Sequence variation in DOCK9 and heterogeneity in bipolar disorder

Background Linkage of bipolar disorder to a broad region on chromosome 13q has been supported in several studies including a meta-analysis on genome scans. Subsequent reports have shown that variations in the DAOA (G72) locus on 13q33 display association with bipolar disorder but these may not account for all of the linkage evidence in the region. Objective To identify additional susceptibility loci on 13q32-q33 by linkage disequilibrium mapping and explore the impact of phenotypic heterogeneity on association. Methods In the initial phase, 98 single nucleotide polymorphism (SNPs) located on 13q32-q33 were genotyped on 285 probands with bipolar disorder and their parents were drawn from families in the NIMH Genetics Initiative consortium for bipolar disorder (NIMH1-4) and two other series. Fine scale mapping using one family series (NIMH1-2) as the test sample was targeted on a gene that displayed the highest evidence of association. A secondary analysis of familial component phenotypes of bipolar disorder was conducted. Results Three of seven SNPs in DOCK9, a gene that encodes an activator of the Rho-GTPase Cdc42, showed significant excess allelic transmission (P=0.0477−0.00067). Fine scale mapping on DOCK9 yielded evidence of association at nine SNPs in the gene (P=0.02−0.006). Follow-up tests detected excess transmission of the same allele of rs1340 in two out of three other sets of families. The association signals were largely attributable to maternally transmitted alleles (rs1927568: P=0.000083; odds ratio=3.778). A secondary analysis of familial component phenotypes of bipolar disorder detected significant association across multiple DOCK9 markers for racing thoughts, psychosis, delusion during mania and course of illness indicators. Conclusion These results suggest that DOCK9 contributes to both risk and increased illness severity in bipolar disorder. We found evidence for the effect of phenotypic heterogeneity on association. To our knowledge this is the first report to implicate DOCK9 or the Rho-GTPase pathway in the etiology of bipolar disorder.


Sequence variation in DOCK9 and heterogeneity in bipolar disorder
Sevilla D. Detera-Wadleigh a , Chun-yu Liu c , Manjula Maheshwari d , Imer Cardona a , Winston Corona a , Nirmala Akula a , C.J.M. Steele a , Judith A. Badner c , Mukta Kundu a , Layla Kassem a , James B. Potash b , NIMH Genetics Initiative for Bipolar Disorder Consortium e,f,g,h,i,j,k , Richard Gibbs d , Elliot S. Gershon c and Francis J. McMahon a Background Linkage of bipolar disorder to a broad region on chromosome 13q has been supported in several studies including a meta-analysis on genome scans. Subsequent reports have shown that variations in the DAOA (G72) locus on 13q33 display association with bipolar disorder but these may not account for all of the linkage evidence in the region.
Objective To identify additional susceptibility loci on 13q32-q33 by linkage disequilibrium mapping and explore the impact of phenotypic heterogeneity on association.
Methods In the initial phase, 98 single nucleotide polymorphism (SNPs) located on 13q32-q33 were genotyped on 285 probands with bipolar disorder and their parents were drawn from families in the NIMH Genetics Initiative consortium for bipolar disorder (NIMH1-4) and two other series. Fine scale mapping using one family series (NIMH1-2) as the test sample was targeted on a gene that displayed the highest evidence of association. A secondary analysis of familial component phenotypes of bipolar disorder was conducted.
Results Three of seven SNPs in DOCK9, a gene that encodes an activator of the Rho-GTPase Cdc42, showed significant excess allelic transmission (P = 0.0477 -0.00067). Fine scale mapping on DOCK9 yielded evidence of association at nine SNPs in the gene (P = 0.02 -0.006). Follow-up tests detected excess transmission of the same allele of rs1340 in two out of three other sets of families. The association signals were largely attributable to maternally transmitted alleles (rs1927568: P = 0.000083; odds ratio = 3.778).
A secondary analysis of familial component phenotypes of bipolar disorder detected significant association across multiple DOCK9 markers for racing thoughts, psychosis, delusion during mania and course of illness indicators.
Conclusion These results suggest that DOCK9 contributes to both risk and increased illness severity in bipolar disorder. We found evidence for the effect of phenotypic heterogeneity on association. To our knowledge this is the first report to implicate DOCK9 or the Rho-GTPase pathway in the etiology of bipolar disorder.

Introduction
Multiple genetic loci have been reported to predispose to bipolar disorder. As with other complex disorders, however, unclear convergence or frank divergence of independent studies has been a recurring problem. This is prominently illustrated by the largely divergent results from three meta-analyses on genome scans for linkage to bipolar disorder (Badner and Gershon, 2002;Segurado et al., 2003;McQueen et al., 2005). Key underlying causes of inconsistent results include genetic and phenotypic heterogeneity, risk-protective allele switches in different samples, gene-gene interaction, underpowered sample size, and contributions of as-yet undetermined environmental factors. could comprise many clinically similar conditions with differing genetic etiologies (Schulze and McMahon, 2004;MacQueen et al., 2005) that could create 'noise' and conceal true findings. Thus, inconsistent genetic findings between independent samples could arise from uneven representations of cases with different, but clinically similar phenotypes.
The poorly understood phenomenon of risk-protective allele switching previously reported in G72 (reviewed in Detera-Wadleigh and McMahon, 2006) and COMT (Funke et al., 2005) can mask association, as can genegene interaction. For example, epistatic effects generated through interaction of variations in serotonin transporter with variations in the dopamine transporter, 5-HT 1B and brain-derived neurotrophic factor have been documented in several studies in mice (Murphy et al., 2003).
We have undertaken a search for susceptibility genes for bipolar disorder on chromosome 13q. Linkage of bipolar disorder and schizophrenia to this region has been strongly supported in several studies (Ginns et al., 1996;Lin et al., 1997;Stine et al., 1997;Blouin et al., 1998;Shaw et al., 1998;Brzustowicz et al., 1999;Detera-Wadleigh et al., 1999;Badenhop et al., 2001;Kelsoe et al., 2001;Faraone et al., 2002;Liu et al., 2003;McInnis et al., 2003;Potash et al., 2003;Shaw et al., 2003;Abecasis et al., 2004;Park et al., 2004). Single nucleotide polymorphism (SNP) screening in the 13q33 region detected association in schizophrenia at the G72/G30 locus (Chumakov et al., 2002). Support for association in bipolar disorder was later found (Hattori et al., 2003) and similar results from various studies in schizophrenia and bipolar disorder further underscored the relevance of G72/G30 variations in disease risk (Detera-Wadleigh and McMahon, 2006). The broadness of the linkage peak suggested that other susceptibility loci may exist in 13q32-q33. To test this possibility we employed a mapping strategy that interrogated SNPs selected to sample common variation across a 7.6 Mb segment of 13q32.1-q33.1. We identified three SNPs associated with bipolar disorder that mapped to DOCK9, a gene that encodes an activator of the RhoGTPase, Cdc42. We further report that fine-scale mapping detected additional SNPs that showed association with various clinical features of the illness. These results implicate a novel pathway in the etiology of bipolar disorder and suggest that more than one gene may account for the genetic linkage of bipolar disorder to chromosome 13q.

Patient samples and phenome file
Family samples that were collected under the auspices of the National Institute of Mental Health (NIMH) Genetics Initiative for Bipolar Disorder, comprising the NIMH Waves 1-2, 3 and 4 were used in this study. Families were ascertained on the basis of a sibling pair affected with bipolar I (BPI) or schizoaffective-bipolar disorder (SA-BP). The genotyped sample contains complete trios, probands with or without an affected sibling, and some families with only one parent and an affected offspring, with or without an affected sibling. Waves 1 and 2 (NIMH1-2) consist of 153 families and were ascertained the earliest by a consortium of four sites (Nurnberger et al., 1997). Wave 3 (NIMH3) and Wave 4 (NIMH4) consist of 221 (Dick et al., 2003) and 275 families, respectively, and were ascertained by a consortium of 10 sites. Additional bipolar disorder family samples included the 'CNG' which consists of 22 multiplex families ascertained by the Clinical Neurogenetics Branch in the 1980s (Berrettini et al., 1991) and 73 multiplex families referred to as 'CHIP' collected by the Departments of Psychiatry at the University of Chicago and Johns Hopkins University and the Genetic Basis of Mood and Anxiety Disorders Unit at the NIMH Intramural Research Program.
Ascertainment and assessment of families were described previously (Berrettini et al., 1991;Simpson et al., 1992;Nurnberger et al., 1997;Dick et al., 2003). In the NIMH and the later portion of the CHIP samples, participants were assessed by use of the Diagnostic Instrument for Genetic Studies (Nurnberger et al., 1994) and diagnosis was based on DSM-IIIR and DSM-IV criteria (American Psychiatric Association, 1987. In the CNG and earlier portions of the CHIP samples, participants were assessed with the Schedule for Affective Disorders and Schizophrenia-Lifetime version  and diagnosed on the basis of Research Diagnostic Criteria . A file of phenotypic variables was generated by collating phenotype information on individual members of the NIMH Waves 1-4 families. This includes data on clinical features of mania and depression, course of illness indicators and comorbid psychiatric conditions. Familiality of variables was evaluated by use of a mixed effects regression model implemented in the MIXEDUP suite (Hedeker and Gibbons, 1996a, b;Schulze et al., 2006). Continuous variables were log-transformed for use in association testing.
(Pyrosequencer PSQHS96, Pyrosequencing/Biotage), fluorescence polarization-TDI (Akula et al., 2002) (Analyst HT, Molecular Devices) and TaqMan allele discrimination assay [Applied Biosystems, (ABI)]. PCR was performed using either HotMaster Taq polymerase (Brinkmann-Eppendorf, New York, USA), FastStart Taq (Roche, Valencia, California, USA) or HotStar Taq (Qiagen, Roche, Indianapolis, Indiana, USA) employing the manufacturer's recommended initial denaturation temperature followed by 40 cycles of denaturation at 941C for 30 s, annealing at 601C for 30 s and 721C for 1 min. At the end of 40 cycles, an additional extension was done at 721C for 7 min followed by a soak of 41C. For some primer pairs, the annealing temperature was set at either 55 or 651C. All genotypes were checked for consistency in duplicate samples and for Mendelian errors.

Association tests
Family-based association test (Laird et al., 2000) was used to analyze genotype data from fine-scale mapping and to assess the effect of familial phenotype variables on association. TDTPHASE (UNPHASED package) (Dudbridge, 2003) was used for analysis of data from the initial screen with 98 SNPs, estimating odds ratios (ORs) and evaluation of transmission of parental alleles. A narrow affection status model (ASM I) that includes BPI and SA-BP as affected was used in the primary analysis. Analysis under a broader model (ASMIII) that includes BPI, SA-BP, BPII and recurrent major depression was also done on four SNPs (rs1340, rs9557134, rs9517575 and rs10492574).
Gene structure determination, polymorphism screening, haplotype block structure and resequencing The predicted exons and splice junctions for KIAA1058 (AB028981) (Kikuno et al., 1999) and NM_015296 ([Zizimin1] Meller et al., 2002) [National Center for Biotechnology Information (NCBI)]; University of California Santa Cruz (UCSC) Genome Browser, May 2004), were sequenced on a panel of 22 cases drawn from the CNG families in an attempt to find new sequence variations. Sequencing was done in-house using the Big Dye terminator kit on an ABI 3100 sequencer. In addition, the entire gene, including all exons and introns of NM_015296 was resequenced on 12 unrelated BPI patients selected from the NIMH families at the Baylor Human Genome Sequencing Center. Additional resequencing of 16.5 kb of selected regions upstream of rs1927568 was done on 24 BPI unrelated probands from the NIMH families (Polymorphic DNA Technologies, Inc., Alameda, California, USA).
To generate the haplotype block structure we used HapMap-derived genotype data from trios of European (CEU) origin (release no. 19/phase II Oct05). Analysis was done using Haploview (version 3.2) (Barrett et al., 2005) on SNPs that had MAFs Z 0.1 using the Gabriel Algorithm and default parameters (Gabriel et al., 2002). Haplotype block structure was derived from the DOCK9 SNP genotype data in NIMH1-2 in a similar manner.

Results
Linkage disequilibrium screening on 13q32-q33 In our previous study we have determined the region with a 95% confidence limit for the location of the susceptibility locus to be within the interval bounded by D13S122 and D13S280 (Liu et al., 2001). Therefore, we targeted this region for our initial SNP screen covering a B7.6 Mb segment from rs1012693 to rs1322713. A set of 98 validated SNPs were used to genotype a sample of 285 families composed of parent-offspring trios, one-parent-one-offspring pairs and affected sib-pairs. This sample included NIMH1-2, CHIP, CNG families and some drawn from the NIMH3 series. SNPs were selected from the Phase I HapMap and other public sources.
In this initial sparse screen, TDT analysis (TDTPHASE) (Dudbridge, 2003) yielded evidence of association with bipolar disorder at rs1927568 (P = 0.00067) ( Table 1). The SNP mapped to the 5 0 flanking region of the DOCK9 gene (RefSeq NM_015296) (Meller et al., 2002). In addition, two of the remaining six SNPs typed on DOCK9 detected association with bipolar disorder: rs2000342 (P = 0.04766) and rs2390129 (P = 0.01996) ( Table 1). No other clusters of significant results were detected in any of the genes sampled in this study (data not shown). On the basis of these results, we targeted DOCK9 for further analysis.

DOCK9 transcripts and haplotype block structure
The DOCK9 gene spans B293 kb and encodes three major transcripts that differ in their amino terminal sequences (UCSC Genome Browser, May 2004; NCBI): KIAA1058 (AB028981) (Kikuno et al., 1999;NCBI), NM_015296 (zizimin 1) (Meller et al., 2002;NCBI) and AK127329 (UCSC Genome Browser, May 2004; NCBI) ( Fig. 1). Our analysis indicated that the nucleotide sequence purportedly coding for the first 27 amino acids in KIAA1058 is part of the 5 0 untranslated region therefore the initiation methionine immediately follows this sequence. AK127329 is an incomplete clone lacking the 3 0 -terminal segment that codes for the C-terminal portion of the protein. DOCK9 has at least 50 exons and three alternative first exons: 1a, 1b and 1c (Fig. 1). Transcription is oriented opposite the genome-wise direction with exon 1a of NM_015296 as the most telomeric exon, B71 kb downstream of exon 1b of AK127329, which in turn, is 37.5 kb distal to exon 1c of KIAA1058 ( Fig. 1) (UCSC Genome Browser).
The haplotype block structure of DOCK9 was calculated over a 320 kb interval delimited by rs1299066 and rs9513550. Genotype data generated by HapMap on 348 SNPs in the CEU including only those SNPs with MAF Z 0.1 was used. Haploview analysis produced 13 haplotype blocks within two major blocks of almost equal size (here designated MHB1 and MHB2), and interrupted by 525 bp of low linkage disequilibrium (LD) in intron 2 (Fig. 1). MHB1 spans about 164 kb, extending from the DOCK9 5 0 flanking region up to exon 2. MHB2 covers B160 kb and incorporates the remaining exons that encode three functional motifs, PH, CZH1 and CZH2 (Meller et al., 2002) (Fig. 1).

Fine-scale linkage disequilibrium mapping
To follow-up on the preliminary evidence of association, fine-scale mapping was done to interrogate the entire length of the DOCK9 gene. Our approach was to individually genotype each of the four sets of bipolar disorder families. NIMH1-2 was used as the primary test sample upon which 40 SNPs were assayed. Follow-up genotyping using 31 of the 40 SNPs were done on NIMH3 and NIMH4. Association was also assessed in the CHIP and CNG family series (combined to achieve reasonable sample size), in which four of the 40 SNPs were analyzed. Haploview-predicted tag SNPs and other SNPs, some of which were detected through resequencing, were used. Six SNPs that deviated from Hardy-Weinberg equilibrium (P < 0.05) were excluded from the final analysis (Table 1, Fig. 1).
At this phase of the study the primary association tests (see below) were conducted using Family-based association test-e (empirical variance option) to give more precise P values in the presence of linkage. In the NIMH1-2, analysis under ASMI detected evidence of association at the P < 0.05 level with nine MHB1 SNPs (smallest nominal P = 0.006) including rs1927568, one of three DOCK9 markers that detected association in the initial screen (Table 1; Figs 1 and 2). These SNPs cluster in a B100 kb block that comprises the 5 0 flanking region, exon 1a, and intron 1 of NM_015296. Intron 1 of NM_015296 includes exon 1b of AK127329 and exon 1c of KIAA1058. This region is enclosed in a large haplotype block in the block structures generated using both the HapMap and NIMH1-2 genotype data (Figs 1 and 2).
Further analysis detected support for association in NIMH3 (rs1340) and CHIP-CNG (rs9517575 and rs9557134) (Table 1; Fig. 2). The same alleles of the same SNPs were transmitted in excess in NIMH1-2, NIMH3 and CHIP-CNG. In contrast, no significant overtransmission of any alleles was detected in NIMH4 (Table 1). The ORs for the over-transmitted allele C, T and C of rs9517575, rs9557134 and rs1927568 were 1.33, 1.80 and 1.93, respectively.
Additionally, we found evidence for a preferential transmission of maternal alleles to affected offspring suggesting that the observed association results were driven largely by maternal alleles. This was most strongly demonstrated by rs1927568 which displayed significant excess transmission of maternal alleles (P = 0.000083,  OR = 3.778), and less over-transmission of paternal alleles (P = 0.0532, OR = 2.0).

Familial component phenotypes of bipolar disorder
To explore the impact of phenotype variability on the association results, we examined individual clinical features in the NIMH families as part of a secondary analysis. Potentially, some entities could impart 'noise' and various combinations of variables could conceal association that does exist. Many variables could be tested, so we evaluated only those that were significantly familial in these samples (Fisfalen et al., 2005;Kassem et al., 2006;Schulze et al., 2006). Although familial variables are not necessarily genetic, those that are not familial are unlikely to have a strong genetic basis. By this criterion, we selected 25 phenotypic variables for analysis. These included symptoms of mania and depression, comorbid mental disorders such as psychosis and panic disorder, and course indicators such as age at onset.

Mania and depression variables
Secondary analyses on mania and depression variables in NIMH1-2 identified the same MHB1 SNPs that displayed association signals under the primary analysis. Interestingly, more SNPs showed significant association signals and P values were in general smaller than in the primary analyses, even though the effective sample sizes were smaller. For example, the variable 'racing thoughts during mania', which is known to be highly diagnostic of bipolar disorder (Goodwin and Jamieson, 1990) showed significant association with 16 SNPs (P = 0.0366-0.0008) ( Fig. 2; Table 2), 'delusions during mania' with 18 SNPs (0.0483-0.0006) (Fig. 2), and 'grandiosity during mania' with 18 SNPs (P = 0.048-0.00072) ( Table 2). In NIMH3, the greatest evidence of association was also detected for 'racing thoughts during mania' (nine SNPs, P = 0.0323-0.007) ( Fig. 2; Table 2). For each associated SNP, the same allele was over-transmitted in both NIMH1-2 and NIMH3.
In familial depressive symptoms, psychomotor retardation detected excess transmission in the highest number of SNPs (16 SNPs, P = 0.0458-0.0049) in NIMH1-2 ( Fig. 2) NIMH3 or NIMH4 did not display association signals with depression variables (data not shown).
Course of illness indicators: mania at onset, age at onset, episode frequency, age at first mania In a recent study on the NIMH dataset, polarity of onset in bipolar disorder has been found to be heritable and this subset of families detected linkage on 16p (Kassem et al., 2006). This analysis on NIMH1-2 showed excess allelic transmission with mania at onset in eight MHB1 and two MHB2 SNPs extending over a larger portion of the gene that includes several DOCK9 exons (Table 3). Surprisingly, NIMH4 that has shown either no or scant evidence of association in prior analyses displayed signals for mania at onset in five MHB1 and four MHB2 SNPs (Table 3). NIMH4 over-transmitted, however, the alternative alleles indicating allele switching, that is, the potential susceptibility allele in NIMH1-2 is the protective allele in NIMH4. Association signals were detectable mostly in NIMH1-2 for variables such as age at first mania (13 MHB1 SNPs, P = 0.037-0.0059), episode frequency (12 MHB1 SNPs, P = 0.0434-0.0034) and age at onset (13 MHB1 SNPs, P = 0.0242-0.0026) (Fig. 3), as well number of manic episodes (Table 2).

Bipolar disorder comorbid psychiatric phenotypes Psychosis
Possible correlation of psychosis with the transmission pattern of DOCK9 SNPs was of particular interest because both bipolar disorder and schizophrenia have been linked to 13q and prior studies have detected suggestive linkage of psychotic bipolar disorder to 13q (reviewed in Detera-Wadleigh and McMahon, 2006). In NIMH1-2, nine of the 10 MHB1 SNPs that detected excess transmission with psychosis at the P < 0.05 level were identical to those that detected association with bipolar disorder (Fig. 4). Similarly, allelic transmission at seven of these SNPs was associated with psychosis in NIMH3. Similar patterns were detected for the more narrowly defined delusions during mania (Fig. 2). It is important to note here that of the total number of ASMI offspring in NIMH1-4 only B6% had SA-BP diagnosis.

Suicide attempts
Suicidal behavior among patients with mood disorders has been well-documented (Angst et al., 2005). In NIMH1-2, a history of suicide attempt(s) was associated only with a single marker, rs1340. This is also the only SNP that showed significant overall association with bipolar disorder in NIMH3 in the primary analysis. By contrast, there was greater evidence of association in NIMH3 involving five MHB1 SNPs (P < 0.05). Covariate analysis (suicidal attempt and bipolar disorder) detected signals in three MHB1 SNPs in NIMH1-2 (Fig. 4) and NIMH3.
In 25-75% of NIMH4 families, excess transmission was displayed by the alternative alleles in three MHB1 and one MHB2 SNPs. Two-trait analysis retained signal in only one of the MHB1 SNPs (data not shown).

Panic disorder, alcoholism and substance abuse
The co-occurrence of panic disorder, alcoholism and substance abuse with bipolar disorder has been welldocumented in epidemiologic and family studies (Regier et al., 1990;Winokur et al., 1996;Freeman et al., 2002;MacKinnon et al., 2002;Doughty et al., 2004). We explored MHB1 SNPs (P = 0.0378-0.0081), respectively (Fig. 4). In NIMH3, covariate analysis detected no association in panic disorder. In alcoholism and substance abuse signals were displayed by one MHB1 SNP and three MHB2 SNPs, respectively.

Discussion
To our knowledge this is the first report to implicate DOCK9 or the Rho-GTPase pathway in the etiology of bipolar disorder. DOCK9 has not been well studied and its function in the brain, where it is highly expressed, remains to be established. DOCK9 has been shown to activate Cdc42, a RhoGTPase (Meller et al., 2002), that has diverse roles including regulation of the actin cytoskeleton, cell migration, axonal guidance, neurite outgrowth and dendritic arbor growth (Threadgill et al., 1997;Li et al., 2000;Luo, 2002;Hall, 2005;Shen et al., 2006). Other DOCK (dedicator of cytokinesis) genes exist that have been better characterized and these may provide clues to the function of DOCK9 (Meller et al., 2005). DOCK10, on 2q36.2, also referred to as dopamine interacting protein 2 (NCBI), and DOCK11, on Xq24, are highly homologous to DOCK9. BLAST analysis also reveals an orthologous rat sequence that has been shown to be down-regulated by thyroid stimulating hormone (Pianese et al., 1994).
Our study unveils associated variants located mostly in the 5 0 flanking region and in intron 1 of NM_015296. The first exons, 1b and 1c, of AK127329 and KIAA1058, respectively, also lie within this intron therefore the associated SNPs might be located in the promoter region of these isoforms. Interestingly, several of the associated variants in intron 1 in the 5 0 flanking region of NM_015296 are evolutionarily conserved suggesting a role in function. Studies have shown active regulatory sequences and enhancers in noncoding regions (Shin et al., 2005;Woolfe et al., 2005;Fisher et al., 2006;Pennacchio et al., 2006) but these remain to be identified in DOCK9. In addition, it is interesting that the ancestral allele in several of the MHB1-associated SNPs is significantly undertransmitted to affected offspring. Resequencing of NM_015296 in bipolar samples identified new polymorphisms (data not shown) but it has not disclosed any common nonsynonymous or splice junction mutations. DOCK9 might confer risk through altered levels of transcription and/or aberrant splicing patterns. Our findings set the stage for future work aimed at uncovering the key functional variation that accounts for the association results.
This study highlights evidence of association between bipolar disorder and several markers in a gene located within a bipolar disorder linkage peak on chromosome 13q that has been supported by several studies (reviewed in Detera-Wadleigh and . The sets of families tested display a heterogeneous pattern of association across individual component phenotypes of bipolar disorder. NIMH1-2 shows the highest evidence for association both in the primary and secondary analyses. For SNPs that detect association signals, the same alleles are over-transmitted in NIMH1-2, NIMH3 and CHIP-CNG. The largest single sample we examined, NIMH4, shows association only with a subset of component phenotypes and in this sample the alternative alleles are over-transmitted. This heterogeneity may reduce the power to replicate our findings in some samples.
Stronger association signals with racing thoughts, delusions during mania, course of illness indicators, and psychosis suggest that DOCK9 may contribute to increased illness severity and imply that future replication attempts should focus on severe cases. This also suggests that DOCK9 variations may have prognostic value.
Switching of over-transmitted (susceptibility) and undertransmitted (protective) alleles is a common phenomenon in complex disease. For example, different alleles of markers in DAOA (G72) are over-transmitted in different samples (reviewed in Detera-Wadleigh and McMahon, 2006), differing alleles of a functional COMT variant have been associated with various neurocognitive phenotypes (Funke et al., 2005), and differing haplotypes of dysbindin have shown association with schizophrenia (Straub et al., 2002;Schwab et al., 2003). Although the possibility of a false positive signal in at least some samples is difficult to rule out, this phenomenon may also reflect true allelic heterogeneity, undetermined influences of other interacting risk loci, or differing allele frequencies and patterns of LD across samples.
The presence of comorbid psychiatric phenotypes possibly reflects the existence of genetic subtypes of bipolar disorder and shared genes for various phenotype presentations. Studies have documented the occurrence of psychotic features in a substantial proportion of bipolar disorder patients (Pope and Lipinski, 1978). The overlap of linkage peaks on 13q32-q33 for bipolar disorder and schizophrenia has led to the speculation of shared gene(s) for psychosis in both disorders (Blouin et al., 1998;Detera-Wadleigh et al., 1999). The association signals we observe with psychosis in both NIMH1-2 and NIMH3 suggest that DOCK9 may contribute to the linkage signals for psychotic disorders detected on chromosome 13q.
Other loci previously implicated in psychosis include G72/G30 in bipolar disorder with persecutory delusions  and in childhood psychosis and schizophrenia (Addington et al., 2004). In contrast, G72/G30 variations that correlated with major mood episodes in schizophrenia and bipolar disorder did not show association with psychosis (Williams et al., 2006). This inconsistency in association findings possibly reflects differences in psychotic phenotypes across various samples. Dysbindin (DTNBP1) variants have been shown also to be nominally associated with psychotic bipolar disorder  and variations in neuregulin (NRG1) have shown correlation with bipolar disorder with mood incongruent psychotic features . Thus, it may be important to analyze DOCK9 in schizophrenia samples as well as other samples of psychotic bipolar disorder.
This study has several limitations, and firm conclusions must await replication studies. The sample sizes were modest, and effective sample sizes were further reduced in the component phenotype analyses, potentially reducing our power to detect true associations. Although pooling all samples together from the start would have increased sample size, this would come at the cost of increased heterogeneity, which could decrease the true power to detect the modest association signals that one expects in complex disease. By treating each individual set of samples, we took into account real differences in the ways the samples were ascertained and evaluated. Our analysis of component phenotypes of bipolar disorder is an attempt to identify specific diagnostic entities that correlate with allelic over-transmission across sample panels, which, to our knowledge, is the first such exploration in bipolar disorder. Genetic analysis of multiple component phenotypes in a complex disease has been reported recently, for example, in migraine (Anttila et al., 2006) and hypertension (Wallace et al., 2006). This approach necessarily entails multiple tests whose mutual dependency is difficult to assess. Certainly, if we were to consider the individual component phenotypes to be independent, then the signals detected in the secondary analyses would not survive correction for multiple testing. Our approach examines variables that are constituent parts of a single overarching phenotype therefore tests on individual components may not be nonindependent, but additional replication studies are needed before firm conclusions can be reached.
The SNPs that consistently display association are clustered in MHB1, one of two CEU major haplotype blocks encompassing DOCK9. LD between SNPs in this region is clearly evident particularly in the analysis of some of the individual variables ( Fig. 2; Table 2). It can be inferred that tests on the MHB1 SNPs are nonindependent; hence may constitute a single test. Given that there are two major haplotype blocks in DOCK9, it may be reasonable to consider only two tests for genewise correction.
The results presented herein highlight some critical elements that contribute to the complexity of bipolar disorder and underscore the need for a refinement of phenotype classifications to improve our ability to uncover genetic and other risk factors. Multiple tests have been performed and therefore some results could have arisen from statistical fluctuations, but findings in the secondary analyses tended to agree with those from the primary tests. Future progress in collecting large, well-phenotyped samples and whole genome association studies should help address some of long-standing issues discussed here. Dissecting the patterns of allelic transmission in DOCK9 may ultimately help untangle salient facets of heterogeneity in bipolar disorder.