Brain volume abnormalities in major depressive disorder: A meta‐analysis of magnetic resonance imaging studies

Abstract

Objective. So far, there have been no attempts to integrate the growing number of all brain volumetric magnetic resonance imaging studies in depression. In this comprehensive meta‐analysis the magnitude and extent of brain volume differences between 2,418 patients with major depressive disorder and 1,974 healthy individuals from 64 studies was determined. Methods. A systematic research was conducted for volumetric magnetic resonance imaging studies of patients with major depressive disorder in relation to healthy control subjects. Studies had to report sufficient data for computation of effect sizes. For each study, the Cohen's d was calculated. All analyses were performed using the random effects model. Additionally, meta‐regression analyses were done to explore the effects of potential sources of heterogeneity. Results. Patients showed large volume reductions in frontal regions, especially in the anterior cingulate and orbitofrontal cortex with smaller reductions in the prefrontal cortex. The hippocampus, the putamen and caudate nucleus showed moderate volume reductions. Conclusions. This is the first comprehensive meta‐analysis in major depressive disorder demonstrating structural brain abnormalities, particularly in those brain areas that are involved in emotion processing and stress‐regulation. Hum Brain Mapp, 2009. © 2009 Wiley‐Liss, Inc.

INTRODUCTION

With an estimated life‐time risk of at least 10% [Kessler et al.,2003; Weissman et al.,1996] major depressive disorder (MDD) is one of the most common psychiatric illnesses. Moreover, depressive symptoms are highly prevalent among other psychiatric disorders such as schizophrenia [Häfner et al.,2005b], alcohol and drug abuse [Kessler et al.,1996; Regier et al.,1990], and post‐traumatic stress disorder [Kilpatrick et al.,2003]. Although brain abnormalities have been identified in MDD, the number of neuroimaging studies in patients with this illness pale in comparison to those performed in schizophrenia (for meta‐analyses see: [Boos et al.,2007; Honea et al.,2008; Steen et al.,2006; Wright et al.,2000]). Despite an incomplete understanding of the neural circuitry underlying MDD, there is growing consensus that several brain areas are involved in depression [Beyer and Krishnan,2002; Campbell and MacQueen,2006; Drevets,2000; Sheline,2003; Videbech,1997]. Although reviews have appeared summarizing the results of neuroimaging studies of the hippocampus [Campbell et al.,2004; Videbech and Ravnkilde,2004] and anterior cingulate cortex [Hajek et al.,2008], there have been no attempts to integrate all volumetric neuroimaging studies in depression to date. Since such an effort could clarify the role of specific brain areas in the pathogenesis of MDD, we conducted a meta‐analysis to determine the magnitude and extent of brain volume differences between patients with MDD and healthy individuals as measured with magnetic resonance imaging (MRI).

METHODS

Data Sources

Magnetic resonance imaging studies that examined differences in brain volumes between patients with MDD and healthy control subjects were obtained through computerized database searches, including PubMED, Embase, Medline, Psychinfo, and the Cochrane library. The keywords used in the computerized search were: (major) depression/‐ive, unipolar, mood (disorder), affective (disorder), MRI, imaging, (brain) volume(s), morphometry, limbic (system) and gray/white matter. Titles and abstracts of the articles were examined to see whether they fulfilled the inclusion criteria. Bibliographies of included articles were checked for primary studies that might be of relevance.

Study Selection

One hundred and four studies were identified as potential candidates for the meta‐analysis. Studies were included if they (1) were MRI studies of brain structures in MDD published before January 2008, (2) compared patients with MDD with a healthy control group, (3) were published in the English language, (4) reported sufficient data to obtain an effect size (means, standard deviations, exact P‐values, or exact F‐values for a 2‐group comparison), (5) reported brain volumes (from multiple slices) rather than an area (from a single slice) and (6) if the mean age in either the MDD or comparison group was above 18 years. Twenty‐one studies had to be excluded from the meta‐analysis because they did not present sufficient data to compute the Cohen's d values [Ballmaier et al.,2004a; Ballmaier et al.,2004b; Bilder et al.,1999; Buchsbaum,1986; Chen et al.,2007; Frodl et al.,2004b; Frodl et al.,2007; Greenwald et al.,1997; Inagaki et al.,2004; Janssen et al.,2007; Kumar et al.,2000; Kumar et al.,2000; Lacerda et al.,2005; Lampe et al.,2003; Lavretsky et al.,2007; Lewine et al.,1995; Munn et al.,2007; Rabins et al.,1991; Rabins et al.,2000; Salloway et al.,1996; Shah et al.,1998]. Five studies were excluded because no healthy control group was included or patients with MDD had a life threatening comorbid disease [Dahabra et al.,1998; Ebmeier et al.,1997; Kumar et al.,1999; Nakano et al.,2002; Simpson et al.,2001]. Finally, 10 studies were excluded because they examined brain volumes in children and/or adolescents [Caetano et al.,2007; Gabbay et al.,2007; MacMaster et al.,2006; MacMaster et al.,2007; MacMaster and Kusumakar,2004; MacMillan et al.,2003; Nolan et al.,2002; Rosso et al.,2005; Steingard et al.,1996; Steingard et al.,2002]. Brain structures were only evaluated when volumes were available in (three or more) independent studies, therefore four studies had to be excluded (overlapping samples: [Frodl et al.,2004a; Lai et al.,2000; Sheline et al.,1999]; less than three studies: [Rubin et al.,1996]), resulting in 64 suitable studies that reported volumes of 130 brain structures in a total of 2,418 patients and 1,974 control subjects. Table I lists the included articles and the brain structures that were analyzed.

Table I.

Summary of 64 studies included in the meta‐analysis

Source	No. of patients	No. of controls	Included brain volumes
Almeida et al. [2003]	51	37	TB, PFC
Ashtari et al. [1999]	40	46	TB, Hip
Axelson et al. [1993]	19	30	TB, Hip
Ballmaier et al. [2004c]	24	19	IC, TB, OFC, ACC
Ballmaier et al. [2007]	14	10	Hip
Botteron et al. [2002]	30	8	TB, ACC
Brambilla et al. [2002]	18	38	ACC
Bremner et al. [2000]	16	16	IC, PFC, Hip, Amyg, Caud
Bremner et al. [2002]	15	20	TB, OFC, ACC
Caetano et al. [2001]	17	39	Thal
Caetano et al. [2004]	31	31	IC, Temp, Hip, Amyg
Caetano et al. [2006]	31	31	ACC
Colla et al. [2007]	24	14	IC, Hip
Coryell et al. [2005]	10	10	ACC
Drevets et al. [1997]	17	21	ACC
Dupont et al. [1995]	30	26	Caud
Frodl et al. [2002b]	30	30	IC, TB, Hip
Frodl et al. [2002a]	30	30	Amyg
Frodl et al. [2003]a	27	27	Amyg
Frodl et al. [2006]	34	34	IC, PFC, Hip
Hannestad et al. [2006]	182	64	Caud
Hastings et al. [2004]	18	18	ACC, Amyg
Hickie et al. [2005]	66	20	IC, Hip
Hickie et al. [2007]	45	16	Amyg, Caud
Husain et al. [1991]	44	44	TB, Puta
Janssen et al. [2004]	28	41	IC, TB, OFC, Hip
Krishnan et al. [1992]	50	50	TB, Caud
Krishnan [1993]	25	20	TB, Thal, Puta, Caud
Kumar et al. [1998]	53	30	IC, TB, PFC, Temp
Lacerda et al. [2003]	25	48	Puta, Caud
Lacerda et al. [2004]	31	34	OFC
Lange and Irle [2004]	17	17	TB, Hip, Amyg
Lavretsky et al. [2004]	41	41	IC, PFC, OFC
Lee et al. [2003]b	41	41	IC, OFC
Lenze and Sheline [1999]	24	24	Puta, Caud
Lloyd et al. [2004]	51	39	Hip
MacQueen et al. [2003]	37	37	Hip
Maller et al. [2007]	45	30	Hip
Mervaala et al. [2000]	34	17	Hip, Amyg
Monkul et al. [2007]	17	17	OFC, ACC, Hip, Amyg
Naismith et al. [2002]	47	20	Caud
Neumeister et al. [2005]	31	57	Hip
O'Brien et al. [2004]	61	40	TB, Hip
Pantel et al. [1997]	19	13	IC, TB, PFC, Temp
Parashos et al. [1998]	32	32	TB, PFC, OFC, Thal, Puta, Caud
Pillay et al. [1997]	38	20	TB
Pillay et al. [1998]	38	20	Caud
Posener et al. [2003]	27	42	TB, Hip
Rusch et al. [2001]	25	15	Hip
Salokangas et al. [2002]	37	19	PFC, Temp
Saylam et al. [2006]	24	24	IC, Hip
Sheline et al. [1996]	10	10	TB, Hip
Sheline et al. [1998]	20	20	TB, Amyg
Sheline et al. [2003]	38	38	Hip
Steffens et al. [2000]	66	18	Hip
Steffens et al. [2003]b	30	40	TB, OFC
Taylor et al. [2005]	135	83	Hip
Taylor et al. [2007]	226	144	OFC
Vakili et al. [2000]	38	20	Hip
Velakoulis et al. [2006]	19	87	Hip, Amyg
von Gunten et al. [2000]	14	14	Hip, Amyg
Vythilingam et al. [2002]	32	14	Hip
Vythilingam et al. [2004]	38	33	Hip
Weniger et al. [2006]	21	23	IC, TB, Hip, Amyg

IC, intracranial; TB, total brain; PFC, prefrontal cortex; ACC, anterior cingulate; Temp, temporal cortex; ofc, orbitofrontal cortex; Hip, hippocampus; Amyg, amygdala; Puta, putamen; Caud, caudate nucleus; Thal, thalamus.

^aOnly patients with recurrent MDD and their matched healthy controls were selected from this study, data regarding the first episode patients was not included.

^bOFC volumes were excluded in the analyses due to overlapping samples with Taylor et al [2007].

Data Extraction

The brain structures that were suitable for analysis included intracranium, total brain, prefrontal cortex, anterior cingulate, temporal cortex, orbitofrontal cortex, hippocampus, amygdala, caudate nucleus, putamen and thalamus. If sufficient data was present, analyses were extended to examine the effect in the two hemispheres separately. When brain volumes were only reported per hemisphere total volume was calculated by summarizing the left and right brain volumes. To obtain the total standard deviation (SD_To), the following formula was used:

(1)

with SD_Le/Ri being the standard deviation of the volume of the particular left/right brain structure and Cor_Le,Ri being the correlation between the left and right volume of the brain structure. Although it is reasonable to assume that left and right brain volumes are (highly) correlated, it is unlikely that correlations are exactly “1” (e.g., due to the influence of handedness on brain volume [Geschwind et al.,2002]). However, to correct for rounding off errors and to be conservative in our estimation, we set the correlation at “1” and hence allowed larger estimated SD's for total volumes.

The key to a meta‐analysis is defining an effect size statistic capable of representing the quantitative findings of a set of research studies in a standardized form that permits meaningful comparison and analyses across the studies [Lipsey and Wilson,2001]. Therefore, for each study in this meta‐analysis, the effect size statistic Cohen's d was calculated. In this analysis, the mean volume of a specific brain structure for patients with MDD was subtracted from the mean volume for comparison subjects and divided by the pooled standard deviation of both. When means and standard deviations were not available, d‐values were calculated from exact P‐values, t‐values, or F‐values. Meta‐analytic methods were applied to obtain a combined effect size, which indicated the magnitude of the association across all studies [Hedges and Olkin,1985]. A t‐test was subsequently performed on the null hypothesis that the d value is 0.00, which we report together with the associated P‐value. According to Cohen [Cohen,1988], d values of 0.2 represent small effects, values between 0.4 and 0.6 moderate effects, and d values of 0.8 or higher large effects.

A second measure of effect size was used to calculate a percentage difference between both groups. We used the ratio of the mean volume in the depression group divided by the mean volume in the comparison group. Specifically, for each region in study i (i = 1,2,3,…k) we used the weighted ratio effect size (EffRW) defined as:

(2)

where pt refers to patients with MDD, nc to the control group, w to the weights per study and M to the group regional volume mean. To control for sample size differences, for each region the EffRW is multiplied with the specific weights per study derived from our meta‐analyses. By adding the separate weights, an average weighted percentage volume difference is obtained.

Next to the effect size the variance between and within studies has to be explored. All analyses were performed with a random‐effects model using the statistical package Comprehensive Meta‐analysis V2 [Borenstein and Rothstein,1999]. If there is significant heterogeneity among the results of the included studies, random effects models will give wider confidence intervals than fixed effect models [DerSimonian and Kacker,2007; DerSimonian and Laird,1986]. For each brain region, a test of homogeneity (Cochran's Q test) was performed to test whether the studies could be assumed to share a common population effect size. A significant Q statistic indicates heterogeneity of the individual study effect sizes, which poses a limitation to a reliable interpretation of the results. Additionally, we also calculated I ² to provide a more interpretable measure of consistency between studies in this meta‐analysis [Higgins et al.,2003; Huedo‐Medina et al.,2006]. The I ² index can be interpreted as the percentage of the total variability in a set of effect sizes due to true heterogeneity, that is, to between‐studies variability. The I ² is placed between 0 and 100% where a value of 0% indicates no observed heterogeneity and larger values imply increasing heterogeneity. Since MDD is a heterogeneous psychiatric illness we expected a significant result in the homogeneity test. Therefore, a meta‐regression analysis was planned to explore the effects of potential sources of heterogeneity, by regressing effect sizes against mean age, gender, duration of illness, illness onset, number of depressive episodes, medication intake or symptom scores. The random effects regression analysis was performed using the unrestricted maximum likelihood model in Comprehensive Meta‐analysis [Borenstein and Rothstein,1999].

To examine the possibility of publication bias, we computed a fail‐safe number of studies [Orwin,1983; Rosenthal,1991]. Publication bias implies that studies with no effect may not be published, posing a threat to the stability of the obtained effect size. The fail‐safe number (NFS) of studies indicates the number of unpublished studies with null effects that must reside in file drawers to reduce the observed effect size to a negligible level. The statistic can be calculated with the use of the formula given by Orwin [1983] and Lipsey and Wilson [2001]:

(3)

with k being the number of studies, ESk, the mean weighted effect size; and ESc, the criterion effect size (which we set at a d value of 0.10).

RESULTS

Meta‐Analyses

As presented in Table II, the results of the meta‐analysis indicate brain volume decreases in patients with MDD as compared with healthy control subjects. The 95% confidence intervals were methodologically stringent for all significant findings. In addition, the fail‐safe number of studies for all analyses was large enough to lend credence to our findings.

Table II.

Brain structures included in Meta‐analysis and Results

Brain structure	No. of studies	No. of patients	No. of controls	Mean weighted effect size Cohen's d (95% CI)	P‐value for d	Average weighted percentage difference in %	Within‐ category homogeneity statistic Q	P‐value for Q	I 2 in %	NFS
Intracranium	14	452	364	0.03 (−0.13 to 0.20)	0.37	NA	17.69	0.17	26.53	10
Total brain	22	703	622	−0.06 (−0.16 to 0.05)	0.29	NA	15.89	0.78	0.00	35
Amygdala
Left	13	321	361	0.07 (−0.28 to 0.43)	0.68	NA	49.53	<0.001	75.72	4
Right	13	321	361	0.14 (−0.11 to 0.40)	0.27	NA	26.00	0.011	53.85	6
Total	14	366	377	0.05 (−0.25 to 0.35)	0.76	NA	42.58	<0.001	69.47	7
Anterior cingulate cortex
Left	8	183	171	−1.11 (−1.88 to −0.34)	0.005	−12.19	74.83	0.005	90.65	97
Right	7	166	150	−0.62 (−0.88 to −0.37)	<.001	−10.71	7.29	0.30	17.65	52
Total	8	181	170	−0.769 (−1.32 to −0.22)	0.006	−11.91	44.58	<0.001	84.30	71
Caudate nucleus
Left	5	285	172	−0.04 (−0.24 to 0.16)	0.67	NA	0.33	0.52	0.00	8
Right	5	285	172	0.00 (−0.20 to 0.20)	0.99	NA	1.27	0.87	0.00	5
Total	10	467	316	−0.31 (−0.58 to −0.04)	0.024	−6.74	26.89	0.001	70.19	41
Hippocampusa
Left	30	1083	914	−0.37 (−0.52 to −0.23)	<0.001	−4.71	65.43	<0.001	55.68	142
Right	30	1083	914	−0.41 (−0.54 to −0.28)	<0.001	−5.12	58.14	0.001	50.12	153
Total	31	1114	991	−0.41 (−0.54 to −0.28)	<0.001	−5.07	60.67	<0.001	50.55	158
Orbitofrontal cortex
Left	5	326	255	−0.52 (−0.85 to −0.18)	0.002	−9.48	10.76	0.029	62.82	31
Right	5	326	255	−0.47 (−0.79 to −0.15)	0.004	−8.71	9.87	0.043	59.48	29
Total	7	373	204	−0.43 (−0.78 to −0.09)	0.014	−9.18	21.95	0.001	72.67	38
Prefrontal cortex
Left	5	157	119	−0.22 (−0.44 to −0.01)	0.045	−2.1	1.55	0.82	0.00	17
Right	5	157	119	−0.22 (−0.44 to 0.00)	0.053	−1.21	1.94	0.75	0.00	16
Total	7	242	181	−0.34 (−0.52 to −0.16)	<0.001	−3.35	4.82	0.57	0.00	31
Putamen
Left	3	90	116	−.32 (−0.88 to 0.23)	0.26	NA	7.53	0.023	73.42	13
Right	3	90	116	−.34 (−0.95 to 0.27)	0.27	NA	9.05	0.011	77.90	14
Total	6	192	184	−0.48 (−0.80 to −0.16)	0.003	−11.28	11.77	0.038	57.52	35
Temporal cortex
Left	3	87	63	0.07 (−0.24 to 0.37)	0.67	NA	0.44	0.80	0.00	2
Right	3	87	63	0.28 (−0.03 to 0.58)	0.076	NA	0.81	0.67	0.00	6
Total	4	140	93	−0.03 (−0.40 to 0.34)	0.88	NA	6.85	0.077	56.20	6
Thalamus	3	74	91	−0.17 (−0.55 to 0.20)	0.37	NA	3.08	0.21	35.10	9

CI, confidence intervals; NFS, fail‐safe number: number of unpublished studies with null effects that must reside in file drawers to reduce the observed effect size to a negligible level; I ²: percentage of the total variability due to true heterogeneity; NA, not applicable.

^aSince exclusion of studies examining medication naive and medication free (at least six weeks of medication) patients did not change the results, the full sample results are displayed.

Areas in the Frontal Lobe

The largest effect was found for the anterior cingulate cortex, with smaller volumes in patients with MDD compared with healthy control subjects (Fig. 1). Eight studies were included with a total group size of 181 patients with MDD and 170 healthy controls, resulting in a combined‐effect Cohen's d of −0.769 (P = 0.006). Excluding the studies that used specific subregions of the anterior cingulate cortex [Botteron et al.,2002; Brambilla et al.,2002; Bremner et al.,2002; Drevets et al.,1997; Hastings et al.,2004] did not change the results. The effect was most pronounced in the left anterior cingulate cortex volume (k = 8; d = −1.11; P = 0.005; right anterior cingulate cortex volume: k = 7; d = −0.624; P < 0.001).

Figure 1.

Mean total anterior cingulate cortex volume. Error bars indicate 95% confidence interval.

Other areas in the frontal lobe that showed reduced volumes in patients with MDD were the orbitofrontal cortex (total: k = 7; d = −0.433; P = 0.014; left: k = 5; d = −0.516; P = 0.002; right: k = 5; d = −0.469; P = 0.004; Fig. 2) and the left (k = 5) and total (k = 7) prefrontal cortex (left: d = −0.223; P < 0.005), total: d = −0.342; P < 0.0001). The right prefrontal cortex volume difference between patients with MDD and healthy controls was significant at trend level (d = −0.216; P = 0.053; Fig. 3).

Figure 2.

Mean total orbitofrontal cortex volume. Error bars indicate 95% confidence interval.

Figure 3.

Mean total prefrontal cortex volume. Error bars indicate 95% confidence interval.

Hippocampus

The hippocampus was the brain structure studied most often (k > 30), with a total group size of 1,114 (left/right: 1,083) patients with MDD and 991 (left/right: 914) control subjects. Reduced hippocampal volumes were found in patients with combined‐effect sizes of −0.373 (P < 0.0001), −0.408 (P < 0.0001) and −0.408 (P < 0.0001) for the left, right and total hippocampus volume, respectively (Fig. 4). Excluding the studies that examined medication naive or medication free (at least 6 weeks free of medication) patients [MacQueen et al.,2003; Neumeister et al.,2005; Saylam et al.,2006; Vythilingam et al.,2004] did not change the results (total: k = 27; d = −0.409; P < 0.001). We also examined hippocampal volume of these studies with untreated patients to elucidate differences as compared with the whole sample with treated patients, but findings in this group are comparable (k = 4; d = −0.362; P = 0.008).

Figure 4.

Mean total hippocampus cortex volume. Error bars indicate 95% confidence interval.

Striatum

Significant volume reductions of the striatum were found in the patient group compared with healthy individuals (Fig. 5, 6). Specifically, a moderate effect was found for the total putamen volume (k = 6; d = −0.48; P = 0.003) and a small effect was found in the total caudate nucleus volume, (d = −0.31, P = 0.024).

Figure 5.

Mean total putamen volume. Error bars indicate 95% confidence interval.

Figure 6.

Mean total caudate nucleus volume. Error bars indicate 95% confidence interval.

Other Brain Areas

Analyses of volumes of the intracranial space, total brain, temporal cortex, (bilateral) amygdala, and thalamus did not show significant differences between the groups.

Meta‐Regression

Significant heterogeneity among studies was detected for several structures that showed significant differences between the groups (Table II lists the Cochran's Q‐coefficients and I ² for the homogeneity tests). A priori it was assumed that inter‐study differences in age, gender, age of onset, duration of illness or symptom scores could explain some of the surplus in variance. Unfortunately, most studies provided insufficient data regarding medication intake or number of episodes; therefore we were unable to perform the meta‐regression with these variables. A significant effect of age (z = −2.38; P = 0.017) was found on total putamen volume (Q _model = 5.33; P = 0.02 vs. Q _residual = 6.73; P = 0.15). None of the other moderator effects reached significance.

DISCUSSION

This meta‐analysis integrated the results of 64 magnetic resonance imaging studies that compared the volumes of various brain structures in patients with MDD (N = 2,418) with those of healthy control subjects (N = 1,974). The results indicate pronounced brain volume reductions in specific brain areas in patients with MDD.

Frontal Lobe

Several lines of evidence suggest the presence of specific neural circuits within the limbic‐cortical system that mediate stress‐responsiveness, mood and emotional regulation [Brody et al.,2001; Seminowicz et al.,2004; Tekin and Cummings,2002]. Interestingly, the results of this meta‐analysis revealed the presence of structural brain abnormalities in patients with MDD in many of the regions involved in emotional processing and stress‐responsiveness. Of particular interest are the prominent volume reductions in frontal regions (anterior cingulate, orbitofrontal and prefrontal cortex) which are known to control emotion regulation by inhibiting the activity of limbic regions such as the hippocampus and the amygdala [Beauregard et al.,2001; Hariri et al.,2000]. Evidence from postmortem studies suggests cell loss and cell atrophy in subgenual prefrontal and orbitofrontal cortex in patients with MDD [Rajkowska,2000]. The results of this meta‐analysis also showed a pronounced volume reduction of the left anterior cingulate cortex relative to the right side. This finding is consistent with previous studies reporting larger left than right subgenual anterior cingulate cortex volumes reductions [Chen et al.,2007; Drevets et al.,1997].

Limbic System and Frontal Lobe

In addition to the decreased volumes of prefrontal regions, hippocampal volumes were also reduced in patients with MDD. Given the putative relationship between increased sensitivity to stress and affective disorders [Swaab et al.,2005], it is important to note that the hippocampus, amygdala and prefrontal cortex are also involved in Hypothalamus‐Pituitary‐Adrenal (HPA)‐axis regulation. This is relevant, since MDD has been linked to disrupted HPA‐axis activity and increased levels of cortisol [Bao et al.,2008; Swaab et al.,2005] which in turn has been postulated to effect hippocampal volume through the inhibition of neurogenesis in this brain structure [Czeh and Lucassen,2007; Henn and Vollmayr,2004; Sapolsky,2000]. Several mechanisms have been proposed to explain how prolonged stress can result in limbic and prefrontal abnormalities, such as decreased dendritic branching [Radley and Morrison,2005], decreased neurogenesis [Duman,2004], loss of neurons, or decreased expression of brain derived neurotrophic factor [Duman et al.,1997; Radley and Morrison,2005]. Evidence for stress‐induced brain abnormalities in MDD is also provided by studies examining genetic variations in the glucocorticoid receptor gene. Especially functional polymorphisms of the NR3C1 gene (Nuclear Receptor Subfamily 3, Group C, Member 1) are associated with increased susceptibility to MDD [van Rossum et al.,2006; van West et al.,2006]. Of particular interest are the findings of a recent study reporting an association of four illness‐related polymorphisms of the NR3C1 gene with overall smaller hippocampal volumes in patients with MDD [Zobel et al.,2008]. This suggests that “at‐risk”‐alleles of the NR3C1 gene influence hippocampal volume.

Amygdala

The meta‐analysis revealed no significant volumetric abnormalities in the amygdala. However, the amygdala findings are highly inconsistent with a broad range of effect sizes. Although the anatomical boundaries of the amygdala are difficult to outline on MRI images, all studies reported high intrarater correlation coefficients suggesting reliably measured volumes. Also, no association was found between choice of segmentation protocol and amygdala volume increases or decreases. An explanation for the discrepancy between studies may relate to genetic differences between subjects and patients samples. For instance, reduced gray matter volume in the amygdala is more pronounced in those subjects carrying the s‐allele of the serotonin transporter promoter polymorphism [Heinz et al.,2005; Pezawas et al.,2005]. Indeed, individuals carrying the s‐allele tend to have increased anxiety‐related temperamental traits, which in turn are related to increased risk for developing depression [Lotrich and Pollock,2004]. Only a few studies examined the association between the serotonin transporter polymorphism and brain structures in patients with MDD, showing mixed results with respect to amygdala, hippocampus and caudate nucleus volumes [Frodl et al.,2004b; Hickie et al.,2007; O'hara et al.,2007; Taylor et al.,2005]. Finally, it is unknown if other factors, such as whether a patient is in a current episode or in remission as well as duration of illness contributes to differences in amygdala volume. In addition, it is unclear to what extent the amygdala is affected by antidepressant medication.

Striatum

The findings from this meta‐analysis support the presence of volume reductions in the striatum, primarily in total putamen volume. The striatum has been associated with mood, cognitive processes, motivation and regulation of movement and is also part of several neuroanatomic circuits that are involved in mood regulation [Alexander et al.,1986; Drevets,2001; Rogers et al.,1998; Tekin and Cummings,2002]. Further evidence of striatal involvement in MDD is provided by a postmortem study showing reduced putamen and pallidum volumes in patients with MDD compared with nonpsychiatric subjects [Baumann et al.,1999]. Interestingly, lesions of the putamen and of the caudate nucleus have been associated with a higher prevalence of MDD and/or depressive symptoms in post‐stroke depression [Starkstein and Robinson,1989; Vataja et al.,2004], Huntington disease [Slaughter et al.,2001a] and Parkinson disease [Slaughter et al.,2001b].

Antidepressant Medication

Most studies included in this meta‐analysis examined brain volumes in patients on antidepressant treatment; this hampered our ability to analyze effects of antidepressant medication on brain volumes in depression. Moreover, definition of medication free status varied greatly among studies. Although the effect of antidepressants on the brain is obviously important, so far only a few studies evaluated this in patients with MDD. In geriatric patients with depression, antidepressant exposure was associated with larger orbitofrontal gray matter volume compared with medication naïve patients [Lavretsky et al.,2005]. In contrast, other studies have reported improved memory and decreased symptom severity, but failed to find an association with hippocampal or orbitofrontal cortex volume [Janssen et al.,2007; MacQueen et al.,2003; Vythilingam et al.,2004]. However, one must bear in mind that with the exception of one study [Vythilingam et al.,2004], all studies were cross‐sectional, and none of these studies corrected for cumulative or life‐time medication intake.

Overlap and Differences With Schizophrenia and Bipolar Disorder

Major depressive disorder, schizophrenia and bipolar disorder share important clinical features, i.e. depressive symptoms, anhedonia, memory deficits, and lack of motivation [Häfner et al.,2005b; Häfner et al.,2005a; Lake,2008]. Moreover, during the course of schizophrenia the prevalence of depression ranges widely, from 6 to 75% [Siris and Bench,2003]. While there is considerable overlap in risk factors and precursors in these disorders, the overlap in brain volume abnormalities is less clear. Recent meta‐analyses indicate mild ventricular enlargement in bipolar disorder [Kempton et al.,2008; McDonald et al.,2004] and reduced cerebral, temporal lobe and amygdala volumes, and enlarged lateral and third ventricles, and basal ganglia volumes in schizophrenia [Boos et al.,2007; Wright et al.,2000] compared with healthy controls. Interestingly, based on our results and the previous meta‐analyses in schizophrenia, patients with MDD and schizophrenia both show reduced hippocampal [Nelson et al.,1998], prefrontal and anterior cingulate cortex volumes [Baiano et al.,2007; Wright et al.,2000], suggesting that brain regions regulating stress response are affected in both disorders. Indeed, psychosocial stress is a well‐established precipitant of depressive episodes as well as psychotic relapses [Walker, 2008]. Moreover, stressors such as life events and high expressed emotion have been found to precede the onset and recurrence of depression [Kessler,1997] and psychotic disorder [Bebbington et al.,1996]. Therefore, it may well be that the phenotypic overlap of reduced brain volumes (on the basis of the previous mentioned meta‐analyses) in MDD and schizophrenia could be explained by a common genetic vulnerability to stress. Indeed, this vulnerability may be phenotypically expressed as a decrease in hippocampal volume. In fact, evidence that schizophrenia and MDD share common genetic background is found in the original DISC1‐gene study (DISC1 = Disrupted in Schizophrenia‐1 family)[Millar et al.,2000]. Usually, DISC1 is considered to be a major risk gene for schizophrenia but affected individuals in this family displayed a range of diagnoses, with the majority being diagnosed with either schizophrenia or MDD [Blackwood et al.,2001; Hashimoto et al.,2006]. In addition, the DISC1‐gene is involved in structural and functional alterations in hippocampal formation and probably adult neurogenesis in the dentate gyrus [Callicott et al.,2005].

Limitations

Some limitations of this meta‐analysis should be noted. First, structures other than those that have been evaluated in this meta‐analysis may also be affected in patients with MDD. For instance, separated gray and white matter volumes of the cerebrum are scarcely reported. Moreover, only two studies [Parashos et al.,1998; Salokangas et al.,2002] examined ventricular volumes, while increased ventricular volume is the most replicated finding in schizophrenia (interestingly, both studies reported volume increases in MDD patients)[Wright et al.,2000]. Second, although we found significant decreases in brain volumes, almost all analyses resulted in significant heterogeneity which hampers a reliable interpretation and may have influenced the results [Hedges and Vevea,1998]. However, all significant findings indicate brain volume reductions with reliable confidence intervals in patients with MDD relative to healthy control subjects. In addition, the meta‐regression analyses did not show differences between studies, except for a small effect of age on the putamen. Unfortunately, many studies included in this meta‐analysis did not provide sufficient data to examine the effects of antidepressant medication. Moreover, most studies lack information on number of depressive episodes and scores on symptom scales. Thus, the possibility that some of the significant effects are confounded by these factors cannot be ruled out.

Future Directions

So far, most studies used a region of interest approach, however, the use of voxel‐based morphometry allows comprehensive and global assessment of brain structures without a priori identification of regions of interest [Ashburner and Friston,2000]. Although, voxel‐based morphometry studies are sparse in MDD [Bell‐McGinty et al.,2002; Chen et al.,2007; Shah et al.,1998; Tang et al.,2007; Vasic et al.,2008], findings are in line with those found in our meta‐analysis. Of interest is also the measurement of cortical thickness, i.e. a measure to determine the thickness of the gray matter of the human cerebral cortex [Davatzikos and Bryan,1996; Fischl and Dale,2000; Kabani et al.,2001; Thompson and Toga,1996]. Cortical thinning is frequently regionally specific and can therefore provide important additional information for characterizing disease‐specific neuroanatomical changes. Thus far, there have been no published studies that measured cortical thickness in MDD.

Relatively new MRI techniques such as diffusion tensor imaging and magnetization transfer imaging are used to study the orientation of white matter tracts in vivo and yield an index of microstructural integrity [Basser et al.,1994; Wolff and Balaban,1989]. Interestingly, findings from these studies in MDD indicate microstructural white matter abnormalities in widespread prefrontal and limbic regions [Alexopoulos et al.,2008; Bae et al.,2006; Gunning‐Dixon et al.,2008; Li et al.,2007; Murphy et al.,2007; Nobuhara et al.,2004; Nobuhara et al.,2006; Taylor et al.,2004; Yang et al.,2007]. Integration of white matter volume measurements and these MRI techniques may improve our understanding of the neural circuitry involved in MDD.

In addition, longitudinal imaging studies can clarify whether the volume reductions are static or progressive over time and to what extent the volume (change) is affected by the effects of antidepressant medication, illness severity or age. Future studies should also include relatives of patients with MDD and (discordant) monozygotic and dizygotic twin‐pairs to examine the relationship between genetic vulnerability to develop the illness and brain morphology. Such studies have proven to be valuable in schizophrenia and bipolar disorder in clarifying some of the underlying mechanisms of the brain volume abnormalities in probands [Baaré et al.,2001; Boos et al.,2007; Brans et al.,2008; Lawrie et al.,2008; van der Schot et al., 2008; Whalley et al.,2007]. Finally, future studies ought to focus on the search for susceptibility genes in relation to brain abnormalities by using linkage and association methods.

CONCLUSION

In summary, our results show that structural brain abnormalities are present in patients with MDD. Since MDD is characterized by abnormalities in emotion regulation and stress‐responsiveness, the majority of studies in our meta‐analysis focused on investigating those areas that are involved in these processes. Our findings indeed provide evidence that many, but not all, of those areas, show volume reductions in patients with MDD. Some of the brain abnormalities in depression are similar to those reported in patients with schizophrenia, such as the declines in hippocampal and frontal volumes with comparable effect sizes. Finally, this meta‐analysis confirms the preservation of global cerebral and temporal cortex volume in MDD patients, which is in line with findings in patients with bipolar disorder, but in contrast to the slight but significant reduction found in schizophrenia patients. Our results strongly suggest that studying brain structure in MDD will contribute to understanding the pathogenesis of this disease.