Molecular Psychiatry (2009) 14, 755–763 & 2009 Nature Publishing Group All rights reserved 1359-4184/09 $32.00 www.nature.com/mp IMMEDIATE COMMUNICATION Genome-wide association study of bipolar disorder in European American and African American individuals EN Smith1,2,24, CS Bloss1,3,24, JA Badner4, T Barrett5, PL Belmonte6, W Berrettini7, W Byerley8, W Coryell9, D Craig10, HJ Edenberg11,12, E Eskin13, T Foroud12, E Gershon4, TA Greenwood14, M Hipolito15, DL Koller16, WB Lawson15, C Liu4, F Lohoff7, MG McInnis17, FJ McMahon18, DB Mirel19, SS Murray1,2,3, C Nievergelt14, J Nurnberger16, EA Nwulia15, J Paschall20, JB Potash6, J Rice21, TG Schulze18, W Scheftner22, C Panganiban10, N Zaitlen13, PP Zandi6,SZo¨llner17, NJ Schork1,2,3 and JR Kelsoe14,23 1Scripps Genomic Medicine, Scripps Translational Science Institute, La Jolla, CA, USA; 2Department of Molecular and Experimental Medicine, The Scripps Research Institute, La Jolla, CA, USA; 3Scripps Health, La Jolla, CA, USA; 4Department of Psychiatry, University of Chicago, Chicago, IL, USA; 5Department of Psychiatry, Portland VA Medical Center, Portland, OR, USA; 6Department of Psychiatry, Johns Hopkins School of Medicine, Baltimore, MD, USA; 7Department of Psychiatry, University of Pennsylvania, Philadelphia, PA, USA; 8Department of Psychiatry, University of California, San Francisco, San Francisco, CA, USA; 9Department of Psychiatry, University of Iowa, Iowa City, IA, USA; 10Neurogenomics Division, The Translational Genomics Research Institute, Phoenix, AZ, USA; 11Department of Biochemistry and Molecular Biology, Indiana University School of Medicine, Indianapolis, IN, USA; 12Department of Medical and Molecular Genetics, Indiana University School of Medicine, Indianapolis, IN, USA; 13Department of Computer Science, University of California, Los Angeles, Los Angeles, CA, USA; 14Department of Psychiatry, University of California, San Diego, La Jolla, CA, USA; 15Department of Psychiatry, Howard University, Washington, DC, USA; 16Department of Psychiatry, Indiana University School of Medicine, Indianapolis, IN, USA; 17Department of Psychiatry, University of Michigan, Ann Arbor, MI, USA; 18Genetic Basis of Mood and Anxiety Disorders Unit, National Institute of Mental Health Intramural Research Program, National Institutes of Health, US Department of Health and Human Services, Bethesda, MD, USA; 19Broad Institute of Harvard University and Massachusetts Institute of Technology, Cambridge, MA, USA; 20National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bethesda, MD, USA; 21Division of Biostatistics, Washington University, St Louis, MO, USA; 22Department of Psychiatry, Rush University, Chicago, IL, USA and 23Department of Psychiatry, VA San Diego Healthcare System, La Jolla, CA, USA

To identify bipolar disorder (BD) genetic susceptibility factors, we conducted two genome-wide association (GWA) studies: one involving a sample of individuals of European ancestry (EA; n = 1001 cases; n = 1033 controls), and one involving a sample of individuals of African ancestry (AA; n = 345 cases; n = 670 controls). For the EA sample, single-nucleotide polymorphisms (SNPs) with the strongest statistical evidence for association included rs5907577 in an intergenic region at Xq27.1 (P =1.6Â 10À6) and rs10193871 in NAP5 at 2q21.2 (P =9.8Â 10À6). For the AA sample, SNPs with the strongest statistical evidence for association included rs2111504 in DPY19L3 at 19q13.11 (P =1.5Â 10À6) and rs2769605 in NTRK2 at 9q21.33 (P =4.5Â 10À5). We also investigated whether we could provide support for three regions previously associated with BD, and we showed that the ANK3 region replicates in our sample, along with some support for C15Orf53; other evidence implicates BD candidate such as SLITRK2. We also tested the hypothesis that BD susceptibility variants exhibit genetic background-dependent effects. SNPs with the strongest statistical evidence for genetic background effects included rs11208285 in ROR1 at 1p31.3 (P =1.4Â 10À6), rs4657247 in RGS5 at 1q23.3 (P =4.1Â 10À6), and rs7078071 in BTBD16 at 10q26.13 (P =4.5Â 10À6). This study is the first to conduct GWA of BD in individuals of AA and suggests that genetic variations that contribute to BD may vary as a function of ancestry. Molecular Psychiatry (2009) 14, 755–763; doi:10.1038/mp.2009.43; published online 2 June 2009 Keywords: ANK3 ; Bipolar Genome Study; genetic background; allelic heterogeneity; GAIN

Correspondence: Dr NJ Schork, Department of Molecular and Introduction Experimental Medicine, The Scripps Research Institute, 10550 North Torrey Pines Avenue, MEM 275, La Jolla, CA 92037, USA. Bipolar disorder (BD) is a paradigmatic complex E-mail: [email protected] 24These authors contributed equally to this work. phenotype with many genetic and nongenetic deter- Received 6 January 2009; revised 3 April 2009; accepted 13 April minants. BD is characterized by episodes of mania 2009; published online 2 June 2009 and depression.1 Onset is usually in late adolescence, GWA study of bipolar disorder EN Smith et al 756 although BD typically recurs and relapses throughout standard single locus analyses within the EA and AA life. BD affects approximately 1% of the world’s groups, we also assessed the extent to which single- population, and carries a lifetime risk for completed nucleotide polymorphisms (SNPs) exhibited evidence suicide as high as 17%. Family, twin and adoption of genetic background-dependent effects or allelic studies all support a substantial genetic component in heterogeneity across the EA and AA individuals. BD.2–4 The sibling recurrence risk is between 7 and 10, Although more than 190 GWA studies have been and heritability is estimated to be about 80%. Although performed to date,12 these studies have virtually all this is consistent with a strong genetic component, the been with EA subjects (NHGRI catalog of published identification of specific genetic variations that influ- GWA studies: http://www.genome.gov/gwastudies/), ence BD susceptibility has been difficult. raising important questions as to the generalizability Although some family studies have suggested that of the results to other populations. We assessed the BD has an autosomal dominant genetic determinant, consistency of the SNPs exhibiting the strongest the vast majority of genetic studies suggest that BD associations with BD in our study with previously has a high level of genetic heterogeneity and a published results of BD GWA studies. We tested substantial polygenic component.5,6 Linkage studies variation at ANK3, which is a candidate that was have identified a number of loci with inconsistent implicated in an earlier GWA study15 and identified replication. Although previous linkage studies were as genome-wide significant in a recent collaborative highly divergent,7 recent meta-analyses of linkage analysis.17 Finally, we genotyped a subset of SNPs in studies have found consistent supportive evidence for the regions exhibiting strongest association in a linkage to a few potential BD susceptibility loci,8–10 replication case–control group and tracked the co- notably 6q, 8q, 13q and 22q. A number of polymor- inheritance of these SNPs with disease in families. phisms in a variety of candidate genes have been tested for association with BD, but most have shown no statistically compelling associations; those that Materials and methods appear to be associated show odds ratios (OR) of 1.1– We provide detailed information in the Supplemen- 1.3, which is again consistent with a polygenic basis tary material regarding study subjects, genotyping 11 for BD. and quality control. The final number of BD cases In the last several years, the development of was 1001 EA subjects and 345 AA subjects; control genome-wide association (GWA) study designs and counts were 1033 EA subjects and 670 AA subjects. analysis methods have made it possible to search for The final data set consisted of 724 067 SNPs in the EA multiple genetic variations underlying a condition data set and 840 730 SNPs in the AA data set. The like BD without a priori assumptions about the genes two data sets had 702 044 SNPs in common, and or genomic regions that might harbor susceptibility these SNPs were used to perform analyses addressing 12 variations. The GWA study approach has revealed SNP and SNP Â genetic background interactions in a number of genetic variations that are unequivocally the combined sample. associated with many different traits and diseases.13,14 The US National Institutes of Mental Health– Statistical analysis sponsored Genetics Initiative for Bipolar Disorder Consortium (Bipolar Consortium) collected over Primary analyses. All genetic analyses were 3500 subjects with BD during 1990–2008. A genetic conducted using PLINK18 versions 1.03 and 1.04. component of the Bipolar Consortium, termed the Our primary analyses tested the association of each ‘Bipolar Genome Study’ (BiGS), was initiated in 2006 SNP (coded additively) with BD using three distinct to conduct a GWA study of BD. The initial BiGS GWA methods that control for genetic background study was funded through the Foundation for the heterogeneity (Table 1): (1) single locus, contingency National Institutes of Health Genetic Information table analysis with genomic control adjustment, (2) Association Network (GAIN) initiative (http://www. logistic regression using Local Ancestry in Admixed genome.gov/19518664) and we report the results of Populations (LAMP19)-derived estimates of ancestry this GWA study here. Of note, many investigators as a covariate and (3) logistic regression using have access to and have utilized the DNA samples the top four multidimensional scaling (MDS) from individuals with BD that were collected as part dimensions as covariates. LAMP ancestry estimates of the Bipolar Consortium. Thus, it is the case that a were also compared to STRUCTURE20 estimates and portion of the samples on which we report here are were similar (see Supplementary methods and not entirely independent of previously published Supplementary Figure S4). Association analyses BD GWA studies15,16 and an ongoing BD GWA study using these three methods were conducted within (Scott et al., submitted). the EA and AA samples separately, as well as with the We performed a GWA study of BD separately in combined sample (see Table 1). For each analysis, the individuals of European ancestry (EA) and of African distribution of expected P-values under the null ancestry (AA) using a variety of methods to control for hypothesis (that is, no association) and the genomic population substructure and admixture among the inflation value (l) were calculated in PLINK using the individuals of AA. Our study is the first GWA study adjust command (Supplementary Figure S1). of BD involving individuals of AA. In addition to Genomic inflation values are reported as both

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 757 Table 1 Most strongly associated SNPs using different methods to control for admixture

Analysis SNP/allele Location MAF Multidimensional LAMP- Genomic scaling adjusted control

OR P-value OR P-value OR P-value

AA only rs2111504/Ta DPY19L3 (19q13.11) 0.23 1.74 1.5 Â 10À6 1.73 1.7 Â 10À6 1.66 5.0 Â 10À6 rs2769605/Tb NTRK2 (9q21.33) 0.22 0.60 4.5 Â 10À5 0.61 4.5 Â 10À5 0.63 1.5 Â 10À4 EA only rs1825828/Ca,c 3q11.2 0.26 0.70 7.0 Â 10À7 0.70 7.6 Â 10À7 0.68 2.9 Â 10À7 rs5907577/Ta Xq27.1 0.31 1.48 1.6 Â 10À6 1.47 2.1 Â 10À6 1.444 4.6 Â 10À6 rs10193871/Gb NAP5 (2q21.2) 0.13 0.65 9.8 Â 10À6 0.65 7.1 Â 10À6 0.65 8.8 Â 10À6 EA/AA combined rs4825220/Ca Xq27.1 0.37 0.71 2.6 Â 10À7 0.70 1.9 Â 10À7 0.61 0.8496d (main effect) EA Pritzker rs6046396/Ga RIN2 (20p11.23) 0.29 1.54 1.43 Â 10À6 1.56 6.2 Â 10À7 1.57 4.93 Â 10À7 non-overlap aOverall top hit. bTop hit in region with at least five SNPs at P <1Â 10À4. cPossibly of poor genotype quality (see text and Supplementary Figure S2). dThis P-value likely reflects the imbalance in the ratio of cases and controls in the EA relative to the AA sample.

Table 2 SNPs showing strongest genetic background-dependent effects in EA/AA combined sample

SNP/allele Location MAF OR SNP Â ancestry interaction P-value rs11208285/A ROR1 (1p31.3) 0.38 2.05 1.4 Â 10À6 rs4657247/T RGS5 (1q23.3) 0.21 2.11 4.1 Â 10À6 rs7078071/T BTBD16 (10q26.13) 0.10 2.85 4.5 Â 10À6

uncorrected and corrected for a study size of 1000 html) with HapMap rel21 phased haplotypes as a cases and 1000 controls.21,22 reference (see Supplementary material). Association was performed with MACH estimated allele dosages using Haplotype tests. Haplotype analyses were performed MDS-adjusted logistic regression in the EA sample. in the EA population using the sliding window approach in PLINK,18 and results for 10-SNP Genetic background-dependent effects. We assessed windows are highlighted. We used OMNIBUS genetic background-specific effects, which could P-values for the primary analysis and individual be due to interactions with background-specific varia- haplotype P-values for the analysis of haplo- tions or alternatively, due to different functional type heterogeneity at ANK3. In these analyses, variations at the locus of interest. We combined the population structure was not taken into account. For EA and AA samples and performed logistic regression the haplotype heterogeneity analysis at ANK3, the analyses with adjustment using a LAMP main proportion of haplotypes that reached a P-value of effect covariate and a LAMP Â SNP interaction term < 0.05 was assessed using the 10-SNP window results. (Table 2). We compared the OR across different cate- For each window, the number of haplotypes with an gories of admixture. We split the AA individuals into individual P-value < 0.05 was counted and a high and low admixture categories by the approxi- proportion calculated for the window. Smoothing mate median %CEU (15%), and calculated the OR was performed over the results for the rest of the and 95% confidence interval within these and the genome, and regions were highlighted that had a EA group (see Figure 1). smoothed proportion greater than or equal to the maximum found at ANK3 (0.33). Regions were Power and statistical significance. In the current delineated by the first and last marker with a study, we have employed a significance threshold smoothed value greater than or equal to 0.33. of 5 Â 10À8, as this threshold has been commonly used in previous GWA studies. As there were no Imputation. Imputation was performed for the SNPs that exceeded this threshold, we report the cleaned EA data set using MACH version 1.0.16 most significant SNP for each sample, as well as (http://www.sph.umich.edu/csg/abecasis/mach/index. regions where there are five SNPs within 100 kb of

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 758 rs11208285 abROR1 (1p31.3) EA + AA SNP*LAMP interaction Low admixture (LAMP = 0−15% CEU) 6 2.5 High admixture (LAMP = 15−75% CEU) 80 European ancestry (LAMP = 90−100% CEU) 2.0 4 60

40 1.5 −log(p) 2 20 1.0 Odds Ratio

0 0 recombination rate (cM/Mb) WTCCC 0.5 STEP−BD EFCAB7 PGM1 ROR1 0.0

ITGB3BP rs11208285 rs4657247 rs7078071 SNP 63800 63900 64000 64100 64200 1 (kb) Figure 1 Top genetic background-dependent single-nucleotide polymorphisms (SNPs). (a) The region around the top background-dependent SNP. The primary SNP is colored in black. Other SNPs are colored according to linkage disequilibrium levels with the primary SNP (r2) (calculated from Phase 3 HapMap data using the CEU population). Recombination rate (HapMap) is shown on the second y axis in blue. RefSeq genes are shown with all possible exons; arrows indicate transcript direction. (b) Odds ratios for the top three background-dependent SNPs within groups showing different degrees of admixture. Low and high admixture groups are all of African ancestry, but are split into two groups by the approximate median percent CEU. Lines indicate 95% confidence intervals.

each other all with P <1Â 10À4. In addition, we BD, we obtained P-values for genotyped markers in list all SNPs reported with a P <1Â 10À4 the WTCCC13 and STEP-BD16 studies; it was the case, (Supplementary Tables S5–S7). The strongest and however, that most of the top SNPs that we describe most compelling tests of association, however, were not genotyped directly within these studies. involve follow-up replication studies, and with the Therefore, we investigated the three regions of data we have available to us, we have attempted to interest identified in a recent collaborative GWA examine the consistency of our results with previous study17 within the EA sample in our study GWA studies of BD and with our own newly (Supplementary Figure S3). We evaluated the top genotyped replication samples (Table 3). SNPs from this study that were directly genotyped in our study, as well as other SNPs in the three regions. Pritzker study overlap and analysis. A concurrent We assessed both the direction of associations study of BD is being undertaken by the Pritzker between the two studies, as well as the strength of Neuropsychiatric Disorders Research Consortium these associations. For comparison, we include (PNDRC; Scott et al., submitted). There is some the results from these studies in the plots of regions overlap between the cases and controls used in our of interest (Supplementary Figure S2). Although study and those used in the Pritzker study (407 EA limited, we further analyzed our results relative to a cases and 357 EA controls). As an extension of our previously published GWA study (Baum et al.15), analyses, we excluded these samples and repeated the which we present in Supplementary Table S4. association tests (see Table 1) so that the results of our Further analysis of the replication within each of study and the study by the PNDRC could be compared the BD GWA studies, however, will require specific on the basis of independent samples. It is also the case delineation of the overlapping individuals within all that a portion of the samples on which we report here the studies and imputation of all genotypes. are not entirely independent of previously published BD GWA studies.15,16 We were not, however, able to perform a similar analysis with the nonoverlapping Replication genotyping samples from these other studies due to restricted A subset of 85 SNPs from the GWA study was selected access to the individual level genotype data. for replication genotyping based on several criteria, with a primary focus on the allelic association P-value Comparison of our results with previous GWA in the larger EA sample. See Supplementary Material studies. Initially, to assess the extent to which our for additional information pertaining to replication results corresponded with previous GWA studies of genotyping.

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 759 Table 3 Replication genotyping results

SNP Chromosome Position Gene MAF Family P-value Replication Meta-analysis GWA study (unaffectedness case–control P-value P-value criteria) P-value

Permissive Strict rs4671825 2 67 762 350 0.47 0.186 0.008 0.639 0.099 0.000296 rs4671826 2 67 762 453 0.47 0.137 0.008 0.640 0.084 0.00012 rs13358880 5 56 845 056 0.14 0.204 0.041 0.196 7.1E-05 0.00012 rs4389663 5 104 451 583 0.17 0.166 0.047 0.918 0.130 0.001364 rs10981870 9 115 462 097 0.34 0.137 0.047 0.449 0.022 0.000218 rs4924281 15 36 742 939 0.27 0.011 0.034 0.363 0.001 0.00066 rs1495186 15 36 778 748 C15ORF53 0.23 0.015 0.043 0.045 0.009 0.000345 rs903530 15 36 786 340 0.23 0.010 0.054 0.033 0.012 0.000475 rs2643217 15 36 791 055 0.23 0.008 0.057 0.033 0.004 7.66E-05 rs2012353 19 37 591 133 DPY19L3 0.17 0.116 0.030 0.349 0.106 0.04628

Results group, we report an intergenic region about 300 kb upstream of NTRK2 (top SNP is rs2769605, Descriptive statistics P = 4.5 Â 10À5). When the EA and AA groups were Demographic variables for EA and AA cases were combined, the most significant SNP was rs4825220 in generally similar (Supplementary Table S1). For the Xq27.1, P = 2.6 Â 10À7, which does not lie near a AA population, controls were less likely to be known gene. We also show QQ plots for each analysis women. In the EA population, gender distributions (Supplementary Figure S1). Within AA and EA between cases and controls were similar. Controls samples, overall P-value inflation was low. However, were much older than cases in both populations we observed high levels of inflation when we (mean age of cases: 18.0 (AA) and 19.3 (EA) vs mean combined the EA and AA data sets. This is due to age of controls: 45.8 (AA) and 52.2 (EA), which the difference in case–control ratios between the should protect against cryptic disease in the control studies, which artificially induces population groups. stratification. Correcting for admixture using either LAMP or MDS covariates effectively removed this Basic association analyses stratification. However, using the genomic control We report the top associated SNP in each ancestry method to adjust the P-values overcorrected the group, as well as the top associated SNP in association (Table 1). the combined sample (Table 1; Supplementary We investigated the consistency of our top Figure S2). hits across each of the different groups in our As discussed in the Materials and methods section, study. Supplementary Table S2 shows to what we utilized different methods to control for genetic extent the top hits in each individual sample (for background heterogeneity and admixture, but ob- example, in the EA group) showed evidence of tained similar results for all the methods. As shown association in the other samples (for example, in the in Table 1, the most significant associations were AA group). Top hits within AA were not significant in different between the EA and AA subjects. For the EA EA and vice versa, suggesting that genetic variation sample, the most significantly associated SNP that contributes to BD may differ between the two was rs1825828 at 3q11.2, P = 7.0 Â 10À7. However, ancestral groups. upon inspection of the genotype intensity plot (see Imputation was performed using MACH (http:// Supplementary Figure S2), rs1825828 appeared to www.sph.umich.edu/csg/abecasis/mach/index.html) be poorly genotyped. The second-best association on the EA data set. Associations with imputed SNPs was rs5907577 in an intergenic region at Xq27.1, generally supported, and were of similar strength as P = 1.6 Â 10À6. For the AA sample, the most the associations with genotyped SNPs (see Supple- significantly associated SNP was rs2111504 in mentary Figures S2 and S3). Near 3q11.2, however, DPY19L3 at 19q13.11, P = 1.5 Â 10À6. In addition, we there were no imputed SNPs that showed association looked for regions where there were multiple SNPs with BD, supporting the argument that the observed with low P-values (P <1Â 10À4) with physical proxi- association is due to a genotyping error. mity to one another (that is, within 100 kb). We report two regions that contain five SNPs that meet this Haplotype analyses criterion. Among the EA group, we report NAP5 (top We performed haplotype-based association tests in SNP is rs10193871, P = 9.8 Â 10À6), and among the AA the EA population, using the sliding-window

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 760 approach on the genotyped SNPs in PLINK. In # haplotypes 0.8 @ p < 0.05 addition to showing 10-SNP haplotype P-values for 0 1 each region where an individual SNP is highlighted, 2 0.6 3 we also describe the most significant haplotype. This 4 haplotype is located on the and is 5 7 Mb away from the SNP with the highest single 0.4 locus association strength (10-SNP OMNIBUS P = 1.9 Â 10À11). This region is about 1 Mb downstream of SLITRK2 (Supplementary Figure S2). 0.2

Comparison of our results with previous GWA studies 17 Proportion of haplotypes @ p < 0.05 0 A recent collaborative GWA study that consisted CDC2 TMEM26 of 4387 cases and 6209 controls pooled from the SLC16A9 RHOBTB1 13 PHYHIPL ANK3 Wellcome Trust Case Control Consortium Bipolar FAM13C1 CCDC6 Analysis, a study by Sklar et al.16 (STEP-BD), and 2365 new samples highlighted three regions of 60500 61000 61500 62000 62500 63000 interest: ANK3 (ankyrin G), CACNA1C (alpha 1C Chromosome 10, kb subunit of the L-type voltage-gated calcium channel) and a region 3.3 kb away from C15Orf53 on chromo- Figure 2 Multihaplotype association in the ANK3 region some 15q14.17 This analysis was restricted to individ- suggests allelic heterogeneity. The proportion of the uals of EA. Therefore, we investigated these regions haplotypes that show a P < 0.05 for a given window is within only the EA sample in our study (Supplemen- plotted on the y axis. The x axis is the physical position of tary Figure S3). When we focused on the top SNPs the midpoint of the haplotype. The points are colored according to the number of haplotypes that show P < 0.05. from this study that were directly genotyped in our The proportion was median smoothed over 101 windows study, only one of the SNPs reached P < 0.05 (ANK3, (blue line). The red dotted line indicates the genome-wide rs1938526/G, P = 0.036, OR = 1.31), although the top mean proportion of haplotypes with P < 0.05. RefSeq genes SNP in the 15q14 region approached P < 0.05 are shown below the x axis. (rs2172835, P = 0.057, OR = 0.88). In both cases, the association was in the same direction and of a similar strength as previously reported. In both of these regions, we saw additional SNPs that showed low (0.01–0.0001) P-values. We further analyzed our results relative to one of We hypothesized that this pattern could indicate the single previously published GWA studies (Baum the presence of allelic heterogeneity, which has also et al.15), which included about half of the probands been suggested by a separate study focusing on two and controls used in the present study, in addition to markers in the region (Schulze et al., in press), and a replication sample that is independent of our which might reflect multiple underlying rare variants. sample. Several of the 88 SNPs with replicated In the presence of allelic heterogeneity, we would association signals in the study by Baum expect that multiple haplotypes would show associa- et al.15 also show nominal evidence of association in tion with BD. Using the 10-SNP window result from the present study (Supplementary Table S4), includ- the haplotype analysis in the EA population, which ing a SNP in ANK3 (that is, rs9804190). The extent to provide P-values for each haplotype in addition to the which this can be considered evidence for a consis- OMNIBUS P-values, we investigated the proportion tent finding, however, is limited given the overlap of haplotypes within each window that had P-values between the samples used in the two studies. < 0.05 (Figure 2). In the ANK3 region, there were many 10-SNP Genetic background  SNP interaction analysis windows that had a high proportion of low P-value Table 2 depicts evidence of SNP associations with haplotypes (up to 75%, genome-wide average 4.7%), genetic background-dependent effects. with 2–5 haplotypes often being implicated. To see if As shown, no SNP showed strong genetic back- this was unexpected in the genome, we fit smooth ground-dependent effects (that is, no P-values curves to the proportion of significant haplotypes <5 10À8). We do, however, report the top three across the genome and looked for other regions that SNPs, which occur in ROR1, RGS5 and BTBD16 (all matched or exceeded the maximum smoothed value P-values < 4.5  10À6). We also show the OR across found in ANK3. We found 10 additional regions different categories of admixture to depict these covering approximately 1.5 Mb or about 0.05% of the results (Figure 1). genome that exceeded this level of heterogeneity, indicating that the proportion of haplotypes that are Comparisons of different methods for determining associated with BD at this locus is relatively high, ancestry compared to the rest of the genome (Supplementary On the basis of LAMP ancestry estimates using all Table S3). This provides additional support that the autosomal SNPs, we observed that the EA set showed region shows allelic heterogeneity. < 1% Yoruban admixture (mean = 0.998, range: 0.944–1,

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 761 s.d. = 0.005), whereas the AA set showed almost 19% Although no single SNP showed significant associa- European admixture (mean = 0.188, range: 0.0–1, tion after correction for genome-wide testing in either s.d. = 0.124). Furthermore, LAMP ancestry estimates of our populations, some noteworthy associations were were significantly correlated with estimates generated observed with BD candidate genes, as well as with with the more traditional STRUCTURE method genes known to be expressed in human brain. Of (r = 0.999, P < 0.001). Including parental allele fre- particular interest are SLITRK2 and NTRK2. SLITRK2 is quencies from the Diversity Panel a member of a family of six genes that are widely (HGDP) subjects instead of the HapMap subjects did expressed in neural tissue,24 producing that not influence ancestry estimates (r = 1.0, P < 0.001). are membrane bound. SLITRK2 regulates neurite out- growth in vitro.Thus,SLITRK2 is a logical BD risk gene. Another member of this gene family, SLITRK1, Analysis of subjects that did not overlap with the has been reported to be associated with Tourette’s PNDRC syndrome.25 NTRK2 (also known as TrkB) is a tyrosine Results of analyses that included the subgroup of EA kinase receptor that binds brain-derived neurotrophic subjects that did not overlap with the Pritzker study factor (BDNF) and possibly other (for revealed a different top hit relative to analyses that review see Reichart26). There is abundant evidence included the entire sample. Specifically, the most from animal models of depression that hippocampal significantly associated SNP in this subsample of EA neurogenesis is decreased during the behavioral syn- subjects was rs6046396, which is upstream of RIN2 at À6 drome, that antidepressants increase neurogenesis, and 20p11.23, P = 1.43  10 (see Table 1). that BDNF has antidepressant-like properties in these animal models.27 BDNF expression is increased in Replication genotyping analyses animals by treatment with antidepressants or lithium, Results from the replication analysis are shown in and BDNF SNPs have repeatedly been implicated in Table 3, with SNPs shown from genomic regions that genetic risk for BD, although the effect sizes are quite demonstrated association (P < 0.03) in the family- limited, for example, ORs of 1.1.28–31 based replication sample. Our genetic background analysis revealed signifi- The analysis of the case–control replication cohort cant levels of admixture among our AA study did not detect significance below the multiple-testing subjects. Given this, we explored the comparability threshold (P < 0.001) among the 85 SNPs genotyped, of different methods to account for admixture in our although 3 of the SNPs in the C15Orf53 region analyses and found that three different methods— demonstrated some evidence of association in the genomic control adjustment, logistic regression using case–control cohort (P = 0.03–0.04) as well as in the a LAMP-generated covariate, and logistic regression family sample (P = 0.008–0.015). The preponderance using MDS-generated covariates—all produced very of association evidence from the family-based analy- similar results within the AA population. Regression sis reported in Table 3 is derived from transmission of with either LAMP- or MDS-based covariates was SNP alleles to affected individuals. These transmis- effective for correcting artificially induced population sions are effectively independent of the population structure when EA and AA samples were combined. association.23 The meta-analysis (Table 3) gave P- Genomic control, however, overcorrected for this, values of 0.004–0.01 for SNPs in this region, and effectively removing any real association that was not reinforced the association evidence for SNP associated with admixture levels. rs13358880 on chromosome 5. Different putative associations were observed among individuals of EA and individuals of AA when analyzed separately, and analyses assessing the extent Discussion to which some SNPs show genetic background- dependent effects highlighted different areas of We conducted two GWA studies, one in a sample of potential association. We believe that this is the first individuals of EA and the second in a sample of report emerging from a GWA to explicitly address the individuals of AA. To account for possible genetic dependency of SNP effects on ancestry, admixture background differences, we (1) considered the analy- and/or genetic background. sis of each sample separately; (2) estimated ancestry We sought to replicate our most strongly associated and genetic background diversity from the genetic SNPs in two independent sets of subjects, one family data and controlled for it in the association studies based and one case–control. Positive results were and (3) looked for evidence of genetic back- seen with rs1495186 in C15Orf53, as well as some ground  SNP interactions. To qualify our results, additional SNPs in that region. When considered we also compared them to previous GWA study cumulatively by meta-analysis with the primary EA results investigating BD, performed replication geno- sample, these results demonstrate consistent support typing of our most strongly associated SNPs on an for association of SNPs in the C15Orf53 region with independent cohort and conducted analysis of only BD and provide additional supportive evidence of the nonoverlapping subjects in our study (with association in other regions. another study’s subjects) to tease out independent We also considered the consistency of our results evidence for associations. with previous BD GWA studies. There have been

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 762 three previous GWA studies of BD,13,15,16 in addition the Tzedakah Foundation, a grant from NIH (R01 to a more recent collaborative GWA meta-analysis MH59553) and a grant from Philip and Marcia Cohen. study.17 The Ferreira et al.17 collaborative GWA meta- Falk Lohoff was supported by the Daland Fellowship analysis study identified a region of strong associa- Award from the American Philosophical Society and tion in ANK3, apparently distinct from that detected by NIH grant K08 MH080372. David Craig and Corrie in study by Baum et al.15 Ferreira et al.17 also found Panganiban acknowledge the Stardust foundation. new evidence for association at 15q14, which is near Follow-up genotyping was performed in the labora- C15Orf53, as well as further support for the pre- tory of HE at Indiana University School of Medicine. viously reported CACNA1C gene; these authors This research was also supported, in part, by the concluded that ion channelopathies may be involved Intramural Research Program of the NIH, National in the pathogenesis of BD. Library of Medicine. Dr Smith, Dr Bloss, Dr Murray In the current study, we found consistent evidence and Dr Schork are supported in part by National of both previously reported ANK3 findings, and Institutes of Health grant NIH 1U54RR025204-01. borderline support for replication of a region char- Data and biomaterials were collected in four projects acterized at 15q14, although we failed to find support that participated in the National Institute of Mental for the finding at CACNA1C.InANK3 and at 15q14, Health (NIMH) Bipolar Disorder Genetics Initiative. multiple SNPs in weak to no-linkage disequilibrium From 1991 to 1998, the principal investigators and with the previously associated SNP showed stronger coinvestigators were: Indiana University, Indianapo- association. Investigation of haplotype-based associa- lis, IN, U01 MH46282, John Nurnberger, Marvin tions in our study provides support for allelic Miller and Elizabeth Bowman; Washington Univer- heterogeneity in the ANK3 region. Allelic heteroge- sity, St Louis, MO, U01 MH46280, Theodore Reich, neity has the potential to have an important function Allison Goate and John Rice; Johns Hopkins in genetically influenced disorders, yet can be University, Baltimore, MD U01 MH46274, J Raymond difficult to detect in population-based samples using DePaulo, Jr, Sylvia Simpson and Colin Stine; NIMH common variants, making it a potential explanation of Intramural Research Program, Clinical Neurogenetics ‘missing heritability’.32 Of note, and as we have Branch, Bethesda, MD, Elliot Gershon, Diane Kazuba previously indicated, the Baum et al.15 and Sklar and Elizabeth Maxwell. Data and biomaterials were et al.16 samples both overlap with the current collected as part of 10 projects that participated in the sample,15,16 thus stringent conclusions pertaining to National Institute of Mental Health (NIMH) Bipolar replication are not warranted. Disorder Genetics Initiative. From 1999 to 2007, the Our GWA study of BD provides some support for principal investigators and coinvestigators were: previous findings that variation in ANK3 and at 15q14 Indiana University, Indianapolis, IN, R01 MH59545, influence BD susceptibility. In addition, regions John Nurnberger, Marvin J Miller, Elizabeth S containing NAP5, NTRK2, SLITRK2 and ROR1 are Bowman, N Leela Rau, P Ryan Moe, Nalini Samavedy, worthy of follow-up studies. As all of these associated Rif El-Mallakh (at University of Louisville), Husseini SNPs and regions have a small effect size, it is likely Manji (at Wayne State University), Debra A Glitz (at that the majority of the genetic variations that Wayne State University), Eric T Meyer, Carrie Smiley, influence BD remain yet to be discovered. Tatiana Foroud, Leah Flury, Danielle M Dick, Howard Edenberg; Washington University, St Louis, MO, R01 MH059534, John Rice, Theodore Reich, Allison Conflict of interest Goate, Laura Bierut; Johns Hopkins University, Baltimore, MD, R01 MH59533, Melvin McInnis, JRK is a founder and holds equity in Psynomics Inc. J Raymond DePaulo, Jr, Dean F MacKinnon, Francis The terms of this arrangement have been reviewed M Mondimore, James B Potash, Peter P Zandi, and approved by UCSD in accordance with its Dimitrios Avramopoulos and Jennifer Payne; Univer- conflict of interest policies. sity of Pennsylvania, PA, R01 MH59553, Wade Berrettini; University of California at Irvine, CA, R01 Acknowledgments MH60068, William Byerley and Mark Vawter; University of Iowa, IA, R01 MH059548, William We thank the participants in the study, as without Coryell and Raymond Crowe; University of Chicago, them this work would not have been possible. IL, R01 MH59535, Elliot Gershon, Judith Badner, For best estimate diagnostic work, we thank Vegas Francis McMahon, Chunyu Liu, Alan Sanders, Maria Coleman, Robert Schweitzer, N Leela Rau and Kelly Caserta, Steven Dinwiddie, Tu Nguyen, Donna Rhoadarmer. For data management, we thank Mariano Harakal; University of California at San Diego, CA, Erpe and for study coordination Carre Fisher RN. This R01 MH59567, John Kelsoe, Rebecca McKinney; Rush work was supported by grants from the NIMH and University, IL, R01 MH059556, William Scheftner, NHGRI to JRK (MH078151, MH081804, MH059567 Howard M Kravitz, Diana Marta, Annette Vaughn- supplement), and by the Genetic Association Infor- Brown and Laurie Bederow; NIMH Intramural mation Network (GAIN). This work was additionally Research Program, Bethesda, MD, 1Z01MH002810- supported by the NIMH Intramural Research Program 01, Francis J McMahon, Layla Kassem, Sevilla Detera- (FJM and TGS). WHB was supported by a grant from Wadleigh, Lisa Austin, Dennis L Murphy.

Molecular Psychiatry GWA study of bipolar disorder EN Smith et al 763 References supports a role for ANK3 and CACNA1C in bipolar disorder. Nat Genet 2008; 40: 1056–1058. 1 Goodwin FK, Jamison KR, Akiskal H, Fawcett J, Ghaemi N, 18 Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender Hammen C et al. Manic-Depressive Illness, 2nd edn. University D et al. PLINK: a tool set for whole-genome association and Press: Oxford, 2005. population-based linkage analyses. Am J Hum Genet 2007; 81: 2 Tsuang MG, Faraone SV. The Genetics of Mood Disorders. Johns 559–575. Hopkins University Press: Baltimore, 1990. 19 Sankararaman S, Sridhar S, Kimmel G, Halperin E. Estimating 3 Gershon ES, Hamovit J, Guroff JJ, Dibble E, Leckman JF, Sceery W local ancestry in admixed populations. Am J Hum Genet 2008; 82: et al. A family study of schizoaffective, bipolar I, bipolar II, 290–303. unipolar, and normal control probands. Arch Gen Psychiatry 1982; 20 Pritchard JK, Stephens M, Donnelly P. Inference of population 39: 1157–1167. structure using multilocus genotype data. Genetics 2000; 155: 4 Rice J, Reich T, Andreasen NC, Endicott J, Van Eerdewegh M, 945–959. Fishman R et al. The familial transmission of bipolar illness. Arch 21 Devlin B, Roeder K. Genomic control for association studies. Gen Psychiatry 1987; 44: 441–447. Biometrics 1999; 55: 997–1004. 5 Spence MA, Flodman PL, Sadovnick AD, Bailey-Wilson JE, Ameli 22 Freedman ML, Reich D, Penney KL, McDonald GJ, Mignault AA, H, Remick RA. Bipolar disorder: evidence for a major locus. Am J Patterson N et al. Assessing the impact of population Med Genet 1995; 60: 370–376. stratification on genetic association studies. Nat Genet 2004; 36: 6 Craddock N, Khodel V, Van Eerdewegh P, Reich T. Mathematical 388–393. limits of multilocus models: the genetic transmission of bipolar 23 Abecasis GR, Cookson WO, Cardon LR. Pedigree tests disorder. Am J Hum Genet 1995; 57: 690–702. of transmission disequilibrium. Eur J Hum Genet 2000; 8: 7 Risch N, Botstein D. A manic depressive history. Nat Genet 1996; 545–551. 12: 351–353. 24 Aruga J, Yokota N, Mikoshiba K. Human SLITRK family genes: 8 Badner JA, Gershon ES. Meta-analysis of whole-genome linkage genomic organization and expression profiling in normal brain scans of bipolar disorder and schizophrenia. Mol Psychiatry 2002; and brain tumor tissue. Gene 2003; 315: 87–94. 7: 405–411. 25 Abelson JF, Kwan KY, O’Roak BJ, Baek DY, Stillman AA, Morgan 9 Segurado R, Detera-Wadleigh SD, Levinson DF, Lewis CM, Gill M, TM et al. Sequence variants in SLITRK1 are associated with Nurnberger Jr JI et al. Genome scan meta-analysis of schizophrenia Tourette’s syndrome. Science 2005; 310: 317–320. and bipolar disorder, part III: bipolar disorder. Am J Hum Genet 26 Reichart LF. -regulated signalling pathways. Philos 2003; 73: 49–62. Trans R Soc Lond 2006; 361: 1545–1564. 10 McQueen MB, Devlin B, Faraone SV, Nimgaonkar VL, Sklar P, 27 Schmidt HD, Duman RS. The role of neurotrophic factors in adult Smoller JW et al. Combined analysis from eleven linkage studies of hippocampal neurogenesis, antidepressant treatments and animal bipolar disorder provides strong evidence of susceptibility loci on models of depressive-like behavior. Behav Pharmacol 2007; 18: 6q and 8q. Am J Hum Genet 2005; 77: 582–595. 391–418. 11 Nurnberger JI, Berrettini W, Niculescu AB. Genetics of psychiatric 28 Neves-Pereira M, Mundo E, Muglia P, King N, Macciardi F, disorders. In: Fatemi SH, Clayton PJ (eds). The Medical Basis of Kennedy JL. The brain-derived neurotrophic factor gene Psychiatry. Humana Press: Totowa, NJ, 2008, pp 487–518. confers susceptibility to bipolar disorder: evidence from 12 Manolio TA, Brooks LD, Collins FS. A HapMap harvest of insights into a family-based association study. Am J Hum Genet 2002; 71: the genetics of common disease. JClinInvest2008; 118: 1590–1605. 651–655. 13 WTCCC. Genome-wide association study of 14,000 cases of seven 29 Sklar P, Gabriel SB, McInnis MG, Bennett P, Lim YM, Tsan G et al. common diseases and 3,000 shared controls. Nature 2007; 447: Family-based association study of 76 candidate genes in bipolar 661–678. disorder: BDNF is a potential risk locus. Brain-derived neutrophic 14 Altshuler D, Daly MJ, Lander ES. Genetic mapping in human factor. Mol Psychiatry 2002; 7: 579–593. disease. Science 2008; 322: 881–888. 30 Lohoff FW, Sander T, Ferraro TN, Dahl JP, Gallinat J, Berrettini 15 Baum AE, Akula N, Cabanero M, Cardona I, Corona W, Klemens B WH. Confirmation of association between the Val66Met poly- et al. A genome-wide association study implicates diacylglycerol morphism in the brain-derived neurotrophic factor (BDNF) gene kinase eta (DGKH) and several other genes in the etiology of and bipolar I disorder. Am J Med Genet B Neuropsychiatr Genet bipolar disorder. Mol Psychiatry 2008; 13: 197–207. 2005; 139B: 51–53. 16 Sklar P, Smoller JW, Fan J, Ferreira MA, Perlis RH, Chambert K 31 Liu L, Foroud T, Xuei X, Berrettini W, Byerley W, Coryell W et al. et al. Whole-genome association study of bipolar disorder. Mol Evidence of association between brain-derived neurotrophic factor Psychiatry 2008; 13: 558–569. gene and bipolar disorder. Psychiatr Genet 2008; 18: 267–274. 17 Ferreira MA, O’Donovan MC, Meng YA, Jones IR, Ruderfer DM, 32 Maher B. Personal genomes: the case of the missing heritability. Jones L et al. Collaborative genome-wide association analysis Nature 2008; 456: 18–21.

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

Molecular Psychiatry