ORIGINAL ARTICLE Dense SNP association study for bipolar I disorder on chromosome 18p11 suggests two loci with excess paternal transmission JG Mulle1, MD Fallin2, VK Lasseter3, JA McGrath3, PS Wolyniec3 and AE Pulver3 1McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins School of Medicine, Baltimore, MD, USA; 2Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA and 3Department of Psychiatry and Behavioral Sciences, Johns Hopkins School of Medicine, Baltimore, MD, USA Parent-of-origin effects have been implicated as mediators of genetic susceptibility for a number of complex disease phenotypes, including bipolar disorder. Specifically, evidence for linkage on chromosome 18 is modified when allelic parent-of-origin is accommodated in the analysis. Our goal was to characterize the susceptibility locus for bipolar I disorder on chromosome 18p11 and investigate this parent-of-origin hypothesis in an association context. This was achieved by genotyping single nucleotide polymorphisms (SNPs) at a high density (1 SNP/5 kb) along 13.6 megabases of the linkage region. To increase our ability to detect a susceptibility locus, we restricted the phenotype definition to include only bipolar I probands. We also restricted our study population to Ashkenazi Jewish individuals; this population has characteristics of a genetic isolate and may therefore facilitate detection of variants for complex disease. Three hundred and forty-four pedigrees (363 parent/child trios) where probands were affected with bipolar 1 disorder were genotyped. Transmission disequilibrium test analysis revealed no statistically significant association to SNPs or haplotypes within this region in this sample. However, when parent-of-origin of transmitted SNPs was taken into account, suggestive association was revealed for two separate loci. Molecular Psychiatry (2007) 12, 367–375. doi:10.1038/sj.mp.4001916; published online 31 October 2006 Keywords: bipolar disorder; genetic association; 18p11; psychiatric genetics; parent of origin; transmission disequilibrium Introduction Bipolar I disorder is a severe mood disorder char- acterized by episodes of both mania and depression. Lifetime prevalence of bipolar I disorder is estimated at 0.4–1.6%; approximately 10% of bipolar I patients commit suicide (Diagnostic and Statistical Manual of Mental Disorder (DSM IV)). Annual costs for bipolar disorder are estimated at $24–45 billion dollars.1 Despite this heavy burden of disease, the underlying pathology of bipolar disorder remains poorly under- stood. A greater understanding of disease etiology is imperative for the development of successful treat- ment and management options. There is substantial evidence that genetic factors contribute to the development of bipolar disorders. Heritability is estimated between 47 and 89%,2 and heritability of specifically bipolar 1 disorder has been estimated to be 93%.3 Despite this high heritability, segregation analyses do not support Mendelian para- digms of inheritance. The complex inheritance of bipolar disorder, as well as phenotypic and etiologic heterogeneity, implies that multiple susceptibility loci may exist. Numerous whole-genome linkage studies have been modestly successful in identifying regions of the genome that may harbor susceptibility loci, including six studies that find significant or suggestive evidence for linkage to chromosome 18p11.4–9 A meta-analysis of 18 whole-genome linkage scans of bipolar disorder identified chromosome 18 as nominally significant;10 another meta-analysis of 11 studies found the 18p region significant at P < 0.003.11 There is also evi- dence for a parent-of-origin effect on linkage evidence at the 18p11 locus.5,9,12 It is important to note that many studies have failed to find linkage evidence to chromosome 18p11. The lack of agreement in findings from genetic studies may be attributable to etiologic heterogeneity of bipolar disorder; it may also be due to lack of consistency in ascertainment or diagnostic methods, or to low statistical power, as many linkage studies have a small number of pedigrees for analysis. A number of candidate genes on chromosome 18p11 have been investigated for association with genetic markers, including NAPG, PACAP, MC5R, IMPA2, NDUFV2, GNAL and CLUL113–21 (Table 1). Received 30 August 2005; revised 24 January 2006; accepted 30 January 2006; published online 31 October 2006 Correspondence: Dr AE Pulver, Department of Psychiatry and Behavioral Sciences, Johns Hopkins School of Medicine, 1820 Lancaster Street, Suite #300, Baltimore, MD 21231, USA. E-mail: aepulver@jhmi.edu Molecular Psychiatry (2007) 12, 367–375 & 2007 Nature Publishing Group All rights reserved 1359-4184/07 $30.00 www.nature.com/mp Some of these studies show promising genetic- association evidence, and await replication and biology-based evidence for involvement in disease. The majority of candidate gene studies have utilized a case–control design; testing of a parent-of-origin hypothesis is not possible in these studies. Lack of consistency among linkage and association findings for bipolar disorder is complicated by the definition of affection status. Many linkage studies do not distinguish between bipolar I, bipolar II, unipolar depression and schizoaffective disorder; often, all are considered affected. A reduction in heterogeneity may be achieved by restricting the investigated phenotype to include a single and more reliable diagnosis. The investigation of disease in an isolated population may also reduce heterogeneity and facilitate identifica- tion of susceptibility variants. Because isolated popula- tions arise from a limited number of founders, genetic variation is reduced and the probability that affected individuals share a common susceptibility locus is increased. The investigation of isolated populations for complex genetic disease has been previously successful for both linkage and association for complex diseases. For example, the EDNRB locus for Hirschsprung disease was discovered by linkage in Mennonite kindreds, and mutations in the gene were also first identified in the Mennonite population.22,23 Subse- quently, it has been shown that mutation in the EDNRB gene is responsible for 5% of Hirschsprung disease in outbred populations.24 Recent studies in the Icelandic population have identified associated regions or spe- cific variants for prostate cancer and type II diabetes; these associations have been replicated in non-Icelan- dic populations.25,26 In the Ashkenazi Jewish popula- tion, variation in the genes MSH2, BRCA1 and BRCA2 contribute to the burden of prostate, breast and colon cancer in the Ashkenazim as well as outbred popula- tions.27–30 We have elected to focus on the Ashkenazi Jewish population, which has arisen from a limited number of founders and where endogamous mating is culturally reinforced. We have previously investigated evidence for linkage in 41 Ashkenazi pedigrees multiplex for bipolar disorder (I and II). The results of this scan were suggestive for linkage to chromosome 18p11; this linkage evidence increased when analysis was restricted to a diagnosis of BPI only (22 pedigrees), and is due to excess sharing of paternal alleles among affected relatives in the 18p11 region.9 Our goal in the present study was to fine map the linkage region with informative SNPs, and further investigate the parent- of-origin effect observed in our linkage study. To this end, we ascertained 344 pedigrees (363 BPI trios) all of Ashkenazi descent. This sample was genotyped for 2777 SNP markers in a 13.6 Mb region on chromo- some 18p11. We find suggestive association evidence for paternally transmitted SNPs. Further investigation and close scrutiny of this region is warranted. Materials and methods Study subjects Ashkenazi Jewish individuals affected with BPI were recruited nationally over a 6-year period through Table 1 Candidate genes studied for bipolar disorder on chromosome 18p11 Candidate gene Position, build 34 Study, year Sample size Parent-of- origin analyzed? No. of markers tested Evidence for association NAPG 10.5 Weller et al.21 460 BPI cases, 191 controls Not possible 5 P = 0.027, single SNP genotypic test, uncorrected ADCYAP1 (PACAP) 0.9 Ishiguro et al.15 86 bipolar (I and II) cases, and 251 controls Not possible 1 No association GNAL 11.7 Tsiouris et al.13 224 trios from 106 families Yes 1 (11 alleles) No association; no parent- of-origin association MC5R 13.8 Lohoff and Berrettini20 345 BPI cases, 275 controls Not possible 3 No association IMPA2 11.9 Sjoholt et al.18 44 Norwegian bipolar cases, 47 controls Not possible 10 0.03, single SNP, uncorrected 92 Arab BPI trios No 10 0.002, single SNP, uncorrected Yoshikawa et al.14 22 bipolar cases; 96 bipolar cases, 59 controls Not possible 3 0.051, uncorrected NDUFV2 9.1 Washizuka et al.;19 Washizuka et al.17 189 bipolar (I and II) Japanese cases, 222 controls Not possible 6 0.0001 (4-SNP haplotype), corrected 104 bipolar (I and II) case–parent trios No 9 0.11 (2-SNP haplotype), corrected CLUL1 0.6 McInnes et al.16 227 (nonindependent) BPI cases, 52 controls No 30 0.01, uncorrected Bipolar I association study on Chr 18p11 JG Mulle et al 368 Molecular Psychiatry advertisements in newspapers and Jewish newslet- ters, talks to community organizations, letters to leaders of the Jewish community and a study Web site (Johns Hopkins Epidemiology/Genetics Program in Psychiatry Home Page, www.hopkinsmedicine. org/epigen). Case–parent trios were eligible for inclu- sion in these analyses if: the proband met DSM-IV criteria for a BPI diagnosis, both parents were available for DNA collection, and all four grand- parents were of Ashkenazi Jewish descent. This recruitment effort led to 344 eligible families for analysis (363 trios), including 24 families (39 trios) from our previous linkage study. Forty-eight percent of affected individuals in our study were male; mean age of onset for BPI was 18.9 years. Probands were assessed for psychiatric illness according to established procedure, as follows: a trained clinical examiner (doctoral-level clinical psychologist) interviewed each study subject in person (usually in their home) using the Diagnostic Interview for Genetic Studies (DIGS), a standard, semi-structured instrument widely used for pheno- typic classification of individuals for psychiatric genetic studies. These interviews were tape-recorded. For each study subject, medical records were also obtained, and collateral interviews with at least one additional informant were conducted. The examiner then completed a written diagnostic workup, includ- ing relevant clinical features and course of illness. All collected clinical information (tape-recorded inter- views, DIGS interview booklets, medical records, collateral interview information and written diagnos- tic workup) was forwarded to two independent, trained clinicians, who reached a consensus diagno- sis of bipolar I disorder before an individual was included for study. Complete details about clinical methods are available in Fallin et al.9 We assessed power in a post hoc manner by first accounting for the number of tests done while incorporating the LD structure of this region in our sample, as follows: all SNPs in an LD block (defined as r2 > 0.8) were regarded as a single test per block (524 blocks, 1883 SNPs). All remaining ‘orphan’ SNPs were considered independent tests (n = 894). By these rules, 1418 independent single SNP tests were conducted in this experiment, requiring a Bonferroni- corrected P-value of 0.000034 for statistical signifi- cance. QUANTO31,32 was then used to calculate power for varying relative risks and disease allele frequencies, given a sample size of 363 trios and an alpha level of 0.000034. Genotyping strategy On chromosome 18p11, 13.6 Mb of the bipolar linkage region was selected for dense SNP genotyping. SNPs were genotyped using the Illumina BeadArray geno- typing technology,33 which has been shown to have a high degree of assay conversion, a low rate of missing data and error rates of less than one error in 1000 genotypes (Altshuler et al.,34 Supplemental Tables 2 and 3). SNP selection. At the time SNPs were selected for this experiment, the HapMap Consortium genotyping project was at an early stage; ‘tagging’ methods had not been developed nor validated. We therefore developed our own criteria for SNP selection, balancing our desire to obtain maximal information with efficiency and cost. As a first step, all SNPs in dbSNP that mapped to the first 13.6 Mb of chromosome 18p11 were vetted using an algorithm proprietary to Illumina, which calculates a ‘design score’ for each SNP (ranging from 0 to 1); this is approximately the probability that an assay will successfully generate data in the Illumina system. SNPs with a design score above 0.6 were considered for assay development. SNPs with reported minor allele frequency below 0.05 were excluded. Of SNPs that met these criteria, those that were in the HapMap or had been deposited in dbSNP by Perlegen were preferentially selected for genotyping at a maximum density of one SNP every 1.8 kb, though there were many regions where large gaps existed. This scaffold was then supplemented from the remaining pool of SNPs at variable density depending on the partial LD information available from HapMap. In regions of low LD, we selected SNPs at a maximum density of one SNP every kb; in regions of high LD maximum selection density was one SNP every 2 kb. This resulted in 3072 SNPs chosen for assay development with an average spacing of one SNP every 4.4 kb. The largest gap between SNPs was 26.5 kb. Seventy-seven genes are cataloged in RefSeq in this region (Pruitt et al.,35 http://www.ncbi.nlm.nih.gov/RefSeq/). One thousand two hundred and thirty-nine SNPs were selected for genotyping in these 77 transcripts. The average SNP density in transcribed sequence was one SNP every 4.2 kb. Analysis Data quality was assessed by (1) consistency of working assays, (2) genotype analysis of replicate samples and (3) the rate of Mendelian errors in families. For (2), genotype analyses of duplicate samples, four CEPH samples were used as controls. These four samples were genotyped on every 96-well plate (13 plates total). To assess the rate of Mendelian errors, family-based study samples were used. Men- delian errors underestimate the true genotyping error rate, because many genotyping errors are consistent with Mendelian inheritance. We therefore assumed that observed Mendelian errors represent one-third of the actual genotyping errors36 and adjusted the estimated error rate in this study accordingly. Hardy– Weinberg equilibrium was also tested for each SNP in the parental population; SNPs out of HWE at P < 0.001 were analyzed separately. Association was investigated in this study via the transmission disequilibrium test (TDT), where trans- mission of alleles from heterozygous parents to affected offspring are tested, and statistical signifi- cance is assessed using the TDT w2 test with 1 d.f.37 Adjusted P-values were assessed by permutation of Bipolar I association study on Chr 18p11 JG Mulle et al 369 Molecular Psychiatry transmitted alleles from heterozygous parents to affected offspring 10 000 times. Haplotypes were inferred using the Stephens– Smith–Donnelly algorithm as modified by Lin et al.,38,39 which accounts for missing data, can exploit inheritance information to generate accurate haplotypes between related individuals, can accu- rately phase between regions of high LD and can handle very large amounts of data.38–40 This algorithm is accurate over long distances (100–300 kb), particu- larly for family-based samples. After haplotypes were generated, ‘exhaustive allelic’ testing was accom- plished by investigating all haplotypes of every possible window size (linear combinations), using the TDT test (EA-TDT38,39). Statistical significance for haplotype-based tests was calculated unadjusted for multiple tests, and was also adjusted for multiple tests via permutation ( > 100 000 permutations). A recent report by Lin et al.41 suggested that evidence for linkage to chromosome 18p11 is en- hanced in later-onset (age of onset X21) bipolar pedigrees. We explored whether late age at onset also led to detectable differences in association evidence by stratifying our sample and conducting associa- tion analysis on trios with probands < 21 (64.8% of probands) and X21 (33.8% of probands) separately. Seven probands lacked information about age-at onset and were excluded from this analysis. Parent-of-origin analyses were conducted in multi- ple ways. To assess whether allelic parent-of-origin was relevant to association results, informative transmissions/nontransmissions were stratified by source parent for all SNPs, and the TDT was recalculated. By this method, four SNPs showed excess paternal transmission at P < 0.0001 (unad- justed), with no excess maternal transmission. Stra- tified TDT can sometimes give spurious results, particularly in the presence of (a) maternal effects or (b) skewed distribution of genotypes between parental classes. We therefore assessed these four SNPs for parent-of-origin association in two additional ways. First, we performed the TAT as suggested by Wein- berg, which is similar to the TDT but excludes double heterozygote mating pairs.42 Next, we formulated a conditional logistic regression matched by parental- child genotype combination, and included an inter- action term for paternal origin of risk allele, such as was used by Le Stunff et al.:43 ln PðtÞ 1 � PðtÞ � � ¼ aA þ bA � F where outcome is probability of transmission, A = 1 if putative risk allele is transmitted for a given parent– child pair, 0 otherwise, and F = 1 if the transmission was paternally informative. Parent-of-origin effects are assessed by a likelihood ratio test between models with and without the interaction term. This analysis has the added advantage of allowing for estimation of the odds ratio for paternal inheritance. Analysis was carried out using standard statistical software (STA- TA/SE 8.2 for windows). Results Three thousand and seventy-two SNP assays were attempted in this study; 183 (5.96%) failed to generate data. Of the remaining 2889 SNPs, 100 were mono- morphic and 12 were out of Hardy–Weinberg equili- brium at P < 0.001, leaving 2777 SNPs for analysis. The estimated error rate for these SNPs, as assessed by presence of Mendelian errors and replicate analysis, is 2.8 and 3.3 errors per 100 000 genotypes, respec- tively. This implies there are between 95 and 112 errors in 3.3 million genotypes in this study. Genotyping was attempted on 1167 unique sam- ples. Genotyping was unsuccessful on four indivi- duals, as assessed by missing data rates and repeated poor clustering of genotype data. One pedigree had > 10% of genotypes with Mendelian errors, and was subsequently removed from analysis due to suspected non-paternity or sample mislabeling (four individuals removed). Analysis We estimated the power of this study in a post hoc way to account for the number of independent tests conducted in this study, given 363 trios tested and the LD structure of our sample. For experiment-wide significance, a P-value of 0.000034 is required (1418 independent tests). This sample has 80% power to detect genetic relative risks X2.2 if disease allele frequency is 10%, given a log-additive genetic model. If disease allele frequency is 50%, this study is powered to detect genetic relative risks X1.8. Transmissions of alleles from heterozygous parents to affected offspring were tabulated, and the TDT statistic was calculated for individual SNPs (Figure 1). The marker with the strongest evidence for association was rs541039, position 4265455 (185 transmitted vs 125 untransmitted, P < 0.0008, permu- tation-adjusted P < 0.86). No SNP in this analysis had an adjusted P-value < 0.05. Details about the 10 SNPs with the highest evidence for association are provided (Table 2). Haplotypes were generated, and transmission of all possible contiguous SNP haplotypes of all possible window sizes was assessed via the TDT. A 14 SNP haplotype of size 120 kb, midpoint at 6121080, was the most strongly associated (59 transmitted: 20 untrans- mitted, unadjusted P < 0.000013; adjusted P < 0.38). No compelling evidence for differential association related to age at onset ( > 20 vs < 21) of bipolar disorder was observed. Parent-of-origin effects were first investigated by stratifying informative transmissions/nontransmis- sions by source parent for all SNPs. This method revealed four SNPs with paternal transmission dis- tortion at P < 0.0001 (unadjusted for multiple tests), and no excess maternal transmission. Three of these SNPs (rs1893157, rs494412, rs620504) map within 13 kb, and are correlated with one another (all r2 > 0.7). For these SNPs, the minor allele is paternally overtransmitted. (These three SNPs are approximately Bipolar I association study on Chr 18p11 JG Mulle et al 370 Molecular Psychiatry equally informative for a single putative susceptibility variant; to avoid redundancy we report detailed results for rs1893157 only.) The remaining SNP (rs789024) is 1.2 Mb telomeric to this locus, and is not correlated with the others (all r2 = 0). We therefore regard rs789024 as an independent putative suscept- ibility locus. These two loci map within D18S59 and D18S63, two microsatellite markers that describe the peak linkage region of paternal allele sharing pre- viously reported in this population9 (Figure 2). The transmission asymmetry test (TAT), as sug- gested by Weinberg, excludes mating pairs of two heterozygotes in order to avoid bias from these matings, which contribute twice to a stratified TDT.42 After this exclusion, excess paternal transmis- sion is still observed for both loci (Table 3a). Next, we formulated a conditional logistic regression matched by parental mating type, and included an interaction term for paternal origin of risk allele, such as was used by Le Stunff et al.43 Likelihood ratio tests for the parent-of-origin interaction term were significant at P < 0.001 for both loci tested (Table 3b). For SNP rs789024, the estimated odds ratio for developing bipolar disorder when the risk allele is paternally transmitted is 4.27 (CI 2.00–9.10), relative to maternal transmission. For SNP rs1893157, estimated odds ratio for bipolar disorder given paternal transmission is 6.53 (CI 2.22–19.20) relative to maternal transmis- sion of the risk allele. SNP rs789024 is out of Hardy–Weinberg equili- brium; there are 11% too many heterozygotes com- pared to what is expected. Excess heterozygosity is consistent for both parent populations. To glean additional information about this SNP, we examined the genotypic distribution in the four HapMap populations. The trend of aberrant heterozygosity is also observed; there are 6.92% too many hetero- zygotes in the CEU population, and 8.66% too few heterozygotes in the CHB population. This trend is not observed for SNP rs1893157 (Figure 3). Discussion The aim of this study was to characterize the linkage region for bipolar disorder on chromosome 18p11, and investigate the paternal parent-of-origin hypoth- esis for susceptibility, as was suggested by an earlier linkage study in the Ashkenazi Jewish population.9 In this study, our sample of 363 trios had 80% power at an alpha of 0.000034 to detect risk ratios of at least 1.8–2.2-fold, at a disease allele frequency of 50–10%, respectively and a log-additive model of genetic risk. Without taking parent-of-origin into account, our inability to detect a statistically significant associa- tion implies that a common allele that confers this magnitude of risk does not exist in the 18p11 region. There is evidence for a parent-of-origin effect at the 18p11 locus from a linkage study in the Ashkenazi Jewish population,9 as well as other populations.5,12 Parent-of-origin effects imply an epigenetic mechanism of inheritance and may be a compelling hypothesis to explain complex genetic disorders such Figure 1 Distribution of association evidence, allelic TDT test. Uncorrected P-values are plotted by chromosomal position for each SNP tested. Highest evidence for asso- ciation is at rs541039, position 4,265,455, 184 T vs 125 U, uncorrected P-value 0.0008, permuted P-value 0.86. Table 2 Ten most associated SNPs, allelic TDT SNP Position T:U w2 P-value rs541039 4265455 184:125 11.265 0.0008 rs11080994 2887993 222:162 9.375 0.0022 rs1893157 2257456 81:47 9.031 0.0027 rs10853370 8960615 204:149 8.569 0.0034 rs551920 4261385 100:63 8.399 0.0038 rs6506386 6437482 177:127 8.224 0.0041 rs16952222 7598051 51:26 8.117 0.0044 rs12607596 1732207 193:141 8.096 0.0044 rs9635859 4747894 91:57 7.811 0.0052 rs1940594 5872643 125:85 7.619 0.0058 Figure 2 TDT results, paternal transmission only. Trans- mission from paternal heterozygotes only was tabulated for each SNP. Unadjusted P-values are plotted for each SNP tested. D18S59 and D18S63 are microsatellite markers that previously showed evidence for excess paternal allele sharing among sibs affected with BPI.9 Bipolar I association study on Chr 18p11 JG Mulle et al 371 Molecular Psychiatry as bipolar disorder.44–46 The paternal allele sharing in the Fallin et al.9 linkage study is described by microsatellite markers D18S59 and D18S63; the two loci (four SNPs) with putative parent-of-origin effects in this study map between these two markers and is therefore consistent with previous results. These loci confer substantial risk for bipolar disorder (fourfold and sixfold respectively, relative to maternal trans- mission). They may represent two different regions that each control the same transcript, two different Table 3a Transmission of alleles to bipolar I offspring, stratified by parent-of-origin of alleles rs789024 rs1893157 T U P-value Tau T U P-value Tau Stratified TDT Paternal 77 35 0.00007 0.69 Paternal 52 19 0.00009 0.73 Maternal 60 57 0.78 0.51 Maternal 24 23 0.88 0.51 Totala 137 92 0.003 0.60 Totala 76 42 0.002 0.64 TAT Paternal 45 24 0.01 0.65 Paternal 43 17 0.0008 0.72 Maternal 32 42 0.24 0.43 Maternal 21 19 0.75 0.53 Total 77 66 0.35 0.54 Total 64 36 0.005 0.64 aExcludes trios where parents and off spring are all heterozygous, as these are not informative. Table 3b Test for parent-of-origin effects Model Estimates �2LL w2 LRTw2 d.f. P-value a (s.e.) b (s.e.) rs789024 Allele effect (A) 0.79 (0.18) 589.32 Allele and PO (A þ A*F) 0.10 (0.26) 1.45 (0.37) 574.90 PO 14.41 1 0.0001 rs1893157 Allele effect (A) 1.15 (0.26) 285.55 Allele and PO (A þ A*F) 0.08 (0.41) 1.87 (0.55) 273.73 PO 11.82 1 0.0006 Figure 3 Genotype frequencies and deviation from expected heterozygosity for two paternally transmitted SNP are shown for Ashkenazi Jewish parents, Ashkenazi maternal and paternal populations separately, and Hapmap populations. Bipolar I association study on Chr 18p11 JG Mulle et al 372 Molecular Psychiatry regions that are related to two different susceptibility genes, or one or both may be false positives. A note about parent-of-origin transmission tests for association: As these methods are only now emerging, the methods to apply empirical P-value estimation in this framework are lacking. The simple permutation of transmission status used for standard TDT analyses would not provide the appropriate null distribution for hypothesis of parent-of-origin. Therefore, we caution the reader to interpret the distribution-based P-values we provide with the understanding that they are based on a w2 assumption. One of the SNPs with excess paternal transmission, rs789024, is out of Hardy–Weinberg equilibrium, with 11% too many heterozygotes observed. This may be due to chance, however, this trend is also observed in two of the four HapMap populations. This aberrant genotypic distribution may be a consequence of selection, implying that rs789024 or a linked variant is involved in susceptibility for bipolar disorder. It is also possible this excess heterozygosity represents genotyping error, perhaps owing to a yet undetected segmental duplication in the human genome. The other SNP with excess paternal transmission, rs1893157, is consistently in Hardy–Weinberg equili- brium, but has a wide spectrum of allele frequencies between populations. This too may represent the action of selection. In this study, we have pursued a family-based study design. Family-based samples are not without draw- backs; they require more collection effort and in- creased genotyping costs compared to case–control designs. However, these drawbacks are balanced by two important advantages: first, it is possible to investigate parent-of-origin hypotheses. Second, fa- mily-based study designs are robust to interference from population stratification. It has been shown that population stratification can mask true positive findings, and cryptic substructure can create false positives.47,48 Parent-of-origin effects have been shown to be important for a number of genetic diseases, usually through an imprinting mechanism. At least eight syndromes have been shown to involve aberrant transmission or regulation of imprinted genes.49 For example, Prader–Willi syndrome50 involves either loss of the paternally transmitted region of chromo- some 15q11–13, or maternal uniparental disomy (UPD) of this same region. Angelman’s syndrome,50 a distinct phenotype, results from loss of maternal 15q11–13 or paternal UPD. Several genes in the critical region on chromosome 15q11–13 including UBE3A, SNRPN, PAR1, PAR5 and PAR7, have shown parent-of-origin-specific expression. Loss of parent- of-origin expression is principally responsible for disease. It is likely that for complex diseases such a bipolar disorder, a similar mechanism contributes to susceptibility. There has not yet been reported evidence for an imprinted region on chromosome 18p11. However, a recent study finds evidence of parent-of-origin- dependent expression for an alternative transcript of GNAL, which encodes Golf, the alpha subunit of a heretotrimeric stimulatory G-protein that may func- tionally associate with the dopamine D1 receptor.51 GNAL has previously been investigated for parent-of- origin effects in bipolar disorder, and no association has been reported; however, this study tested a single marker in a small number of trios.13 GNAL maps more than 10 Mb distal to the loci reported in this study. It is unlikely that the loci we report here are responsible for control of GNAL expression; 800 kb is the greatest distance yet reported between a regulatory element and its target gene.52 The mitochondrial subunit gene NDUVF2 has been reported to be associated with both bipolar disorder and schizophrenia,17,19,53 as has IMPA2,14,18 though these loci are also located 8–10 Mb distal to the loci reported herein. Parent-of-origin associations have not been investigated at these loci. Of candidate genes investigated, CLUL and ADCYAP (PACAP) are located within 1 Mb of the paternally transmitted SNPs we report herein. These loci have been investigated previously in a case–control study design, and evidence for association has been margin- al (CLUL116) or nonexistent (PACAP15). It is possible that accounting for parent-of-origin would strengthen association evidence. The results of the current study suggest a reexamination of these loci; of particular interest is whether parent-of-origin-specific expres- sion of these genes can be detected. The results of this study suggest that parent-of- origin of transmission plays a significant role in susceptibility to bipolar disorder for variants on chromosome 18p11. This statistical evidence must be supplemented with biological information about candidate loci near SNPs reported in this paper; further experimentation in this region is warranted. References 1 Kleinman L, Lowin A, Flood E, Gandhi G, Edgell E, Revicki D. Costs of bipolar disorder. Pharmacoeconomics 2003; 21: 601–622. 2 McGuffin P, Rijsdijk F, Andrew M, Sham P, Katz R, Cardno A. The heritability of bipolar affective disorder and the genetic relation- ship to unipolar depression. Arch Gen Psychiatry 2003; 60: 497–502. 3 Kieseppa T, Partonen T, Haukka J, Kaprio J, Lonnqvist J. High concordance of bipolar I disorder in a nationwide sample of twins. Am J Psychiatry 2004; 161: 1814–1821. 4 Berrettini WH, Ferraro TN, Goldin LR, Weeks DE, Detera-Wadleigh S, Nurnberger Jr JI et al. Chromosome 18 DNA markers and manic- depressive illness: evidence for a susceptibility gene. Proc Natl Acad Sci USA 1994; 91: 5918–5921. 5 Stine OC, Xu J, Koskela R, McMahon FJ, Gschwend M, Friddle C et al. Evidence for linkage of bipolar disorder to chromosome 18 with a parent-of-origin effect. Am J Hum Genet 1995; 57: 1384–1394. 6 Lin JP, Bale SJ. Parental transmission and D18S37 allele sharing in bipolar affective disorder. Genet Epidemiol 1997; 14: 665–668. 7 Detera-Wadleigh SD, Badner JA, Berrettini WH, Yoshikawa T, Goldin LR, Turner G et al. A high-density genome scan detects evidence for a bipolar-disorder susceptibility locus on 13q32 and other potential loci on 1q32 and 18p11.2. Proc Natl Acad Sci USA 1999; 96: 5604–5609. 8 Bennett P, Segurado R, Jones I, Bort S, McCandless F, Lambert D et al. The Wellcome trust UK–Irish bipolar affective disorder Bipolar I association study on Chr 18p11 JG Mulle et al 373 Molecular Psychiatry sibling-pair genome screen: first stage report. Mol Psychiatry 2002; 7: 189–200. 9 Fallin MD, Lasseter VK, Wolyniec PS, McGrath JA, Nestadt G, Valle D et al. Genomewide linkage scan for bipolar-disorder susceptibility loci among Ashkenazi Jewish families. Am J Hum Genet 2004; 75: 204–219. 10 Segurado R, Detera-Wadleigh SD, Levinson DF, Lewis CM, Gill M, Nurnberger Jr JI et al. Genome scan meta-analysis of schizophrenia and bipolar disorder, part III: bipolar disorder. Am J Hum Genet 2003; 73: 49–62. 11 Badner JA, Gershon ES. Meta-analysis of whole-genome linkage scans of bipolar disorder and schizophrenia. Mol Psychiatry 2002; 7: 405–411. 12 Nothen MM, Cichon S, Rohleder H, Hemmer S, Franzek E, Fritze J et al. Evaluation of linkage of bipolar affective disorder to chromosome 18 in a sample of 57 German families. Mol Psychiatry 1999; 4: 76–84. 13 Tsiouris SJ, Breschel TS, Xu J, McInnis MG, McMahon FJ. Linkage disequilibrium analysis of G-olf alpha (GNAL) in bipolar affective disorder. Am J Med Genet 1996; 67: 491–494. 14 Yoshikawa T, Padigaru M, Karkera JD, Sharma M, Berrettini WH, Esterling LE et al. Genomic structure and novel variants of myo- inositol monophosphatase 2 (IMPA2). Mol Psychiatry 2000; 5: 165–171. 15 Ishiguro H, Ohtsuki T, Okubo Y, Kurumaji A, Arinami T. Association analysis of the pituitary adenyl cyclase activating peptide gene (PACAP) on chromosome 18p11 with schizophrenia and bipolar disorders. J Neural Transm 2001; 108: 849–854. 16 McInnes LA, Service SK, Reus VI, Barnes G, Charlat O, Jawahar S et al. Fine-scale mapping of a locus for severe bipolar mood disorder on chromosome 18p11.3 in the Costa Rican population. Proc Natl Acad Sci USA 2001; 98: 11485–11490. 17 Washizuka S, Kakiuchi C, Mori K, Kunugi H, Tajima O, Akiyama T et al. Association of mitochondrial complex I subunit gene NDUFV2 at 18p11 with bipolar disorder. Am J Med Genet B Neuropsychiatr Genet 2003; 120: 72–78. 18 Sjoholt G, Ebstein RP, Lie RT, Berle JO, Mallet J, Deleuze JF et al. Examination of IMPA1 and IMPA2 genes in manic-depressive patients: association between IMPA2 promoter polymorphisms and bipolar disorder. Mol Psychiatry 2004; 9: 621–629. 19 Washizuka S, Iwamoto K, Kazuno AA, Kakiuchi C, Mori K, Kametani M et al. Association of mitochondrial complex I subunit gene NDUFV2 at 18p11 with bipolar disorder in Japanese and the National Institute of Mental Health pedigrees. Biol Psychiatry 2004; 56: 483–489. 20 Lohoff FW, Berrettini WH. Lack of association between variations in the melanocortin 5 receptor gene and bipolar disorder. Psychiatr Genet 2005; 15: 255–258. 21 Weller AE, Dahl JP, Lohoff FW, Ferraro TN, Berrettini WH. Analysis of variations in the NAPG gene on chromosome 18p11 in bipolar disorder. Psychiatr Genet 2006; 16: 3–8. 22 Puffenberger EG, Hosoda K, Washington SS, Nakao K, deWit D, Yanagisawa M et al. A missense mutation of the endothelin-B receptor gene in multigenic Hirschsprung’s disease. Cell 1994; 79: 1257–1266. 23 Puffenberger EG, Kauffman ER, Bolk S, Matise TC, Washington SS, Angrist M et al. Identity-by-descent and association mapping of a recessive gene for Hirschsprung disease on human chromosome 13q22. Hum Mol Genet 1994; 3: 1217–1225. 24 Chakravarti A. Endothelin receptor-mediated signaling in hirsch- sprung disease. Hum Mol Genet 1996; 5: 303–307. 25 Amundadottir LT, Sulem P, Gudmundsson J, Helgason A, Baker A, Agnarsson BA et al. A common variant associated with prostate cancer in European and African populations. Nat Genet 2006; 38: 652–658. 26 Grant SF, Thorleifsson G, Reynisdottir I, Benediktsson R, Mano- lescu A, Sainz J et al. Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nat Genet 2006; 38: 320–323. 27 Struewing JP, Hartge P, Wacholder S, Baker SM, Berlin M, McAdams M et al. The risk of cancer associated with specific mutations of BRCA1 and BRCA2 among Ashkenazi Jews. N Engl J Med 1997; 336: 1401–1408. 28 Neuhausen SL, Godwin AK, Gershoni-Baruch R, Schubert E, Garber J, Stoppa-Lyonnet D et al. Haplotype and phenotype analysis of nine recurrent BRCA2 mutations in 111 families: results of an international study. Am J Hum Genet 1998; 62: 1381–1388. 29 Foulkes WD, Thiffault I, Gruber SB, Horwitz M, Hamel N, Lee C et al. The founder mutation MSH2*1906G– > C is an important cause of hereditary nonpolyposis colorectal cancer in the Ashkenazi Jewish population. Am J Hum Genet 2002; 71: 1395–1412. 30 Lynch HT, Coronel SM, Okimoto R, Hampel H, Sweet K, Lynch JF et al. A founder mutation of the MSH2 gene and hereditary nonpolyposis colorectal cancer in the United States. JAMA 2004; 291: 718–724. 31 Gauderman WJ, Morrison JM. QUANTO 1.1: a computer program for power and sample size calculations for genetic-epidemiology studies, http://hydra.usc.edu/gxe, 2006. 32 Gauderman WJ. Sample size calculations for matched case–control studies of gene-environment interaction. Statistics in Medicine 2002; 21: 35–50. 33 Oliphant A, Barker DL, Stuelpnagel JR, Chee MS. BeadArray technology: enabling an accurate, cost-effective approach to high-throughput genotyping. Biotechniques Suppl 2002; 32: S56–S61. 34 Altshuler D, Brooks LD, Chakravarti A, Collins FS, Daly MJ, Donnelly P. A haplotype map of the human genome. Nature 2005; 437: 1299–1320. 35 Pruitt KD, Tatusova T, Maglott DR. NCBI Reference Sequence (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res 2005; 33(Database issue): D501–D504. 36 Douglas JA, Skol AD, Boehnke M. Probability of detection of genotyping errors and mutations as inheritance inconsistencies in nuclear-family data. Am J Hum Genet 2002; 70: 487–495. 37 Spielman RS, McGinnis RE, Ewens WJ. Transmission test for linkage disequilibrium: the insulin gene region and insulin- dependent diabetes mellitus (IDDM). Am J Hum Genet 1993; 52: 506–516. 38 Lin S, Chakravarti A, Cutler DJ. Exhaustive allelic transmission disequilibrium tests as a new approach to genome-wide associa- tion studies. Nat Genet 2004; 36: 1181–1188. 39 Lin S, Chakravarti A, Cutler DJ. Haplotype and missing data inference in nuclear families. Genome Res 2004; 14: 1624–1632. 40 Lin S, Cutler DJ, Zwick ME, Chakravarti A. Haplotype inference in random population samples. Am J Hum Genet 2002; 71: 1129– 1137. 41 Lin PI, McInnis MG, Potash JB, Willour VL, Mackinnon DF, Miao K et al. Assessment of the effect of age at onset on linkage to bipolar disorder: evidence on chromosomes 18p and 21q. Am J Hum Genet 2005; 77: 545–555. 42 Weinberg CR. Methods for detection of parent-of-origin effects in genetic studies of case–parents triads. Am J Hum Genet 1999; 65: 229–235. 43 Le Stunff C, Fallin D, Bougneres P. Paternal transmission of the very common class I INS VNTR alleles predisposes to childhood obesity. Nat Genet 2001; 29: 96–99. 44 Petronis A. Epigenetics and bipolar disorder: new opportunities and challenges. Am J Med Genet C Semin Med Genet 2003; 123: 65–75. 45 Bjornsson HT, Fallin MD, Feinberg AP. An integrated epigenetic and genetic approach to common human disease. Trends Genet 2004; 20: 350–358. 46 Jiang YH, Bressler J, Beaudet AL. Epigenetics and human disease. Annu Rev Genomics Hum Genet 2004; 5: 479–510. 47 Marchini J, Cardon LR, Phillips MS, Donnelly P. The effects of human population structure on large genetic association studies. Nat Genet 2004; 36: 512–517. 48 Campbell CD, Ogburn EL, Lunetta KL, Lyon HN, Freedman ML, Groop LC et al. Demonstrating stratification in a European American population. Nat Genet 2005; 37: 868–872. 49 Morison IM, Ramsay JP, Spencer HG. A census of mammalian imprinting. Trends Genet 2005; 21: 457–465. 50 OMIM Online Mendelian Inheritance in Man, OMIM (TM) McKusick–Nathans Institute for Genetic Medicine, Johns Hopkins University (Baltimore, MD) and National Center for Biotechnology Bipolar I association study on Chr 18p11 JG Mulle et al 374 Molecular Psychiatry Information (Bethesda, MD), world wide web URL: http:// www.ncbi.nlm.nih.gov/omim. 51 Corradi JP, Ravyn V, Robbins AK, Hagan KW, Peters MF, Bostwick R et al. Alternative transcripts and evidence of imprinting of GNAL on 18p11.2. Mol Psychiatry 2005; 10: 1017–1025. 52 Lettice LA, Horikoshi T, Heaney SJ, van Baren MJ, van der Linde HC, Breedveld GJ et al. Disruption of a long-range cis-acting regulator for Shh causes preaxial polydactyly. Proc Natl Acad Sci USA 2002; 99: 7548–7553. 53 Washizuka S, Kametani M, Sasaki T, Tochigi M, Umekage T, Kohda K et al. Association of mitochondrial complex I subunit gene NDUFV2 at 18p11 with schizophrenia in the Japanese population. Am J Med Genet B Neuropsychiatr Genet 2006; 141: 301–304. Bipolar I association study on Chr 18p11 JG Mulle et al 375 Molecular Psychiatry Dense SNP association study for bipolar I disorder on chromosome 18p11 suggests two loci with excess paternal transmission Introduction Materials and methods Study subjects Genotyping strategy SNP selection Analysis Results Analysis Discussion References