Journal of Affective Disorders 134 (2011) 483–487

Contents lists available at ScienceDirect

Journal of Affective Disorders

j ourna l homepage: www.e lsev ie r.com/ locate / j ad

Brief report

Hippocampal atrophy in first episode depression: A meta-analysis ofmagnetic resonance imaging studies

James Cole, Sergi G. Costafreda⁎, Peter McGuffin, Cynthia H.Y. FuInstitute of Psychiatry, King's College London, London, UK

a r t i c l e i n f o

⁎ Corresponding author at: Institute of Psychiatry,De Crespigny Park, PO Box 68, London SE5 8AF, UK. Tefax: +44 203 228 2016.

E-mail address: sergi.1.costafreda@kcl.ac.uk (S.G. C

0165-0327/$ – see front matter © 2011 Elsevier B.V.doi:10.1016/j.jad.2011.05.057

a b s t r a c t

Article history:Received 28 March 2011Received in revised form 30 May 2011Accepted 30 May 2011Available online 13 July 2011

Background: Reduced hippocampal volume has been consistently observed in majordepressive disorder. Hippocampal volume loss is particularly evident in patients withrecurrent and chronic depression. However, the reports in first episode depression havebeen mixed.

Methods: We performed a random effects meta-analysis to establish whether hippocampalatrophy exists from disease onset. We included magnetic resonance imaging studies ofhippocampal volume in patients with first episode major depressive disorder and matchedhealthy controls.

Results: A total of 7 studies met our inclusion and exclusion criteria, representing independentobservations in a total sample of 191 patients and 282 healthy controls. The cumulativeanalysis revealed hippocampal volume loss in patients with first episode depression relative tocontrols in both the left (standardised mean difference, SMD=−0.41, 95% ConfidenceInterval: [−0.78;−0.03], z=−2.14, p=0.0321) and right (SMD=−0.53[−0.98;−0.09], z=−2.38, p=0.0173) hippocampi. The average volume reduction was −4.0% in the left and−4.5% in the right hippocampus.

Conclusions: Hippocampal volume loss in first episode depression is consistent with aneurodevelopmental model of depression, advocating hippocampal structure as a potentialdiagnostic neurobiomarker for depression.

© 2011 Elsevier B.V. All rights reserved.

Keywords:DepressionStructural magnetic resonance imagingHippocampusMeta-analysis

1. Introduction

Initial reports of hippocampal volume in depressionsuffered from limited resolution of early neuroimagingmethods leading to poor separation of the hippocampus andadjacent structures (reviewed in: Fu et al., 2003). More recentmeta-analyses have convincingly established that reducedhippocampal volume is a neurobiological feature of majordepressive disorder, particularly in illnesses characterised byrecurrent episodes and in chronic depression (Campbell andMacQueen, 2004; McKinnon et al., 2009; Videbech and

King's College Londonl.: +44 203 228 3052;

ostafreda).

All rights reserved.

Ravnkilde, 2004). However, the stage at which hippocampalatrophy begins in depression is unclear. In first episodedepression, the evidencehas been equivocal,with some studiesobserving decreased hippocampal volume (Cole et al., 2010;Frodl et al., 2002), while others have reported no significantdifference (Eker et al., 2010; Kronmüller et al., 2009).

The presence of hippocampal atrophy from disease onsetwould have significant etiological and clinical implications as itwould support a model of depression in which structuralabnormalities appear earlier than previously believed withpotentially deleterious effects on clinical outcomeand responseto antidepressant therapy (MacQueen and Frodl, 2010). In thepresent study, we sought to establish whether hippocampalatrophy exists from disease onset. We performed a randomeffects meta-analysis of magnetic resonance imaging studies ofhippocampal volume in first episode depression.

484 J. Cole et al. / Journal of Affective Disorders 134 (2011) 483–487

2. Methods

2.1. Literature search

MEDLINE, EMBASE, Scopus and PsychINFO electronicdatabases were queried with the following search terms:“depression”, “major depressive disorder”, “magnetic reso-nance imaging”, “hippocampus”, and abbreviations “MDD”,“MRI”, “unipolar” for papers published between January 1990and February 2011. References from the retrieved papers andreviews of hippocampal and regional brain volumes in majordepressive disorder (MDD) were also inspected for relevantarticles.

Inclusion criteriawere: 1) patientswith a primary diagnosisof first-episode major depressive disorder (MDD) assessedusing international diagnostic criteria; 2) a healthy comparisongroup; 3) clinical sampleswere comprised of patients withfirstand/or recurrent MDD; 4) participants were screened forneurological and medical disorders that could affect brainstructure, including alcohol and substance abuse; 5) magneticresonance imaging (MRI) was the primary measurement tool;6) a continuous measure of hippocampal volume as thedependent variable; 7) distinct measures of the hippocampus,rather than combined measures with adjacent regions such asthe amygdala. In samples from which more than onepublication from the same research group and overlapsbetween samples were not explicitly described, authors werecontacted to establish whether these samples could beconsidered independent. In cases in which an overlap wasestablished, the most extensive sample was included. If papersdid not report sufficient data, but satisfied all other inclusioncriteria, requests for the requisite data were made.

The literature search identified 8 published researchreports meeting inclusion and exclusion criteria (Cole et al.,2010; Eker et al., 2010; Frodl et al., 2002; Kaymak et al., 2010;Kronmüller et al., 2009; MacQueen et al., 2003; Meisenzahl etal., 2010; van Eijndhoven et al., 2009).We confirmedwith theauthors of two of the studies (Frodl et al., 2002; Meisenzahlet al., 2010) an overlap in their samples and therefore includedonly the most recent one, presenting the largest number ofsubjects. This strategy led to a total of 7 independent studies.

2.2. Data extraction and analysis

For each study, the following data were extracted for thepatients and controls samples: number of participants, meanhippocampal volumes and their standard deviations, meanpatient age at scanning, mean disease duration, genderdistribution and medication status of the patient sample.Left and right volumes were assessed separately and Hedges'adjusted g was calculated for each study as the effect sizeestimator (Hedges, 1981). One major source of inconsistencyacross studies could be the divergent protocols used to definehippocampal boundaries. The main point of contentionbetween protocols is the inclusion of white matter areas(i.e. the alveus and fimbria) adjacent to the hippocampusproper. Based on Konrad et al.'s (2009) assessment, each ofthe current studies was categorised as including white matteror not (Supplementary Table A). Where studies were notexplicitly considered by Konrad et al. (2009), the history ofthe tracing protocol was examined. All studies were found to

derive their protocol from one assessed by Konrad et al.(2009).

Effect sizes were combined using the DerSimonian–Lairdmethod for random effects models (DerSimonian and Laird,1986) to generate a pooled standardised mean difference(SMD), and 95% confidence intervals (95% CI). For the studieswhich reported a number of findings from more than oneexperimental group with first episode depression (Kronmülleret al., 2009; van Eijndhoven et al., 2009), separate measure-ments were aggregated into one overall summary effect size toensure independence of study estimates, by taking theweighted mean and pooled variance of all within-studymeasurements, with the weight being the sample size in eachexperimental group. The random effects model for meta-analysis was selected over the fixed effects approach asprevious analyses have indicated considerable between-studyheterogeneity. The investigation of the effects of categoricalmoderators was also performed by using random effects meta-regression (Thompson and Higgins, 2002). Where possiblepotential moderators were kept as continuous variables:patient age, percentage of males in the sample and diseaseduration; while medication status was investigated as acategorical variable: unmedicated versus medicated patientsamples, as well as whether the delineation protocol wasinclusive of white matter in the hippocampal definition. One ofthe studies was identified as reporting outlier results (Kaymaket al., 2010), and we therefore conducted a sensitivity analysisby recomputing the random effects estimates after exclusion ofthat study to ensure that the findings were not biased by suchoutlier measurements. Analyses were conducted using stan-dard packages for meta-analysis of the R statistical software(http://www.r-project.org/).

3. Results

A total of 7 independent studies (Cole et al., 2010; Eker et al.,2010; Kaymak et al., 2010; Kronmüller et al., 2009; MacQueenet al., 2003;Meisenzahl et al., 2010; vanEijndhoven et al., 2009)met our inclusion and exclusion criteria, reporting measure-ments of 191 patients with a first episode of depression(Table 1, additional methodological study characteristics inSupplementary TablesAandB). Briefly, all studies had recruitedadult patients (mean age: 37.6 years, weighted by sample size;64.3% females). Patients suffered from moderate depression(mean HRSD score 21.0) with average illness duration of14.4 months (range of 4.7–25.2 months). Approximately half ofthe patients were either drug naïve (n=33) or drug free(n=65), while the remainder were taking antidepressantmedication. All control samples were matched by age andgender, with the exception of one study in which controls werematched only by gender (Meisenzahl et al., 2010). One studyemployeda3Tesla scanner (Kaymaket al., 2010),while all otherswere conducted on 1.5 Tesla systems, and image slice thicknessequal or less than 2 mm in all cases. Manual segmentationprotocols with high intra and intra and inter-rater reliability(N0.90 in most cases) were employed throughout. There wassubstantialheterogeneity inhippocampalvolumemeasurements(Fig. 1), justifying the use of a random-effects analysis approach.

Thecumulativeanalysis showedatrophy infirst episodeMDDpatients relative to controls for both left (standardised meandifference, SMD=−0.41, 95% Confidence Interval: [−0.78;

Table 1Demographic and clinical characteristics of first episode depression studies included in meta-analysis.

Study Year Participants Age, years % female Mood atscan

Moodmeasure

Disease duration,months

Medication status Dx

MacQueen 2003 20 1st episode MDD 28.4 (11.8) 65% n/s HDRS;BDI

25.2 (n/s) Some 1st episodepatients taking ADM

SCID20 matched controls 28.4 (11.5) 65%

vanEijndhoven

2009 20 depressed 1st episode 34.1 (11.6) 65% Depressed/remitted HDRS 7.1 (5.5) Medicationnaïve/medicationfree

SCID; MINI20 remitted 1st episode 35.8 (11.7) 70% 33.7 (17.3)20 controls 37.3 (12.7) 65%

Kronmuller 2009 13 1st episode MDDmale 38.1 (11.9) 0% Depressed HDRS 10.6 (13.7) All taking ADM SCID13 1st episode MDD female 41.5 (16.7) 100% 10.6 (13.7)11 male controls 42.0 (11.3) 0%19 female controls 42.7 (14.0) 100%

Eker 2010 25 1st episode MDD 32.1 (9.3) 72% Depressed HDRS 7.1 (7.0) 13 drug naïve, all 4weeks drug free

SCID22 controls 29.7 (6.4) 77%

Kaymak 2010 20 1st episode MDD 32.0 (8.5) 100% Depressed HDRS 4.7 (3.6) All drug naïve SCID15 controls 29.3 (5.8) 100%

Cole 2010 13 1st episode MDD 38.1 (7.9) 85% Depressed HDRS n/s All 2 weeks drugfree

SCID37 controls 42.2 (9.0) 76%

Meisenzahl 2010 47 1st episode MDD 41.8 (13.5) 55% Depressed HDRS 19.2 (37.2) Most on ADM. SCID138 controls 33.3 (12.2) 43%

MDD = patients with Major Depressive Disorder; HDRS = Hamilton Depression Rating Scale; BDI = Beck Depression Inventory; SCID = Structured ClinicalInterview for DSM-IV; MINI = Mini-International Neuropsychiatric Interview; ADM = anti-depressant medication; n/s = data not stated.

485J. Cole et al. / Journal of Affective Disorders 134 (2011) 483–487

−0.03], z=−2.14,p=0.0321) and right (SMD=−0.53[−0.98;−0.09],z=−2.38,p=0.0173) hippocampi (Fig. 1).The averagevolume reduction infirst episodedepression relativeto healthy controls, weighted by each study sample size, was−4.0% in the left and −4.5% in the right hippocampus. None ofthe clinical or methodological moderating values demonstratedstatistically significant effects, although unmedicated patientsamples tended to report larger right hippocampal volumereductions than medicated ones (z=1.72,p=0.0849).

Given the relatively small number of studies available for thismeta-analysis and the presence of a study with outlier results ofextreme atrophy (Kaymak et al., 2010, reporting −19.0% and−20.0% atrophy for left and right hippocampi, respectively), weconducted an additional sensitivity analysis to investigatewhether the summary findings were unduly determined bythis particular study. Following its exclusion, the magnitude ofatrophy was reduced bilaterally (weighted volume reduction,left: −2.2%, right: −2.6%), but these differences remainedstatistically significant (SMD =−0.23[−0.42;−0.02],z=−2.17,p=0.0295) and right (SMD =−0.30[−0.56;−0.04],

Fig. 1. Cumulative analysis of hippocampal volume differences between patients witplot presents the standardised mean difference in hippocampal volume between paNegative SMD represents volume loss in patients relative to controls. The diamond-present for first episode patients bilaterally (left p=0.0321; right p=0.0173).

z=−2.24,p=0.0248). In this sensitivity analysis the associationbetween medication status and hippocampal volume reductionwas not significant (z=1.14,p=0.2547).

4. Discussion

The present meta-analysis establishes that hippocampalvolume loss is not only a feature of recurrent and chronicdepression but is also evident in adult patients with a firstepisode of the illness. Hippocampal volume reduction in firstepisode patients has also been recently demonstrated usingwhole-brain voxel-based morphometry (Zou et al., 2010).These findings can be linked with recent studies reportingsmaller hippocampal volume in subjects at high risk fordepression due to a family history of depression or earlychildhood adversity (Chen et al., 2010; Rao et al., 2010) and inadolescents with early-onset depression (MacMaster andKusumakar, 2004). Taken together, these results are consis-tent with a neurobiological model whereby volume loss is not

h a first episode of depression and healthy controls. For each study, the foresttients and controls, along with its 95% confidence interval (horizontal line).shape random effects summary shows that bilateral hippocampal atrophy is

486 J. Cole et al. / Journal of Affective Disorders 134 (2011) 483–487

just a “scar” of depression, but may be a marker of riskpreceding and possibly predisposing to depression.

The existence of hippocampal atrophy from disease onsethas significant clinical implications. Deficits in autobiographical(Williams and Scott, 1988) and episodic memory (Ilsley et al.,1995), are a common featureof depression, and the volume lossin the hippocampus may parallel the memory impairmentexperienced by patients. Depression is also associated withgreater than a 50% increased risk for dementia (Saczynski et al.,2010), which may reflect a progression of the hippocampalatrophy with onset in first episode depression and is alsoapparent with subsequent episodes (McKinnon et al., 2009).

Disturbed hypothalamic pituitary adrenal (HPA) axis func-tion is a feature of depression, and it has been proposed thatadrenal hypersecretion of glucocorticoids, in particular cortisol,is linked to hippocampal atrophy in depression (Sapolsky,2001) as cortisol leads to neural atrophy and inhibition ofneurogenesis in the hippocampus (McEwen, 1999). Hippocam-pal volumetric reductions may therefore be a keymarker of theHPA disturbances linking early adversity, stress and depression(Frodl et al., 2008b). A key prediction from the proposedneurodevelopmental model of depression is that hippocampalatrophy should also exist in patients with a first episode ofadolescent depression. As all of the studies identified by ourstrategy were composed of adult patients, we were unable inthepresentmeta-analysis to test this predictionwhichwouldbean important hypothesis for future research.

Hippocampal atrophy has also been shown to predict poortreatment response and increased risk of relapse of depres-sion (Frodl et al., 2008a; MacQueen and Frodl, 2010). The firstepisode of this illness may thus be a critical point for assertiveintervention in which effective treatments may palliate orreverse structural changes, resulting in improved long-termclinical outcomes. Hippocampal atrophy, alongside otherstructural or functional abnormalities, may contribute togenerate new clinical biomarkers for diagnosis and prognos-tic prediction in depression (Costafreda et al., 2011; Nour-etdinov et al., 2010). Three-dimensional shape analysis mayidentify the initial development of subregional volumetriclosses and establish whether there is diagnostic specificity ofabnormalities in depression (Cole et al., 2010) that aredistinct from other disorders with hippocampal abnormali-ties, such as schizophrenia (Narr et al., 2004).

One of the studies identified by the meta-analysis reportedfindings thatwere anoutlier,withup to20% reduction betweenpatients and controls in hippocampal volume, more thandouble than those reported by the other studies (Kaymaket al., 2010). Further examination of the patient sampleindicated that these subjects had a similar severity of illnessof a moderate to severe range (Hamilton Depression RatingScale mean score of 23.10±4.20), but the patients had markedneuropsychological impairments over a range of tasks inexecutive function, memory and language, which may belinked to theirmarked hippocampal atrophy.We thus repeatedthe analysis, excluding this report, and demonstrated that thefindings of hippocampal volume reduction were robust to theinclusion or exclusion of this study.

Potential moderating effects of age, gender, or diseaseduration did not show any significant effects on hippocampalvolume. The range of mean ages of the patients included in themeta-analysis thoughwas limited (range of 28.4 to 42.7 years),

reducing the ability to detect effects that may be apparent inolderpatients. Significant effects of gender anddisease durationhave been demonstrated in some individual studies with firstepisodepatients (Frodl et al., 2002; vanEijndhovenet al., 2009),but not all (Cole et al., 2010).Age anddisease durationhave alsobeen shown to have significant effects in a recentmeta-analysisinvolving samples of both recurrent and first episode depres-sionpatients (McKinnonet al., 2009).While all the studies in thepresent meta-analysis employed manual segmentation tomeasure the hippocampal volume, studies differed with regardto theirdefinitionof thehippocampus. Inparticular, three studies(Eker et al., 2010; Kaymak et al., 2010; van Eijndhoven et al.,2009) includedwhitematter areas (i.e. the alveus andfimbria) intheir computation of volume. While these methodologicaldifferences are likely to have contributed to the heterogeneityof results across studies, we did not find a significant effect onhippocampal volume. Therewas amarginally significant effect ofmedication, in which unmedicated samples tended to reportlarger right hippocampal volume reduction consistent withreports of a neurotrophic benefit of antidepressants on thehippocampus (Boldrini et al., 2009; Duman et al., 2001). Asmorestudies in first episode patients become available, it would beinteresting to revisit potential moderating effects in first episodedepression with increased statistical power.

In summary, significant hippocampal volume loss is evidentin the first episode of depression. The findings from the presentmeta-analysis are consistent with a neurodevelopmentalmodel of depression and indicate the utility of the hippocampalstructure as a biomarker for depression. The clinical implica-tions underline the importance of early interventions to help toreduce the deleterious effects of recurrent episodes.

Supplementary materials related to this article can befound online at doi:10.1016/j.jad.2011.05.057.

Role of funding sourceFunders had no role in the design and conduct of the study; collection,

management, analysis, and interpretation of the data; and preparation,review, or approval of the manuscript.

Conflict of interestAll authors declare that they have no conflicts of interests.

AcknowledgementsJC was supported by a Medical Research Council PhD studentship. SGC

was supported by the National Institute for Health Research (NIHR)Specialist Biomedical Research Centre for Mental Health award to theSouth London and Maudsley NHS Foundation Trust and the Institute ofPsychiatry, King's College London.

References

Boldrini, M., Underwood, M.D., Hen, R., Rosoklija, G.B., Dwork, A.J., Mann, J.J.,Arango, V., 2009. Antidepressants increase neural progenitor cells in thehuman hippocampus. Neuropsychopharmacology 34, 2376–2389.

Campbell, S., MacQueen, G., 2004. The role of the hippocampus in thepathophysiology of major depression. J. Psychiatry Neurosci. 29,417–426.

Chen, M., Hamilton, J., Gotlib, I., 2010. Decreased hippocampal volume inhealthy girls at risk of depression. Arch. Gen. Psychiatry 67, 270–276.

Cole, J., Toga, A., Hojatkashani, C., Thompson, P., Costafreda, S., Cleare, A.,Williams, S., Bullmore, E., Scott, J., Mitterschiffthaler, M., et al., 2010.Subregional hippocampal deformations in major depressive disorder. J.Affect. Disord. 126, 272–277.

Costafreda, S.G., Dinov, I.D., Tu, Z., Shi, Y., Liu, C., Kloszewska, I., Mecocci, P.,Soininen, H., Tsolaki, M., Vellas, B., Wahlund, L., Spenger, C., Toga, A.W.,Lovestone, S., Simmons, A., 2011. Automated hippocampal shape

487J. Cole et al. / Journal of Affective Disorders 134 (2011) 483–487

analysis predicts the onset of dementia in mild cognitive impairment.Neuroimage 56, 212–219.

DerSimonian, R., Laird, N., 1986. Meta-analysis in clinical trials. Control. Clin.Trials 7, 177–188.

Duman, R.S., Nakagawa, S., Malberg, J., 2001. Regulation of adult neurogenesisby antidepressant treatment. Neuropsychopharmacology 25, 836–844.

Eker, C., Kitis, O., Taneli, F., Eker, O.D., Ozan, E., Yucel, K., Coburn, K., Gonul, A.S.,2010. Correlation of serum BDNF levels with hippocampal volumes in firstepisode, medication-free depressed patients. Eur. Arch. Psychiatry Clin.Neurosci. 260, 527–533.

Frodl, T., Meisenzahl, E., Zetzsche, T., Born, C., Groll, C., Jager, M., Leinsinger,G., Bottlender, R., Hahn, K., Moller, H., 2002. Hippocampal changes inpatients with a first episode of major depression. Am. J. Psychiatry 159,1112–1119.

Frodl, T., Jäger,M., Smajstrlova, I., Born, C., Bottlender, R., Palladino, T., Reiser,M.,Möller, H., Meisenzahl, E., 2008a. Effect of hippocampal and amygdalavolumes on clinical outcomes in major depression: a 3-year prospectivemagnetic resonance imaging study. J. Psychiatry Neurosci. 33, 423–430.

Frodl, T.S., Koutsouleris, N., Bottlender, R., Born, C., Jäger,M., Scupin, I., Reiser,M.,Möller, H., Meisenzahl, E.M., 2008b. Depression-related variation in brainmorphology over 3 years: effects of stress? Arch. Gen. Psychiatry 65,1156–1165.

A guide to neuroimaging in psychiatry. In: Fu, C.H., Walsh, N.D., Drevets, W.C.(Eds.), Chapter Neuroimaging Studies of Mood Disorders. Martin Dunitz,UK, pp. 131–169.

Hedges, L., 1981. Distribution theory for Glass's estimator of effect size andrelated estimators. J. Educ. Behav. Stat. 6, 107–128.

Ilsley, J., Moffoot, A., O'Carroll, R., 1995. An analysis of memory dysfunction inmajor depression. J. Affect. Disord. 35, 1–9.

Kaymak, S.U., Demir, B., Senturk, S., Tatar, I., Aldur, M.M., Ulug, B., 2010.Hippocampus, glucocorticoids and neurocognitive functions in patientswith first-episode major depressive disorders. Eur. Arch. Psychiatry Clin.Neurosci. 260, 217–223.

Konrad, C., Ukas, T., Nebel, C., Arolt, V., Toga, A.W., Narr, K.L., 2009. Definingthe human hippocampus in cerebral magnetic resonance images–anoverview of current segmentation protocols. Neuroimage 47,1185–1195.

Kronmüller, K., Schröder, J., Köhler, S., Götz, B., Victor, D., Unger, J., Giesel, F.,Magnotta, V., Mundt, C., Essig, M., Pantel, J., 2009. Hippocampal volumein first episode and recurrent depression. Psychiatry Res. 174, 62–66.

MacMaster, F., Kusumakar, V., 2004. Hippocampal volume in early onsetdepression. BMC Med. 2, 2.

MacQueen, G., Frodl, T., 2010. The hippocampus in major depression:evidence for the convergence of the bench and bedside in psychiatricresearch? Mol. Psychiatry 16, 252–264.

MacQueen, G., Campbell, S., McEwen, B., Macdonald, K., Amano, S., Joffe, R.,Nahmias, C., Young, L., 2003. Course of illness, hippocampal function, andhippocampal volume in major depression. Proc. Natl. Acad. Sci. U.S.A.100, 1387–1392.

McEwen, B., 1999. Stress and hippocampal plasticity. Annu. Rev. Neurosci. 22.McKinnon, M., Yucel, K., Nazarov, A., MacQueen, G., 2009. A meta-analysis

examining clinical predictors of hippocampal volume in patients withmajor depressive disorder. J. Psychiatry Neurosci. 34, 41–54.

Meisenzahl, E.M., Seifert, D., Bottlender, R., Teipel, S., Zetzsche, T., Jager, M.,Koutsouleris, N., Schmitt, G., Scheuerecker, J., Burgermeister, B., Hampel, H.,Rupprecht, T., Born, C., Reiser, M., Moller, H.J., Frodl, T., 2010. Differences inhippocampal volume between major depression and schizophrenia: acomparative neuroimaging study. Eur. Arch. Psychiatry Clin. Neurosci. 260,127–137.

Narr, K.L., Thompson, P.M., Szeszko, P., Robinson, D., Jang, S., Woods, R.P.,Kim, S., Hayashi, K.M., Asunction, D., Toga, A.W., Bilder, R.M., 2004.Regional specificity of hippocampal volume reductions in first-episodeschizophrenia. Neuroimage 21, 1563–1575.

Nouretdinov, I., Costafreda, S.G., Gammerman, A., Chervonenkis, A., Vovk, V.,Vapnik, V., Fu, C.H.Y., 2010. Machine learning classification with confi-dence: application of transductive conformal predictors to MRI-baseddiagnostic andprognosticmarkers indepression. Neuroimage56, 809–813.

Rao, U., Chen, L., Bidesi, A.S., Shad, M.U., Thomas, M.A., Hammen, C.L., 2010.Hippocampal changes associatedwith early-life adversity and vulnerabilityto depression. Biol. Psychiatry 67, 357–364.

Saczynski, J., Beiser, A., Seshadri, S., Auerbach, S., Wolf, P., Au, R., 2010.Depressive symptoms and risk of dementia: the Framingham heartstudy. Neurology 75, 35–41.

Sapolsky, R., 2001.Depression, antidepressants, and the shrinking hippocampus.Proc. Natl. Acad. Sci. U.S.A. 98, 12320–12322.

Thompson, S.G., Higgins, J.P.T., 2002. How should meta-regression analysesbe undertaken and interpreted? Stat. Med. 21, 1559–1573.

van Eijndhoven, P., van Wingen, G., van Oijen, K., Rijpkema, M., Goraj, B.,Verkes, R., Voshaar, R., Fernández, G., Buitelaar, J., Tendolkar, I., 2009.Amygdala volume marks the acute state in the early course ofdepression. Biol. Psychiatry 65, 812–818.

Videbech, P., Ravnkilde, B., 2004. Hippocampal volume and depression: ameta-analysis of MRI studies. Am. J. Psychiatry 161, 1957–1966.

Williams, J., Scott, J., 1988. Autobiographical memory in depression. Psychol.Med. 18, 689–695.

Zou, K., Deng, W., Li, T., Zhang, B., Jiang, L., Huang, C., Sun, X., Sun, X., 2010.Changes of brain morphometry in first-episode, drug-naive, non-late-lifeadult patients with major depression: an optimized voxel-basedmorphometry study. Biol. Psychiatry 67, 186–188.