Molecular Psychiatry (2013) 18, 195–205 & 2013 Macmillan Publishers Limited All rights reserved 1359-4184/13 www.nature.com/mp ORIGINAL ARTICLE Genome-wide association study meta-analysis of European and Asian-ancestry samples identifies three novel loci associated with bipolar disorder DT Chen1, X Jiang1, N Akula1, YY Shugart1, JR Wendland1, CJM Steele1, L Kassem1, J-H Park2, N Chatterjee2, S Jamain3, A Cheng4, M Leboyer3, P Muglia5, TG Schulze1,6, S Cichon7,MMNo¨then7, M Rietschel8, BiGS9 and FJ McMahon1 1Human Genetics Branch, National Institute of Mental Health, Intramural Research Program, National Institutes of Health, US Department of Health and Human Services, Bethesda, MA, USA; 2Division of Cancer Epidemiology and Genetics, NCI, NIH, DHHS, Rockville, MA, USA; 3Inserm U955, Department of Psychiatry, Groupe Hospitalier Henri Mondor-Albert Chenevier, AP-HP, Universite´ Paris Est, Fondation FondaMental, Cre´teil, France; 4Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan; 5Department of Psychiatry, University of Toronto, Toronto, ON, Canada; 6Section on Psychiatric Genetics, Department of Psychiatry and Psychotherapy, University Medical Center, Georg-August-Universita¨t, Go¨ttingen, Germany; 7Institute of Neuroscience and Medicine, Juelich, Germany and Department of Genomics, Life and Brain Center, University of Bonn, Bonn, Germany and 8Department of Genetic Epidemiology in Psychiatry, Central Institute of Mental Health, University of Mannheim, Mannheim, Germany

Meta-analyses of bipolar disorder (BD) genome-wide association studies (GWAS) have identified several genome-wide significant signals in European-ancestry samples, but so far account for little of the inherited risk. We performed a meta-analysis of B750 000 high-quality genetic markers on a combined sample of B14 000 subjects of European and Asian-ancestry (phase I). The most significant findings were further tested in an extended sample of B17 700 cases and controls (phase II). The results suggest novel association findings near the TRANK1 (LBA1), LMAN2L and PTGFR. In phase I, the most significant single nucleotide polymorphism (SNP), rs9834970 near TRANK1, was significant at the P = 2.4 Â 10À11 level, with no heterogeneity. Supportive evidence for prior association findings near ANK3 and a on 3p21.1 was also observed. The phase II results were similar, although the heterogeneity test became significant for several SNPs. On the basis of these results and other established risk loci, we used the method developed by Park et al. to estimate the number, and the effect size distribution, of BD risk loci that could still be found by GWAS methods. We estimate that > 63 000 case–control samples would be needed to identify the B105 BD risk loci discoverable by GWAS, and that these will together explain < 6% of the inherited risk. These results support previous GWAS findings and identify three new candidate genes for BD. Further studies are needed to replicate these findings and may potentially lead to identification of functional variants. Sample size will remain a limiting factor in the discovery of common alleles associated with BD. Molecular Psychiatry (2013) 18, 195–205; doi:10.1038/mp.2011.157; published online 20 December 2011 Keywords: ANK3; bipolar disorder; LBA1; meta-analysis; TRANK1; 3p21

Introduction heritability estimated by previous twin studies.1–11 With a lifetime prevalence worldwide between 0.5 The genetic basis of bipolar disorder (BD) is still and 1.5%, BD is characterized clinically by often largely unknown despite robust evidence of high disabling fluctuations of mood and behavior, com- monly developing in late adolescence to early adult- hood. Although the pathogenesis of BD remains Correspondence: Dr DT Chen, Human Genetics Branch, National Institute of Mental Health, Intramural Research Program, National unclear, genome-wide association studies (GWAS) Institutes of Health, US Department of Health and Human have so far identified and replicated a few risk loci Services, 35 Convent Drive, Room 1A-208, Bethesda, MD 20892, (near the genes DGKH, ANK3 and CACNA1C),12–16 USA. along with a locus on chromosome 3p21.1 that harbors E-mail: [email protected] 17 9 anumberofgenes. Together, these loci account for The Bipolar Genome Study (BiGS) authorship list is shown in the little of the BD heritability, suggesting that additional Appendix. Received 31 May 2011; revised 7 October 2011; accepted 17 risk loci remain undiscovered. The total BD GWAS October 2011; published online 20 December 2011 sample size studied, so far, remains low compared Three novel loci associated with bipolar disorder DT Chen et al 196 with many other common traits studied, such as type 2 flagged as having different alleles than in HapMap diabetes, height, serum lipids, colorectal cancer and CEU or as monomorphic were reviewed, after which rheumatoid arthritis.18–23 Some of the missing herit- they were recoded for the reverse-strand (flipped) or ability may be explained by additional risk loci that dropped. SNPs flagged for allele frequencies markedly can only be identified in larger sample sizes.24 different from HapMap CEU were also reviewed. Psychiatric disorders such as BD pose statistical Palindromic SNPs whose allele frequencies were challenges when it comes to very large sample sizes. consistent with reversed coding were flipped. SNPs Phenotyping by direct diagnostic evaluation is expen- with unexpected allele frequencies were dropped. sive. Reliance on physician or hospital-assigned PLINK (version1.4) was used to flip and drop SNPs.34 diagnoses can save money, but introduces potential After all allele-coding, monomorphism and palin- biases, like changing diagnostic-criteria, which can drome issues were resolved, imputation was run be difficult to correct.25 Increasing sample size by again. SNPs in the result files were dropped if the combining data across studies, can be fruitful. Meta- minor allele frequency (MAF) in cases or controls was analysis is an efficient and largely unbiased way < 0.05, or if the error rate (in the .erate output file) was to increase effective sample size by systematically > 0.01. The imputed data were then formatted into combining association signals across studies. As most PLINK binaries for analysis. Supplementary Table 1 common genetic variation is ancient and widespread, provides detailed description regarding genotyping some risk alleles may be shared across continental and imputation for the Taiwan, Wellcome Trust Case– populations. It is possible to use meta-analysis to Control Consortium, STEP-BD,35 FondaMental Bipolar combine study samples of differing ancestry, as long and GlaxoSmithKline (GSK) samples. as appropriate ancestry-matched controls are used within each study.26–28 Meta-analysis In this study, we have sought to identify novel risk PLINK output (assoc) files were modified with alleles for BD by meta-analysis of world-wide BD columns for direction-of-association, sample size GWAS, comprising case–control samples of both and strand. For most samples, sample size equaled European and Asian ancestry. The combined sample the sum of cases and controls included in the final size of 17 656 is the largest so far in BD, to our analysis, after the quality-control steps were com- knowledge. The results suggest significant novel plete. For STEP-BD, sample size was set to equal the association signals near the genes TRANK1 (LBA1), number of cases plus controls that did not overlap LMAN2L and PTGFR, and provide supportive evi- with those in the NIMH-GAIN or TGEN. This was dence for the previously reported association signals done to avoid over-weighting the results from the near ANK3 and within the 3p21.1 locus. Largely con- NIMH-control sample, overlapping portions of sistent signals were observed in both the European which were included in both the NIMH-GAIN, TGEN ancestry and Asian-ancestry samples. Based on these and STEP-BD. Modified files were loaded into findings and discoveries to date, we also present a METAL (November 2010 version), then processed GWAS discovery trajectory for BD. using the GENOMICCONTROL option, which applies a genomic control36 correction in samples where the genomic inflation factor is > 1.0. METAL weights Materials and methods each sample based on the square root of the sample Study samples size. Care was taken to avoid mis-assigning alleles The samples used in the meta-analysis have been when combining results from different samples and described previously, and details are provided in platforms.37 Using the ‘STRANDLABEL’ and ‘USES- Table 1 and Supplementary Table 1.12,15,17,29–32 For TRAND ON’ commands in METAL, these SNPs were phase I, we obtained five European and one Asian- recoded to ensure consistent allele coding across the ancestry sample, totaling B14 000 cases and controls. samples analyzed. Because the German sample was The most significant hits (P <4Â 10À3) were tested in genotyped on the Illumina platform (San Diego, CA, an extended sample that included phase I plus two USA) that contains no palindromic SNPs, we used independent European ancestry samples (B3800 that sample as the gold standard for our study. cases/controls). We refer to this as the phase II sample. Results were combined under a fixed-effects model, using METAL. For initial discovery purposes, the Imputation fixed-effects model is more powerful than the tradi- Genotype data from the NIMH-GAIN, German and tional random-effects model, and Pereira et al.24 TGEN samples were used to impute data on 2.1 suggest the fixed-effects model is preferable, espe- million HapMap phase2 single-nucleotide polymorph- cially when the cumulative sample size is in the range ism (SNPs), by use of the program Markov Chain of 2000–20 000. Haplotyping (MACH version1.0; http://www.sph. After selected results were confirmed, heterogene- umich.edu/csg/abecasis/MACH/download/).33 MACH ity statistics were calculated, using Comprehensive uses Markov chain haplotyping to resolve haplotypes, Meta-analysis version 2.0. When heterogeneity tests and thereby missing genotypes, from observed geno- are significant, assumptions of the fixed-effects model types in unrelated individuals. We used the ‘greedy’ are violated, and the results can be anti-conserva- algorithm, as recommended by the authors. SNPs tive.24 To address this, we also analyzed selected

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 197 Table 1 Descriptive statistics for the samples analyzed

Sample Cases Case diagnosis Controls Platform Total subjects

Phase I samples NIMH bipolar 1001 Bipolar I, 1033 Affymetrix 6.0 2034 schizoaffective bipolar German 645 Bipolar I 1310 Illumina 1955 HumanHap550 TGEN 1201 Bipolar I, 401 Affymetrix 6.0 1602 schizoaffective bipolar WTCCC 1856 Bipolar I, bipolar II, 2945 Affymetrix 500K 4801 schizoaffective bipolar STEP-BD 955 Bipolar I, bipolar II, 1498 (466) Affymetrix 500K 2453 (1421) schizoaffective bipolar Taiwan 1000 Bipolar I 1000 Illumina 2000 HumanHap550 Phase I sample 6658 8187 (7155) 14 845 (13 813) subtotal Phase II samples FondaMental 484 Bipolar I, bipolar II, 1823 Illumina 2307 bipolar bipolar NOS HumanHap550, Illumina HumanHap300, Illumina human 610-Quad, Illumina Infinium II GSK 631 Bipolar I, bipolar II 905 Illumina 1536 HumanHap550 Sample total 7773 10 915 18 688 (17 656) (9883)

Phase I samples include: NIMH Bipolar, NIMH bipolar disorder sample; German, German bipolar disorder sample; TGEN, ‘Wave 5’ collection of the Bipolar Consortium; WTCCC, Wellcome-Trust Case Control Consortium bipolar disorder sample; STEP-BD, STEP-BD bipolar disorder sample; Taiwan, Taiwan Bipolar Consortium. Phase II samples include: FondaMental Bipolar, French national network for mental health (Fondation FondaMental) in three university-affiliated psychiatry departments (Paris-Cre´teil, Bordeaux and Nancy); GSK, the Institute of Psychiatry in London, UK, the Centre for Addiction and Mental Health in Toronto, Canada and the University of Dundee, UK. Counts refer to subjects who passed all quality- control filters (see Materials and methods). Note: Value in parenthesis represent non-overlapping (with NIMH Bipolar) STEP- BD controls, as well total subjects using the non-overlapping STEP-BD controls.70 Presented GSK cases are non-overlapping with WTCCC. Overlapping GSK cases were excluded from this analysis.

SNPs under a random-effects model. As the tradi- Power analysis tional random-effects method assumes a markedly Power Analysis was done with genetic power calcu- conservative null-hypothesis model,38 we used a lator.45 We assumed a trait prevalence of 2%, minor novel method, implemented in Metasoft (vers3.1),39 allele frequency of 0.2, alpha of 5 Â 10À8 and a marker- designed to increase power when there is true allele D0 value of 0.9. The phase I sample had > 80% heterogeneity. power to detect an allele that confers a genotype- We used a threshold of genome-wide significance relative risk of 1.25 under a log-additive model. of P =5Â 10À8 derived from a published, genome- wide simulation of common variants in samples Drug treatment of cell cultures and quantitative of European ancestry.40–42 Marginally significant real-time PCR (qPCR) (P <1Â 10À6) SNPs are also reported. Random-effects HeLa, SH-SY5Y and HEK293 cells were grown in P-values were used when the Q-test of heterogeneity dulbecco’s modified Eagles’s medium with 4.5 gm l–1 d- was significant at P < 0.1.43 Although we had full glucose supplemented with 2 mML-glutamine and 10% access to complete genome-wide results for all phase I fetal calf serum. Lithium, valproic acid (VPA), dex- samples, we had access to the GSK and FondaMental amethasone and triamcinolone were purchased from results for only the most significant SNPs identified Sigma-Aldrich (St Louis, MO, USA). Drug treatments in phase I. As joint analysis has been shown to be were conducted in HeLa, SH SY5Y and HEK293 more powerful than replication analysis in situations cells.46–49 Total RNA was extracted using RNeasy plus like this,44 the most significant markers from phase I Mini-kits (QIAGEN, Valencia, CA, USA). First-strand (P <4Â 10À3) were combined with the additional cDNA was synthesized with First Strand Superscript III phase II samples for the joint analysis. kits (Invitrogen, Carlsbad, CA, USA). mRNA levels of

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 198 selected genes were determined by using Roche Light- top hits were examined for cis-effects in SNPExpress, Cycler 480 (F.Hoffmann-La Roche Ltd, Basel, Switzer- and in Gibbs et al. Bonferroni correction was applied land) and the Roche Universal Probe Library System. for the total number of top hits tested. To help in clarifying the genetic role of TRANK1, we examined expression in cells after drug treat- Results ment.47,48 SH-SY5Y cells were treated with lithium carbonate (1 or 2 mM), VPA (0.5 or 5 mM), dexametha- Overall, 748 555 SNP markers were consistently sone (200 nM or 2 mM) or triamcinolone (3,10, or 100 mM) scored across the six phase I samples. The genome- for 24 h. VPA treatment was also given at 0.5 mM or wide mean Z-score was < 0.004, approximating the 5mM for 75 h. Total RNA was isolated and quantitative theoretical mean of zero. This indicates an unbiased real-time PCR (qPCR) was performed using LightCycler experiment. 480 (Roche). The DDCT method was used to quantify Analysis of the combined European and Asian relative mRNA levels. Any differences were tested ancestry phase I sample detected genome-wide signi- with the Student’s t-test. Data represent mean±s.e.m. ficant evidence of association between BD and three from three independent experiments. SNPs located near two different genes (Figures 1 and 2). At rs9834970, the C-allele was consistently Analysis of eQTL from human postmortem brain more common in cases than controls (P = 2.41 Â 10À11 sample data sets Table 2a; Supplementary Tables 2, and 3A). Several Postmortem brain tissues from two collections,50–52 nearby SNPs in linkage disequilibrium with rs9834970 totaling 243 samples, were used to look at the were significant at the P <4Â 10À3 level (Figure 2). potential role of the top-hit SNPs in nearby gene These SNPs are clustered within and around the gene regulation. Gibbs et al.50 and Heinzen et al.51 describe TRANK1 (LBA1) on chromosome 3p22.2. these sample –data sets in detail. Gibbs et al. provide Two additional SNPs, rs2271893 and rs6746896, both the uncorrected and empirical P-value. For also returned genome-wide significant evidence of Heinzen et al. in SNPExpress, significance threshold association with BD (P <4Â 10À8). Several nearby was calculated on the basis of the total number of SNPs in linkage disequilibrium with rs2271893 and association tests conducted within the analyses. The rs6746896 were significant at the P <4Â 10À3 level.

ANCESTRY: European and Asian ANCESTRY: European SAMPLE TYPE: PHASE I SAMPLE TYPE: PHASE I (13,813 cases and controls) (11,813 cases and controls) Significance SNP List/Total Number Significance SNP List/Total Number Level Level P< 5x10–8 LMAN2L - rs2271893, rs67468962 P< 5x10–8 LMAN2L - rs2271893, rs67468962 TRANK1 - rs9834970 TRANK1 - rs9834970

P< 1x10–6 PTGFR, C1orf87/NFIA, SCP2, IRF4, P< 1x10–6 PTGFR, C1orf87/NFIA, SPRED2, CADM2, 3p21 Region, CYP7A1 CADM2, 3p21 Region

P< 4x10–3 7,417 Total SNPs P< 4x10–3 5,828 Total SNPs

Meta-analysis combining initial Phase I samples with Phase II samples (FondaMental with 484 cases and 1823 controls; GSK with 631 cases and 905 controls)

ANCESTRY: European and Asian ANCESTRY: European SAMPLE TYPE: PHASE I AND PHASE II SAMPLE TYPE: PHASE I AND PHASE II (17,656 cases and controls) (15,656 cases and controls) Significance SNP List Significance SNP List Level Level P< 5x10–8 PTGFR - rs4650608 P< 5x10–8 PTGFR - rs4650608 LMAN2L - rs2271893 LMAN2L - rs2271893, rs67468962 3p21 Region - rs7618915 3p21 Region - rs7618915 TRANK1 - rs9834970 TRANK1 - rs9834970 ANK3 - rs4948418 ANK3 - rs4948418

Figure 1 Meta-analysis flow-diagram. Described on the left is the world-wide discovery collection consisting of European plus Asian-ancestry meta-analysis, both at the phase I sample analysis level as well at the phase I sample and phase II sample joint analysis level. On the right of the diagram, a similar European-ancestry only meta-analysis, is likewise described. At both the level of phase I analysis and phase I sample and phase II sample joint analysis, genome-wide significant (P <5Â 10À8) SNPs are described as well as those which were found to be marginally significant (P <1Â 10À6; for detailed list, lease refer to Table 2 and Supplementary Tables 2 and 3). For easy reference, names of nearest genes are given.

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 199 a European plus Asian-ancestry Phase I sample 11 C 10

9

8 B

7

6

5

4 –log10 (P-value)

3

2

1

0 Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20 Chr21 Chr22

bcrs2271893 (CEU) rs9834970 (CEU)

80 Recombination rate (cM/Mb) 12 80

rs2271893 rs9834970 Recombination rate (cM/Mb) 8 P=1.08e-08 0.8 P=2.41e-11 0.8 0.5 60 9 0.5 60 6 2 2 r 40 6 r 40 4

2 20 3 20 observed (–logP) observed (–logP) 0 0 0 0 KIAA1310 LMAN2L CNNM4 CNNM3 DCLK3 LBA1 EPM2AIP1 FER1L5 MLH1

96600 96700 96800 36800 36900 37000 position (hg18) (kb) position (hg18) (kb)

Linkage Disequilibrium Linkage Disequilibrium for the CEPH (CEU) for the CEPH (CEU) from phased genotypes from phased genotypes

LD for the Han Chinese + LD for the Han Chinese + Japanese from Tokyo Japanese from Tokyo (JPT+CHB) from (JPT+CHB) from phased genotypes phased genotypes

Figure 2 (a) Manhattan plot of the European plus Asian-ancestry phase I sample meta-analysis results, generated by Haploview 4.2. Physical position is shown along the X-axis with each chromosome shown in distinct color; Àlog (meta- P-value) is shown along the Y-axis. The red guideline indicates the threshold of genome-wide significance (5 Â 10À8). The blue line indicates suggestive (P <1Â 10À6) results. (b) Detail of the associated region for rs2271893, generated by SNAP 2.2. Physical position and gene annotations (1000 Genome Pilot I) are shown along the X-axis, Àlog (meta-P value) is shown on the left Y-axis, recombination rate (CEU) on the right Y-axis; Below, linkage disequilibrium (r2) as estimated from HapMap 3 phased genotypes, generated by UCSC Genome Browser. Darker red indicates higher values. Recombination rates (CHB/JPT) are not significantly different. (c) Similarly, detail of the associated region for rs9834970, generated by SNAP 2.2. Physical position and gene annotations (1000 Genome Pilot I) are shown along the X-axis, Àlog (meta-P value) is shown on the left Y-axis, recombination rate (CEU) on the right Y-axis; Below, linkage disequilibrium (r2) as estimated from HapMap 3-phased genotypes, generated by UCSC Genome Browser. Darker red indicates higher values. Again, recombination rates (CHB/JPT) are not significantly different.

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 200 These SNPs are located on chromosome 2q11.2 in the corticosteroids dexamethasone and triamcinolone, vicinity of LMAN2L (Figure 2). sometimes known to provoke manic episodes in BD An additional 36 SNPs were significant at the patients.54 We tested TRANK1 mRNA expression in P <10À6 level. These include SNPs near ANK3 and the three separate cell lines. As shown in Figure 3 and 3p21.1 locus, reported in previous studies.12,13,17 Supplementary Figure 2, VPA markedly increased These and the three genome-wide significant SNPs TRANK1 mRNA expression in a dose- and time- were all taken forward into the phase II analysis. dependent manner, increasing expression from very After addition of the two additional samples in low baseline levels. Lithium, dexamethasone, and the phase II meta-analysis, the overall picture triamcinolone had no measurable effect on the low was similar. Genome-wide significant association levels of TRANK1 expression in the cell lines we signals remained at rs9834970 (random effects tested. P = 1.48 Â 10À12) and rs2271893 (random effects We also queried this data set for previously P = 2.2 Â 10À10), although the heterogeneity test published genome-wide significant BD markers revealed significant results (Table 2b). At these two (Supplementary Table 4).12,13,17,55–57 All published, SNPs, the direction of the association signal was BD-associated markers were available in our meta- consistent in the FondaMental sample, but reversed analysis except rs10994336. rs2251219 was signi- (non-significantly) in the GSK sample (Supplemen- ficant at P < 1.84 Â 10À7 level. None of the other SNPs tary Tables 2, and 3B). Additional genome-wide were significant beyond the P <3Â 10À4. significant signals were detected at rs4650608 near PTGFR; at rs4948418, near ANK3; and at rs7618915, Discussion within the 3p21 locus. Analysis of only the European-ancestry samples We describe, in a worldwide collection, genome-wide gave similar results. SNP rs9834970 near TRANK1 significant evidence in support of three novel genetic remained genome-wide significant (P = 5.65 Â 10À10), markers associated with BD. The results suggest novel along with rs2271893 and rs6746896 near LMAN2L association findings near the genes TRANK1 (LBA1), (P = 1.21 Â 10À9, 1.66 Â 10À9, respectively). Several LMAN2L and PTGFR. We show that VPA markedly regions contained SNPs with marginally significant increased TRANK1 mRNA expression in vitro in a associations (5 Â 10À8 < P <1Â 10À6), including a set of dose- and time-dependent manner, supporting a SNPs in linkage disequilibrium at 3p21.1 (Figure 1; functional role for TRANK1 in BD treatment. We also Table 2c; Supplementary Table 3C). The phase II find evidence consistent with previous studies, analysis similarly identified the same genome-wide supporting association between BD and SNPs at a significant SNPs near TRANK1, LMAN2L and PTGFR chromosome 3p21 locus and near ANK3. These (Figure 1; Table 2d; Supplementary Table 3D). results are also broadly consistent with those based Heterogeneity analysis was significant (P < 0.1) for on other large samples (Cichon et al.;55 Psychiatric several SNPs (Table 2; Supplementary Figure 1). The GWAS Consortium,58), although individual markers Asian-ancestry sample did not significantly contri- in each study vary in significance between the bute to heterogeneity as measured by I2. We observed P <10À5 and P <10À9 levels. As discussed below, no significant heterogeneity for rs9834970 within much larger sample sizes may be needed before the phase I samples (Q = 3.95, P = 0.267; I2 = 23.98; consistent genome-wide significance is attained for Table 2). However, the heterogeneity test became signi- any one marker. ficant in the phase II analysis (Q = 10.95, P =0.052; How many more markers like those already found I2 = 54.34; Table 2), owing to the GSK sample. might eventually be identified by GWAS methods? We explored the possible role of the top-hit SNPs in We used the method by Park et al.59 to estimate the regulating expression of nearby genes (Supplemen- number of underlying susceptibility loci that are tary Table 5).50–52 SNP rs9834970 was not available in likely to reside within the range of effect sizes either of the brain-expression data sets we queried. observed in the current and in previous GWAS.59 The SNPs rs2251219, rs4650608 and rs6746896 were We selected the five novel SNPs that showed found to show significance after Bonferroni correction association with BD at P <5Â 10À7 in the combined in one data set.51 None of the top hits nor their proxies phase I European and Asian ancestry samples. A bias- with r2 > 0.4 were represented in Gibbs et al.50 correction method, described by Ghosh et al.,60 was Interestingly, rs4650608 was not associated with applied to the corresponding odds ratio estimates to expression of the closest gene, PTGFR, but rather address the ‘Winner’s Curse’.60 We also included five with IFI44L, an interferon-induced with independent, previously-identified SNPs (or proxies, possible increased expression in the hypothalamus.53 r2 > 0.8) with association P-values < 4 Â 10À3 in phase I. SNP rs9834970 is located about 12 kb distal to the As the previously identified SNPs showed genome- 30 UTR of TRANK1 (hg18), recently known as LBA1. wide significant associations in other studies, odds To explore a possible functional role of TRANK1 in ratio estimates for these SNPs were taken directly BD, we assessed TRANK1 expression after exposing from the phase I results without correction, under the cells to established treatments for BD. TRANK1 assumption that phase I serves as a replication expression was measured in vitro after treatment with sample. Once power for detection was calculated the mood stabilizers lithium and VPA, as well as the based on the design of the current study, the number

Molecular Psychiatry Table 2 Meta-analysis results by ancestrya

SNP Gene symbol Allele n Direction Fixed Random I2 Q Q test effects P-value effects P-value test P-value

(a) European and Asian-ancestry phase I samples Direction (W/SC/ST/GGT) rs4650608 PTGFR T 13 813 þþþþ 1.20EÀ07 9.25E–06 52.26 6.28 0.099 rs2271893 LMAN2L G 13 813 ÀÀÀÀ 1.08EÀ08 5.20E–10 43.83 5.34 0.148 rs6746896 LMAN2L A 13 813 þÀþ þ 3.57E–08 4.19E–07 52.07 6.26 0.100 rs2875907 CADM2 A 13 813 þþþþ 7.44E–08 5.84E–08 38.50 4.88 0.181 rs17023019 CADM2 A 13 813 þþþþ 8.37E–08 3.63E–07 63.83 8.29 0.040 rs2251219 3p21 region T 13 813 þþþþ 6.35E–07 5.63E–07 0 2.82 0.421 rs7618915 3p21 region G 12 392 ÀÀ?À 8.92E–07 2.04E–07 0 0.29 0.866 rs9834970 TRANK1 C 13 813 ÀÀÀÀ 2.41E–11 3.03E–13 23.98 3.95 0.267

(b) European and Asian-ancestry phase I & phase II samples Direction (W/SC/ST/F/G/GGT) rs4650608 PTGFR T 17 656 þþþþþþ 8.35E–09 1.06E–06 39.04 8.20 0.145 rs2271893 LMAN2L G 17 656 ÀÀÀÀ þ À 5.21E–09 2.20E–10 52.31 10.49 0.063 rs6746896 LMAN2L A 17 656 þÀþ þÀþ 1.59E–08 1.78E–07 57.92 11.88 0.036 rs2251219 3p21 region T 17 656 þþþÀþþ 1.84E–07 1.63E–07 32.90 7.45 0.189 rs7618915 3p21 region G 16 235 ÀÀ?ÀÀÀ 1.64E–09 2.88E–10 0 1.95 0.744 rs9834970 TRANK1 C 17 656 ÀÀÀÀ þ À 4.71E–10 1.48E–12 54.34 10.95 0.052 rs4948418 ANK3 T 17 656 þþþþþþ 8.93E–09 3.71E–10 80.77 26.01 8.90E–05 rs10848642 CACNA1C G 17 656 ÀÀÀÀ þ À 2.80E–07 1.61E–05 56.12 11.39 0.044 rs6079468 MACROD2, C20orf133 C 17 656 ÀÀÀÀÀÀ 4.50E–07 2.11E–05 33.08 7.47 0.188

(c) European-ancestry phase I samples disorder bipolar with Chen associated DT loci novel Three Direction (W/ST/GGT) rs4650608 PTGFR T 11 813 þþþ 2.10E–07 2.29E–05 61.92 5.25 0.072 rs2271893 LMAN2L G 11 813 ÀÀÀ 1.21E–09 1.73E–10 16.16 2.39 0.303

rs6746896 LMAN2L A 11 813 þþþ 1.66E–09 2.62E–08 0 0.55 0.761 al et rs17023019 CADM2 A 11 813 þþþ 2.97E–07 3.30E–06 68.15 6.28 0.043 rs2251219 3p21 region T 11 813 þþþ 2.81E–07 2.36E–07 0 0.69 0.708 rs9834970 TRANK1 C 11 813 ÀÀÀ 5.65E–10 8.03E–12 48.59 3.89 0.143

(d) European-ancestry phase I and phase II samples Direction (W/ST/F/G/GGT) rs4650608 PTGFR T 15 656 þþþþþ 1.56E–08 2.71E–06 44.18 7.17 0.127 rs2271893 LMAN2L G 15 656 ÀÀÀ þ À 9.81E–10 1.34E–10 51.27 8.21 0.084 rs6746896 LMAN2L A 15 656 þþþÀþ 1.35E–09 3.23E–08 43.88 7.13 0.129 rs2251219 3p21 region T 15 656 þþÀþþ 1.03E–07 8.39E–08 29.50 5.67 0.225 rs7618915 3p21 region G 14 235 À?ÀÀÀ 1.79E–08 8.19E–09 0 1.88 0.598 rs9834970 TRANK1 C 15 656 ÀÀÀ þ À 1.03E–08 3.08E–11 63.40 10.93 0.027 rs4948418 ANK3 T 15 656 þþþþþ 2.45E–08 2.99E–10 82.87 23.35 1.08E–04 rs17138171 LOC646089, ODZ4 C 15 656 ÀÀÀÀÀ 6.36E–07 1.28E–07 0 1.89 0.755 rs6079468 MACROD2, C20orf133 C 15 656 ÀÀÀÀÀ 7.52E–07 0.00014 0 3.95 0.413

Abbreviations: Allele, the allele predominantly more common in cases than controls; F, FondaMental; G, GSK; Gene Symbol, nearest gene to the SNP; GGT, Gain/ German/TGEN; SC, Taiwan; ST, STEP-BD; W, WTCCC. a À6

oeua Psychiatry Molecular Meta-analysis main results featuring genome-wide significant SNPs and other unique top hits (Supplementary Table 3 lists the complete top hit results for P <10 from METAL). SNPs listed in chromosomal order. For easy reference, names of nearest genes are given. Direction is with respect to that allele ( þ , same allele; À, other allele; ?, missing), and is shown for each sample, from left to right. I2, Q test, and Q test P-value refer to tests of heterogeneity (see Materials and methods). Meta- analysis P-values < 5 Â 10À8 are shown in bold (fixed effects when Q-test P-value X0.1, random effects otherwise). 201 Three novel loci associated with bipolar disorder DT Chen et al 202 of underlying loci—and the trajectory for their The BD discovery trajectory has several interesting discovery as sample sizes grow—was estimated by properties (Figure 4): (1) Assuming equal numbers of an inverse-power weighted approach.59 case–controls, we estimated that a total sample size of B16 000 is needed to discover B10 independent- susceptibility loci at the P <5Â 10À7 level (Table 2b; Supplementary Tables 3B and 6); (2) We observed that B B 14 10 susceptibility loci account for 1% of total ** genetic variance; (3) About 19 000 additional indivi- 12 24h treatment duals may be needed to discover markers explaining the next 2% of genetic variance; (4) The identification 75h treatment of SNPs that account for the total B5.5% of genetic 10 variance would require at least 63 000 total indivi- duals; (5) The sample size required to discover half 8 the total number of susceptibility loci—a value we designate the ND50—is about 35 000–45 000. We 6 would like to point out that our confidence in the discovery trajectory is limited by the fact that only 10 4 ** Fold change of TRANK1 change of Fold mRNA in SH SY5Y cells observed BD susceptibility loci could be employed in * estimation of the model; the new discovery trajectory 2 will become more reliable as additional susceptibility loci are identified.59 0 The total sample size used in this meta-analysis, Control VPA 0.5mM VPA 5mM although large, still limited the power to detect novel risk loci. Previous studies of meta-analysis 4 ** for discovery purposes in common traits have approximated 66 000 for case–control studies of low- prevalence/high-heritability diseases, compared with 3 a 2.4-fold-increase in samples required for high- prevalence/low-heritability diseases.12,61–63 For analy- sis of the population’s quantitative traits, sample size 2 *

6

mRNA in HeLa cells 1 Fold change of TRANK1 change of Fold 5 63K

0 ControlVPA 0.5mM VPA 5mM 4 45K

5 ** 3 35K

4 2 26K Sample size required Sample size 3 1 16K * 2 0 Cumulative expected variance explained (%) explained variance expected Cumulative 0

mRNA in HEK293 cells 0 20 40 60 80 100 Fold change of TRANK1 change of Fold 1 Cumulative expected number of loci

0 Figure 4 Estimated novel locus discovery (ND) trajectory ControlVPA 0.5mM VPA 5mM for BD with additional individuals (AI) collected. The filled marker indicates present level of observed data. Cumulative Figure 3 Dose-and time-dependent changes in TRANK1 expected loci estimate is depicted on the X-axis. The expression after valproic acid (VPA) treatment in SHSY5Y left Y-axis describes the percentage of genetic variance (a), HeLa (b) and HEK293 (c) cells. Significance was deter- accounted for by these cumulative expected loci. The right mined by using Student’s t test. Data are mean±s.e.m. from Y-axis shows estimated sample size required for the three independent experiments. *P < 0.05; **P < 0.01 com- expected number of loci and variance explained. (1) With pared with control. As shown, VPA increased TRANK1 B16 000 total individuals (TI), assuming equal case– mRNA expression in a dose- and time-dependent manner in controls, B10ND is observed; (2) > 63 000 TI is estimated the three cell lines. for discovery of the cumulative expected number of loci.

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 203 would need to exceed 120 000 to achieve equivalent was significant. Further testing in large samples will discovery power. Our own estimates describe similar be needed before we can claim replication. increases in discovery of novel loci as sample sizes The Wellcome Trust Case–Control Consortium pre- grow (Figure 4). viously suggested association with BD for rs9834970 On the basis of the cumulative GWAS findings to (P = 4.50 Â 10À7). It is interesting to note that associa- date, we estimate that a B2-fold-increase in total tion between BD and rs984790 was recently reported sample size will be needed to reach the ND50 for at experiment-wide significance in an additional, BD—a challenging task, but not impossible. We independent sample collected in Brazil.66 We did estimate a minimum of B105 common BD suscept- not include this sample in our main meta-analysis as ibility loci could be found by GWAS. With the genome-wide data were not available. The effect size identification of uncommon and rare variants, for was quite large in this small family sample (odds example, by large-scale sequencing, the total genetic ratio = 2.64), and the direction of association was variance explained should increase, along with the consistent with our findings. When the Brazilian data number of risk loci.59 were combined with the European and Asian-ances- Findings in the combined European and Asian- try samples, the overall significance increased further, ancestry samples largely agreed with those in the along with the evidence for heterogeneity of effect European-only analysis. Such concordant findings, sizes (fixed-effects P = 1.54 Â 10À14, random-effects spanning continental populations, may suggest P = 1.61 Â 10À14, Q = 16.47, P = 0.011, I2 = 63.58; Sup- shared risk alleles.64 In 1904, Emil Kraepelin con- plementary Table 1). This illustrates an interesting ducted a comparative psychiatric study in the asylum property of the heterogeneity tests commonly of Buitenzorg, in Java, most of whose indigenous employed in meta-analysis: Heterogeneity can people were then relatively isolated from European increase when a sample shows a much larger effect influences. Kraepelin65 wrote that he saw ‘yno size than other samples in the analysis. compelling grounds for assuming that the natives of Taken together, these data support TRANK1 as a Java suffer from new and unknown forms of insanity.’ novel risk locus for BD. TRANK1, formerly named, In 2010, Pritchard et al.,64 using HapMap data to KIAA0342 and LBA1, stands for tetratricopeptide estimate the influence of natural selection on genetic repeat and ankyrin repeat containing 1. It is a roughly adaptation, observed few fixed, evolutionarily- 34 kb (14 total exon) gene on the reverse strand of selected alleles, instead finding steady, gradual chromosome 3p22.2 that encodes a P-loop containing variation in allele frequencies across continents. This nucleoside triphosphate hydrolase, associated with gradual divergence in allele frequencies between DNA/ATP binding or DNA repair, with significant people of West-Eurasian and East-Asian origins may expression in the brain.53 The dose and time-depen- have occurred as recently as 60 000 YBP. Our results dent increase in TRANK1 mRNA expression that we are broadly consistent with these ideas. observed in response to VPA treatment in three We did detect significant heterogeneity, although separate cell lines provides an independent line of this appeared to be driven in this study by the GSK supportive evidence, suggesting that the causal sample. Power is low with such small samples, and it variant(s) in TRANK1 lead to a loss of function. is difficult to draw strong conclusions concerning Future expression studies of TRANK1 as well as heterogeneity. It may be important that the GSK sequencing may be fruitful future directions. sample is largely non-familial, whereas most of the Genome-wide significant signals were also obser- other samples consist of probands collected with ved for SNPs near LMAN2L, PTGFR, the 3p21 locus family-based ascertainment. Further studies are and ANK3, but with significant evidence of hetero- needed to understand all the sources of heterogeneity geneity in the final analysis. LMAN2L, also known that may contribute here. as lectin, mannose-binding 2-like, located on chro- It has been typical in the meta-analysis literature to mosome 2, is hypothesized to be involved in the restrict fixed-effects models to situations where the regulation of export from the endoplasmic reticulum heterogeneity test is not significant. However, Pereira of a subset of glycoproteins, and may also serve as a et al.24 have shown that random-effects models are regulator of ERGIC-53.53,67 PTGFR on chromosome much less powerful when the total sample size is 1p31.1 encodes the prostaglandin F receptor, a below 20 000. They conclude that ‘y fixed effects member for the G-protein-coupled receptor family.67,68 may be preferable for the purposes of initial dis- Both LMAN2L and PTGFR are highly expressed in the covery, if the aim is simply to screen and identify as brain.53,67 Further replication and expression studies many of the true variants as possible,’ consistent with will be needed to validate these findings. the objective of our study. They further note that These results support a few of the previous GWAS ‘ywith fixed-effects, the rate of false positives findings and add three new candidate genes to a increases substantially as more data accumulate’ growing list of BD risk loci. It will be necessary to and recommend that ‘associations that pass desired increase the set of robustly identified risk loci further significance thresholds with fixed-effects calcula- if we are to succeed in using these findings to tions, but not with random-effects calculations, may triangulate the biochemical pathways involved in require further replication.’ For this reason, we used a the etiology of BD. Sample size will remain a limiting random-effects test whenever the heterogeneity test factor in the discovery of common alleles associated

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 204 with BD by GWAS. Methods that interrogate rarer detects a risk locus for major mood disorders on chromosome alleles, such as next-generation sequencing, could 3p21.1. Nat Genet 2010; 42: 128–131. help fill the remaining heritability gap, as suggested 18 Zeggini E, Scott LJ, Saxena R, Voight BF, Marchini JL, Hu T et al. 69 Meta-analysis of genome-wide association data and large-scale by recent studies. Clearly, there are many BD risk replication identifies additional susceptibility loci for type 2 loci yet to discover. As they accumulate, they will diabetes. Nat Genet 2008; 40: 638–645. shed light on the biology of BD and help lead us 19 Zeggini E, Weedon MN, Lindgren CM, Frayling TM, Elliott KS, toward better treatments. Lango H et al. Replication of genome-wide association signals in The full set of results is available for download at UK samples reveals risk loci for type 2 diabetes. Science 2007; 316: 1336–1341. http://mapgenetics.nimh.nih.gov. 20 Weedon MN, Lango H, Lindgren CM, Wallace C, Evans DM, Mangino M et al. Genome-wide association analysis identifies 20 loci that influence adult height. Nat Genet 2008; 40: 575–583. Conflict of interest 21 Willer CJ, Sanna S, Jackson AU, Scuteri A, Bonnycastle LL, Clarke R et al. Newly identified loci that influence lipid con- The authors declare no conflict of interest. centrations and risk of coronary artery disease. Nat Genet 2008; 40: 161–169. 22 Houlston RS, Webb E, Broderick P, Pittman AM, Di Bernardo MC, Lubbe S et al. Meta-analysis of genome-wide association data References identifies four new susceptibility loci for colorectal cancer. Nat Genet 2008; 40: 1426–1435. 1 Bertelsen A, Harvald B, Hauge M. A Danish twin study of manic- 23 Raychaudhuri S, Remmers EF, Lee AT, Hackett R, Guiducci C, depressive disorders. Br J Psychiatry 1977; 130: 330–351. Burtt NP et al. Common variants at CD40 and other loci confer risk 2 Kendler KS, Pedersen N, Johnson L, Neale MC, Mathe AA. A pilot of rheumatoid arthritis. Nat Genet 2008; 40: 1216–1223. Swedish twin study of affective illness, including hospitaland 24 Pereira TV, Patsopoulos NA, Salanti G, Ioannidis JP. Discovery population-ascertained subsamples. Arch Gen Psychiatry 1993; properties of genome-wide association signals from cumulatively 50: 699–700. combined data sets. Am J Epidemiol 2009; 170: 1197–1206. 3 Kendler KS. Twin studies of psychiatric illness: an update. Arch 25 Moreno C, Laje G, Blanco C, Jiang H, Schmidt AB, Olfson M. Gen Psychiatry 2001; 58: 1005–1014. National trends in the outpatient diagnosis and treatment 4 Cardno AG, Marshall EJ, Coid B, Macdonald AM, Ribchester TR, of bipolar disorder in youth. Arch Gen Psychiatry 2007; 64: Davies NJ et al. Heritability estimates for psychotic disorders: the 1032–1039. Maudsley Twin Psychosis Series. Arch Gen Psychiatry 1999; 56: 26 Pritchard JK, Coop G, Pickrell JK, Novembre J, Kudaravalli S, Li J 162–168. et al. The role of geography in human adaptation. PLoS Genet 5 Edvardsen J, Torgersen S, Roysamb E, Lygren S, Skre I, Onstad S. 2009; 5: 1–16. Heritability of bipolar spectrum disorders. Unity or heterogeneity? 27 de Bakker PI, Neale BM, Daly MJ. Meta-analysis of genome-wide J Affect Disord 2008; 106: 229–240. association studies. Cold Spring Harb Protoc 2010; doi:10.1101/pdb. 6 Smoller JW, Finn CT. Family, twin, and adoption studies of bipolar top81; http://cshprotocols.cshlp.org/content/2010/6/pdb.top81.full. disorder. Am J Med Genet C Semin Med Genet 2003; 123C: 48–58. 28 de Bakker PI, Ferreira MA, Jia X, Neale BM, Raychaudhuri S, 7 Wendland JR, McMahon FJ. Genetics of Bipolar Disorder. In: Manji Voight BF. Practical aspects of imputation-driven meta-analysis HK and Zarate Jr CA (eds). Behavioral Neurobiology of Bipolar of genome-wide association studies. Hum Mol Genet 2008; 17: Disorder and its Treatment. Springer: London, UK, 2011 pp 19–31. R122–R128. 8 Gershon ES, Hamovit J, Guroff JJ, Dibble E, Leckman JF, Sceery W 29 Smith EN, Bloss CS, Badner JA, Barrett T, Belmonte PL, Berrettini et al. A family study of schizoaffective, bipolar I, bipolar II, W et al. Genome-wide association study of bipolar disorder in unipolar and normal control probands. Arch Gen Psychiatry 1982; European American and African American individuals. Mol 39: 1157–1167. Psychiatry 2009; 14: 755–763. 9 Kieseppa¨ T, Partonen T, Haukka J, Kaprio J, Lo¨nnqvist J. High 30 Lee MTM, Chen CH, Lee CH, Chen CC, Chong MY, Ouyang WC concordance of bipolar I disorder in a nationwide sample of twins. et al. Genome-wide association study of bipolar I disorder in Han Am J Psychiatry 2004; 161: 1814–1821. Chinese Population. Mol Psychiatry 2010; 16: 548–556. 10 Goodwin FK, Jamison K. Manic-Depressive Illness: Bipolar 31 Etain B, Dumaine A, Mathieu F, Chevalier F, Henry C, Kahn JP Disorders and Recurrent Depression, 2nd Edn. Oxford Univ Press: et al. A SNAP25 promoter variant is associated with early-onset New York, USA, 2007. bipolar disorder and a high expression level in brain. Mol 11 Craddock N, Jones I. Genetics of bipolar disorder. J Med Genet Psychiatry 2010; 15: 748–755. 1999; 36: 585–594. 32 Sebastiani P, Solovieff N, Puca A, Hartley SW, Melista E, 12 Ferreira MA, O’Donovan MC, Meng YA, Jones IR, Ruderfer DM Andersen S et al. Genetic signatures of exceptional longevity in et al. Collaborative genome-wide association analysis supports a humans. Science 2010; e-pub ahead of print 11 November 2010; role for ANK3 and CACNA1C in bipolar disorder. Nature Genet doi:10.1126/science.1190532. 2008; 40: 1056–1058. 33 Li Y, Abecasis GR. MACH 1.0: Rapid haplotype reconstruction and 13 Baum AE, Akula N, Cabanero M, Cardona I, Corona W, Klemens B missing genotype inference. Am J Hum Genet 2006; S79: 2290. et al. A genome-wide association study implicates diacylglycerol 34 Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D kinase eta (DGKH) and several other genes in the etiology of et al. PLINK: a tool set for whole-genome association and population- bipolar disorder. Mol Psychiatry 2007; 13: 197–207. based linkage analyses. Am J Hum Genet 2007; 81:559–575. 14 Schulze TG, Detera-Wadleigh SD, Akula N, Gupta A, Kassem L, 35 Browning BL, Browning SR. A unified approach to genotype Steele J et al. Two variants in Ankyrin 3 (ANK3) are independent imputation and haplotype phase inference for large data sets of trios genetic risk factors for bipolar disorder. Mol Psychiatry 2009; 14: and unrelated individuals. Am J Hum Genet 2008; 84: 210–223. 487–491. 36 Devlin B, Roeder K. Genomic control for association studies. 15 Scott LJ, Muglia P, Kong XQ, Guan W, Flickinger M, Upmanyu R Biometrics 1999; 55: 997–1004. et al. Genome-wide association and meta-analysis of bipolar 37 Ustun TB, Ayuso-Mateos JL, Chatterji S, Mathers C, Murray CJ. disorder in individuals of European ancestry. Proc Natl Acad Sci Global burden of depressive disorders in the year 2000. Br J USA 2009; 106: 7501–7506. Psychiatry 2004; 184: 386–392. 16 Sklar P, Smoller JW, Fan J, Ferreira MA, Perlis RH, Chambert K 38 Han B, Eskin E. Random-effects model aimed at discovering et al. Whole-genome association study of bipolar disorder. Mol associations in meta-analysis of genome-wide association studies. Psychiatry 2008; 13: 558–569. Am J Hum Genet 2011; 88: 586–598. 17 McMahon FJ, Akula N, Schulze TG, Pierandrea M, Tozzi F, Detera- 39 Metasoft vo3.1 [Internet]. [cited 2011 July 7] [Place Unknown] Wadleigh SD et al. Meta-analysis of genome-wide association data UCLA, Available from: http://genetics.cs.ucla.edu/meta 2011.

Molecular Psychiatry Three novel loci associated with bipolar disorder DT Chen et al 205 40 Pe’er I, Yelensky R, Altshuler D, Daly MJ. Estimation of the 56 Wellcome Trust Case Control Consortium. Genome-wide associa- multiple testing burden for genome-wide association studies of tion study of 14 000 cases of seven common diseases and 3000 nearly all common variants. Genet Epidemiol 2008; 32: 381–385. shared controls. Nature 2007; 447: 661–678. 41 Dudbridge F, Gusnanto A. Estimation of significance thresholds for 57 O’Donovan MC, Craddock N, Norton N, Williams H, Peirc T, genome-wide association scans. Genet Epidemiol 2008; 32: 227–234. Moskvina V et al. Identification of loci associated with schizo- 42 Manolio TA, Brooks LD, Collins FS. A HapMap harvest of in phrenia by genome-wide association and follow-up. Nat Genet sights into the genetics of common disease. J Clin Invest 2008; 118: 2008; 40: 1053–1055. 1590–1605. 58 Psychiatric GWAS Consortium Bipolar Working Group, Sklar P, 43 Ioannidis JP. Interpretation of tests of heterogeneity and bias in Ripke S, Scott L, Andreassen OA, Cichon S et al. Large-scale meta-analysis. J Eval Clin Pract 2008; 14: 951–957. genome-wide association analysis of bipolar disorder identifies a 44 Skol AD, Scott LJ, Abecasis GR, Boehnke M. Joint analysis is more new susceptibility locus near ODZ4. Nat Genet 2011; 43: 977–983. efficient than replication-based analysis for two-stage genome- 59 Park JH, Wacholder S, Gail MH, Peters U, Jacobs KB, Chanock SJ wide association studies. Nat Genet 2006; 38: 209–213. et al. Estimation of effect size distribution from genome-wide 45 Purcell S, Cherny SS, Sham PC. Genetic Power Calculator: design association studies and implications for future discoveries. Nature of linkage and association genetic mapping studies of complex Genet 2010; 42: 570–575. traits. Bioinformatics 2003; 19: 149–150. 60 Ghosh A, Zou F, Wright FA. Estimating odds ratios in genome 46 Phiel CJ, Zhang F, Huang EY, Guenther MG, Lazar MA, Klein PS scans: an approximate conditional likelihood approach. Am J et al. Histone deacetylase is a direct target of valproic acid, a Hum Genet 2008; 82: 1064–1074. potent anticonvulsant, mood stabilizer, and teratogen. J Biol Chem 61 Wray NR, Pergadia ML, Blackwood DHR, Penninx BWJH, Gordon 2001; 276: 36734–36741. SD, Nyholt DR et al. Genome-wide association study of major 47 Schu¨le C, Baghai TC, Eser D, Ha¨fner S, Born C, Herrmann S et al. depressive disorder: New results, meta-analysis, and lessons 2009) The combined dexamethasone/CRH test (DEX/CRH test) and learned. Mol Psychiatry 2010; e-pub ahead of print 2 November prediction of acute treatment response in major depression. PLoS 2010. doi: 10.1038/mp.2010.109. ONE 2009; 4: e4324. 62 Green EK, Grozeva D, Jones I, Jones L, Kirov G, Caesar S et al. The 48 Barbier E, Wang JB. Anti-depressant and anxiolytic like behaviors bipolar disorder risk allele at CACNA1C also confers risk of in PKCI/HINT1 knockout mice associated with elevated plasma recurrent major depression and of schizophrenia. Mol Psychiatry corticosterone level. BMC Neurosci 2009; 10: 132. 2010; 15: 1016–1022. 49 Gilmor ML, Skelton KH, Nemeroff CB, Owens MJ. The effects of 63 Yang J, Wray NR, Visscher PM. Comparing apples and oranges: chronic treatment with the mood stabilizers valproic acid equating the power of case–control and quantitative trait associa- and lithium on corticotropin-releasing factor neuronal systems. tion studies. Genet Epidemiol 2010; 34: 254–257. J Pharmacol Exp Ther 2003; 305: 434–439. 64 Pritchard JK, Pickrell JK, Coop G. The genetics of human 50 Gibbs JR, van der Brug MP, Hernandez DG, Travnor BJ, Nalls MA, adaptation: hard sweeps, soft sweeps, and polygenic adaptation. Lai SL et al. Abundant quantitative trait loci exist for DNA Curr Biol 2010; 20: R208–R215. methylation and in human brain. PLoS Genet 65 Kraeplin E. Comparative Psychiatry. In: Hirsch SR (eds). Themes 2010; 6: e1000952. and Variations in European Psychiatry. The University Press of 51 Heinzen E, Ge D, Cronin KD, Maia J, Shianna KV, Gabriel W et al. Virginia: Charlottesville, pp 3–30. Tissue specific genetic control of gene expression and alternative 66 Secolin R, Banzato CEM, Oliveira MCM, Bittar MFR, Santos ML, splicing: Implications for the study of human complex traits. PLoS Dalgalarrondo P et al. Family-based association study for bipolar Biol 2008; 6:e1. affective disorder. Psychiatr Genet 2010; 20: 126–129. 52 SNPExpress [Internet]. Duke University 2011. Available from: 67 Yanai I, Benjamin H, Shmoish M, Chalifa-Caspi V, Shklar M, Ophir http://people.genome.duke.edu/Bdg48/SNPExpress/. R et al. Genome-wide midrange transcription profiles reveal 53 Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, expression level relationships in human tissue specification. Zahler AM et al. The browser at UCSC. Genome Bioinformatics 2005; 5: 650–659. Res 2002; 12: 996–1006. 68 GeneCards [Internet]. [cited 11 May 2011] Weizmann Institute of 54 David AS, Fleminger S, Kopelman MD, Lovestone S, Mellers JDC Science 2011. Available from: http://www.genecards.org/cgi-bin/ (eds). Lishman’s Organic Psychiatry: A Textbook of Neuropsychia- carddisp.pl?gene = TRANK1. try, 4th edn. Wiley-Blackwell: Chichester, 2009. 69 Lee SH, Wray NR, Goddard ME, Visscher PM. Estimating missing 55 Cichon S, Mu¨ hleisen TW, Degenhardt FA, Mattheisen M, Miro´ X, heritability for disease from genome-wide association studies. Am Strohmaier J et al. Genome-wide association study identifies J Hum Genet 2011; 88: 294–305. genetic variation in neurocan as a susceptibility factor for bipolar 70 Littell JH, Corcoran J, Pillai VK. Systematic Reviews and Meta- disorder. Am J Hum Genet 2011; 88: 372–381. Analysis. Oxford University Press: New York, USA, 2008.

Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)

Appendix The Bipolar Genome Study (BiGS) Authorship List University California, San Diego: John R. Kelsoe, Washington University: John Rice; University Cali- Tiffany A. Greenwood, Caroline M. Nievergelt, fornia, San Francisco: William Byerley; National Rebecca McKinney, Paul D. Shilling; Scripps Transla- Institute Mental Health: Francis J. McMahon, tional Science Institute: Nicholas J. Schork, Erin N. David T Chen; University of Pennsylvania: Wade H. Smith, Cinnamon S. Bloss; Indiana University: John I. Berrettini; Johns Hopkins: James B. Potash, Peter P. Nurnberger, Jr., Howard J. Edenberg, Tatiana Foroud, Zandi, Pamela B. Mahon; University of Michigan: Daniel L. Koller; University of Chicago: Elliot S. Melvin G. McInnis, Sebastian Zo¨llner, Peng Zhang; Gershon, Chunyu Liu, Judith A. Badner; Rush Uni- The Translational Genomics Research Institute: David versity Medical Center: William A. Scheftner; Howard W. Craig, Szabolcs Szelinger; Portland Veterans University: William B. Lawson, Evaristus A. Nwulia, Affairs Medical Center: Thomas B. Barrett; Georg- Maria Hipolito; University of Iowa: William Coryell; August-University Go¨ttingen: Thomas G. Schulze.

Molecular Psychiatry