Associations of the GSK3β rs6438552 A allele with reduced regional gray matter (GM) volume differences in patients with major depressive disorder (MDD). A, The GSK3β single-nucleotide polymorphism (SNP) rs6438552 showed significant associations with structural variation in patients with MDD in the right and left superior temporal gyrus (STG) (green clusters) and the right hippocampus (red cluster) (PFWE < .001, PFWE = .02, and PFWE = .02, corrected for whole-brain search, respectively; FWE indicates familywise error). The Montreal Neurological Institute coordinates are for top (46, −14, 2) and bottom images (32, −26, −2). B, The effect of the GSK3β SNP rs6438552 on GM volume differences in individual patients with MDD in which the A allele is associated with lower GM volume in the right STG are shown (similar plots were revealed for left STG and right hippocampus; data not shown).
Associations of the GSK3β rs12630592 G allele with reduced regional gray matter (GM) volume differences in patients with major depressive disorder (MDD) showing overlap with associations for GSK3β rs6438552. A, Significant clusters, corrected for whole-brain search, for high linkage disequilibrium GSK3β single-nucleotide polymorphisms rs6438552 (shown in red) and rs12630592 (shown in green), with cluster colocalization shown in brown. The Montreal Neurological Institute coordinates are for top (46, −14, 2) and bottom images (32, −26, −12). B, The effect of GSK3β rs12630592 on GM volume differences in individual patients with MDD in which the G allele is associated with lower GM volume in the right superior temporal gyrus (STG) is shown (similar plots were revealed for left STG and right hippocampus; data not shown).
Individual plots for patients with major depressive disorder (green) and healthy control subjects (blue) of the significant voxel-level interaction effect in the right hippocampus (A) and the right superior temporal gyrus (STG) (B). GM indicates gray matter.
Inkster B, Nichols TE, Saemann PG, Auer DP, Holsboer F, Muglia P, Matthews PM. Association of GSK3β Polymorphisms With Brain Structural Changes in Major Depressive Disorder. Arch Gen Psychiatry. 2009;66(7):721-728. doi:10.1001/archgenpsychiatry.2009.70
Copyright 2009 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2009
Indirect evidence suggests that the glycogen synthase kinase-3β (GSK3β) gene might be implicated in major depressive disorder (MDD).
We evaluated 15 GSK3β single-nucleotide polymorphisms (SNPs) to test for associations with regional gray matter (GM) volume differences in patients with recurrent MDD. We then used the defined regions of interest based on significant associations to test for MDD × genotype interactions by including a matched control group without any psychiatric disorder, including MDD.
General linear model with nonstationary cluster-based inference.
Patients with recurrent MDD (n = 134) and age-, sex-, and ethnicity-matched healthy controls (n = 143).
Main Outcome Measures
Associations between GSK3β polymorphisms and regional GM volume differences.
Variation in GM volume was associated with GSK3β polymorphisms; the most significant associations were found for rs6438552, a putative functional intronic SNP that showed 3 significant GM clusters in the right and left superior temporal gyri and the right hippocampus (P < .001, P = .02, and P = .02, respectively, corrected for multiple comparisons across the whole brain). Similar results were obtained with rs12630592, an SNP in high linkage disequilibrium. A significant SNP × MDD status interaction was observed for the effect on GM volumes in the right hippocampus and superior temporal gyri (P < .001 and P = .01, corrected, respectively).
The GSK3β gene may have a role in determining regional GM volume differences of the right hippocampus and bilateral superior temporal gyri. The association between genotype and brain structure was specific to the patients with MDD, suggesting that GSK3β genotypes might interact with MDD status. We speculate that this is a consequence of regional neocortical, glial, or neuronal growth or survival. In considering core cognitive features of MDD, the association of GSK3β polymorphisms with structural variation in the temporal lobe and hippocampus is of particular interest in the context of other evidence for structural and functional abnormalities in the hippocampi of patients with MDD.
Indirect evidence has implicated the glycogen synthase kinase-3β gene (GSK3β; OMIM 605004) in major depressive disorder (MDD).1 Serotonin-related dysfunction provides a well-developed unifying hypothesis to explain the pathogenesis and therapeutic responses in MDD.2 Serotonin receptor–mediated signaling modulates GSK3β activity.3 Inhibition of GSK3β activity is a proposed mechanism of action of lithium in mood disorders.4,5 Furthermore, the antidepressants fluoxetine and imipramine increase inhibitory Ser phosphorylation of GSK3β in the mouse brain.3
Identifying genes associated with MDD will provide new insight into disease pathophysiology. The relatively recent introduction of array-based genotyping technologies allow association analyses at the whole-genome level, with no need for a priori hypotheses. However, the large number of tests involved, together with the anticipated small effects of common genetic variants, require the use of very large samples to identify genetic risk factors for disease.6,7 The clinical heterogeneity of MDD adds to the complexity of identifying genes associated with the disorder. This has prompted efforts to explore the use of alternative phenotypic markers for genetic association studies that are more closely related to the underlying neurobiology of the disease.8
While there is no clear consensus on how specific regional brain structural variations are associated with MDD, the concept that differences in brain structure are associated with MDD is supported by several studies.9,10 This, combined with evidence that several brain structural measures are highly heritable,11 suggests that neuroimaging is a useful endophenotype for genetic association studies of psychiatric disorders.7,12,13
In this study, we tested for associations between regional gray matter (GM) volume estimated from magnetic resonance imaging (MRI) and 15 GSK3β single-nucleotide polymorphisms (SNPs) using a nonstationary cluster-based morphometric analysis in a large sample of well-characterized patients with recurrent MDD and matched controls. We then tested for disease × genotype interactions within the masks derived from significant association clusters in the MDD group and age-, sex-, and ethnicity-matched healthy controls to identify structural variation distinguishing the MDD group.
The subjects used for the imaging study were a subset of a larger cohort of 1022 patients with recurrent MDD and 1000 healthy controls recruited for a genetic association study. A detailed description of the recruitment procedure has been previously described.14 Briefly, subjects were recruited and clinically assessed at the Max Planck Institute of Psychiatry, Munich, Germany, except for 2 patients, who were recruited at satellite clinical sites (Bezirkskrankenhaus, Augsburg, and Klinikum, Ingolstadt, Germany). The recruiting hospital obtained approval by the research ethical board to conduct the study, and all individuals gave written informed consent. Patients were diagnosed according to the DSM-IV or International Statistical Classification of Diseases and Related Health Problems (ICD-10), which was established after the administration of the semistructured interview Schedules for Clinical Assessment in Neuropsychiatry. This interview was administered by experienced research assistants who received training at World Health Organization Training and Research Centres. All patients included in the study had a diagnosis of recurrent MDD. Patients with bipolar disorder, mood-incongruent psychotic symptoms, a lifetime history of intravenous drug use, a diagnosis of drug dependency, depression secondary to alcohol, or substance abuse or depression as consequence of medical illnesses or use of medication were not included in the study. Each subject who participated in this study completed a questionnaire regarding their demographics, family, individual, and medical history, and ethnicity. All patients and controls referred to themselves as white with parents of Northern European origin. All control subjects filled in the questionnaires while supervised by a study nurse. The demographic and main clinical characteristics for all subjects included in this study are summarized in Table 1.
High-resolution T1-weighted MRIs were collected at the Max Planck Institute of Psychiatry, for an initial study population of 392 participants (193 patients with MDD and 199 healthy controls). Magnetic resonance images were screened for quality control and analysis at the Clinical Imaging Centre, GlaxoSmithKline. After quality control procedures for both imaging and genetics, 278 subjects were entered in the final voxel-based morphometry (VBM) and genetic analysis (eTablehttp://www.archgenpsychiatry.com). Details on subjects excluded for phenotypical-, imaging- and gene-related criteria are described in the eTable. In brief, subjects were excluded due to (1) detection of gross brain pathology (2 patients with MDD; 3 controls); (2) problems involving data transfer (12 patients with MDD; 5 controls); (3) poor image quality due to head coil instability, excessive motion artifact, or failure to acquire a full image (22 patients with MDD; 3 controls); (4) anatomical deviations (eg, enlarged ventricles) that prevented appropriate cortical segmentation or spatial normalization (7 patients with MDD; 3 controls), poor segmentation in the basal ganglia region or misclassification of dura mater as GM (5 patients with MDD; 2 controls); (5) genotyping exclusions (see below) (eTable).
All GSK3β SNP genotypes were extracted from genotyping obtained using the HumanHap550 BeadChip platform (Illumina Inc, San Diego, California). The whole-genome scan genotyping, which was available for the full sample of cases and controls collected for genetic studies, followed extensive quality control procedures, as described in detail previously.16 For the individuals available with imaging data, 23 subjects (2 patients with MDD; 21 controls) were excluded because DNA samples were of poor quality and were not sent for genotyping. Of the total DNA samples sent for genotyping, 30 subjects (11 patients with MDD; 19 controls) for SNP rs6438552 and 27 subjects (9 patients with MDD; 18 controls) for all remaining SNPs failed to produce genotyping data owing to low-concentration or poor-quality DNA (eTable). The whole-genome association analysis of the full sample of cases and controls produced a genomic control17 of λ = 1.002,16 which suggested the absence of major population structure and that our cases and controls were relatively homogenous in terms of genetic structure. Therefore, any large distortion in our results caused by residual substructure can be excluded with reasonable confidence.
Two-channel signal intensity data corresponding to the 2 alleles at each SNP were evaluated using the software Beadstudio 3.1 (Illumina Inc). The initial genotype calls were generated using the cluster file provide by Illumina. None of the GSK3β genotypes deviated from Hardy-Weinberg equilibrium for patients with MDD and healthy controls (all P > .13). The genotypes for the most significant SNP (rs6438552) association showed a cluster separation of 0.3. The raw intensity data for rs6438552 genotypes from the full sample used in the whole-genome scan analysis were visually inspected to further ensure accuracy (eFigure 1).
High-resolution T1-weighted MRIs were acquired on a 1.5-T General Electric scanner (Signa, later upgraded to Signa Excite; Waukesha, Wisconsin); inversion recovery prepared spoiled gradient echo recalled with a field of view of 22 × 22 cm2, a matrix of 256 × 256, 124 sagittal slices, and a resulting voxel size of (1.2 − 1.4) × 0.9 × 0.9 mm3 (depending on brain size) (time to repetition, 10.3 milliseconds; echo time, 3.4 milliseconds; flip angle, 20°).
Simultaneous segmentation and combined linear/nonlinear intersubject registration and normalization to the Montreal Neurological Institute (MNI) atlas was performed using statistical parametric mapping 5 (SPM5 [http://www.fil.ion.ucl.ac.uk/spm, version 573, last updated 25-07-2006]; Wellcome Trust Centre for Neuroimaging, London, England) procedures. Gray matter segmented images in atlas space were modulated to generate maps with volume per voxel of atlas space. Isometric 10-mm Gaussian smoothing was applied.
All MRI scans were subjected to full radiological reporting supervised by board-certified neuroradiologists to detect relevant brain pathology. Furthermore, image acquisition artifacts, including motion artifact and anatomical factors preventing automated morphometry, were assessed prior to analysis (eTable). Following this, GM segmented data were assessed for accuracy, especially in subcortical areas and for dura that was misclassified as GM. The atlas space GM data were checked to assure good cortical and ventricular alignment. After a basic model was fit with no genetic variables, subject outlier detection was performed with SPMd (http://www.sph.umich.edu/ni-stat/SPMd; Wellcome Trust Center for Neuroimaging). Data that showed uncorrectable VBM segmentation-related preprocessing errors in 2 subjects due to enlarged ventricles and severe cortical atrophy were excluded (eTable). For the final subjects included in all analyses, there were no significant differences in total GM volume between healthy controls (mean [SD], 0.61 [0.07] L) and patients with MDD (mean [SD], 0.61 [0.07] L).
A separate SPM analysis was performed for each SNP. A general linear model was performed at each voxel and included as a group effect and a group-specific SNP effect as well as nuisance effects of age, sex, total GM volume (used to discount global variation in GM affected by head size), medication, and timing with respect to scanner upgrade. The group effect had 3 levels for controls and for patients with MDD with and without comorbid anxiety. This subdivision of MDD was used because evidence for GM volume differences involving comorbid anxiety was found in a previous study (Becky Inkster, DPhil, et al, unpublished data, November 2008). However, when testing for interactions, we averaged the 2 MDD subgroups, as there was no interest in the anxiety effect specifically. The SNP effect was split by the 3 groups, fitting SNP × group interactions. Unlike the approach of most genetic studies that fit an additive effect of allele dose, in our study for SNPs with more than 10% of subjects with the rarest genotype (ie, minor allele frequency > √0.10 = 0.3162), we used a genotypic model parameterized with orthogonal polynomials, providing inference on an additive effect while providing a full 2-df18 fit. For example, if there were equal numbers of subjects with the 2 homozygote genotypes, the additive predictor would have values −1, 0, 1 for the 3 genotypes and the nonadditive predictor would have values −1/2, 1, −1/2. More generally, the 2 predictors begin coded as above (−10−1, −1/211/2), but the additive predictor is centered and the nonadditive predictor is centered and orthogonalized with respect to the additive predictor. For SNPs with a minor allele frequency less than 0.3162, a recessive model was used, merging the rare homozygous and heterozygous groups.
Standard cluster-based methods15,19 assume stationary noise (constant smoothness throughout the image), and VBM data has been observed to have variable smoothness.20,21 While most studies avoid using cluster-based inference owing to problems with VBM's variable smoothness, we used a novel approach of nonstationary cluster-size inference that allows for valid inference on clusters while accounting for heterogeneous smoothness (ie, as spatially extended effects were expected, cluster-based inference was used to obtain optimal sensitivity22). We therefore used nonstationary cluster-size inference23 with a cluster-defining threshold of α = .001. We report familywise error (FWE)–corrected cluster P values, which control for the chance of 1 or more false positives. P values corrected for searching over image space are denoted PFWE(S), and the minimum nonstationary cluster P value corrected for searching over imaging space was used as a summary F statistic, reflecting the overall strength of association. Anatomical locations of clusters were established using the MRIcro AAL (Anatomical Automatic Labeling) template (http://www.sph.sc.edu/comd/rorden/mricro.html).24 Cluster locations are defined in MNI standard space (as defined by the MNI-space image shipped with SPM).
Major depressive disorder × rs6438552 genotype interactions (where the group is either control or MDD) were conducted using a 1-dfF test for any direction of SNP association between groups. First, interaction testing was restricted to the region of interest24 masks defined by the F-test contrast of rs6438552 genotype effect on GM volume differences in patients with MDD only. Because cluster inference can have reduced sensitivity in small search regions, voxelwise inference (FWE-corrected) was used within the 3 region of interest (ROI) masks.
Each SNP was calculated in a separate model and Bonferroni correction was applied for testing multiple GSK3β SNPs. The Bonferroni correction is valid for any set of tests and, though it is known to be very conservative when there are many tests or severe dependence; with only 15 SNPs it should operate satisfactorily. The SPM-corrected cluster P values (PFWE(S)) were then corrected for multiple SNP testing and P values are denoted as PFWE(S,S). The SNPs tested in this study that were not independent of other SNPs (ie, in high or complete linkage disequilibrium) were still included for analysis because a replication would confirm that no experimental genotyping errors or analysis errors occurred (the extent of linkage disequilibrium between the SNPs is shown in eFigure 2A).
We tested 15 SNPs at the GSK3β locus for association with GM variation using VBM in our sample of patients with MDD. Nonstationary cluster-based morphometric analysis revealed significant GM variation associations with 2 SNPs, rs6438552 and rs12630592, that are highly correlated (r2 = 0.951, which is similar to the 0.950 observed for white individuals' reference data from HapMap data). The most significant GM volume associations were observed for rs6438552. After correcting for multiple comparisons across the whole brain, 3 statistically significant clusters of GM volume differences were identified in the right and left superior temporal gyri (STG) and the right hippocampus (PFWE(S) < .001, P =.02, and P = .02, respectively) (Table 2; Figure 1A). The A allele was associated with lower GM volume in all 3 clusters (Figure 1B). Relative decreases in the GM volume of 7% to 11% were found for AA compared with GG homozygotes in the cluster ROIs (Table 3). The SNP rs12630592 showed colocalized GM association clusters (Figure 2A) with similar load-dependent decreases in GM volume for the G allele, with an effect magnitude similar to rs6438552 (Figure 2B). The results for all SNP associations are summarized in eFigure 2B. Post hoc analyses were performed to examine rs6438552 genotype differences with respect to episode severity, age, or sex between genotype groups, but no significant results were revealed (P = .32, P = .99, and P = .34, respectively, data not shown). We also performed a post hoc analysis with all 15 SNPs included as covariates into 1 model and gained no significant result (eFigure 3).
After correcting for whole-brain search and for the number of SNPs tested, the cluster in the right STG remained significant for rs6438552 (PFWE(S,S) < .001) and a trend was found for rs12630592 (P = .06) (Table 2).
To evaluate whether these associations are specific to MDD or brain structural variation more generally, we tested for interaction between GM volumes within masks derived from the 3 significant clusters for rs6438552 SNP across patients with MDD and matched healthy controls. Significant interactions were revealed within the right hippocampus ROI mask (Table 4); the mean (SD) change in modulated GM (at voxel 30, −26, −12) was 2.5% (0.037%) lower in MDD AA homozygotes relative to healthy controls' AA homozygotes (Figure 3A). When searching within the right STG ROI mask, significant interactions were also revealed (Table 4); the mean (SD) change in modulated GM (at voxel 48, −16, 0) was 5.5% (0.038%) lower in MDD AA homozygotes relative to healthy control AA homozygotes (Figure 3B). No significant interactions were revealed when searching within the left STG cluster. The interaction effects in the right hippocampus and STG remained significant after correcting for searching across multiple masks (Table 4). There were no significant genotype effects when restricting the ROI analysis to controls only.
Here we have shown that variation in GM volume in the right hippocampus and bilateral STG is associated with a common intronic polymorphism (rs6438552) at the GSK3β locus in patients with MDD. Previous in vitro data has demonstrated that this intronic polymorphism regulates the selection of splice acceptor sites and thus alters GSK3β transcription.25 The polymorphism might therefore define functional differences in GSK3β expression. An additional similar association between genotype and local GM volume was also detected for rs12630592, an SNP in high linkage disequilibrium with rs64385522.
The regional specificity of our findings, particularly in the right hippocampus, is notable when considering meta-analytical evidence from MRI studies for a role of hippocampal integrity in depression.26 In addition, GSK3β is highly expressed in the hippocampus and temporal cortex in the human and mouse brain (Allen Mouse Brain Atlas: http://www.brain-map.org). Furthermore, STG abnormalities have been described in subjects with bipolar disorder who are not taking medication relative to healthy controls,27 and our group has reported higher STG GM volumes in a subset of the same group of patients with MDD with comorbid anxiety.
There is growing evidence to suggest that inhibition of GSK3β activity might play a role in the therapeutic effects of antidepressants and lithium in patients with refractory MDD.28 Lithium inhibits the enzyme encoded by GSK3β4 and has been shown to regulate GSK3β expression in human peripheral blood mononuclear cells.29 The activity of GSK3β is inhibited by the antidepressants fluoxetine and imipramine in mice.3 Pharmacogenetic support from a recent study showed that several common GSK3β polymorphisms were associated with selective serotonin reuptake inhibitor therapeutic effects in a large sample of patients with MDD.30 Furthermore, a postmortem study that examined GSK3β activity in the ventral prefrontal cortical tissue of patients with MDD compared with matched controls found increased GSK3β activity in patients with MDD.31 The potential role that GSK3β plays in bipolar disorder has also been examined extensively. Associations have been described between a functional GSK3β polymorphism, −50 T/C (rs334558) and age of onset,32,33 therapeutic response to lithium salts,34 response to lithium augmentation treatment,35 and therapeutic response to total sleep deprivation.33 In a combined sample of patients with bipolar disorder and patients with MDD, an association between the −50 T/C promoter SNP and delusional symptoms and personality traits related to delusions was observed.36 There are, however, some studies that have failed to replicate associations between GSK3β and age of onset36 as well as response to lithium.37 No evidence of association with lithium response was revealed for another GSK3β promoter polymorphism that was examined.38 A recent study reported that a copy number variation disrupts several exons of GSK3β, and could therefore have adverse effects on GSK3β expression, was more frequent in bipolar patients.18
Genotyping for the −50 T/C promoter SNP as well as other SNPs previously described in the literature (except rs6438552) were not available for our study. The functional promoter polymorphism (−50 T/C) has not been genotyped in Centre d’Etudes du Polymorphisme Humain (CEPH) families (HapMap), so it was not possible to establish the linkage disequilibrium between this SNP and the SNPs analyzed in our study.
The MDD-specific effect found in our study suggests that MDD status, disease-associated genetic risk factors, or both might be interacting with GSK3β in determining the regional GM differences observed in this study. It remains unclear whether GSK3β genotype-dependent differences in brain morphology develop throughout the course of MDD or if these differences exist premorbidly. Major depressive disorder status is often associated with an hypothalamic-pituitary-adrenal axis dysfunction, which produce consequential effects on growth and resilience of specific neuronal and glial populations.39 Furthermore, environmental factors such as life events that have been suggested to have an effect in increasing the risk for MDD might be, in part, responsible for the hippocampal morphological differences observed in patients with MDD.40 Alternatively, or in conjunction, MDD susceptibility genes may interact with GSK3β to determine the structural associations we have observed in our patients with MDD. Although speculative, the GSK3β associations found in this study may relate to previously described genetic associations with hippocampal volume changes in patients with MDD. For example, the association of the 5-HTTLPR polymorphism with hippocampal volume in patients with MDD41 could be related to the role of GSK3β activity in modulation of serotonergic signaling.3,42
While our results are novel, the VBM methodology used for the between-groups contrast is well established.21,43 Identification of significant associations between brain structural variation and single gene polymorphisms is also well precedented.44- 47 Some brain regions might be more strongly genetically determined than others. For example, there is higher heritability for middle and superior frontal, sensorimotor, paralimbic and, consistent with our results, temporal structures.48- 50 In principle, therefore, differences in the heritability of different cortical regions might limit the potential for this kind of imaging genetic association study to genes whose variation determines specific brain regions.
There are some additional limitations of our study. The association demonstrated here is robust statistically, but we also have shown considerable variation in the magnitude of the effect across the population of patients with MDD that we studied. This heterogeneity implies that large samples might be required to replicate our findings.6,7 In addition, as the size of the effect we have observed is likely to be overestimating the real contribution of GSK3β to brain morphological differences in the overall population of patients with MDD, significantly larger samples than the one used in our study might be required to replicate this finding.51 We previously reported no significant GM differences in the hippocampus or temporal cortex between MDD cases and controls (unpublished data, November 2008) when genetic factors were not included in the analysis, yet this result was based on a larger cohort than the subset for the GSKβ SNP-GM association analysis. Furthermore, it remains possible that the lack of an effect in the healthy control population, and thus the specificity of the association for MDD, reflects limitations of study power rather than lack of true association. Owing to potential disease heterogeneity and reversion to the mean, replication may demand a substantially larger sample size. Finally, VBM-based GM volume differences cannot be related specifically to neuroanatomical features, as differences can arise from relative changes either in cortical folding or thickness.52 Cluster detection, in which significance is determined by the spatial contiguity of suprathreshold voxels, is biased toward larger local regions of change. The study might not be adequately powered to detect additional smaller regions. Additional imaging techniques (eg, functional MRI) will provide further tests for the observed associations. Furthermore, these associations need to be examined in other relevant psychiatric diseases such as schizophrenia and bipolar disorder. Additionally, there is an opportunity to test additional candidate genes, but this is beyond the scope of the current investigation.
Our results suggest that in MDD, a proportion of bilateral STG and right hippocampal structural variation can be explained by genetic variation at the GSK3β locus in ways distinguishable from healthy controls. Based on these results, however, it is not possible to differentiate whether this GSK3β polymorphism contributes to the risk of developing MDD or occurs as a consequence of major depression. They are equally consistent with a genetic influence on a predisposing brain structural trait, a higher independent risk of developing brain changes in the course of the disease, or development of these changes in interaction with other genetic or environmental risk factors. In support of the growing evidence suggesting a role for GSK3β in MDD,53 our study is the first to suggest directly that GSK3β has a role in determining any aspect of brain structure or function related to MDD.
Correspondence: Paul M. Matthews, DPhil, FRCP, GlaxoSmithKline Clinical Imaging Centre, Hammersmith Hospital, Du Cane Road, London W12 0NN, England (firstname.lastname@example.org).
Submitted for Publication: April 29, 2008; final revision received November 28, 2008; accepted December 3, 2008.
Financial Disclosure: Drs Inkster, Nichols, and Saemann are employees of GlaxoSmithKline.
Funding/Support: This study was supported by GlaxoSmithKline.
Additional Contributions: We would like to acknowledge the patients and control subjects who have participated in this study. We would also like to thank the staff at the Max Planck Institute of Psychiatry, Munich, Germany, who contributed to the success of this study and to the numerous colleagues at GlaxoSmithKline, in particular, Anil Rao, PhD, Khanum Ridler, DPhil, Federica Tozzi, MD, and Emilio Merlo-Pich, MD, for their contributions to the overall conduct of the study.
This article was corrected online for typographical errors on 7/6/2009.