Genome-Wide Association Study of Bipolar Disorder Accounting for Effect of Body Mass Index Identiﬁes a New Risk Allele in TCF7L2

ORIGINAL ARTICLE Genome-wide association study of bipolar disorder accounting for effect of body mass index identiﬁes a new risk allele in TCF7L2

SJ Winham1, AB Cuellar-Barboza2,3, A Oliveros4, SL McElroy5,6, S Crow7, C Colby1, D-S Choi3,4, M Chauhan8,MFrye3 and JM Biernacka1,3

Bipolar disorder (BD) is associated with higher body mass index (BMI) and increased metabolic comorbidity. Considering the associated phenotypic traits in genetic studies of complex diseases, either by adjusting for covariates or by investigating interactions between genetic variants and covariates, may help to uncover the missing heritability. However, obesity-related traits have not been incorporated in prior genome-wide analyses of BD as covariates or potential interacting factors. To investigate the genetic factors underlying BD while considering BMI, we conducted genome-wide analyses using data from the Genetic Association Information Network BD study. We analyzed 729 454 genotyped single-nucleotide polymorphism (SNP) markers on 388 European- American BD cases and 1020 healthy controls with available data for maximum BMI. We performed genome-wide association analyses of the genetic effects while accounting for the effect of maximum BMI, and also evaluated SNP–BMI interactions. A joint test of main and interaction effects demonstrated signiﬁcant evidence of association at the genome-wide level with rs12772424 in an intron of TCF7L2 (P = 2.85E − 8). This SNP exhibited interaction effects, indicating that the bipolar susceptibility risk of this SNP is dependent on BMI. TCF7L2 codes for the transcription factor TCF/LF, part of the Wnt canonical pathway, and is one of the strongest genetic risk variants for type 2 diabetes (T2D). This is consistent with BD pathophysiology, as the Wnt pathway has crucial implications in neurodevelopment, neurogenesis and neuroplasticity, and is involved in the mechanisms of action of BD and depression treatments. We hypothesize that genetic risk for BD is BMI dependent, possibly related to common genetic risk with T2D.

Molecular Psychiatry (2014) 19, 1010–1016; doi:10.1038/mp.2013.159; published online 10 December 2013 Keywords: bipolar disorder; body mass index; genome-wide association study; interaction; network analysis

INTRODUCTION as several studies have shown that bipolar patients with comorbid Bipolar disorder (BD) has a high heritability estimated between obesity have a higher psychiatric and general illness burden and 9–12 60% and 85%.1,2 To identify common genetic risk factors for BD, increased mood-episode recurrence. For example, we found multiple genome-wide association (GWA) studies have been greater rates of binge eating, higher degrees of suicidality and conducted. Although allele risk variants have been identified in greater general medical burden in subjects with BD and higher 12 various genes, including ANK3, CACNA1C and ODZ4,3–6 these BMI, indicating that this phenotype defines a clinically important variants vary widely between studies and explain a very limited subgroup. Furthermore, BMI is a heritable complex trait13 that proportion of the heritability of BD. is also associated with other psychiatric disorders including Complex phenotypes, including psychiatric disorders like BD, schizophrenia, major depressive disorder and binge eating are hypothesized to involve a large number of environmental disorder,14,15 motivating the inclusion of BMI in the investigation and genetic components interacting to confer illness. In addition, of genetic risk factors for metabolic and psychiatric phenotypes. they might share genetic risk variants with other phenotypic The positive association between BD and BMI (as a measure of traits.7 Hence, models involving relevant phenotypic traits as obesity), together with the heritability of both BD and BMI, covariates and interacting risk factors should be considered in the suggests a potential interaction in the genetic mechanisms search for genetic factors contributing to complex phenotypes. underlying those traits. In prior GWA studies of type 2 diabetes In particular, BD is positively associated with many phenotypic (T2D) and insulin resistance, accounting for differences in BMI16 traits, including obesity (usually defined as body mass index (BMI) and single-nucleotide polymorphism (SNP)–BMI interactions17 ⩾30 kg/m2) and therefore higher BMI, with one cross-sectional proved beneficial. The rationale for these studies incorporated study showing an obesity rate of 41% in BD compared with 27% in the mediating quality of adiposity as measured by BMI in the those without BD.8 This comorbidity has strong clinical implications heterogeneous etiology of phenotypes, such as T2D and insulin

1Department of Health Sciences Research, Mayo Clinic Rochester, Rochester, MN, USA; 2Department of Psychiatry, University Hospital, Universidad Autonoma de Nuevo Leon, Monterrey, Mexico; 3Department of Psychiatry and Psychology, Mayo Clinic Rochester, Rochester, MN, USA; 4Department of Molecular Pharmacology and Experimental Therapeutics, Mayo Clinic Rochester, Rochester, MN, USA; 5Lindner Center of HOPE, Mason, OH, USA; 6University of Cincinnati, College of Medicine, Cincinnati, OH, USA; 7Department of Psychiatry, University of Minnesota, Minneapolis, MN, USA and 8Department of Psychiatry and Psychology, Mayo Clinic Health Systems Austin, Austin, MN, USA. Correspondence: Dr JM Biernacka, Department of Psychiatry and Psychology, Mayo Clinic, 200 First St SW, Rochester, MN 55905, USA. E-mail: [email protected] Received 12 July 2013; revised 8 October 2013; accepted 10 October 2013; published online 10 December 2013 Bipolar GWAS accounting for BMI identifies TCF7L2 SJ Winham et al 1011 resistance.18 By accounting for BMI effects, these studies found logistic regression model including only the log-additive SNP effect and significant risk loci. In analyses of T2D in the Wellcome Trust Case max BMI, and conducted a one-degree-of-freedom test of the SNP effect. Control Consortium cohort stratified by obesity status, Timpson As previously mentioned, the analyses were restricted to the EA sample et al.16 identified a variant in TCF7L2—a known, strong T2D to avoid confounding by population structure. Moreover, we employed susceptibility gene—not previously identified in the full cohort. In principle components analysis as a safeguard to further correct for population stratification.21 We evaluated the top four principle compo- addition, Manning et al. examined SNP effects on fasting insulin – nents (PCs), and determined that none were associated with either BD or while accounting for the effect of BMI as well as potential SNP max BMI (P>0.5). Therefore, the reported analyses are not adjusted for PCs. BMI interactions, and observed six novel genome-wide significant QQ plots were used to measure the degree of inflation of association loci (Po5E − 8) that were also associated with higher triglyceride statistics, and sensitivity analyses were conducted correcting the top and lower high-density lipoprotein cholesterol levels, suggesting genome-wide associations for the first PC. involvement in insulin resistance pathways.16,17 All analyses were performed in R Statistical Software, version 2.14.0 Despite the known correlations between BD and obesity, BMI (http://www.r-project.org/). We used a genome-wide statistical significance has not previously been incorporated in GWA studies of BD. threshold of Po5E − 8. Following a rationale similar to the one used in T2D research, we propose to reduce BD phenotypic heterogeneity by accounting Network analysis for differences in BMI between BD patients and healthy – Following GWA analyses, we implemented Ingenuity Pathway Analysis individuals, as well as potential SNP BMI interactions on BD (IPA) using SNP P-values from the two degree-of-freedom tests to aid in disease susceptibility. Utilizing publicly available data from a the interpretation of the results (www.ingenuity.com). SNPs were assigned previous GWA study carried out by the Genetic Association to genes within 10 kb using NCBI Build 37, and were pre-filtered 4 Information Network (GAIN), we reevaluate genetic associations to P ⩽ 0.05; this resulted in over 11 000 top SNPs mapped to genes with BD while incorporating data on maximum lifetime BMI. In considered for network and pathway analysis. particular, we conducted GWA analyses adjusted for BMI and With the use of a functional analysis algorithm, IPA constructed examined potential SNP–BMI interaction effects. networks of direct and indirect interactions between genes and molecules in our data on the basis of published, peer-reviewed literature within the curated Ingenuity Knowledge Base. In addition, IPA was used to identify gene sets with enriched association signals, among gene sets representing MATERIALS AND METHODS predefined processes of biological functions, disease states and canonical pathways. Gene set enrichment was tested using a 2 × 2 contingency right- Study description tailed Fisher’s exact test applied to counts of genes with P-values o0.001, The Bipolar Disorder Genome Study Consortium (BiGS) conducted a GWA where gene P-values were defined by the minimum P-value for SNPs study of BD; the data is available as part of the GAIN BP data in dbGaP19 mapping to the gene. and has been previously described by Smith et al,4 including descriptions of study subjects and genotyping and quality control procedures. Although the BiGS/GAIN study included data from both European- Genome-wide imputation American (EA; N = 2035) and African-American (AA; N = 1015) subjects, Analyses were repeated with imputed SNP data. Haplotype phasing was we restricted our analyses to the EA subjects to avoid population performed with SHAPEIT,22 and genome-wide imputation was conducted stratification. on the phased data using Impute 2.2.2,23 using the 1000 Genomes Phase 1 Over the 18-year recruitment period, 1001 BD subjects were enrolled at data as the reference panel (all populations). SNPs with poor imputation multiple institutions. All cases met the criteria for DSM IV-defined bipolar I quality (dosage R2o0.3) or low minor allele frequency (o0.01) were disorder. However, subjects recruited at different times were interviewed removed, leaving 8 466 825 SNPs for analysis. with either the Diagnostic Interview for Genetic Studies 2, 3, or 4 (DIGS 2, 3, 4). Current height (inches) and maximum lifetime body weight (pounds) were collected only in DIGS 4, and therefore self-reported maximum BMI (max BMI) was available only in a subset of 388 of the EA cases. Current RESULTS body weight and other weight measures in addition to lifetime maximum BD cases had a significantly higher max BMI than controls (median body weight were not available. 2 − Control subjects completed a separate psychiatric questionnaire, and of 31.3 vs 29.3 kg/m ; P = 3.96E 6 from Wilcoxon test). In the were excluded if they met diagnostic criteria for depression or had a two-degree-of-freedom GWA analysis that tested the SNP effects history of BD or psychosis. Controls were matched to cases for both gender while incorporating possible SNP–BMI interaction effects, the top and ethnicity.4 Current height and maximum lifetime body weight were ranking variant was rs12772424 in an intron of TCF7L2, which collected in the control questionnaire, and self-reported max BMI data achieved statistical significance at the genome-wide level were available for 1020 of the EA controls. (P = 2.85E − 8, Table 1, Figure 1c). Further investigation of the main and interaction effects shows a significant interaction between BMI and this SNP, where the minor allele has a stronger Statistical analysis protective effect as BMI decreases (Figure 1b). We conducted GWA analyses of 729 454 SNP markers on the 388 EA Other top association signals included rs17541406 in the bipolar cases and 1020 EA mentally healthy control subjects with available pseudogene SEPHS1P2, downstream of MCTP2 (P = 1.21E − 6), with BMI data. GWA analyses were conducted by comparing cases with controls to evaluate the genetic association with BD while allowing for the the same pattern of genetic effect that is dependent on BMI. possibility that this effect may be modified by max BMI. For each SNP, we Variants in the protein-coding regions of WW domain-containing fit a logistic regression model including the SNP effect (assuming a log- oxidoreductase (WWOX)(rs11150078, P = 1.73E − 6) and CDH23 additive genetic model), max BMI and the SNPxBMI interaction. To (rs1625975, P = 4.16E − 6) were also highly ranked and demon- evaluate the genetic effect of each SNP, we conducted a joint two-degree- strated evidence of interaction effects (Table 1, Figures 1b and c). of-freedom likelihood ratio test involving both the marginal SNP effect and The minor allele of rs7690204, in a protein-coding region of 20 the SNPxBMI interaction. The joint two-degree-of-freedom test has been GYPA, exhibited a marginal protective effect (odds ratio = 0.65, shown to be a powerful method to detect genetic associations because it P = 1.11E − 6; Table 1, Figure 1a) but not of a SNP–BMI interaction. allows for either a marginal or interaction effect. However, the joint test Similarly, two SNPs residing in an intergenic region upstream of does not characterize the nature of this association. Therefore we also MARCKS, rs6934970 and rs17074473 (in linkage disequilibrium), performed genome-wide analyses of the marginal genetic effect adjusted − − for the effect of BMI, and the SNP–BMI interaction effect separately. To are also highly ranked (P = 7.08E 6 and P = 8.23E 6, respectively; evaluate the SNP–BMI interaction effect, we conducted a one-degree-of- Figure 1c) in the analysis of marginal genetic effect after adjusting freedom test of the interaction effect in the logistic regression model. To for BMI, but showed no evidence of an interaction with BMI. examine the marginal genetic effect after adjusting for max BMI, we fita Specifically, the minor allele of rs6934970 is associated with an

Table 1. Top SNP results for the two-degree-of-freedom (DF) test of genetic effect, allowing for SNPxBMI interaction

Two-DF testa Adjusted for BMIb SNPxBMI interactiona

rsID Gene Chr Type X2 P OR P OR P

rs12772424 TCF7L2 10 Protein coding 34.75 2.85E −08 0.78 5.87E − 03 1.07 6.16E − 07 rs17541406c SEPHS1P2 15 Pseudo 27.25 1.21E − 06 0.94 4.77E − 01 1.06 8.37E − 07 rs11150078 WWOX 16 Protein coding 26.53 1.73E − 06 0.89 7.51E − 01 0.56 5.26E − 03 rs1043607 TMEM132B 12 Protein coding 24.93 3.86E − 06 0.62 1.53E − 05 1.04 2.55E − 02 rs1625975 CDH23 10 Protein coding 24.78 4.16E − 06 0.58 2.85E − 03 1.10 2.67E − 04 rs7690204 GYPA 4 Protein coding 24.33 5.21E − 06 0.65 1.11E − 06 1.00 8.40E − 01 rs6934970d LOC100287612 6 miscRNA 23.72 7.08E − 06 1.83 8.48E − 07 1.01 7.70E − 01 rs2699783 MIR4778 2 miscRNA 23.63 7.41E − 06 0.50 1.55E − 04 0.93 1.41E − 02 rs10070308 FBXL17 5 Protein coding 23.44 8.12E − 06 0.54 5.20E − 06 1.01 3.76E − 01 rs17074473 LOC100287612 6 miscRNA 23.41 8.23E − 06 1.82 9.45E − 07 1.00 9.78E − 01 Abbreviations: BMI, body mass index; OR, odds ratio; SNP, single-nucleotide polymorphism. aFrom a logistic regression model including the SNP effect, maximum BMI and the SNPxBMI interaction. bFrom a logistic regression model including only the SNP effect and maximum BMI. cDownstream of MCTP2. dIntergenic region upstream of MARCKS.

Figure 1. Manhattan plots for GWA analyses. For each SNP, − log10(P-value) (y-axis) is plotted against chromosomal position (x-axis) for (a) the marginal SNP effect adjusted for BMI (no SNP–BMI interaction), (b) the SNP–BMI interaction effect and (c) the joint genetic test accounting for max BMI and a potential SNPxBMI interaction.

increased odds of BD after accounting for BMI (odds ratio = 1.83, The highest ranked canonical pathways based on IPA included P = 8.5E − 7; Table 1, Figure 1a). ‘synaptic long-term depression’, ‘G-protein αq signaling’, The association test statistics in the GWA analyses were not and ‘synaptic long-term potentiation’ (Table 2). Top networks strongly inflated (λ = 1.04 for the two-degree-of-freedom test include molecules involved in cellular growth and development, allowing for a BMI–SNP interaction), suggesting no substantial as well as cardiovascular system development and signaling population stratification. There is little change in the results after (Figure 2). adjusting for the first PC, although the effect sizes of some SNPs Imputed results were similar to the results for the observed are attenuated (Supplementary Table 1). The QQ plots for the genotypes, with denser peaks in the top ranking regions unadjusted and PC-adjusted analyses are shown in Supplementary (Supplementary Figure S3 shows the results for the two-degree- Material (S1 and S2). of-freedom analysis). In the two-degree-of-freedom analysis of

Molecular Psychiatry (2014), 1010 – 1016 © 2014 Macmillan Publishers Limited Bipolar GWAS accounting for BMI identiﬁes TCF7L2 SJ Winham et al 1013 SNP effect including BMI–SNP interaction, a new peak was lifetime max BMI with BD; furthermore, we also allowed SNP observed on chromosome 3 (Supplementary Table 2). This peak effects on BD risk to depend on BMI (i.e., SNP–BMI interaction is in an intergenic region downstream of TRIM42, and includes effects). Most prior genome-wide analyses for BD have imple- SNPs rs72977016, rs182827559 and rs60035994, all of which mented univariate, one-SNP-at-a-time analyses, overlooking com- represent rare variation (minor allele frequency o0.01). Because plex models including relevant covariates and interactions— the minor allele frequencies of these SNPs are low, and thus critical factors for complex traits. The reduction of variation and imputation quality is poor, this new peak should be interpreted confounding effects through covariate adjustment and incorpora- with caution. tion of interactions was central to the current study. By investigating max BMI as an interacting factor, we were able to detect novel genetic variants correlated with susceptibility to DISCUSSION BD. Our top SNP resides in a gene of biological plausibility for both BD- and BMI-related phenotypic traits, which may be an important Recognizing the relationship between BD and BMI, in the current component in the biological mechanisms linking BD and its study we used a genome-wide approach to identify novel BD- metabolic impact. We observed strong evidence of a SNP–BMI susceptible SNP variants after accounting for the association of interaction involving a variant in TCF7L2 at the genome-wide level. When the associations between rs12772424 and BD or BMI are analyzed separately, there is no evidence for a marginal SNP effect on either trait; yet when examined collectively, we observe a Table 2. Highest ranking canonical pathways via IPA SNP–BMI interaction on BD risk where the protective effect of the minor allele becomes stronger as BMI decreases. Pathway Number of genes Po0.001/ 24–26 number of total genes (%) TCF7L2 is the strongest genetic risk factor for T2D and notably, a risk locus of T2D in a study stratifying subjects by 16 Synaptic long-term depression 9/142 (6.3%) BMI. TCF7L2 codes a T-cell factor/lymphoid enhancer factor, that G-protein αq signaling 8/157 (5.1%) upon β-catenin translocation to the nucleus activates the Synaptic long-term potentiation 7/117 (6.0%) expression of target genes downstream of the Wnt/β-catenin canonical pathway.27,28 This signaling transduction cascade has

Figure 2. BMI–gene interaction network implicated in BD disease risk. Highest scoring network including many of the genes from the ‘Cellular Growth and Proliferation, Cardiovascular System Development and Function, Organismal Development’ canonical pathway and fourth highest scoring network including many of the genes from the ‘Cellular Development, Cellular Growth and Proliferation, Skeletal and Muscular System Development and Function’ canonical pathway are shown together to illustrate the interaction between top genes in our analysis. The color of the genes indicates the degree of association in our analyses: red = lowest P-value, pink = low/moderate P-value, gray = moderate P-value and white = P>0.001. Shapes indicate the gene/molecule type (cytokine, transporter, kinase, so on). Solid lines indicate direct relationships and dashed lines indicate indirect relationships. Note: Wnt, Wnt receptor frizzled, co-receptor LPR1, cytoplasmic signal transduction protein disheveled, and B-catenin destruction complex components, AXIN1 and APC, were added to the network to illustrate the Wnt canonical pathway upstream of genes CTNNB1, GSK3B and TCF7L2 shown in our analysis.

© 2014 Macmillan Publishers Limited Molecular Psychiatry (2014), 1010 – 1016 Bipolar GWAS accounting for BMI identifies TCF7L2 SJ Winham et al 1014 been linked to many diseases—most notably cancer—and is Table 3. Comparison of top-reported SNPs in Smith et al. and our critically involved in the development of many organisms, analyses including humans.29 The Wnt/β-catenin canonical pathway is crucial to different 30 31–33 Smith et al. BMI adjusted aspects of neuroplasticity and adult neurogenesis. The (unadjusted)a (1 df)b action of mood stabilizers on this signaling pathway is regarded as indirect evidence of its involvement in BD pathophysio- CHR Gene rsID OR P OR P logy.34 Lithium antidepressant effects on behavior, for example, have been hypothesized to be secondary to inhibition of the 3 — rs1825828 0.68 1.98E − 07 0.80 1.87E − 02 β-catenin inactivator GSK-3β35,36 through direct37,38 and indirect 23 — rs5907577 1.44 3.36E − 06 1.45 7.54E − 04 mechanisms39; β-catenin has similar effects to lithium in rodent 2 NCKAP5 rs10193871 0.65 6.60E − 06 0.55 1.99E − 05 models.40 Valproate has also been suggested to inhibit GSK-3β 11 — rs4936819 0.50 1.15E − 06 0.55 2.96E − 03 and additionally to increase mRNA Wnt/β-catenin canonical 9 DBC1 rs11789399 1.34 4.24E − 06 1.33 9.83E − 04 41 4 GYPA rs7690204 0.75 6.73E − 06 0.65 1.11E − 06 pathway targets in the hippocampus. Recently, neurons − − differentiated from induced pluripotent stem cells of schizophre- 12 ITPR2 rs4964010 0.73 3.98E 06 0.68 5.81E 05 nic patients have shown altered expression of many components Abbreviations: BMI, body mass index; df, degree of freedom; OR, odds in this pathway, including different TCF/LF isoforms.42 Further- ratio; SNP, single-nucleotide polymorphism. aFrom a logistic regression more, multiple animal findings have associated several steps of model including only the SNP effect. bFrom a logistic regression model this cascade to stress induced depressive and antidepressant including the SNP effect and maximum BMI (no SNPxBMI interaction). response.43 TCF/LF involvement in both neuropsychiatric disorders and T2D makes it a feasible candidate for complex shared genetic risk. Moreover, the variant rs1625975 in CDH23, part of the cadherin superfamily of cell–cell adhesion glycoproteins, was highly ranked current height and maximum lifetime weight, which restricted our in our analysis. Several cadherins have been shown to suppress analyses to only those subjects, limiting the available sample size β-catenin activity,44 and hence are negatively regulated by Wnt and effectively reducing power; yet even with a small sample size, signaling.45 Also, several cadherin and adhesion glycoprotein- phenotypic heterogeneity was reduced through the inclusion of coding genes have previously been associated with BD in genetic BMI, and we were able to uncover novel genetic associations of studies.46–50 genome-wide significance. Furthermore, the lifetime maximum Rs11150078 in WWOX presented an inverse relationship to that weight used in the analyses was based on self-report, which may observed with TCF7L2, suggesting a protective effect of the minor underestimate BMI and potentially bias results. Finally, other allele on BD risk as BMI increases. The WWOX gene codes a protein measures of BMI (such as current BMI) were not available in this that sequesters transcription factors involved in neoplasias,51 and sample, limiting the scope of our conclusions to the lifetime defects have been widely studied in cancer literature. WWOX measure of max BMI and hindering comparisons to the general interacts with multiple signaling pathways including Wnt,52 but to population. However, as opposed to current BMI, lifetime BMI may our knowledge has not been previously related to psychiatric be more relevant with regard to obesity and cardiovascular disorders. comorbidities, and may be a more relevant phenotype for genetic IPA network analysis defined an interactome that links two of association studies. our top genes, TCF7L2 and WWOX, together and emphasized It is important to recognize that BMI may represent a more additional genes of potential relevance via interaction with intermediary measure related to other underlying conditions. For β-catenin promoters Jun and Rap1 and inhibitor GSK-3β: growth instance, many of the most effective medications to treat BD, such factor downstream signaling cascade molecules Akt and Erk as neuroleptics, are associated with weight gain. Our analyses do (Figure 2). Furthermore, an enriched ‘cell death and survival’ not account for the current treatment that each case was network involves BD risk genes CACNA1C,6 CACNA2D1 and receiving or the weight gain that may have been associated with CACNB3. that treatment, and hence our results may be confounded with This study incorporated additional information on max BMI into medication type or treatment response. Although basic informa- the genome-wide analyses of BD risk, identifying novel genetic tion on lifetime antipsychotic use is available, detailed information variants with biological relevance. Interestingly, we observed about current medication usage, duration and dosage, treatment genetic associations with greater statistical significance than the response and associated weight gain were not; rarely is such original analyses, despite the fact that our analyses were restricted reliable information available in a retrospective study, making to those with available max BMI data (i.e., 388 cases/1020 controls accounting for confounding effects of medication type or vs 1001 cases/1034 controls). These results suggest that we treatment response challenging. Nevertheless, an examination of attained higher power to detect genetic effects through the the impact of medication use and treatment-associated weight reduction of heterogeneity due to integration of a crucial gain on the associations observed here is an important area of covariate. In general, the top ranking SNPs in our analyses were future research. quite different from those reported for European Americans in Furthermore, in this study, we did not have data on T2D Smith et al.4 Many of the estimated odds ratios for the previously history and thus could not investigate whether BMI may be a reported top SNPs are attenuated in our analyses, indicating that surrogate for T2D predisposition. In addition to T2D, high BMI is BMI may be a confounding factor (Table 3). However, rs7690204 in strongly associated with numerous diseases and disorders, GYPA was highly ranked in both studies. In fact, both rs7690204 including other psychiatric disorders and comorbidities (such as and rs10193871 in NCKAP5 (a.k.a. NAP5) demonstrated stronger binge eating disorder), as well as metabolic disorders and heart protective effects in our analyses accounting for BMI (Table 3). disease, further complicating the interpretation of our results. Although there are clear advantages of our approach to identify Determining genetic factors correlated with BD and BMI is only genetic risk factors by integrating relevant covariates and the beginning in understanding the causal biological and genetic interacting factors into analyses, there are also limitations. In the mechanisms. In fact, this study suggests a potential for shared study from which the data were drawn, subsets of bipolar cases genetic risk factors between psychiatric traits and metabolic received diagnostic interviews based on different versions of DIGS. disorders, such as BD and T2D, which have exhibited significant Only subjects who were interviewed with DIGS 4 were asked their genetic overlap in previous network analyses.7 Although not

Molecular Psychiatry (2014), 1010 – 1016 © 2014 Macmillan Publishers Limited Bipolar GWAS accounting for BMI identifies TCF7L2 SJ Winham et al 1015 directly investigated here, a simultaneous examination of com- 9 Cerimele JM, Katon WJ. Associations between health risk behaviors and symp- mon genetic risk factors for multiple comorbidities—pleiotropic toms of schizophrenia and bipolar disorder: a systematic review. Gen Hosp effects—could shed further light on the meaning behind the Psychiatry 2012; 35:16–22. identified associations of this study. Follow-up cohort and 10 McElroy SL, Keck PE Jr. Obesity in bipolar disorder: An overview. Curr Psychiatry 14 – functional studies are necessary to decipher the complexity of Rep 2012; :650 658. these associations and to understand the relationship between 11 McElroy SL, Kotwal R, Malhotra S, Nelson EB, Keck PE, Nemeroff CB. Are mood disorders and obesity related? A review for the mental health professional. BMI and the genetic predisposition to BD. 65 – fi J Clin Psychiatry 2004; :634 651, quiz 730. To our knowledge, the current study is the rst to investigate 12 McElroy SL, Crow S, Biernacka JM, Winham S, Geske J, Barboza AB et al. Clinical genome-wide genetic risk factors for BD and obesity-related phenotype of bipolar disorder with comorbid binge eating disorder. J Affect measures simultaneously. In this study, we identified many Disord 2013; 150: 981–986. biologically relevant genetic variants that are associated with 13 Yang J, Ferreira T, Morris AP, Medland SE, Madden PA, Heath AC et al. Conditional BD, after accounting for the effect of max BMI. Further and joint multiple-SNP analysis of GWAS summary statistics identifies additional investigation is needed to replicate and validate these relation- variants influencing complex traits. Nat Genet 2012; 44: 369–375. ships and to determine if the identified variants or genes have a 14 Lopresti AL, Drummond PD. Obesity and psychiatric disorders: Commonalities in dysregulated biological pathways and their implications for treatment. functional role in disease susceptibility. Investigating complex 45C – genetic relationships involving covariates and genetic interactions Prog Neuropsychopharmacol Biol Psychiatry 2013; :92 99. 15 Kessler RC, Berglund PA, Chiu WT, Deitz AC, Hudson JI, Shahly V et al. The pre- is necessary to better understand the genetic etiology of BD and valence and correlates of binge eating disorder in the World Health Organization obesity. World Mental Health Surveys. Biol Psychiatry 2013; 73: 904–914. 16 Timpson NJ, Lindgren CM, Weedon MN, Randall J, Ouwehand WH, Strachan DP et al. Adiposity-related heterogeneity in patterns of type 2 diabetes susceptibility CONFLICT OF INTEREST observed in genome-wide association data. Diabetes 2009; 58:505–510. Dr McElroy is a consultant to or member of the scientific advisory boards of Alkermes, 17 Manning AK, Hivert M-F, Scott RA, Grimsby JL, Bouatia-Naji N, Chen H et al. fi Bracket, Corcept, MedAvante, Shire, and Teva. She is a principal or co-investigator on A genome-wide approach accounting for body mass index identi es genetic fl studies sponsored by Alkermes, AstraZeneca, Cephalon, Eli Lilly and Company, Forest, variants in uencing fasting glycemic traits and insulin resistance. Nat Genet 2012; 44 – Marriott Foundation, Orexigen Therapeutics, Inc., Pfizer, Shire, Takeda Pharmaceutical :659 681. Company Ltd, and Transcept Pharmaceutical, Inc. She is also an inventor on United 18 Wang X, Ding X, Su S, Spector TD, Mangino M, Iliadou A et al. Heritability of insulin States Patent No. 6,323,236 B2, Use of Sulfamate Derivatives for Treating Impulse sensitivity and lipid profile depend on BMI: evidence for gene-obesity interaction. Control Disorders, and along with the patient’s assignee, University of Cincinnati, Diabetologia 2009; 52: 2578–2584. Cincinnati, Ohio, has received payments from Johnson & Johnson, which has 19 Mailman MD, Feolo M, Jin Y, Kimura M, Tryka K, Bagoutdinov R et al. The NCBI exclusive rights under the patent. Dr Crow has received research grants from Shire, dbGaP database of genotypes and phenotypes. Nat Genet 2007; 39: 1181–1186. Alkermes and Phillips Respironics. Dr Frye has had grant support from Pfizer and 20 Kraft P, Yen YC, Stram DO, Morrison J, Gauderman WJ. Exploiting gene-environment Myriad, and has served as an unpaid consultant for Allergan, Myriad, Sunovion and interaction to detect genetic associations. Hum Hered 2007; 63:111–119. Teva Pharmaceuticals. The remaining authors declare no conflict of interest. 21 Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D. Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet 2006; 38:904–909. ACKNOWLEDGMENTS 22 Delaneau O, Zagury JF, Marchini J. Improved whole-chromosome phasing for disease and population genetic studies. Nat Methods 2013; 10:5–6. This study was supported by funding from the Marriott Family Foundation and Mayo 23 Howie B, Fuchsberger C, Stephens M, Marchini J, Abecasis GR. Fast and accurate ’ Clinic s Center for Individualized Medicine. Funding support for the Whole Genome genotype imputation in genome-wide association studies through pre-phasing. Association Study of Bipolar Disorder was provided by the National Institute of Nat Genet 2012; 44: 955–959. Mental Health (NIMH) and the genotyping of samples was provided through the 24 Jin T, Liu L. The Wnt signaling pathway effector TCF7L2 and type 2 diabetes Genetic Association Information Network (GAIN). The data sets used for the analyses mellitus. Mol Endocrinol 2008; 22:2383–2392. described in this manuscript were obtained from the database of Genotypes and 25 Saxena R, Elbers CC, Guo Y, Peter I, Gaunt TR, Mega JL et al. Large-scale gene- Phenotypes (dbGaP) found at http://www.ncbi.nlm.nih.gov/gap through dbGaP centric meta-analysis across 39 studies identifies type 2 diabetes loci. Am J Hum accession number phs000017.v3.p1. Samples and associated phenotype data for the Genet 2012; 90:410–425. Collaborative Genomic Study of Bipolar Disorder were provided by the NIMH 26 Nobrega MA. TCF7L2 and Glucose Metabolism: Time to Look Beyond the Genetics Initiative for Bipolar Disorder. Pancreas. Diabetes 2013; 62:706–708. 27 Behrens J, vonKries JP, Kuhl M, Bruhn L, Wedlich D, Grosschedl R et al. Functional interaction of beta-catenin with the transcription factor LEF-1. Nature 1996; 382: REFERENCES 638–642. 1 Lichtenstein P, Yip BH, Bjork C, Pawitan Y, Cannon TD, Sullivan PF et al. Common 28 Cadigan KM, Waterman ML. TCF/LEFs and Wnt signaling in the Nucleus. Cold genetic determinants of schizophrenia and bipolar disorder in Swedish families: a Spring Harb Perspect Biol 2012; 4: a007906. population-based study. Lancet 2009; 373:234–239. 29 Saito-Diaz K, Chen TW, Wang XX, Thorne CA, Wallace HA, Page-McCaw A et al. 2 McGuffin P, Rijsdijk F, Andrew M, Sham P, Katz R, Cardno A. The heritability of The way Wnt works: Components and mechanism. Growth Factors 2013; 31: bipolar affective disorder and the genetic relationship to unipolar depression. 1–31. Arch Gen Psychiatry 2003; 60: 497–502. 30 Budnik V, Salinas PC. Wnt signaling during synaptic development and plasticity. 3 Burton PR, Clayton DG, Cardon LR, Craddock N, Deloukas P, Duncanson A et al. Curr Opin Neurobiol 2011; 21: 151–159. Genome-wide association study of 14,000 cases of seven common diseases and 31 Jang MH, Bonaguidi MA, Kitabatake Y, Sun J, Song J, Kang E et al. Secreted 3,000 shared controls. Nature 2007; 447:661–678. frizzled-related protein 3 regulates activity-dependent adult hippocampal neuro- 4 Smith EN, Bloss CS, Badner JA, Barrett T, Belmonte PL, Berrettini W et al. Genome- genesis. Cell Stem Cell 2013; 12: 215–223. wide association study of bipolar disorder in European American and African 32 Seib DR, Corsini NS, Ellwanger K, Plaas C, Mateos A, Pitzer C et al. Loss of American individuals. Mol Psychiatry 2009; 14: 755–763. Dickkopf-1 restores neurogenesis in old age and counteracts cognitive decline. 5 Ferreira MA, O'Donovan MC, Meng YA, Jones IR, Ruderfer DM, Jones L et al. Cell Stem Cell 2013; 12:204–214. Collaborative genome-wide association analysis supports a role for ANK3 and 33 Wexler EM, Geschwind DH, Palmer TD. Lithium regulates adult hippocampal CACNA1C in bipolar disorder. Nat Genet 2008; 40: 1056–1058. progenitor development through canonical Wnt pathway activation. Mol Psy- 6 Psychiatric GWAS Consortium Bipolar Disorder Working Group. Large-scale chiatry 2008; 13:285–292. genome-wide association analysis of bipolar disorder identifies a new suscepti- 34 Valvezan AJ, Klein PS. GSK-3 and Wnt Signaling in Neurogenesis and Bipolar bility locus near ODZ4. Nat Genet 2011; 43:977–983. Disorder. Front Mol Neurosci 2012; 5:1. 7 Rzhetsky A, Wajngurt D, Park N, Zheng T. Probing genetic overlap among com- 35 Klein PS, Melton DA. A molecular mechanism for the effect of lithium on deve- plex human phenotypes. Proc Natl Acad Sci USA 2007; 104: 11694–11699. lopment. Proc Natl Acad Sci USA 1996; 93: 8455–8459. 8 Sicras A, Rejas J, Navarro R, Serrat J, Blanca M. Metabolic syndrome in bipolar 36 O'Brien WT, Harper AD, Jove F, Woodgett JR, Maretto S, Piccolo S et al. Glycogen disorder: a cross-sectional assessment of a Health Management Organization synthase kinase-3 beta haploinsufficiency mimics the behavioral and molecular database. Bipolar Disord 2008; 10:607–616. effects of lithium. J Neurosci 2004; 24: 6791–6798.

© 2014 Macmillan Publishers Limited Molecular Psychiatry (2014), 1010 – 1016 Bipolar GWAS accounting for BMI identifies TCF7L2 SJ Winham et al 1016 37 Zhang F, Phiel CJ, Spece L, Gurvich N, Klein PS. Inhibitory phosphorylation 45 Heuberger J, Birchmeier W. Interplay of cadherin-mediated cell adhesion and of glycogen synthase kinase-3 (GSK-3) in response to lithium - Evidence for canonical Wnt signaling. Cold Spring Harb Perspect Biol 2010; 2: a002915. autoregulation of GSK-3. J Biol Chem 2003; 278: 33067–33077. 46 Blair IP, Chetcuti AF, Badenhop RF, Scimone A, Moses MJ, Adams LJ et al. 38 O'Brien WT, Huang J, Buccafusca R, Garskof J, Valvezan AJ, Berry GT et al. Positional cloning, association analysis and expression studies provide conver- Glycogen synthase kinase-3 is essential for beta-arrestin-2 complex formation and gent evidence that the cadherin gene FAT contains a bipolar disorder lithium-sensitive behaviors in mice. J Clin Invest 2011; 121: 3756–3762. susceptibility allele. Mol Psychiatry 2006; 11:372–383. 39 Beaulieu JM, Marion S, Rodriguiz RM, Medvedev IO, Sotnikova TD, Ghisi V et al. 47 Soronen P, Ollila HM, Antila M, Silander K, Palo OM, Kieseppa T et al. Replication of A beta-arrestin 2 signaling complex mediates lithium action on behavior. Cell GWAS of bipolar disorder: association of SNPs near CDH7 with bipolar disorder 2008; 132: 125–136. and visual processing. Mol Psychiatry 2010; 15:4–6. 40 Gould TD, Einat H, O'Donnell KC, Picchini AM, Schloesser RJ, Manji HK. 48 Lydall GJ, Bass NJ, McQuillin A, Lawrence J, Anjorin A, Kandaswamy R et al. Beta-catenin overexpression in the mouse brain phenocopies lithium-sensitive Confirmation of prior evidence of genetic susceptibility to alcoholism in a Behaviors. Neuropsychopharmacology 2007; 32: 2173–2183. genome-wide association study of comorbid alcoholism and bipolar disorder. 41 Zhou R, Yuan P, Wang Y, Hunsberger JG, Elkahloun A, Wei Y et al. Evidence for Psychiatr Genet 2011; 21:294–306. selective microRNAs and their effectors as common long-term targets for the 49 Redies C, Hertel N, Hubner CA. Cadherins and neuropsychiatric disorders. Brain actions of mood stabilizers. Neuropsychopharmacology 2009; 34: 1395–1405. Res 2012; 1470: 130–144. 42 Brennand KJ, Simone A, Jou J, Gelboin-Burkhart C, Tran N, Sangar S et al. 50 Cichon S, Muhleisen TW, Degenhardt FA, Mattheisen M, Miro X, Strohmaier J et al. Modelling schizophrenia using human induced pluripotent stem cells. Nature Genome-wide association study identifies genetic variation in neurocan as a 2011; 473: 221–225. susceptibility factor for bipolar disorder. Am J Hum Genet 2011; 88:372–381. 43 Voleti B, Duman RS. The roles of neurotrophic factor and Wnt signaling in 51 Del Mare S, Salah Z, Aqeilan RI. WWOX: Its Genomics, Partners, and Functions. depression. Clin Pharmacol Ther 2012; 91: 333–338. J Cell Biochem 2009; 108: 737–745. 44 Gottardi CJ, Wong E, Gumbiner BM. E-cadherin suppresses cellular transformation 52 Bouteille N, Driouch K, Hage PE, Sin S, Formstecher E, Camonis J et al. Inhibition of by inhibiting beta-catenin signaling in an adhesion-independent manner. J Cell the Wnt/beta-catenin pathway by the WWOX tumor suppressor protein. Onco- Biol 2001; 153: 1049–1059. gene 2009; 28: 2569–2580.

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