OUP user menu

The neuropathology of primary mood disorder

Paul J. Harrison
DOI: http://dx.doi.org/10.1093/brain/awf149 1428-1449 First published online: 1 July 2002


The biological mechanisms proposed to underlie primary mood disorder do not usually include a neuropathological component. However, a significant MRI literature attests to structural abnormalities in regions and has encouraged neuropathological investigations from which candidate histological correlates have begun to emerge. In particular, there are several reports of cytoarchitectural alterations in anterior cingulate and prefrontal cortices, characterized by a decrease in the number or density of glia. Reductions in the size and density of some neuronal populations have also been described, accompanied by alterations in indices of synaptic terminals and dendrites. This form of pathology putatively reflects aberrant neurodevelopment or impaired cellular plasticity. A separate pathological process is suggested by the excess of subcortical focal lesions seen on MRI, especially in elderly patients; these probably reflect white matter damage of vascular origin. Both types of pathology have been observed, to a greater or lesser extent, in unipolar as well as bipolar mood disorders. None of the findings appear attributable to treatment with antidepressants, mood stabilizers or electroconvulsive therapy (ECT). However, all findings remain preliminary due to a lack of unequivocal replication and the failure to control fully for other potential confounders and co‐morbid conditions. There are also basic questions to be answered concerning the clinical correlates, magnitude, progression and heterogeneity of the pathology. Nevertheless, it must now be considered likely that changes in brain structure, both macroscopic and microscopic, are a feature of primary mood disorder, a fact to be taken into account when interpreting functional imaging, neuropsychological and neurochemical data. The neuropathology is postulated to contribute to the pathophysiology and dysfunction of the neural circuits which regulate mood and its associated cognitions, behaviours and somatic symptoms.

  • Keywords: affective disorder; bipolar disorder; cytoarchitecture; connectivity; depression; morphometry
  • Abbreviations: ECT = electroconvulsive therapy; GFAP = glial fibrillary acidic protein; WMH = white matter hyperintensities


Mood disorders exemplify the enduring dichotomy in psychiatric classifications between organic (secondary) and functional (primary). Hence, an organic mood disorder has a ‘presumed direct causation by a cerebral or other physical disorder’(World Health Organization, 1992). The Diagnostic and Statistical Manual of Mental Disorders (DSM‐IV), used more commonly for research, adopts a similar approach, under the rubric of ‘Mood disorder due to a general medical condition’ (American Psychiatric Association, 1994). The organic category includes mood disorders associated with overt neuropathologies, such as degenerative, neoplastic, infective and inflammatory processes. There are separate categories for substance‐induced mood disorders. Although both classification systems emphasize that functional disorders are not without an organic component, they are defined by a lack of demonstrable features of this kind. From a neuropathological perspective, there has been little reason to question this basic assumption: if schizophrenia has been a neuropathological graveyard, primary mood disorders have remained an uncharted wilderness (Jeste et al., 1988; Guze and Gitlin, 1994). However, in the past 5 years or so, the situation has begun to change, with the first noteworthy, albeit preliminary, histological correlates being described.

Primary mood disorders are classified according to the nature and severity of symptoms during each episode, and by the course of the illness (Gelder et al., 2001). A basic distinction is drawn between unipolar (depressive) disorder and bipolar disorder (manic depression). Within the unipolar mood disorder category, major depression is the main subtype, and the only one yet to be studied neuropathologically. It is characterized by the occurrence of one or more episodes of low mood and/or anhedonia, together with a range of cognitive and somatic symptoms, such as fatigue, loss of appetite, sleep disturbance, impaired concentration and negative thoughts of guilt, worthlessness and death. There may or may not be full recovery between episodes. In bipolar disorder, depressed and euthymic periods are interspersed with manic episodes, when an abnormally elevated mood is accompanied by associated behaviours and cognitions (e.g. grandiosity, irritability, disinhibition). Mood‐congruent delusions and hallucinations can occur during severe mood swings in either direction.

Review coverage

This review includes all known post‐mortem neuropathological data papers on primary mood disorder published in English in peer‐reviewed journals since 1980 which have a sample size of at least four patients and an appropriate comparison group. It also covers relevant methodological and interpretational issues. ‘Neuropathological’ is defined here to encompass studies using synaptic and dendritic markers as well as morphometric and immunocytochemical measurements of neurones and glia, but not studies of receptors and other neurochemical aspects of pathology. Investigations of suicide victims without a clear diagnosis of mood disorder are excluded. Literature was located in three ways: from weekly searching of Reference Update deluxe edition disks from 1990 onwards, from databases (Medline, PsychLit) and from perusal of reference lists in the author’s reprint collection. The final searches were performed in December 2001.

Background to neuropathological studies of mood disorder

Two factors explain the current neuropathological interest in mood disorder and provide the context in which the studies can best be understood. First, they reflect an increasing neurobiological emphasis towards psychiatric disorders in general, and, more specifically, the recent progress in elucidating the neuropathology of schizophrenia, which has encouraged an equivalent approach. Secondly, MRI evidence for structural brain abnormalities in mood disorder has steadily accumulated over the past few years, giving impetus to the search for their histological and cellular correlates. The MRI studies have also influenced the choice of brain areas investigated neuropathologically. Positive volumetric MRI findings in mood disorder have been mainly in the frontal lobe, medial temporal lobe (hippocampal formation and amygdala) and striatum. (For meta‐analyses, see Elkis et al., 1995; Videbech, 1997; Hoge et al., 1999; for narrative reviews, see Soares and Mann, 1997; Pearlson, 1999; Drevets, 2000). These regions overlap with those implicated by neuropsychological, functional imaging and neurochemical studies, as well as with the location of lesions associated with secondary mood disorders. Together, these considerations have led to a basic consensus that the pathophysiology of mood disorder reflects dysfunction of cortico‐limbic and cortico‐striatal networks. It is within this conceptualization of mood disorder that the neuropathological findings, potentially revealing a structural component to such a process, are first reviewed and then interpreted.

The frontal lobes

Anterior cingulate cortex

A major role for the anterior cingulate cortex in mood disorders is apparent from a wealth of neuropsychological, neuroanatomical and functional imaging data (Ebert and Ebmeier, 1996), consistent with the increasingly sophisticated models which place it at the interface of emotion, cognition, drive and motor control (Devinsky et al., 1995; Carter et al., 1999; Paus, 2001). In keeping with its functional complexity, the anterior cingulate cortex is heterogeneous in terms of its cytoarchitecture (Vogt et al., 1995) and afferent and efferent connections (see Öngür and Price, 2000; Paus, 2001). Such details are relevant when considering the precise localization and clinical correlates of the neuropathological changes to be described in this region in mood disorder (Fig. 1).

Fig. 1 Frontal lobe areas and the neuropathology of mood disorder. (A) Lateral, (B) medial and (C) coronal (at the level of the dashed line) views of the cerebral cortex showing Brodmann areas of the frontal cortex currently implicated in the neuropathology of mood disorder. Subdivisions of the anterior cingulate cortex (areas 24, 25, 32 and 33) are also shown (see Vogt et al., 1995). Delineation of Brodmann areas is from Perry (1993). All markings are approximate, because boundaries are defined by cytoarchitecture not by surface anatomy, and there is considerable individual and hemispheric variability in size of some areas (e.g. Rajkowska and Goldman‐Rakic, 1995; Paus et al., 1996; Ide et al., 1999).

Neuropathological studies of mood disorder were stimulated by an MRI report that a specific part of the anterior cingulate gyrus lying ventral to the genu of the corpus callosum, the subgenual region sg24 (also called subgenual prefrontal cortex; Fig. 1; Carmichael and Price, 1994), was 40% smaller in a group of 38 subjects with familial mood disorder compared with 21 controls (Drevets et al., 1997). The volume decrease was left lateralized and was seen in both unipolar and bipolar disorders. Hirayasu et al. (1999) also found left sg24 volume reductions (of 25%) in 14 subjects with first‐episode affective psychosis (severe mood disorder) and a positive family history; lesser and non‐significant trends were seen in the right hemisphere and in patients without a family history.

Drevets and colleagues went on to investigate the cellular correlates of the sg24 volume reduction. The resulting paper is noteworthy, being the first large, well‐conducted neuropathological study of mood disorder (Öngür et al., 1998). In a small initial sample, there was a trend towards a decreased volume of left sg24, and for a reduced density and number of glia therein (Table 1). These data led to a larger study using brain tissue from the Stanley Foundation Neuropathology Consortium, which comprises 60 individuals in four groups: major depression, bipolar disorder, schizophrenia and controls, and is the first significant collection of tissue from mood disorder subjects available for contemporary research. The diagnostic groups are demographically matched, relatively young (mean age ∼45 years) and have good clinical documentation including family, medication and substance misuse histories (Torrey et al., 2000). In this series, a decrease of glial density and glial number in sg24 in major depression and bipolar disorder was confirmed, being significant only in the subset of cases with a family history. The decrease was observed through the depth of the cortex, and was also said to occur in orbitofrontal cortex but not in somatosensory cortex, though these data were not presented. There were no changes in glial size, or in neuronal density, number or size, and no glial deficits in the schizophrenia group. No analyses of possible hemispheric differences (cf. the imaging studies) were reported. There was no evidence that the results were due to medication or substance misuse.

View this table:
Table 1

Morphometric studies of the anterior cingulate cortex in mood disorder

Study and areaCases/controlsMethods and parametersMain findings in mood disorder+
Öngür et al. (1998) BA24 (subgenual), BA3ba. 4 BD, 4 MD/5 CONa. Nissl stain; volume, neurone and gliadensity and number; optical disectora. Trend for decreased volume. Decreased glial number anddensity in MD and familial BD
b. 14 BD, 9 MD/14 CON*b. Nissl stain; volume, neurone and gliadensity and number, optical disector androtatorb. Trend for decreased volume. Decreased glial number anddensity in MD (mainly due to fMD). Decreased glial number anddensity in fBD. Glial size unchanged. No differences in BA3b
Benes et al. (2000) BA24(pregenual)5 BD/12 CON*Nissl stain; neuronal sizeNo differences in pyramidal or non‐pyramidal neurones
Cotter et al. (2001a) BA24(supragenual)15 MD, 15 BD/15 CON*Nissl stain; glial and neurone density byoptical disector; neuronal size bynucleator In MD, decreased glial density (–22%), and decreasedneurone size (–18%) in VI, with same trend in V.Neurone density unchanged. No changes in BD
Benes et al. (2001) BA24(pregenual)10 BD/12 CON*Nissl stain; neuronal density and size;glial density Reduced density of non‐pyramidal neurones in II.Trend for lower density of pyramidal neurones in deeper laminae.Pyramidal neurones larger in II; non‐pyramidal neurones larger in II and III.No change in glia
Bouras et al. (2001) BA24(supra‐ and subgenual), BA18 20 MD, 21 BD/55 CON*Nissl stain; laminar and cortical depth;neuronal density and size In BD: decreased depth of III, V and VI, and reduced neuronal densityin these laminae, in subgenual BA24. Neuronal size: no significant changes.No differences in supragenual 24, or in BA18, or in either area in MD
Cotter et al. (2002b) BA24(supragenual)15 MD, 15 BD/15 CON*CB‐, PV‐ and CR‐immunoreactiveneurone densities and distributionIn BD: decreased density of CB neurones in II, and increasedclustering of PV neurones. No differences in MD.

BA = Brodmann area; BD = bipolar disorder; CB = calbindin; CON = control; CR = calretinin; f = familial; MD = major depression; PV = parvalbumin. *Schizophrenia comparison group studied as well (data not shown). +Roman numerals refer to cortical laminae. Includes the five subjects studied in Benes et al. (2000).

Another morphometric study of sg24 in mood disorder has been reported recently, in a large series of brains collected over a 50‐year period (Bouras et al., 2001). As well as sg24, these authors examined the dorsal part of area 24, and area 18 (visual cortex), all in the left hemisphere. The main analyses were carried out on 21 patients with bipolar disorder, 20 patients with major depression and 55 normal controls. There were two positive findings, both in sg24 in the bipolar disorder group: the depths of laminae III, V and VI were reduced by ∼20%, and neuronal density in these laminae was decreased by ∼30%. No glial data were presented.

Like Öngür et al. (1998), Cotter et al. (2001a) used the Stanley Foundation tissue to measure anterior cingulate cortex neurones and glia, but the two studies differ in several ways. Cotter and colleagues investigated the supragenual part (area 24b; Fig. 1) and used a more sophisticated data and statistical analysis. The price of the latter is a complex data set and difficulty drawing direct comparisons with the earlier study. The main positive findings were in major depression, with decreased glial density in lamina VI, and reduced neuronal size in laminae Vb and VI (Table 1). Neuronal density was unaltered. There were similar but non‐significant trends in bipolar disorder. No investigation of the effect of family history was carried out, but a subsequent analysis of this factor was negative (D. Cotter, personal communication, July 2001). Overall, the study partially replicated the results of Öngür et al. (1998), providing support for a glial pathology, as well as for neuronal size reductions seen in other prefrontal regions to be described in the next section.

The remaining morphometric studies of anterior cingulate cortex are by Benes’ group, who have examined the rostral (pregenual) part of area 24 in subjects with bipolar disorder (Table 1). They found a decreased density of interneurones in lamina II, and a trend towards a lower density of pyramidal neurones (Benes et al., 2001), but no differences in glial density (Benes et al., 2001) or neuronal size (Benes et al., 2000).

As well as investigating cell bodies, a comprehensive analysis of the cytoarchitecture requires, amongst other things, evaluation of synapses and dendrites. Direct visualization of these structures is problematic in post‐mortem tissue, and measurement of gene products localized to these cellular compartments has become a widely used alternative (Masliah et al., 1990; Honer et al., 2000). The approach is now being applied to mood disorders to inform about possible alterations in neural connectivity (Table 2). In area 24b of the Stanley Foundation brain series, we observed decrements in bipolar disorder for three synaptic proteins: synaptophysin, complexin II and growth‐associated protein‐43 (GAP‐43), but no change in a fourth protein, complexin I (Eastwood and Harrison, 2001). Complexin II was also reduced in major depression. The data suggest a reduced density and perhaps plasticity of some synaptic populations in this area in mood disorder, although this interpretation must be made with caution (Harrison and Eastwood, 2001). The only data pertaining to dendrites in area 24 are those reported by Bouras et al. (2001), who, in a small subset of cases (three in each diagnostic category), found reduced amounts of the dendritic microtubule‐associated proteins MAP1b and MAP2 in bipolar disorder but not in major depression. In total, these initial findings are consistent with the presence of a synaptic pathology in the anterior cingulate cortex to accompany the glial and neuronal alterations in mood disorder, especially bipolar disorder.

View this table:
Table 2

Axonal, synaptic and dendritic studies of mood disorder

Cases/controlsMethods and parametersMain findings in mood disorder
Cortical areas
Honer et al. (1999)Anterolateral prefrontal 11 MD/10 CON*ELISA; synaptophysin,MBP, GAP‐43MBP decreased. Synaptophysin and GAP‐43 unchanged
Eastwood and Harrison (2001)BA24 (supragenual)15 BD, 15 MD/15 CON*Immunoblotting; synaptophysin, GAP‐43, complexin I and IISynaptophysin, GAP‐43 and complexin II decreased in BD.Complexin II also decreased in MD
Hippocampal formation
Dowlatashi et al. (2000)Mossy fibre pathway13 BD, 14 MD/15 CON*Timm’s stainIncreased supragranular layer staining in BD
Eastwood and Harrison (2000) 15 BD, 15 MD/15 CON*ISHH; complexin I and II mRNAsBoth mRNAs decreased in CA4, subiculum andparahippocampal gyrus in BD. No changes in MD
Rosoklija et al. (2000)Subiculum, fusiform gyrus6 (4 BD, 1 MD, 1 ODS)/8 CON*Rapid Golgi stain; dendritic arborization and spine density on pyramidal neuronesDecreased subicular apical dendrite arborization andspine density. No difference in spine density of basilardendrites. Fusiform gyrus unaffected
Fatemi et al. (2001)13 BD, 12 MD/15 CON*SNAP‐25‐IRDecreased in BD in stratum oriens. Increased in MDin stratum moleculare and presubiculum
Müller et al. (2001)Hippocampus 2 BD, 13 MD/16 CON+Synaptophysin‐IR and GAP‐43‐IR Synaptophysin unchanged. GAP‐43‐IR reduced in CA2
Webster et al. (2001)4 BD, 6MD/10 CON*ISHH: synaptophysin and GAP‐43 mRNAsSynaptophysin mRNA decreased. GAP‐43 mRNA unchanged

BD = bipolar disorder; CON = controls; ELISA = enzyme‐linked immunosorbent assay; GAP‐43 = growth‐associated protein‐43; IR = immunoreactivity; ISHH = in situ hybridization histochemistry; MBP = myelin basic protein; MD = major depression; ODS = organic depressive syndrome; SNAP‐25 = synaptosome‐associated protein of 25 kDa. *Schizophrenia comparison group studied as well (data not shown). +Also included corticosteroid‐treated comparison group.

Other areas of prefrontal cortex

Orbital and dorsolateral regions of the prefrontal cortex have also been implicated in mood disorder on functional and structural grounds (Goodwin, 1997; Elliott, 1998; Merriam et al., 1999; Lai et al., 2000; Öngür and Price, 2000). The subsequent neuropathological investigations have revealed findings broadly similar to those described in the anterior cingulate cortex.

The key studies have been performed by Rajkowska and colleagues (Table 3). The first report was of major depression, with measurements in dorsolateral (area 9), rostral orbital (area 10/47) and caudal orbital (area 47) prefrontal cortices (Fig. 1). In all three areas, there was decreased glial density and reduced size of neurones in one or more laminae (Rajkowska et al., 1999). Similar alterations have been described since in bipolar disorder in area 9, with lamina‐specific reductions in glial and pyramidal neurone density, as well as alterations in glial shape and size (Rajkowska et al., 2001). In a follow‐on study of the major depression subjects, glial fibrillary acidic protein (GFAP) was used as a marker of astrocytes to see if a loss of this glial subtype explained the earlier observations. GFAP staining and GFAP‐positive cell counts were unaltered in the whole sample, but there was a decrease in the younger (30–45 years) major depression cases compared with a subgroup of age‐matched controls (Miguel‐Hidalgo et al., 2000). Corroborative evidence for astrocytic involvement comes from a proteomics study of the Stanley Foundation series which found significant reductions of GFAP isoforms in the prefrontal cortex in bipolar disorder and major depression (Johnston‐Wilson et al., 2000).

View this table:
Table 3

Morphometric studies of other prefrontal cortical areas in mood disorder

Study and areaCases/controlsMethods and parametersMain findings in mood disorder+
Rajkowska et al. (1999)Left BA9, 10/47, 4712 MD/12 CONNissl stain; cortical thickness, neurone densityand size; glial density and glial nuclear sizeBA9: decreased neurone size in III and IV (–5% and –7%).Decreased density of large neurones in II, III and IV (by 20–60%),with more small neurones (+40%). 20–30% lower glial density in IIIand IV. More glia with large nuclei in III (+60–120%).
BA47: decreased neurone size in II; decreased density of large neuronesin IIIa and Va. Overall glial density reduced (–15%). Decreased densityof medium and large sized glia in V and VI.
BA10–47: reduced cortical thickness (–12%).Decreased neurone size in II and III. Decreased neurone density in II, III and IV (20–60%).Increase in density of small neurones in II (+30–70%).Trend for overall decrease in glial density, and fewer glia withmedium or large nuclei in IIIa and IV.
Miguel‐Hidalgo et al. (2000)Left BA914 MD/15 CONGFAP immunoreactivity; areal fraction anddensity of GFAP +ve cells. No overall differences. Reduced areal fraction in III–V in subgroup ofyoung cases. Trend decreases in density of medium and large glia
Lewis et al. (2001)Left BA920 MD/20 CON*PV‐immunoreactive varicosities(a marker of thalamic axon terminals)No differences.
Rajkowska et al. (2001)Left BA9 10 BD/11 CONNissl stain; cortical and laminar thickness;neuronal density and size; glial density andnuclear sizeCortical thickness unchanged. Pyramidal and total neurone densityreduced in II, III and V by 9–30%; no change in non‐pyramidalneurone density. Neurone size unchanged. Glial density decreased inIIIc and Vb. Glial size increased in I and IIIc.
Cotter et al. (2002a) BA915 BD, 15 MD/15CON*Nissl stain; neuronal and glial densityand size, and K‐function analysis ofspatial clustering of cells In BD: neuronal size decreased in V (–14%) and VI (–18%).In MD: glial density decreased in V (–30%) and neuronal sizedecreased in VI (–20%). No alteration in glial clustering aroundneurones, or in neuronal density.

BA = Brodmann area; BD = bipolar disorder; CON = control; GFAP = glial fibrillary acidic protein; MD = major depression; PV = parvalbumin. *Schizophrenia comparison group studied as well (data not shown). +Roman numerals refer to cortical laminae.

Independent support for morphometric alterations in area 9 in mood disorder comes from a study by Cotter et al. (2002a). Neuronal size was decreased in both bipolar disorder and major depression, and glial density was reduced in the latter. Neuronal density showed only trend reductions in either group. The spatial arrangement of cells was also measured, with particular reference to whether the normal clustering of glia (especially oligodendrocytes) around neurones differed in mood disorder. The negative result led the authors to infer that oligodendrocytes are unlikely to be responsible for the glial deficits, although the conclusion contrasts with a preliminary report of a loss of oligodendrocytes in the same tissue (Orlovskaya et al., 2000).

There has been only one synaptic protein study, and none of dendrites, in the prefrontal cortex. In patients with major depression dying by suicide, Honer et al. (1999) found no change in synaptophysin or GAP‐43. They did report a reduction of myelin basic protein which may indicate altered myelination and of possible relevance to the white matter findings to be discussed. In major depression, unlike in schizophrenia, there is no evidence for an involvement of thalamocortical axon terminals in the dorsolateral prefrontal cortex (Lewis et al., 2001).

In summary, in several areas of the frontal lobe, a number of groups have reported decreases in the density or number of glia, and the density and size of some neurones, in mood disorder. In this respect, there is a consistency and robustness to the observations, and the rudiments of a neuropathology of mood disorder. However, important uncertainties and discrepancies remain. For example, Öngür et al. (1998) found glial pathology but no neuronal changes, Benes et al. (2001) found the opposite, and Rajkowksa et al. (1999, 2001) and Cotter et al. (2001a, 2002a) found both; moreover, the laminar distribution of the alterations varies between studies (Tables 1 and 3). There are also inconsistencies as to whether it is bipolar disorder or major depression which shows the greater differences. These variable results are likely to be due to a combination of the anatomical, demographic and methodological issues to be discussed below.

Hippocampal formation

The hippocampal formation (dentate gyrus, Ammon’s horn, subiculum and parahippocampal cortex) has been implicated in mood disorder for two main reasons. First, several, though by no means all, MRI studies have found smaller hippocampal volumes in major depression (Sheline et al., 1996; Shah et al., 1998; Bremner et al., 2000) and bipolar disorder (Strakowski et al., 1999; Altshuler et al., 2000); there is also a report of decreased weight of the parahippocampal gyrus in elderly depressed patients (Bowen et al., 1989). Secondly, there is a well‐studied model linking depression, via growth factors and second messengers, to the atrophic effects of glucocorticoids and stress on hippocampal pyramidal neurones and their dendrites (Duman et al., 1997; Brown et al., 1999; Sapolsky, 2000); the model now also encompasses impaired neurogenesis (Jacobs et al., 2000). The model is based, in part, on the role of the hippocampus in regulating the hypothalamo‐pituitary–adrenal axis (HPA), thereby relating it to the hypercortisolaemia and other signs of HPA axis dysfunction which occur in mood disorders (Checkley, 1996: Plotsky et al., 1998) and which may be a risk factor for them (Starkman et al., 1999; Goodyer et al., 2000; Harris et al., 2000).

Despite these considerations, there have been very few neuropathological studies of the hippocampal formation in primary mood disorder (Table 4). Beckmann and Jakob (1991) described dysplasia and heterotopias in the entorhinal cortex (anterior parahippocampal cortex) in four cases of bipolar disorder, as they had reported previously in schizophrenia. A similar, though less dramatic finding in a mixed group of mood disorder subjects was reported by Bernstein et al. (1998a). If confirmed, these alterations would have major aetiological implications, being strongly suggestive of an early developmental anomaly. However, they have not been replicated consistently in schizophrenia (see Harrison, 1999a), and their occurrence in mood disorder must be viewed as highly speculative at this stage.

View this table:
Table 4

Morphometric studies of the hippocampal formation in mood disorder

Study Cases/controlsMethods and parametersMain findings in mood disorder
Beckmann and Jakob (1991)4 BD/4 CONNissl stain; laminae II and III neurone clusters inentorhinal cortexHeterotopic clusters
Benes et al. (1998) 4 BD/11 CON*Nissl stain; number and density of neurones inpyramidal layer of CA1–4Decreased number, density and size of non‐pyramidalneurones in CA2. No differences in pyramidal neurones
Bernstein et al. (1998a)7 (3 MD, 2 BD, 2 SA)/45 CON*Nissl stain; heterotopically displaced laminae II and III neurone clusters in entorhinal cortexMore heterotopias in the right entorhinal cortex
Lucassen et al. (2001)2 BD, 13 MD/16 CON+Nissl stain; markers of apoptosis (ISEL)and cellular stress (hsp70 and NF‐κB);qualitative analysisApoptotic cells seen in one or more subfields in11/15 MD cases versus 1/16 controls.No obvious morphological or neuronal density differencebetween groups. No displaced or irregularly oriented neurones.
Damadzic et al. (2001)14 MD, 13 BD/15 CON*GFAP‐IR; density of GFAP‐positive astrocytes,in entorhinal cortexNo differences between groups
Müller et al. (2001)2 BD, 13 MD/16 CON+Nissl (qualitative analysis); GFAP‐IR for astrocytesand fibresNo neuronal morphometric differences.Decreased GFAP‐IR in CA1 and CA2

BD =  bipolar disorder; CON = controls; GFAP = glial fibrillary acidic protein; hsp70 = heat shock protein‐70; IR = immunoreactivity; ISEL = in situ end labelling; MD = major depression; NF‐κB = nuclear transcription factor‐κB; SA = schizoaffective. *Schizophrenia comparison group studied as well (data not shown). +Also included corticosteroid‐treated comparison group.

Lucassen et al. (2001) reported the first neuropathological examination of the hippocampal stress–atrophy hypothesis. The glucocorticoid‐related neurotoxicity predicted by the model would be expected to be apoptotic, and associated with evidence of cellular stress. The authors therefore used a range of immunocytochemical markers of these processes, comparing mood disorder (mostly major depression) subjects with normal controls and steroid‐treated medical patients. In the mood disorder group, only minor increases in apoptotic cells were seen, and cell stress markers were negative. Moreover, the positive findings were not in the hippocampal subfields at risk for glucocorticoid damage (Brown et al., 1999; Sapolsky, 2000). In a further study of the same brains, a detailed, though qualitative, analysis found no evidence of differences in neuronal density, orientation or clustering, and only minor, localized alterations of GFAP immunoreactivity (Müller et al., 2001; Table 5). The latter findings also applied to the subgroup of six cases which previously had been shown to have neuropathological signs of HPA axis overactivity (Raadsheer et al., 1994, 1995; Purba et al., 1996; Table 5). The essentially negative data in the studies of Lucassen et al. (2001) and Müller et al. (2001) illustrate that the presence and nature of neuropathology should not be assumed on the basis of a plausible model but must be demonstrated empirically.

View this table:
Table 5

Morphometric studies of subcortical areas in mood disorder

Cases/controlsMethods and parametersMain findings in mood disorder
 Raadsheer et al. (1994) Paraventricular nucleus3 BD, 3 MD, 2 DNOS/15CONCRH‐IR neuronesIncreased 4‐fold in MD and BD. Cell size unaltered
 Raadsheer et al. (1995) Paraventricular nucleus7 MD, 2 DNOS/10 CON+CRH mRNAIncreased in MD
 Purba et al. (1996) Paraventricular nucleus3 BD, 3 MD, 2 DNOS/8 CONAVP‐IR and OXT‐IR neuronesAVP‐IR neurones increased (+56%);OXT‐IR neurones increased (+23%)
 Bernstein et al. (1998b) Paraventricular nucleus2 BD, 4MD, 2 SA/13 CON* NOS‐IR neuronesDecreased number and density (–40%)
 Zhou et al. (2001) Suprachiasmatic nucleus8 BD, 3MD/11 CONAVP‐IR neuronesIncreased number (+78%)
 Klimek et al. (1997)9 MD/9 CONNeuromelanin‐containing cell numberNo differences
Locus coeruleus
 Baumann et al. (1999a)6 MD, 6 BD/12 CONNeuromelanin‐containing cell numberNo differences from controls,though more neurones in BD than MD
 Baumann et al. (1999c)6 MD, 6 BD/12 CONTH‐IR neuronesFewer TH‐IR neurones in depressed subjectsnot dying by suicide
 Underwood et al. (1999)5 MD/6 CONVolume of dorsal raphe;PH8‐positive neuronesNo difference in volume. Neurone density andnumber increased (∼33%)
Raphe nuclei
 Baumann and Bogerts (2001)6 MD, 6 BD/12 CONNissl; neuronal numberReduced number of ovoid and round neurones(reduced by ∼25%) in mood disorder groups
 Lohr and Jeste (1986) 5 BD, 7 MD/37 CON* Purkinje cell and dentate multipolarcell densitiesNo differences
 Helmkamp et al. (1999) 12 BD, 15 MD/15 CON* Vermal Purkinje cell density and position No differences
Other regions
 Nasrallah et al. (1983) Corpus callosum7 BD/11 CON*Fibre and glial densityNo differences
 Baumann et al. (1999b) Basal ganglia8 MD/8 CONPlanimetrySmaller external pallidum (–20%),left nucleus accumbens (–32%), right putamen (–15%)

BD = bipolar disorder; CON = controls; DNOS = depression not otherwise specified; IR = immunoreactive; MD = major depression; NOS = nitric oxide synthase; PH8 = phenylalanine hydroxylase antibody (marks serotonergic neurones); SA = schizoaffective; TH = tyrosine hydroxylase. *Schizophrenia group included as well (data not shown). Subgroup analysis of MD cases (t test, P = 0.027), calculated from data in their Table 1. Statistics not given in paper, but result significant by t‐tests from data presented in their Figure 5. +Alzheimer’s disease group also studied (data not shown).

Preliminary studies suggest, as in the prefrontal cortex, the presence of synaptic and dendritic pathology in the hippocampal formation, especially in bipolar disorder (Table 2). Rosoklija et al. (2000) found decreased arborization of apical dendrites, and a reduced density of dendritic spines, on subicular pyramidal neurones in six mood disorder subjects. Dendritic changes are suggestive of decreased afferent synaptic innervation or activity, a possibility supported by a reduced expression of synaptic protein genes in bipolar disorder, especially in the subiculum (Eastwood and Harrison, 2000; Fatemi et al., 2001; Webster et al., 2001). There is no clear evidence for hippocampal presynaptic pathology in major depression (Eastwood and Harrison, 2000; Fatemi et al., 2001; Müller et al., 2001).

Brainstem nuclei and other subcortical regions

Alterations in adrenergic (Ressler and Nemeroff, 1999) and serotonergic (Maes and Meltzer, 1995) systems remain the predominant neurotransmitter theories of depression. Morphometric studies have now begun into their cell bodies of origin, in the locus coeruleus and raphe nuclei, respectively. As shown in Table 5, there is no evidence for loss or atrophy of neurones in the locus coeruleus (Klimek et al., 1997; Baumann et al., 1999a); indeed, the latter study of bipolar disorder pointed in the opposite direction. The raphe data are contradictory, with a report of increased numbers of serotonergic neurones (Underwood et al., 1999), but also a loss of total neurones with no change in serotonergic neurones (Baumann and Bogerts, 2001). Whatever cellular pathology of monoaminergic nuclei may be identified in the future, it is also possible that there are pathological alterations of the fibre projections and their pattern of terminations (cf. Akil et al., 1999). Moreover, the raphe and locus coeruleus data raise the question of whether changes in neuronal activity may lead to altered numbers of neurones considered positively labelled, and thus counted, in immunocytochemical or in situ hybridization studies which target molecules related to neurotransmission and signal transduction (e.g. Baumann et al., 1999c). Such differences may or may not be accompanied by an alteration in total neurone number. The same issue pertains to the positive findings concerning neuropeptide‐ and nitric oxide synthase‐expressing neurones in the hypothalamus (Table 5), and reflects more generally the ‘state or trait’ question as to the nature of the neuropathology, discussed below.

Subcortical white matter

In addition to the regional brain abnormalities mentioned, other MRI studies have shown a strong association between mood disorder and the number and severity of focal signal hyperintensities on T2‐weighted images. These white matter hyperintensities (WMH) occur particularly in the deep subcortical white matter and to a lesser extent in the basal ganglia and periventricularly. They are seen in excess in bipolar and unipolar mood disorder, with an odds ratio of ∼3 to 7 (Videbech, 1997; Bearden et al., 2001). In major depression, WMH are particularly common in elderly subjects, where they are linked to risk factors for, and the presence of, vascular disease (O’Brien et al., 1996). This finding is consistent with a robust epidemiological association between the two conditions (Alexopoulos et al., 1997), and with the increased atheromatous disease found in late‐life depression (Thomas et al., 2001). WMH confer a poor prognosis in major depression (Hickie et al., 1995; O’Brien et al., 1998) and bipolar disorder (Moore et al., 2001).

The links between mood disorder and vascular disease imply that WMH reflect focal pathology due to ischaemia and infarction, as is the case in other situations (Awad et al., 1986; Chimowitz et al., 1992; Fazekas et al., 1993). A recent diffusion tensor imaging study supports the view that WMH indicate damage to white matter tracts (Taylor et al., 2001). The clinical consequences for mood and other symptoms of mood disorder, notably the cognitive slowing seen in elderly depressed subjects, are thought to arise from the consequent interruption of axonal pathways, especially fronto‐subcortical connections (Greenwald et al., 1998; MacFall et al., 2001). However, a small study found no evidence for greater vascular pathology (or, by inference, white matter damage) in depressed subjects who were cognitively impaired compared with those who were not (O’Brien et al., 2001). Furthermore, to date, post‐mortem investigations of white matter lesions in mood disorder are limited to two case reports (Lloyd et al., 2001), and a systematic study is essential to confirm their neuropathological basis.

Neuropathological effects of mood disorder treatments

Most subjects with major depression studied post‐mortem will have been treated with antidepressants. Some will also have received electroconvulsive therapy (ECT), lithium, antipsychotics or minor tranquillizers. Bipolar disorder patients often receive these treatments too, as well as other mood stabilizers such as sodium valproate and carbamazepine. The potential therefore exists for these treatments to have caused, enhanced, ameliorated or obscured the reported neuropathological alterations.

Electroconvulsive therapy

A comprehensive review concluded that ECT produces no demonstrable neuropathological effects (Devenand et al., 1994). Subsequent data support this view. A proton magnetic resonance spectroscopy study using N‐acetyl aspartate, a composite marker of neuronal structural and functional integrity, found no changes following a course of ECT (Ende et al., 2000). Another study found no ECT‐associated changes in CSF markers of neuronal and glial damage (Zachrisson et al., 2000). There are reports of alterations after electroconvulsive shock in rats, including increased expression of GFAP (Orzi et al., 1990; Steward, 1994) and synaptic (Jorgensen and Bolwig, 1979) and dendritic (Pei et al., 1998) markers, as well as enhanced hippocampal neurogenesis (Madsen et al., 2000; Scott et al., 2000). Although the interpretation and clinical relevance of these data is questionable in light of the negative human ECT data, it would be prudent to bear them in mind if a subject included in a morphometric study had received ECT in the weeks prior to death.


Lithium overdose causes an acute neurotoxicity (Akai et al., 1977; Schneider and Mirra, 1994), but no neuropathological effects of long‐term therapeutic levels of lithium (∼0.4–1.0 mmol/l) have been described. An MRI study reported that 4 weeks lithium treatment increases cortical grey matter volume (Moore et al., 2000a) and N‐acetyl aspartate signal (Moore et al., 2000b), suggesting that lithium is neurotrophic (Manji et al., 2000). Lithium may also enhance neurogenesis and inhibit apoptosis (Chen and Chuang, 1999; Chen et al., 2000), although it is a matter of conjecture whether these various effects are linked, and what their functional significance might be. The one morphometric study carried out so far found no change in cortical neurone number, density and size in rats after 30 weeks lithium administration (Licht et al., 1994), and so the histological correlates of the in vivo findings remain unclear. One group has reported increased GFAP (Rocha and Rodnight, 1994) and astrocytosis (Rocha et al., 1998) in the hippocampus of rats after 4 weeks lithium treatment. As with ECT, the significance here is that lithium may impinge upon the morphometric alterations reported in mood disorder subjects. Weak correlational evidence for this is seen in the study of Rajkowska et al. (2001).

Other drugs used in mood disorder

There are no neuropathological studies of the effects of other mood stabilizers, antidepressants or minor tranquillizers. A small experimental literature suggests that antidepressants may affect neuronal morphology (Smialowska et al., 1988), regenerate monoaminergic axons (Nakamura, 1990; Kitayama et al., 1997), promote neurogenesis (Malberg et al., 2000) and prevent the loss of dendritic spines seen in some animal models of depression (Norrholm and Ouimet, 2001). It is unknown whether any of these processes occur in patients. In contrast, the neuropathological effects of antipsychotic drugs have been relatively well studied in humans and experimental animals, and comprise alterations in synaptic and neuronal morphology, particularly in the caudate–putamen (Harrison, 1999b; Konradi and Heckers, 2001). In addition, increased glial density has been reported in the prefrontal cortex of monkeys treated chronically with antipsychotics (Selemon et al., 1999). With regard to the glial deficits reported in mood disorder, this finding, like the reports of increased GFAP expression after ECT and lithium, emphasizes that treatments have the potential to mask as well as produce positive findings.

Other methodological issues

Neuropathological studies of psychiatric disorders are complicated by many other variables beyond treatment effects (Harrison and Kleinman, 2000; Lewis, 2002). Overcoming, or at least minimizing, these issues requires careful experimental and statistical design, replications and relevant parallel animal studies. To date, none of the reported positive findings in mood disorder fully meets all these criteria.

One problem is alcohol and substance misuse, which is common in mood disorder. For example, it occurs in 30–50% of subjects in epidemiological surveys of bipolar disorder, and was a factor in five of the 12 major depression cases studied by Rajkowska et al. (1999). Recognizing and dealing with such co‐morbidity is difficult but important, because alcohol, and potentially some illicit drugs, may produce neuropathological effects which overlap with those in mood disorder. Notably, alcoholics are reported to have fewer glia, both astrocytes and oligodendrocytes, in the hippocampus (Korbo, 1999), as well as neuronal morphometric differences (Harding et al., 1997; Kril and Halliday, 1999).

The relationship between mood disorder and suicide is also problematic since many individuals in post‐mortem studies of one meet criteria for the other (Bachus et al., 1997; Mann, 1998). Suicide could complicate matters for three reasons: (i) it may have its own neuropathological associations (Baumann et al., 1999c; Bown et al., 2000; Salib and Tadros, 2000; Rubio et al., 2001); (ii) it may indicate a more severe or otherwise atypical subtype of mood disorder (Ahearn et al., 2001); and (iii) and it may produce artefacts secondary to the mode of death. For example, brain pH is higher on average after suicide than after deaths from natural causes, presumably because of shorter average agonal phases (Harrison et al., 1995). Brain pH can affect cell density (Cotter et al., 2001a, 2002a), glial size (Cotter et al., 2002a) and protein and mRNA levels (Harrison et al., 1995). Experimental data also indicate that hypoxia and acidosis may influence glial counts (Bondarenko and Chesler, 2001).

Even after controlling as far as possible for confounders, the nature and magnitude of the changes being sought in mood disorder mean that the methods used must be particularly sensitive and reliable. The relative merits of different quantitative morphometric techniques are controversial, especially concerning the application and interpretation of stereology (Hyman et al., 1998; Everall and Harrison, 2002). Both false‐positive and false‐negative results may occur. For instance, unknown reference volumes or tissue shrinkage, especially in the presence of cell size changes, can lead to cell density data which are biased or hard to interpret (West, 1999); on the other hand, some stereological strategies may not be optimal for identification of localized and subtle alterations (Guillery and Herrup, 1997; Benes and Lange, 2001). A related point concerns the identification of the cells being counted. A Nissl stain, used in most studies, does not permit unequivocal distinction of glia from neurones, let alone one glial type from another. It is thus possible that an altered appearance of small neurones or glia, or variability in how different researchers classify them, could contribute to the discrepancies noted in mood disorder. Whilst this may be unlikely, wider use of more specific neuronal and glial markers will be valuable in the future.

Conceptual issues and interpretation of the findings

What are the clinical correlates of the neuropathology?

Primary mood disorders are syndromes of unknown aetiology and validity. Neuropathology has the potential to advance understanding of mood disorder and help identify meaningful boundaries and subdivisions, as it has done in dementia and epilepsy. Equally, the current uncertainties cause difficulties determining appropriate study design and analysis strategies.

Most studies have been based on the bipolar/unipolar distinction, but the results have not established clearly the neuropathological commonalities and differences between them. The basic form of pathology appears similar, in terms of the glial and neuronal morphometric differences in the prefrontal cortex, but, as Tables 15 show, beyond this point the data are conflicting as to the extent to which clear differences in nature, location or severity of the abnormalities have been identified. The default interpretation would be that no good evidence yet exists for a neuropathological separation of bipolar from unipolar mood disorder; views on this point are affected by whether the null hypothesis is that they are one disorder or two. Given the uncertainties, it is worth considering other categorizations of mood disorder that might be valuable neuropathologically; notably, further investigation of the familial versus sporadic mood disorder concept (Öngür et al., 1998), as well as of the putative ‘white matter mood disorder’ subtype underlying late‐onset major depression and bipolar disorder.

Rather than being linked to any diagnostic category, different elements of the neuropathology might map onto specific symptoms, many of which are shared by mood disorder subsyndromes. For example, the ventral tegmentum, ventral striatum and medial prefrontal cortex may be especially relevant for anhedonia, whereas the amygdala may be more important for anxiety symptoms and depressive ruminations (Drevets, 2001). Furthermore, agitation, a feature of severe depression, is linked to a greater neurofibrillary tangle burden in the left anterior cingulate and orbitofrontal cortices in Alzheimer’s disease (Tekin et al., 2001). Although such clinico‐pathological correlations are inevitably crude, they can provide hypotheses for future investigation. Testing them will require more extensive clinical documentation of cases, preferably from prospective studies, than has occurred to date.

A final possibility is that the clinical correlates of the neuropathology are the neuropsychological characteristics of mood disorders, not their symptoms. Since post‐mortem studies include patients dying at all phases of illness, including some who had been euthymic for months or even years, it is more likely a priori that the neuropathological alterations are primarily trait rather than state related. As Ebert and Ebmeier (1996) pointed out, ‘depressive episodes as reversible mental states are likely to be associated with reversible brain states’. It is therefore notable that residual neuropsychological abnormalities, affecting discrete domains of attention and memory performance, are seen during remission (Kessing, 1998; Van Gorp et al., 1998; Ferrier et al., 1999; Rubinsztein et al., 2000; Austin et al., 2001). Functional imaging data, though complex and incomplete, also indicate that there are persistent abnormalities in relevant brain areas after treatment and recovery (Drevets et al., 1992; Goodwin et al., 1993; Mayberg et al., 2000). A similar coupling between specific neuropsychological deficits and particular elements of pathology may be envisaged as was mentioned for symptoms (Bench et al., 1992; Dolan et al., 1994; Mayberg 1997; Elliott, 1998).

The argument that it is trait phenomena which are the most plausible clinical correlates of the neuropathology applies to other pervasive characteristics associated with mood disorder as well, such as neuroticism, susceptibility to emotional dysregulation and impulsivity. Neuroticism is also a major risk factor for depression (Kendler et al., 1993), raising the possibility that the neuropathology might be related to the vulnerability to mood disorder as well as to the disorder itself; schizophrenia provides a precedent for this suggestion (Harrison, 1999c).

Is the neuropathology diagnostically specific?

A degree of neuropathological continuity across major psychiatric phenotypes is to be expected, since it is also observed in other respects (e.g. genetic predisposition, MRI findings, treatment response) but, by the same token, it is unlikely that the alterations reported in mood disorder will prove to have no diagnostic specificity at all. Although far from conclusive, the available data support this basic assumption.

It is schizophrenia with which mood disorder most usefully can be compared neuropathologically, because several of the key mood disorder studies also included a schizophrenia group (denoted by an asterisk in Tables 15), and there is also a sizeable separate schizophrenia literature. Features reported in schizophrenia as well as in mood disorder include decreased neuronal size in prefrontal cortex, reduced neuronal density in anterior cingulate cortex, reduced synaptic and dendritic markers in prefrontal cortex and hippocampus, and glial deficits (for schizophrenia references see Harrison, 1999a; Honer et al., 2000). On the other hand, the changes are by no means identical in the two disorders, leading some authors to emphasize the differences rather than the similarities (Benes et al., 1998, 2001; Baumann and Bogerts, 1999; Rajkowska et al., 2001). Certainly the glial changes appear more prominent in mood disorder, although it would be premature to argue that this (yet) allows a discrimination from schizophrenia. Rather, there may well be a neuropathological continuum between these conditions, just as there is clinically and probably aetiologically.

MRI findings in obsessive–compulsive disorder (Saxena et al., 1998) and post‐traumatic stress disorder (Bremner, 2001) implicate some of the same brain regions and circuits as those affected in mood disorder, but as yet there are no neuropathological data for these or other related psychiatric syndromes.

What is the distribution of pathology?

Within the cerebral cortex, a uniform pathology of mood disorder is unlikely given the negative results in sensory cortices (Öngür et al., 1998; Bouras et al., 2001; Tables 13). However, the regional distribution and the pattern of alterations outside the cortex (Tables 4 and 5) have not been well investigated. The amygdala and basal ganglia are priority areas for study since both are strongly implicated in mood disorder on several other grounds (Austin et al., 1995; Rogers et al., 1998; Mayberg et al., 2000; Drevets, 2001); glial changes in the amygdala have already been described in an abstract (Bowley et al., 2000).

At the next level of anatomical resolution, it will be important to establish whether pathology is uniform or focal within a given area, since adjacent subregions have differing cytoarchitecture and connections (e.g. Carmichael and Price, 1994; Hof et al., 1995; Freedman et al., 2000) and putative functions (Paus, 2001). This is especially pertinent for the anterior cingulate cortex, given the evidence that there may be a selective subgenual involvement (Öngür et al., 1998; Bouras et al., 2001). Conversely, anatomically discrete areas of the frontal lobe share many connections and functional roles (Duncan and Owen, 2000) and might also share a common vulnerability to the kind of pathology envisaged in mood disorder.

A further aspect of regional localization is hemispheric asymmetry. The possibility that changes in the anterior cingulate cortex, and perhaps elsewhere, are lateralized has been raised by some of the findings in mood disorder (Drevets et al., 1997; Hirayasu et al., 1999; Botteron et al., 2002), as well as by neuropsychological theories (see Drevets, 2000; Liotti and Mayberg, 2001) and the demonstration that the region is structurally asymmetrical (Paus et al., 1996; Ide et al., 1999). Moreover, Cotter et al. (2001a) found a hemispheric difference, and a hemisphere by diagnosis interaction, for some of their glial density data.

Within a given area or subfield, the populations of neurones, glia and synapses affected must be identified. In the anterior cingulate cortex, for example, one can advocate involvement of inhibitory interneurones and their synapses (Benes et al., 2000, 2001), aberrant monoaminergic innervation (Rajkowska, 2000b) or excitatory connections (Eastwood and Harrison, 2001). Characteristic human cytoarchitectural features are also candidate elements worthy of investigation (Schlaug et al., 1995; Nimchinsky et al., 1999; Hof et al., 2001). As yet, few data are available to inform such speculation, and locating the circuits will not be simple. Even in well‐studied neuropsychiatric conditions such as Alzheimer’s disease, it has not been a trivial process to determine the regional (Pearson et al., 1985; Van Hoesen and Solodkin, 1994), cellular (Morrison et al., 1998) or synaptic (Masliah et al., 1990; DeKosky et al., 1996) distribution of pathology. Despite the fundamental differences between Alzheimer’s disease and mood disorder, progress in the latter may be facilitated by awareness of how the question was tackled in the former, including the integration of neuropathological with functional and longitudinal approaches (Mielke et al., 1996; Kanne et al., 1998; Nagy et al., 1999; Rose et al., 2000; Grady et al., 2001; Silverman et al., 2001).

What do the glial deficits mean?

Glia are usually of interest to neuropathologists in the context of gliosis—the proliferation and hypertrophy of glia, especially astrocytes—because it is basic evidence for some form of degenerative or inflammatory process (Norenberg, 1994; Kreutzberg et al., 1997). Against this background, the finding of fewer glia in mood disorder was unexpected. Beyond showing that these disorders are therefore not classically neurodegenerative, the question arises as to the causes and consequences of the glial reduction. Answering this question requires knowledge of which glial type is involved. Given the numerical predominance of astrocytes in the grey matter, it is likely that they are the glial population primarily affected, though microglia (Bayer et al., 1999) and oligodendrocytes (Orlovskaya et al., 2000; Uranova et al., 2001) should not be neglected. Astrocytes are increasingly recognized to have many functions, with roles in neuronal migration, synaptogenesis, neurotransmission and synaptic plasticity (Araque et al., 1999; Barres, 1999; Coyle and Schwarcz, 2000; Bezzi and Volterra, 2001; Oliet et al., 2001; Parri et al., 2001). It is of note that these roles include maintenance of neuronal structure (Ullian et al., 2001). Hence a model can be proposed in which glial deficits are the central pathological event, with the alterations in neuronal, synaptic and dendritic morphology being downstream (Rajkowska, 2000a). Such models can help in developing a conceptual framework, but it is also clear that there are several other plausible explanations and sequences of events to be considered (Cotter et al., 2001b) and all theorizing is largely premature until key neuropathological findings are replicated and better characterized.

Whatever their pathogenic significance, it is worth noting that astrocytes are major contributors to PET and fMRI signals (Magistretti, 2000). Hence glial deficits may explain some of the differences in functional imaging parameters observed in mood disorder (Nikolaus et al., 2000; Videbech, 2000). Astrocytes also express many neurotransmitter receptors and transporters (Porter and McCarthy, 1997; Gallo and Ghiani, 2000) and may thus contribute to the alterations reported in neurochemical and ligand‐based neuroimaging studies of mood disorder, as well as providing a link between monoaminergic and glial aspects of pathology (Griffith and Sutin, 1996; Khan et al., 2001). For example, 5‐HT1A receptor binding is decreased in mood disorder (Drevets et al., 1999; Sargent et al., 2000), and 5‐HT1A receptors are expressed by astrocytes (Azmitia et al., 1996; Cohen et al., 1999) as well as by pyramidal neurones (Burnet et al., 1995). A reduced glial density, as well as smaller and/or fewer neurones, could therefore underlie, at least partly, the imaging findings. Future neurochemical studies of mood disorder, both in vivo and in vitro, will benefit from consideration of the cytoarchitectural differences which may be present in the patient group.

When does the neuropathology of mood disorder arise?

Another key question concerns the timing of the neuropathological alterations and whether they are static, progressive or reversible. Post‐mortem studies of first‐episode cases are, in practice, impossible. Instead, the occurrence of neuropathology at that time must be assessed indirectly through imaging. For example, the sg24 volume reduction is present in first‐episode cases (Hirayasu et al., 1999) and adolescent depression (Botteron et al., 2002), and preliminary magnetic resonance spectroscopy data also suggest cingulate neuronal and neurochemical alterations relatively early in the disease process (Auer et al., 2000; Winsberg et al., 2000). These findings make it reasonable to infer that the associated glial and neuronal pathology also exists at or soon after the onset of illness. In concert with other data (Van Os et al., 1997; Sigurdsson et al., 1999; Heim and Nemeroff, 2001), this raises the possibility that the neuropathology of mood disorder reflects, at least partly, a neurodevelopmental process of some kind. Indirect support for this view comes from another inferential source, in which aberrant expression of developmental genes is viewed as a persistent sign of a prior abnormality. Notably, expression of reelin, a gene involved in neuronal migration, cortical lamination and synaptogenesis, is decreased in mood disorder (Fatemi et al., 2000; Guidotti et al., 2000); there are also changes in developmentally regulated isoforms of nerve cell adhesion molecule in bipolar disorder (Vawter et al., 2000). Such interpretations, although justified, must be made with caution since an increasing number of genes initially thought to be exclusively ‘developmental’ turn out to be expressed normally in the adult brain, wherein they may have different functions (Benson et al., 2000; Rodriguez et al., 2000; Costa et al., 2001).

Whatever the magnitude of early pathological changes, their subsequent evolution during the course of illness must also be considered. The lack of gliosis, or of any signs of neuronal cytoskeletal pathology, does mean that a conventional neurodegenerative process is unlikely to underlie mood disorder, for the reasons noted with regard to schizophrenia (Roberts and Harrison, 2000). However, their absence does not preclude other forms of ongoing pathological process which might influence and be influenced by the clinical course of mood disorder, as suggested by some MRI findings (Sheline et al., 1999) and as postulated by the stress–neurotoxicity model mentioned earlier. Furthermore, glial and neuronal number and morphology are affected by many environmental factors, both early and late (Hawrylak and Greenough, 1995; Kolb and Whishaw, 1998; Soffié et al., 1999; Esiri and Pearson, 2000; Fuchs and Gould, 2000; Lee et al., 2000), emphasizing that, as with the gene expression data, the morphometric changes could have arisen at any stage. To date, however, there are few empirical data about the timing and progression of neuropathology in mood disorder, other than a correlation observed between the duration of illness and the extent of synaptic protein reduction in the anterior cingulate cortex (Eastwood and Harrison, 2001). Again, larger and better characterized samples, preferably in concert with longitudinal imaging studies, will be needed to assess this issue.

Neuropathological features demonstrable only after the onset of a disorder cannot unequivocally be related to its aetiology, but neither should they be considered merely epiphenomena. Emergent (or reversible) changes could be an integral part of the progressive (or recurrent) pathophysiology of mood disorder, affecting circuits which underlie both the high vulnerability to relapse and the persisting deficits. Thus, even if some of the neuropathological abnormalities reported in mood disorder turn out to be secondary rather than primary, they could still have significant pathogenic and therapeutic implications.

Is mood disorder a neuropathology of connectivity?

Recent neuropsychological and imaging reviews conceptualize mood disorder in terms of networks and circuits, and implicate many of the structures discussed here (Austin and Mitchell, 1995; Mayberg, 1997; Elliott, 1998; Frith and Dolan, 1998; Rogers et al., 1998; Bearden et al., 2001; Drevets, 2001). As the imaging methods and statistical analyses become increasingly sophisticated, more specific predictions can be made as to the critical connections and the nature of their involvement (e.g. Friston et al., 1993; Meyer‐Lindenberg et al., 2001). It is therefore pertinent to ask whether the neuropathology of mood disorder is the structural basis of the dysfunctional circuits, a disorder of connectivity in an anatomical sense. There is a danger that such a proposal is either meaninglessly vague or tautological, and it is derivative of a similar conceptualization of schizophrenia, with similar limitations (Friston, 1998; Harrison, 1999a). Nevertheless, it is reasonable to view the various abnormalities of neurones, synapses and glia as together indicating cytoarchitectural differences which will have consequences for the activity and plasticity of the affected neural circuits (Duman et al., 2000; Manji et al., 2001), and it is parsimonious to link these with the evidence for functional abnormalities referable to the same areas and circuits. Certainly, in so far as there is now a neuropathology of mood disorder to explain, it is hard to envisage in what way the reported alterations could be of significance other than in terms of aberrant connectivity of some kind.


A range of neuropathological abnormalities have been reported in recent studies of primary mood disorder. They are mainly in the prefrontal cortex and are cytoarchitectural in nature. A loss of glia is the most notable finding, along with a reduced size and density of some neurones (Tables 1 and 3). There are also alterations described in the hippocampal formation (Table 4) and subcortical structures (Table 5) and concerning synaptic terminals and dendrites (Table 2). Together, the changes are suggestive of a difference in the cellular composition and circuitry of these regions in mood disorder. The findings provide sufficient reason to investigate the field further—and there are probably more studies in progress than the total reported to date—but the available data provide more questions than answers. Most importantly, there is a pressing need to establish which findings are robust. Ideally, the studies will not only attempt direct replication, but will be of appropriate size and design to cross diagnostic boundaries and allow variables implicated by the recent work, such as mood disorder subtype, clinicopathological correlations and anatomical heterogeneity, to be investigated.

The main conceptual and practical issues at this point are reminiscent of those which corresponding studies of schizophrenia have had to address. For example, in both disorders, the pathology is clearly not of a gross nature, with an absence of evidence of specific histopathological features detectable using routine stains; rather, there are quantitative alterations in cytoarchitectural parameters and, by inference, an altered organization of specific neuronal and glial circuits. Equally, the appropriate investigative tools are sensitive morphometric and molecular techniques. A combination of these methods, applied with increasing judiciousness over the past 15 years, has slowly produced convergent evidence of a molecular neuropathology of schizophrenia. A similarly multifaceted and sustained approach to primary mood disorder is now necessary.


The author wishes to thank David Cotter and Guy Goodwin for discussions, Rebecca Gittins for help compiling the tables, and Margaret Cousin for secretarial assistance. Supported by a Stanley Foundation Research Centre award, and a Medical Research Council ‘Neurobiology of Mood Disorders’ Co‐operative Group grant.


View Abstract