leone et al 1996) (Savitz et al 2004) (McDonald et al 2004b).
The physiological mechanisms underlying these effects have been extensively debated. Explanations generally circle around the theme of a positive emotion-producing, appetitive, reward-seeking circuit in the left PFC which interacts in a dialectical manner with a mirror neural network in the right-hemisphere (Flor-Henry 1979) (Sackeim et al 1982) (Savitz et al 2007a). Conceivably, a disruption of the regulatory capacity of these circuits, such as disinhibition of the right hemisphere after left hemisphere damage, may lead to affective illness.
Consistent with the above hypothesis, differential hemispheric functional activity of the mPFC has been reported in the literature. Nevertheless, caution should be exerted in interpreting these data because the PET data are too low in resolution to accurately resolve left from right in regions close to the midline.
Since the vast majority of the studies samples were limited to currently depressed patients, it is unclear whether the possible dichotomy in hemispheric activity is a stable trait which applies to euthymic populations. There is some evidence that it is more likely to be a state-dependent phenomenon. In both the (Drevets et al 1997) and the (Blumberg et al 2000) cohorts, the manic subjects had greater rCBF or metabolism in the ACC then in depressed or remitted individuals (in contrast, the volumetric deficit in the sgACC persisted across illness phases). Further, (Liotti et al 2002) reported that after the execution of a sadness-induction paradigm, rCBF was higher in the sgACC in healthy controls compared to remitted depressives.
It should be noted that a minority of researchers have reported the diametric pattern of right-sided hypometabolism or volume loss and left-sided hypermetabolism in MDD (Drevets et al 1992) (Wu et al 1999) (Kennedy et al 2001) and BD (Kegeles et al 2003) (Sharma et al 2003). In the case of the latter study, however, the sample was euthymic while the (Kegeles et al 2003) cohort was composed of both BD and MDD patients with comorbid anxiety disorders and was characterized by benzodiazepine (BZ) use at the time of the scanning. Further, these researchers reported right-sided hypometabolism of the pregenual as opposed to the sgACC and this may have accounted for the discrepancy in results.
Nevertheless, (Davidson and Irwin 1999) warn that most claims about hemispheric asymmetry of function are methodologically flawed because they are based on the observation that voxel-based activity in only one hemisphere reaches statistical significance. It does not, however, necessarily follow from this that the two hemispheres are significantly different from each other. This hypothesis needs to be tested by examining group or condition × hemisphere interactions (Davidson and Irwin 1999).
(Fornito et al 2007) favor a developmental rather than a stress-driven explanation for the sgACC volumetric changes believed to be characteristic of affective illness. These authors examined the folding patterns of the ACC, and reported a lower incidence of the adjacent paracingulate sulcus (PCS) in BD subjects compared with controls. Nevertheless, given the fact that the PCS is absent in many healthy individuals, the functional significance of this pattern of cortical folding is unclear.
Autonomic, endocrine and metabolic (McIntyre et al 2007) (Rottenberg 2007) (Smolin et al 2007) (Harvey 2008) (Surtees et al 2008) abnormalities are over-represented in individuals with affective illness. The ventromedial PFC, with its connections to the hypothalamus and brain-stem, modulates visceral output in response emotional stimuli (Price 2007). Conceivably, therefore, damage to the ventromedial PFC may not only dysregulate emotional function, but may play a pathoetiological role in the development of comorbid medical conditions.
Hypometabolism and to a lesser extent GM volume decrements of particularly the left DLPFC have been widely reported in people suffering from depression. These data are consistent with the executive and working memory deficits characteristic of many such patients (Savitz et al 2005). They are also congruent with fMRI analyses that have reported altered BOLD signal in the DLPFC of MDD patients challenged with cognitive tasks that require executive control (Harvey et al 2005) (Matsuo et al 2007) (Siegle et al 2007).
As is the case with other neural networks, one complication of this literature is the potentially lateralized function of the PFC. Based on our rudimentary understanding of how emotional processing is lateralized in the brain, one would expect a preponderance of left dorsal cortex hypoactivation in depression. We are aware of at least 9 studies that have recorded left-sided hypofunction or GM loss of the dorsal PFC in MDD (Bench et al 1995) (Shah et al 1998) (Kennedy et al 2001) (Drevets et al 2002a) and BD (Baxter et al 1989) (Martinot et al 1990) (Lopez-Larson et al 2002) (Frangou 2005) (Dickstein et al 2005). Furthermore, right hemisphere hypoactiviation in MDD responders to sleep deprivation has been recorded (Clark et al 2006). Congruent with these reports, (Beauregard et al 2006) found that those individuals who experienced difficulty actively down-regulating sad thoughts showed a greater (presumably compensatory) response in the left mPFC (BA 10). More recently, (Grimm et al 2007) reported left-sided hypoactivity and right-sided hyperactivity in MDD patients, who as a group, tended to judge pictures as more negatively valenced than healthy controls.
This hypothetical lateralization of function could theoretically explain the minority of studies detailing increased activity of the DLPFC in MDD (Brody et al 2001) (Holthoff et al 2004). Nevertheless, none of these studies reported exclusive hypermetabolism of the right hemisphere. However, in one study the patients were only withdrawn from benzodiazepines three days prior to scanning, potentially confounding the metabolic measures (Holthoff et al 2004). Decreased activity of the right (Hurwitz et al 1990) DLPFC in depressed MDD patients has also been noted but again the withdrawal of benzodiazepines one day prior to scanning is a significant limitation.
Consistent with the notion of GM atrophy, histological changes to the DLPFC have been reported. Although no significant differences in cortical density were apparent across their MDD and control groups, (Rajkowska et al 1999) noted a depression-associated decrease in neuronal size with a concomitant decline in large cell density and increase in small neuronal cell density. Further, suicide has been associated with microgliosis of the DLPFC in a post-mortem schizophrenia and MDD sample (Steiner et al 2008)
BD samples have tended to display evidence of dorsal PFC GM volume loss, and these putative changes are consistent with reports of aberrant dorsal PFC gene expression in patients with BD (Nakatani et al 2006) (Beneyto and Meador-Woodruff 2007) (Kanazawa et al 2007) (Pennington et al 2007) (Pillai 2008) (Shao and Vawter 2008) (Silberberg et al 2008) – although see (Ryan et al 2006) for a negative report. The mechanism by which these putative molecular changes impinge on the development of gross anatomical structures such as the DLPFC is not yet understood.
The functional consequences of callosal WM abnormalities are unclear. One proposal is that a disruption to inter-hemispheric transmission impairs cognitive function in people with mood disorders (Brambilla et al 2004). Damage to the CC, however, may retard the transfer of visual, motor, linguistic and somatosensory information in a regionally specific manner, and moreover, there may be some redundancy in these information highways, both within the CC itself, and subcortical circuitry (Gazzaniga 2005). Given the absence of salient deficits in these domains in affective illness, it seems likely that any lesions of the CC would preferentially affect cognitive or emotional processing. One possibility is that volume loss of the CC is secondary to abnormalities of the surrounding GM. Further work is needed to clarify the relationship between potential demyelination of the CC and cognitive and/or emotional impairment.
The heritability of BD converges on the 60–80% range (McGuffin et al 2003) (Kieseppa et al 2004) while the heritability score for MDD is closer to the 40% mark (Agrawal et al 2004) (Boomsma et al 2005). Thus although genetic factors play an important etiological role in affective disorders, the importance of environmental variables should not be discounted. In fact, the average concordance rate among monozygotic (MZ) and dizygotic (DZ) twins stands at approximately 40% and 10%, respectively, for BD (McGuffin et al 2003) (Kieseppa et al 2004), and approximately 30% and 20% for MDD (Lyons et al 1998) (Kendler and Prescott 1999).
One potentially important environmental variable is exposure to adversity: stressful life events are known to precipitate bouts of affective illness in both BD (Kessing et al 2004); (Johnson 2005) and MDD (Kendler et al 1999) (Hettema et al 2006) populations.
Current models of pathological change in affective illness are deeply rooted to these environmental contingencies, as functional and anatomical change in depression is often viewed as degenerative: stress-induced, glutamate-mediated excitotoxicity leads to metabolic changes or decreases in GM volume. In contrast, the role of genetically-mediated neuroplasticity is not usually explicitly addressed in contemporary models of illness-associated neuropathology. We will first discuss the neurodegenerative hypothesis, which conceptualises GM or WM volume loss as a downstream effect of an unspecified environmental pathogen, such as psychological stress.
The degenerative model is based on the premise that psychological stress or some other pathophysiological insult is a necessary determinant of affective illness. Given the clustering of MDD and BD in families, and some reports that GM volume loss only occurs in familial cases (Drevets et al 1997) (Hirayasu et al 1999a) (Ongur et al 1998) (Brambilla et al 2001a) (Kegeles et al 2003) (Lacerda et al 2005) (Kogelnik et al 2006), one possibility is that psychological stress must also be familial in order for this conceptualization of illness to be valid.
Actually, there may be a familial or genetic component to the experience of life stress because of a phenomenon known as gene-environment correlation (Bergeman et al 1988); (Rutter 2007). (Rowe 1981) first reported this counter-intuitive notion by showing that adolescent twins’ reports of their parents’ levels of accepting and rejecting behaviour were under genetic influence; a finding that has been extended to retrospective measures of family warmth and parental control (Plomin et al 1988) as well as family cohesion and encouragement of growth (Bouchard et al 1990).
There are at least three ways in which gene-environment correlation might apply in the case of the degenerative hypothesis. The parental genotype or affective disorder may exert an effect on parental behaviour such that their children are reared in a high-stress environment. Here there is a correlation between passing on “stress-provoking” genes and providing a stressful family environment.
Secondly, it can be argued that people are selecting and shaping their environmental experiences on the basis of their genetic heritage, leading to preferential exposure to significantly stressful events and depression-associated neuroplastic changes in a subset of the population. This phenomenon may be related to the way in which individuals perceive or process information in their environment; an intrinsic bias often described as temperament.
Certainly, so-called dysthymic or anxiety-related personality traits have been widely described in MDD individuals, while BD is characterized by various combinations of dysthymic, cyclothymic-unstable and hypomanic traits (Evans et al 2005) (Savitz and Ramesar 2006) (Savitz et al 2008a) (Savitz et al 2008b). Since these traits are likely underpinned by genetic factors (reviewed in (Savitz and Ramesar 2004) (Ebstein 2006)), temperament may mediate the impact of genes on environmental experiences (Savitz and Ramesar 2006).
Yet a third possibility is that genetic effects play no role in influencing exposure to stressors but moderate the physiological effect of these events on neural tissue. One class of proteins potentially involved in this type of gene-environment interaction is the neurotrophins.
As reviewed by (Poo 2001), one of these enzymes, brain-derived neurotrophic factor (BDNF) , increases forebrain serotonin fibre density and neurogenesis, prevents spontaneous and neurotoxin-induced cell death, and modulates the formation of synaptic connections, particularly in the PFC and hippocampus.
Recent studies have suggested that the low expression (met) allele of a functional single nucleotide polymorphism (SNP) (Val66Met) of the BDNF gene may increase the probability of developing depression (Kaufman et al 2006) and cognitive impairment (Savitz et al 2007c) after exposure to childhood maltreatment. Perhaps through its reduced ability to protect against neurotoxicity, the met allele has also been reported to increase the risk of developing depression after stroke (Kim et al 2007).
Another potential moderator of the stress-response is central serotonergic activity. As reviewed by (Drevets 2000b), the binding of serotonin to post-synaptic 5-HT1A receptors not only enhances the negative feedback inhibition of cortisol release but also prevents dendritic cytoskeletal breakdown by catalyzing the release of the neurotrophic factor, S100β, and indirectly inhibiting protein kinase-induced apoptosis (Szatmari et al 2007). The regulation of 5-HT1A receptors in the raphe is at least partly controlled by functional variants of the HTR1A  (−1019C/G; rs6295) (Lemonde et al 2003) (Parsey et al 2006) and the SLC6A4  (promoter region length polymorphism) (David et al 2005) genes, respectively.
In contradistinction to the degenerative model, the developmental model advocates that neuroanatomical changes precede the onset of affective illness. An interesting set of animal experiments has lent credence to this hypothesis: A line of rats, genetically-bred to suffer from learned helplessness display baseline hypometabolism of the amygdalae, BG, VTA, dorsal-frontal, medial-OFC and ACC, but increased metabolism of the infraradiata (sgACC), hippocampi and habenula (Shumake et al 2000); (Shumake et al 2002). To further control for the effects of early-life stress, (Shumake et al 2004) examined the brains of this genetic line of rats at birth and again found hypo or hypermetabolism of most of these regions. Moreover, the midbrain and brain-stem regions were found to be disconnected from limbic forebrain regulation, suggesting to the authors that the fundamental disturbance in depression is one of top-down regulatory control (Shumake et al 2004).
In humans one way of examining this issue is to compare the degree of variation in regions of interest across the life-span. (Lupien et al 2007) found that there was just as much variability in the hippocampal volumes of healthy young adults as in older individuals, implying that volume decrements attributed to aging or stress could be reflective of neurodevelopmental differences. Specifically, a quarter of their subjects in the 18–24-year age group had hippocampal volumes as small as the average hippocampal size in their 60–75-year-old sample, and the mean difference in hippocampal volumes between the upper and lower quartiles of the young age group (12–16%) was greater than volumetric reductions typically seen in depressed samples (Lupien et al 2007). These data are congruent with an earlier study (Gilbertson et al 2002) which reported an association between post-traumatic stress disorder (PTSD) and smaller hippocampal volume in war veterans. Intriguingly, the MZ twin brothers of the PTSD cohort who did not serve in the military also presented with smaller hippocampi than the PTSD group, raising the possibility that reduced hippocampal volumes are a contributing cause rather than an effect of PTSD.
Recent findings from the emerging field of imaging genomics also emphasize the importance of genetic influences. A variant of the neuregulin 1 (NRG1)  gene, which is involved in the myelination process, and has been implicated in both BD and schizophrenia, may be associated with WM density and integrity of the internal capsule (McIntosh et al 2007). In another study, the short (s) allele of the serotonin transporter (5-HTT) gene (SLC6A4)  promoter polymorphism has been shown to be associated with increased resting CBF in the amygdala and decreased perfusion of the ventromedial PFC in healthy individuals (Rao et al 2007).
The neurodevelopmental hypothesis can also be evaluated by searching for neural changes in unaffected family members who presumably share a genetic diathesis for the disorder with their ill relatives. (Noga et al 2001) compared MZ pairs discordant for BD with a control group of unaffected twins, and found that the right hippocampus was smaller in the affected twins, but that both ill and well twins had larger caudate nuclei than the control pairs. (McDonald et al 2004a) reported that genetic risk for BD was associated with reduced volume of the right anterior cingulate gyrus and ventral striatum. Similarly, a compensatory hypermetabolic response to a sadness induction paradigm was observed in the medial PFC of healthy BD relatives compared with background controls (Kruger et al 2006). Conversely, (McIntosh et al 2006) and (Munn et al 2007) failed to detect any significant neural changes in at-risk BD relatives.
Concerning MDD, (Monk et al 2007) reported that the pediatric offspring of parents with MDD displayed greater amygdala and accumbens area activity in response to fearful faces, and lower accumbens area activation in response to happy faces than a low-risk control group. A small number of serotonin depletion experiments have succeeded in inducing depressive symptoms in otherwise healthy relatives of MDD probands. Six out of 20 healthy males with a family-history of affective illness but 0 out of 19 male controls displayed a lowering of mood in response to tryptophan depletion (Benkelfat et al 1994). More recently, (van der Veen et al 2007) reported a greater lowering of mood, and stronger amygdala response to fearful faces under tryptophan depletion in healthy individuals with a family history of depression compared with controls. (Neumeister et al 2002) showed that this effect may be mediated by the SLC6A4 promoter length polymorphism: family-history-positive heterozygotes showed a greater mood-lowering effect than their heterozygote counterparts without a history of depression.
These potential neuroimaging endophenotypes are most likely genetically-driven. For example, serotonergic activity not only plays an important role in modulating the impact of stressful events, but is a key regulator of neural development influencing neurogenesis, apoptosis and dendritic growth (Gaspar et al 2003). Fluoxetine-induced suppression of the serotonin transporter during development has been shown to result in abnormalities of emotional behaviour in mice (Ansorge et al 2004). Recent data have also shown the direct effect of the 5-HTT polymorphism on neural tissue with increased hemodynamic response of the amygdala (Hariri et al 2002); (Heinz et al 2005); (Pezawas et al 2005); (Dannlowski et al 2006) to fearful faces (versus geometric images or neutral faces) in s allele carriers.
The distinction between early, developmental, and later-onset stress-induced pathology may be partly artificial. Genetically-determined subtleties of brain-wiring may sensitise the individual to the effect of ubiquitous, relatively mild stressors.
The s allele of the above-mentioned 5-HTT gene insertion/deletion polymorphism, first associated with anxiety-related personality traits more than a decade ago (Lesch et al 1996), is now one of the prototypical examples of a risk variant that interacts with adverse life experiences to predispose to psychopathology (Caspi et al 2003); (Kaufman et al 2004).
(Pezawas et al 2005) investigated the neurobiology behind this process. The authors demonstrated that healthy carriers of the s allele of the polymorphism display reduced functional connectivity between the amygdala and the sgACC. The latter structure exerts an inhibitory effect on the amygdala3 and thus a genetically-determined attenuation of this negative-feedback loop may increase sensitivity to environmental adversity and by implication, lead to maladaptive neuroplastic changes (Pezawas et al 2005).
A nascent trend in the study of psychiatric disorders is a recognition of the potentially important role of epigenetic mechanisms in disease causation (see (Mill and Petronis 2007)). Epigenetic inheritance refers to a regulated pattern of gene expression which is transmitted intact from one or other of the parents to their offspring. The process is mediated by the methylation and histone acetylation of cytosine residues and chromatin, respectively, leading to the activation or silencing of particular genes. The phenomenon is epigenetic because it results in phenotypic traits that are inherited independently of the informational content of DNA.
(Meaney and Szyf 2005) review rodent studies which demonstrate that stress sensitivity in rat pups is modulated by parental grooming behavior that exerts its effect through a histone modification-driven regulation of glucocorticoid receptor gene expression. If these biological mechanisms generalize to humans then exposure to adversity may modify gene expression in pathways that impact neuroplasticity.
(1). A common confounding variable is medical treatment. As (Drevets 1998) notes, rCBF and metabolism of various neuroanatomical regions may be reduced by antidepressants, anti-psychotics and anxiolytics. Further, psychotropic drugs have been shown to alter both the behavioral and the neurophysiological response to the environmentally valenced stimuli used as neurocognitive probes in fMRI studies.
Recent studies have detailed the neurotrophic effects of lithium and mood stabilizers on hippocampal and other neural tissue. In fact, there has been some suggestion that AD may also exert a neurotrophic effect in particular regions of the brain (Stewart and Reid 2000) (Rocher et al 2004) (Duman and Monteggia 2006). These data are supported by an fMRI study which showed that depressed patients suffer from reduced functional coupling of the amygdala with diverse brain regions including the hippocampus, caudate, putamen, and ACC; an effect reversed by treatment with fluoxetine (Chen et al 2007b). Similarly, a decrease in left amygdala activity to normal has been observed after antidepressant treatment (Drevets et al 2002a). Interestingly, a decrement in amygdala activity in response to aversive stimuli also been observed in healthy individuals treated with citalopram (Harmer et al 2006) and reboxetine (Norbury et al 2007).
Neuroleptics, which are commonly used to treat BD, may also exert a neurotrophic effect, with reports of post-treatment increases in basal ganglia volumes (or shape) in patients with BD (Hwang et al 2006) and schizophrenia (Chakos et al 1994) (Gur et al 1998) (Massana et al 2005) (Glenthoj et al 2007) (Okugawa et al 2007).
The association between imaging changes and medication is rendered even more complicated by the possibility that treatment may be a proxy variable for genetic etiology. In other words, the hippocampal or sgACC volumetric changes associated with a drug like lithium, for example, could theoretically be characteristic of a particular subtype of BD that just happens to be responsive to lithium (Moore et al In press). This is one example of etiological heterogeneity that may be a significant contributor to inter-study variability.
(2). Another example of disease heterogeneity is the difference in the pathophysiological mechanisms underlying early-onset and late-onset depression discussed above. Other hidden etiological differences may be an even more pernicious source of bias. In genetic studies, a distinction is often made between those patients who are recurrently ill and those subjects who have experienced one life-time episode of depression (Zubenko et al 2002), but this is rarely seen in the imaging literature. Further, the degree to which MDD or BD runs in the families of recruited subjects or whether MDD patients are recruited from BPD families is not always detailed. Differences between psychotic and non-psychotic “subtypes” of BD may also contribute to inter-study variability in findings. Approximately 50% of BD I patients experience psychosis (Keck et al 2003) (Ketter et al 2004) and recent evidence suggests that working memory deficits may be specific to those patients with a history of psychosis (Glahn et al 2007) (Savitz et al in press).
Information about obstetric or other developmental problems is rarely reported. The diagnosis of BD in children has increased sharply (mostly in the Unites States) in recent years, yet it is unclear as to whether children diagnosed with BD actually suffer from the same disorder as adults (Duffy 2007). This is particularly true when the broader phenotype of severe mood dysregulation (irritability and hyperarousal), which overlaps with ADHD, is used (Duffy 2007). A narrower pediatric clinical phenotype may more faithfully represent adult nosological categories (Brotman et al 2007). Caution should therefore be exerted in extrapolating neuroimaging findings from these pediatric populations to adults and vice versa, particularly when broad diagnostic criteria are used. This caveat is illustrated by (Biederman et al 2007) who reported that pediatric BD patients with co-morbid ADHD show additive patterns of MRI-evinced volume loss characteristic of both disorders when studied independently.
(3). State versus trait outcome differences do not always receive the attention that they merit: identification of the neuroanatomical or functional changes associated with the acute effects of depression and mania may serve as a useful means of evaluating the efficacy of treatments. Conversely, mood-independent abnormalities may be endophenotypes that can be exploited for genetic studies.
(4). While functional and morphometric analyses of affective illness often receive separate treatment in the literature, these two factors cannot be divorced from each other on either theoretical or methodological grounds. As alluded to above, GM volume loss may occur secondary to glutamate-driven excitotoxicity, a phenomenon which presumably correlates with elevated glucose metabolic activity (Shulman et al 2004). Volume loss may however, lead to an underestimation of metabolic activity because of partial voluming effects: the inclusion of hypometabolic WM or CSF in putatively GM-containing voxels.
An example of this potential confound can be found in our own work. Initially, we found evidence of reduced metabolic activity in conjunction with reduced volume of the subgenual ACC in the depressed phase of both MDD and BD (Drevets et al 1997) (Drevets et al 2002a). Compatible with this hypothesis, depressed BD samples treated chronically with lithium indeed showed elevated sgACC metabolism irrespective of correction for partial volume effects (Bauer et al 2005) (Mah et al 2007), consistent with evidence that lithium treatment is associated increase in the sgACC GM volume (Moore et al In press).
Another example can be found in studies of the amygdala. Despite reports of volume reductions, hypermetabolism of the amygdala is a consistent feature of the literature, which suggests that the increase in activity, if genuine, may be substantial – on the order of 70%, even though state-of-the-art PET imaging technology detects this as only a 6% difference between depressives and controls due to spatial resolution (i.e., partial volume) limitations (Drevets et al 1992). The partial voluming effect can be attenuated to some extent by region of interest (ROI) based analyses of a small central region of the ROI as implemented by the Drevets group, presumably facilitated their finding of substantially increased amygdalar activity (Drevets et al 2002b).
Limitations in spatial resolution also have impinged on the efficacy of volumetric MRI measures. Further, most of the older and even current imaging platforms do not have the necessary spatial resolution to accurately detail volumetric changes in dimunitive anatomical structures such as individual amygdalar nuclei. This limitation may lead to the implicit reification of neuroanatomical function; that is, the notion that gross anatomical structures should be uniformally affected by mood disorders simply because they are imagable units.
If one takes this argument to its logical extreme it becomes questionable as to what constitutes a replication. In other words, how similar do the functional or structural changes reported across studies have to be in order to establish that the identical functional units are affected?
(5). A comparison of the analytical methods used in imaging analysis is beyond the scope of this review. Nevertheless, a brief discussion of the 2 primary analytical approaches to imaging data analysis bears mentioning. Voxel-based analysis (as implemented in imaging analysis software such as the Statistical Parametric Mapping [SPM] series) is based on a mass-univariate statistical approach that compares the differences between two or more subject groups at each individual voxel in the brain (Ashburner and Friston 2000). As such, Gaussian random field theory is used to correct for the multiple dependent comparisons that would, if uncorrected, lead to false positive results (Ashburner and Friston 2000).
Voxel-wise analyses, however, are also highly sensitive to Type II errors because spatial normalization algorithms cannot precisely overlay small structures like the amygdala and sgACC across subjects, exaggerating the statistical variance. Possibly therefore, voxelwise analyses should not be the method of choice when diminutive, subcortical structures are the subject of study (Bookstein 2001) (Nugent et al 2006). Despite this limitation, voxel-wise analysis affords the advantage of allowing agnosticism with respect to the pathophysiology of the disorder under investigation.
On the other hand, the region-of-interest (ROI) approach relies on extant empirical or theoretical data to identify candidate regions that may show inter-group differences. Thus, assuming the veridicality of the disease model, Bayesian inference suggests that the probability of a significant ROI result being a true positive, is higher than in cases where convergent a priori data are absent. Nonetheless, the ROI approach is time and resource intensive especially when subject groups are large. The strength of the method is critically dependent on the precision of ROI segmentation, which depends heavily on the reproducibility of the anatomical landmarks chosen to delineate the target structure.
To our knowledge, very few comparisons of ROI and voxel-wise approaches have been made in psychiatrically ill samples. In one such study, (Kubicki et al 2002) that the voxel-wise approach for comparing cerebral volumes, termed “voxel-based morphometry (VBM)”, produced an analogous finding (volume of STG) to their previously published ROI study, but also showed changes in other regions, one of which had not been previously implicated in schizophrenia. Later schizophrenia studies produced reasonably convergent results across ROI and VBM analyses in cortical regions (Job et al 2002) (Giuliani et al 2005). (Douaud et al 2006) extended the correspondence in VBM-ROI results to the striatum in a case-control study of Huntington’s disease; although it is unclear whether a VBM analysis would be able to detect the more subtle subcortical changes characteristic of affective illness. In at least one case, however, PTSD-associated hippocampal changes have been obtained in the same sample using both ROI and VBM approaches (Emdad et al 2006).
We did not identify a clear difference in the results of VBM and ROI analyses in the BD and MDD literature (Tables 1 through through25),25), possibly because there are very few published VBM studies of subcortical regions.
(6). Genetic association studies are plagued by false positive results: with approximately 20, 000 genes, and multiple variants within each gene of interest, the a priori probability of a true association is low. Sample sizes in analyses which combine genetic and imaging data are by nature small, although this is offset to some extent by the relative precision of the phenotypic data that are collected. The problem can be partially ameliorated by targeted hypotheses.
Top-down and bottom-up disruptions to cortico-striatal-limbic circuits are the most straightforward method of describing the pathophysiological and symptomatological changes associated with affective illness. We have not made any attempt to distinguish between MDD and BD, here. A graphical representation is provided in Figure 1.
As discussed above, projections from the orbital and medial PFC to the amygdala and its associated limbic and brainstem nuclei form a “visceromotor network” that modulates endocrine, autonomic, and behavioral aspects of emotion (Ongur et al 2003).
In the top-down model, impaired PFC function, or cortical-subcortical “disconnection” disinhibits downstream limbic projections, altering emotional behavior. For example, disinhibition of the amygdala projections to the bed nucleus of the stria-terminalus (BNST), hypothalamus and periaqueductal gray matter (PAG) (Behbehani 1995) (Sah et al 2003) may increase cortisol releasing hormone (CRH) release and anxiety symptoms. Disinhibition of projections from the amygdala to the nucleus basalis, locus ceruleus and ventral tegmental area (VTA) (Davis and Whalen 2001) (Sah et al 2003) could account for the alterations in cholinergic (ACh), noradrenergic (NE) and dopaminergic (DA) transmission which may affect mood and attention. Finally, disinhibition of amygdala projections to the ventral striatum (Cardinal et al 2002) would attenuate reward-seeking and goal-directed behavior, potentially contributing to the anhedonia and amotivation characteristic of depression. According to the bottom-up model, a functional hypersensitivity of limbic nuclei such as the amygdala, raphe and parahippocampus would predispose to dysregulation of PFC-mediated regulatory mechanisms.
In reality, the distinction between bottom-up and top-down models is artifical. If MDD and BD are polygenic disorders underpinned by many genes of small effect size, affected individuals are likely to possess many different risk variants affecting multiple neuroanatomical pathways and a single point of origin is unlikely.
The etiology of the suspected hypersensitivity or lesions4 is unclear but most likely involves complex gene-environment interactions. For example, the s allele of the 5-HTT promoter variant may predispose to reduced functional connectivity between the amygdala and perigenual PFC (Pezawas et al 2005), which may be maladaptive under stress. Similarly, the NRG1 gene may impact the myelination process and therefore predispose to WM lesions and a disruption to cortical-subcortical connections in the presence of cardiovascular risk factors. Two nonsynonymous SNPs of another gene, proline dehydrogenase (oxidase) 1 (PRODH)  have also been associated with frontal WM volume reductions in schizophrenia (Zinkstok et al 2007).
Other potential examples are the −1019C/G single nucleotide polymorphism (rs6295) of the HTR1A gene promoter which disrupts a glucocorticoid binding site and modulates raphe-PFC serotonergic transmission (Lemonde et al 2003), and the val66met change in the BDNF gene which appears to impact hippocampal volumes in schizophrenic (Szeszko et al 2005) (Ho et al 2006), healthy (Pezawas et al 2004) (Bueller et al 2006), BD (Chepenik et al 2008) and depressed (Frodl et al 2007) individuals.
Many other similar examples are likely to emerge over time and hold out great promise for disinterring the latent pathophysiological basis of affective illness.
1Perhaps because of inadequate differentiation between non-human primate and human neuroanatomy, the literature is characterized by inconsistent usage of Brodmann map areas which make up the sgACC. We recommend the human anatomical schema of Price and colleagues which can be found in Figure 3. Nevertheless, in Tables 21 through through2424 we have cited the authors’ original anatomical descriptions, verbatim.
2Oligodendrocytes have a high density of glutamate receptors and also are particularly vulnerable to glutamate-induced excitotoxicity.
3The regulation of the amygdala by the sgACC is proving to be increasingly complicated. Recent evidence suggests that although area 25 does indeed exert inhibitory effects, projections from area 24 actually have an excitatory effect on the amygdala. Vidal-Gonzalez I, Vidal-Gonzalez B, Rauch SL, Quirk GJ (2006): Microstimulation reveals opposing influences of prelimbic and infralimbic cortex on the expression of conditioned fear. Learn Mem 13:728–733.
4Here we use the term “lesion”, loosely, to signify any illness predisposing change in neurochemical function or neuroanatomical structure.