Gene-based SNP mapping of a psychotic bipolar affective disorder linkage region on 22q12.3: Association with HMG2L1 and TOM1

Authors


  • Please cite this article as follows: Potash JB, Buervenich S, Cox NJ, Zandi PP, Akula N, Steele J, Rathe JA, Avramopoulos D, Detera-Wadleigh SD, Gershon ES, NIMH Genetics Initiative Bipolar Disorder Consortium, DePaulo JR Jr, Feinberg AP, McMahon FJ. 2007. Gene-Based SNP Mapping of a Psychotic Bipolar Affective Disorder Linkage Region on 22q12.3: Association With HMG2L1 and TOM1. Am J Med Genet Part B 147B:59–67.

Abstract

Genetic linkage studies in both bipolar affective disorder (BPAD) and schizophrenia have implicated overlapping regions of chromosome 22q. We previously reported that BPAD pedigrees containing multiple members with psychotic symptoms showed suggestive linkage to chromosome 22q12.3. Now we have tested 189 single nucleotide polymorphisms (SNPs) spanning a 3 Mb region around the linkage peak for association with BPAD in 305 families, unrelated cases, and controls. SNPs were selected in or near genes, resulting in coverage at a density of 1 SNP per 6.7 kb across the 22 annotated genes in the region. The strongest signal emerged from family-based association analysis of an 11-SNP, 54 kb haplotype straddling the gene HMG2L1 and part of TOM1. A 3-marker haplotype of SNPs within TOM1 was associated with BPAD (allele-wise P = 0.0011) and with psychotic BPAD (allele-wise P = 0.00049). As hypothesized, the mean odds ratio for the risk alleles across the region was 1.39 in the psychotic but only 0.96 in the non-psychotic subset. Genotype-wise analyses yielded similar results, but the psychotic/non-psychotic distinction was more pronounced with mean odds ratios of 1.91 versus 0.8. Permutation of genotype-wise results for rs2413338 in HMG2L1 showed an empirical P = 0.037 for the difference between subsets. HMG2L1 is a negative regulator of Wnt signaling, a pathway of interest in psychotic BPAD as it is activated by both mood stabilizer and anti-psychotic medications. Further work is needed to confirm these results and uncover the functional variation underlying the association signal. © 2007 Wiley-Liss, Inc.

INTRODUCTION

Bipolar affective disorder (BPAD), a psychiatric illness characterized by episodes of depression and mania, has a lifetime prevalence of 1–2% [Weissman et al., 1988]. Family, twin, and adoption studies have established a substantial genetic contribution, with heritability estimates averaging about 70% [Smoller and Finn, 2003]. The high heritability and ready availability of multiplex families have provided a strong rationale for genetic linkage studies.

Several BPAD linkage scans have implicated a region on chromosome 22q12-13. Kelsoe et al. [2001] found a genome-wide significant signal on 22q12.3, while investigators using samples from the National Institute of Mental Health (NIMH) Genetics Initiative, and the Clinical Neurogenetics pedigrees reported suggestive evidence of linkage [Edenberg et al., 1997; Detera-Wadleigh et al., 1999; Kelsoe et al., 2001]. The 22q12 region was first implicated in major mental illness susceptibility through linkage studies of schizophrenia [Pulver et al., 1994; Gill et al., 1996]. A meta-analysis examined 11 BPAD genome scans and found that 22q12-13 was the second strongest linkage region in the genome [Badner and Gershon, 2002]. When this same study analyzed 18 genome scans in schizophrenia samples along with the BPAD scans, 22q12 emerged as the strongest linkage region in the combined data. A second meta-analysis yielded no findings for BPAD and modest findings for schizophrenia on 22q12 [Lewis et al., 2003; Segurado et al., 2003], with the divergence from the earlier meta-analysis perhaps attributable to differences in which studies were included and how they were weighted.

Overlapping linkage signals in BPAD and schizophrenia may reflect overlapping susceptibility genes [Berrettini, 2000], and these may manifest in overlapping phenotypic features. The psychotic symptoms, hallucinations, and delusions, constitute one important area of phenotypic overlap, as they occur in many cases of BPAD and virtually all cases of schizophrenia. If these features have a genetic basis, they should show evidence of familial aggregation. Previously we showed, in two distinct sets of families, that psychotic features do indeed aggregate in a subset of BPAD pedigrees [Potash et al., 2001; Potash et al., 2003a], a finding that is supported by independent studies in other samples [Omahony et al., 2002; Schurhoff et al., 2003]. We have also found that BPAD families in which at least three relatives had psychotic features show suggestive evidence of linkage to 22q12.3 that is not detected in an unselected sample [Potash et al., 2003b]. Taken together, the results published to date support the presence of a gene in this region that contributes to psychotic BPAD.

Now we have tested 189 single nucleotide polymorphisms (SNPs) spanning a 3 Mb region on 22q12.3 for association with BPAD in a large sample of families, unrelated cases, and controls. The region was chosen based on the 95% confidence interval for the Potash et al. [2003b] psychotic BPAD linkage result computed by GENEFINDER [Liang et al., 2001], and on the overlap of this region with the localization of prior findings in BPAD and schizophrenia mentioned above [Pulver et al., 1994; Gill et al., 1996; Edenberg et al., 1997; Detera-Wadleigh et al., 1999; Kelsoe et al., 2001; Badner and Gershon, 2002]. We hypothesized that one or more genes in this region would show association with BPAD, and that the strength of the association would be greatest in the subset of cases with psychotic BPAD. Our results demonstrate an association between BPAD and haplotypes straddling the genes HMG2L1 and TOM1 that is attributable to cases of psychotic BPAD in this sample.

MATERIALS AND METHODS

Subject Ascertainment and Assessment

BPAD family samples were drawn from three sources: the NIMH Bipolar Disorder Genetics Initiative, the Johns Hopkins University family collection, and the Clinical Neurogenetics pedigrees. For the NIMH Genetics Initiative sample, the ascertainment criteria focused on families with a bipolar I disorder (BPI) proband and at least one sibling with either BPI or the closely related diagnosis of schizoaffective disorder, bipolar type (SABP). Detailed ascertainment information has been published elsewhere [NIMH Genetics Initiative Bipolar Group, 1997]. Interviews were conducted using the Diagnostic Interview for Genetic Studies (DIGS), either version 1.0 or 2.0 [Nurnberger et al., 1994]. Diagnoses were made using a best estimate procedure and employing the criteria in the Diagnostic and Statistical Manual (DSM), III-R. The NIMH Genetics Initiative sample was collected in four waves, with the waves I and II samples used in the current study collected by Johns Hopkins, Indiana University, Washington University in St. Louis, and the NIMH intramural program. For the Johns Hopkins sample, the ascertainment criteria required a treated BPI proband and two additional relatives with a major mood disorder (See Simpson et al. [1992] for details). Probands and their relatives were interviewed using the Schedule for Affective Disorders and Schizophrenia-Lifetime Version (SADS-L) [Endicott and Spitzer, 1978]. Diagnoses were made using a best estimate procedure and employing the Research Diagnostic Criteria (RDC) [Spitzer and Endicott, 1975]. The criteria for BPI are almost the same in the RDC as in the DSM-III-R and DSM-IV with the exception that some people diagnosed with BPI by the DSM criteria would have been diagnosed with schizoaffective disorder by RDC. The methods employed for the Clinical Neurogenetics sample were similar to those for the Johns Hopkins sample (see Berrettini et al. [1991] for description).

Informed consent for participation in genetic studies was obtained from all subjects included in this report.

Subject Selection for Genotyping

The detection of the kind of modest genetic effects expected in complex disorders requires large samples to achieve adequate power. This consideration led us to perform our primary analysis in a combined set of several BPAD family samples. Subjects were drawn from the NIMH Genetics Initiative waves I and II samples, the Johns Hopkins sample, and the Clinical Neurogenetics sample. Probands, all available parents, and up to two affected siblings were selected. Siblings with BPI were preferred but when they were unavailable siblings with SABP (N = 18) were also included. From the NIMH Genetics Initiative waves I and II samples, there were 183 families, 298 parents, and 306 affected probands and siblings. From the Johns Hopkins sample, there were 103 families, 178 parents, and 187 affected probands and siblings. From the Clinical Neurogenetics sample, there were 19 families, 30 parents, and 36 affected probands and siblings. In the total sample there were thus 305 families encompassing 506 parents and 529 affected probands and siblings. Power was estimated by use of PBAT, which takes into account the variable family structures in the sample [Lange and Laird, 2002]. Under a multiplicative model with K = 0.02, ρ = 0.2, and D′= 0.875, the family sample had 80% power to detect a variant conferring a relative risk of at least 1.5.

In order to assess baseline patterns of inter-marker linkage disequilibrium (LD), corroborate allele frequency estimates, and exclude transmission ratio distortion in the family-based analyses, we also genotyped random control subjects. These were drawn from two sources: (1) 129 unrelated founders in the Centre d'Etude du Polymorphisme Humain (CEPH) Utah and French pedigrees (excluding pedigree 880 because of its known history of BPAD); and (2) 200 unrelated, apparently healthy, self-declared Caucasians from the Coriell Human Diversity panel. For case-control association analyses, we compared these 329 controls to 305 cases comprising the probands of the 305 families in the initial BPAD sample described above.

Psychosis Variable

The determination of psychosis was made based on a history of hallucinations and/or delusions during a depressive or manic episode. If such symptoms were present only while the subject was intoxicated, in alcohol or drug withdrawal, or during delirium, the subject was not considered to have had psychotic mood disorder. In the NIMH Genetics Initiative sample, the determination of psychosis was based on questions in the DIGS about the presence of hallucinations and delusions during severe episodes of depression and mania, and on additional questions about such symptoms throughout the lifetime. In the Johns Hopkins and the Clinical Neurogenetics samples, the data were based on responses to questions in the SADS-L about whether these symptoms were present during the most severe mania and/or the most severe depression. In these latter samples, family informant data and medical record data were also used to supplement the interview. The rate of psychotic symptoms among affected probands and siblings was 58.8% (N = 311/529) in the initial combined sample. We note that the reliability of the assessment of hallucinations or delusions has been found to be high across many prior studies [Wing et al., 1967; Endicott and Spitzer, 1978]. We did not examine any other variables as covariates in this study, since we set out to test a single hypothesis about psychosis in BPAD.

SNP Selection

A gene-based rather than a map-based approach was chosen to perform an initial screen of the region. This was decided based on the completed state of sequencing and the advanced state of annotation in the region at the time the study was conceived. The Sanger track from the University of California, Santa Cruz human genome browser, June, 2002 assembly was used to determine likely genes in the region. A total of 189 SNPs were chosen based on location within 10 kb of one of the 22 genes and ESTs in region (see Table I). This distance was chosen because it would likely include regulatory sequences and SNPs in LD with regulatory sequences or coding regions. We did not attempt to genotype SNPs in the remaining 1.67 Mb (56%) of this region. Priority was given to: (a) coding SNPs since these had the highest likelihood of being functionally relevant; (b) SNPs that had been validated since these had a higher likelihood of being true polymorphisms; (c) SNPs found in Caucasians, because in our predominantly Caucasian sample these were most likely to be informative.

Table I. Gene-Based SNP Coverage on 22q12.3
GeneGap (kb)*Size (kb)SNPsCoverage (kb/SNP)
Pre-screenPost-screenPre-screenPost-screen
  • *

    Gap between most centromeric SNP assayed in gene region and most telomeric. SNP assayed in prior gene region.

Large 667,338312521.526.7
CN13G6297.223,692982.63.0
Aconitase367.83,286113.33.3
dJ288.3108.629,237664.94.9
dJ288.19.730,097764.35.0
HS371476114.65,288331.81.8
SIX3L165.835,909874.55.1
Bk714b732.034,455993.83.8
HM2L186.452,835866.68.8
TOM13.667,52718153.84.5
HMOX13.124,9111124.924.9
MCM53.730,775744.47.7
RASD242.940,947994.64.6
MB19.439,728984.45.0
APOL615.133,866655.66.8
APOL531.621,486544.35.4
RBM971.442,878528.621.4
APOL3278.552,86812114.44.8
APOL47.138,522984.34.8
APOL24.935,020553.93.9
APOL16.217,503662.92.9
MYH94.080,13415135.36.2
Total (22 genes)1,673.61,408.31891616.16.7

Genotyping

SNP genotyping in the initial phase of the study was performed by Illumina using their BeadArray platform [Gunderson et al., 2004]. This platform employs hybridization of template-generated oligonucleotides labeled with allele-specific fluorophores to generate signals read by an automated detection system. Because for logistical reasons a subset of the sample was sent to Illumina before the remainder of the sample, we had the opportunity to deselect a portion of the SNPs to conserve resources. We chose to deselect, for the second portion of Illumina genotyping, and for all analyses, all monomorphic SNPs, one of a pair in perfect LD, those in Hardy–Weinberg disequilibrium in controls, and those with no difference in allele-frequency between controls and cases, and controls and psychotic cases (P > 0.9).

Analysis

Haploview was employed to measure LD by r2 and to visualize the results graphically [Barrett et al., 2005]. Haploview also provided a measure of Hardy–Weinberg equilibrium.

Family-based association testing was carried out with FBAT [Laird et al., 2000]. To test the null hypothesis of no association in the presence of linkage, we employed the empirical variance option, which provides an unbiased test for association in the presence of linkage by accounting for the correlation among sibling marker genotypes. Because the FBAT program cannot run a genotypic model with an empirical variance option, the GENASSOC program was used to determine family-based genotypic relative risks and odds ratios. The TDTphase component of the UNPHASED package [Cordell and Clayton, 2002] was used to calculate allelic odds ratios. Case-control association testing was performed (using the proband from each family as a case) with the COCAPHASE component of the UNPHASED package [Cordell and Clayton, 2002]. The EM algorithm was used to provide maximum-likelihood estimates of haplotype frequencies. Since the EM algorithm does not accurately estimate haplotype frequencies below 1%, haplotypes with estimated frequencies <1% were excluded.

RESULTS

Genotype Quality and SNP Screening

Out of 190 SNPs initially sent to Illumina, 189 yielded genotypes. Only 0.1% of 51,597 possible genotypes were missing. A significant deviation from Hardy–Weinberg equilibrium was observed in controls for three SNPs. Following a preliminary analysis of the first phase Illumina data we deselected 13 monomorphic SNPs, five in perfect LD with adjacent SNPs, five in Hardy–Weinberg disequilibrium in controls, and five with no difference in allele-frequency between controls and cases, and controls and psychotic cases (P > 0.9). This resulted in a total of 28 being deselected and 161 being carried forward to the second phase of Illumina SNP genotyping (see Table I). This second phase genotyping was part of a larger Illumina SNP genotyping project that yielded only 0.02% missing data, and genotypes of sufficient quality for analysis from 97% of markers. A significant deviation from Hardy-Weinberg equilibrium was observed in controls for two SNPs in the second phase. Identity-by-state analysis with Graphical Representation of Relatedness [Abecasis et al., 2001] identified two pairs of control samples who were apparently identical. In each case, the one with the least complete data was dropped, prior to the analysis. The remaining samples all had identity-by-state values consistent with their stated relatedness.

Linkage Disequilibrium

The LD pattern across the region, as measured by r2, is displayed graphically in Figure 1a. The graphic illustrates that many of the SNPs assayed were in LD with each other, as would be expected given that SNPs were selected in clusters centered on genes. There were 34 haplotype blocks identified using this LD approach, while 64 SNPs did not fall into blocks. Thus a total of 98 distinct LD regions were sampled. The largest haploblock identified extended across 11 SNPs from rs2413338-rs138784, at a distance of 53.7 kb. This corresponds closely to a haploblock previously identified, using a haplotype diversity approach, [Dawson et al., 2002] and posted on the UCSC Genome Browser as CHR22_A_49, which extends 65.7 kb and includes the gene HMG2L1 and part of TOM1 (see Fig. 1).

Figure 1.

LD pattern across region, and genes within the largest LD block. A: This graphic, generated by the program HAPLOVIEW, depicts intermarker LD, as measured by r2, between the 161 SNPs assayed. Black represents perfect LD, with an r2 = 1, while shades of gray represent lesser values of r2. White indicates no LD, or an r2 of 0. Triangular lines indicate haploblocks, as determined using the program's Gabriel method algorithm. B: The 11-SNP haploblock identified in HAPLOVIEW corresponds to one previously identified [Dawson et al., 2002] and included in the UCSC genome browser. All of HMG2L1 is contained within it as well as the promoter region and the first exon of TOM1. Markers across this haploblock showed modest but replicated association with BPAD, which is strongest in the psychotic subset of subjects. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Association Results

The strongest signal in the family-based analysis emerged from the 11 SNP HMG2L1/TOM1 haploblock mentioned above. P-values for all SNPs in the region were <0.05 in the BPAD allele-wise analysis (see Table II); the most significant result was for rs138784 within TOM1 (P = 0.0048). When only psychotic BPAD was considered, rs138784 was still the most significantly associated marker. When a three-marker sliding window was used, the haplotype of markers rs4511-rs138784-rs739015 yielded results with a P = 0.0013 for BPAD and P = 0.00049 for psychotic BPAD. The application of a Bonferroni correction for having tested 98 LD regions resulted in a corrected P = 0.048 for the three-marker haplotype using the psychotic BPAD phenotype.

Table II. Results of Allele-Wise Analysis Using Family-Based and Case-control Samples
SNPGeneFamily-based (FBAT) P-valuesCase-control (COCAPHASE) P-values
Over/under transmitted allele1-marker1-marker psychotic3-marker3-marker psychoticOver/under represented allele1-marker1-marker psychotic3-marker3-marker psychotic
  1. 1-marker = single-point analysis with all BPAD subjects as affecteds; 1-marker psychotic = same as 1-marker but using only those BPAD subjects with psychotic symptoms; 3-marker = 3-marker sliding window with indicated marker first of the three; 3-marker psychotic = same as 3-marker but using only those BPAD subjects with psychotic symptoms. Dashed lines indicate that only 1 or 2 of the markers that comprise the 3-marker window are present in the table.

Region 1
 rs2157220dJ288L1.3T/C0.360.610.0750.11C/T0.690.330.290.13
 rs2015181dJ288L1.3A/G0.160.300.150.18G/A0.490.240.450.21
 rs1540300dJ288L1.3G/C0.540.420.0640.14G/C0.070.020.330.11
 rs2097368dJ288L1.3A/G0.540.710.0130.016G/A0.690.660.580.48
 rs2157221dJ288L1.3T/G0.390.71G/T0.250.11
 rs137252dJ288L1.1G/T0.010.025G/T0.430.73
Region 2
 rs2413338HMG2L1C/T0.0460.0100.0450.012C/T0.150.120.350.27
 rs713770HMG2L1A/G0.0490.0180.0470.015A/G0.170.190.170.18
 rs932290HMG2L1G/A0.0480.0160.0410.013G/A0.170.170.170.18
 rs2038010HMG2L1G/A0.0410.0150.0450.018G/A0.170.190.170.19
 rs4462TOM1A/G0.0370.0130.0380.011A/G0.170.190.160.19
 rs138764TOM1T/C0.0370.0170.0430.013T/C0.170.210.140.19
 rs138773TOM1C/G0.040.0090.030.0076C/G0.180.190.0640.15
 rs4465TOM1G/A0.030.0090.00670.0033G/A0.140.160.0730.15
 rs138781TOM1G/A0.0470.0130.00670.0033G/A0.130.140.0560.15
 rs4511TOM1C/G0.00660.0030.00130.00049G/C0.720.990.130.091
 rs138784TOM1A/G0.00480.0020.0210.0043G/A0.80.940.150.25
 rs739015TOM1A/G0.130.0730.160.14A/G0.0170.0320.890.27
 rs4467TOM1C/T0.8510.000.150.13T/C0.680.750.510.94
 rs762967TOM1A/G0.120.150.10.12A/G0.590.870.240.40
 rs743810TOM1A/C0.420.430.0470.12A/C0.910.740.320.25
 rs5750102TOM1G/A0.340.280.060.09A/G0.630.340.0660.17
 rs11703672TOM1G/A0.0420.10A/G0.0340.076
 rs11089741TOM1A/G0.710.12G/A0.510.23

In the allele-wise analysis, the mean odds ratio for the risk alleles across the region was 1.39 in the psychotic but only 0.96 in the non-psychotic subset (see Fig. 2). To further characterize this result, we performed a genotype-wise analysis on these markers. This yielded similar results (data not shown), but the psychotic/non-psychotic distinction was more pronounced with mean odds ratios of 1.91 versus 0.8 under a recessive model. We sought to determine an empirical statistical significance for the difference between the results in the subset with psychotic features as compared to the set without them. The psychotic status was permuted for rs2413338 in HMG2L1 in 10,000 simulations. For the allele-wise analysis, a P = 0.01 (the significance of the allele-wise result for the psychotic subset) was obtained 860 times, so that the empirical P = 0.086 for the difference between subsets. For the genotype-wise analysis, a P = 0.01 (the significance of the genotype-wise result for the psychotic subset) was obtained 368 times for an empirical P = 0.037.

Figure 2.

Family-based odds ratios for allele-wise and genotype-wise analyses in psychotic versus non-psychotic BPAD subjects across the HMG2L1/TOM1 region. The association signal from the family-based allele-wise analysis was entirely accounted for by the subset of the BPAD sample with psychotic features, where the mean odds ratio for disease conferred by a risk allele in the HMG2L1/TOM1 haplotype block was 1.39. The comparable odds ratio in the subset without psychotic features was about 0.96, indicating no contribution to disease risk. The difference was more pronounced in the genotype-wise analysis with the mean odds ratio difference being 1.91 versus 0.8 for the two groups under a recessive model. Permuting the psychotic status across affected subjects 10,000 times for rs2413338 in HMG2L1 in the allele-wise analysis yielded an empirical P = 0.086 for the difference between the psychotic and non-psychotic subsets, whereas performing the same procedure in the genotype-wise analysis provided an empirical P = 0.037 for the difference between subsets.

In contrast to the family-based results, the case-control results within the 11-SNP region did not show any nominally significant findings (see Table II), however, for each SNP the odds ratios were similar to those observed in the family-based analysis, being in the range of 1.2–1.3 across the HMG2L1/TOM1 region. Further, results for an adjacent TOM1 SNP, rs739015, were significant both in the full sample and in the psychotic case subset.

A second region with positive results covered 32 kb, from rs1540300 to rs137252, encompassing Sanger Center-annotated genes dJ288L1.3 and dJ288L1.1. In the FBAT analysis, one SNP, rs137252, was nominally significant, with P = 0.01 for BPAD, while the three-SNP haplotype of rs2097368-rs2157221-rs137252 yielded a P = 0.0045 for BPAD. Association results were not stronger in the psychotic BPAD subset. In the case-control analysis rs1540300 yielded results with a P = 0.02 for BPAD, and P = 0.0065 for psychotic BPAD, while for rs2157221, analyses showed P = 0.09 for BPAD and P = 0.04 for psychotic BPAD.

DISCUSSION

Our study found evidence for association in a large BPAD family sample across an 11-SNP, 53-kb haploblock on chromosome 22q12.3 containing HMG2L1 and part of TOM1. The 22q12.3 chromosomal region has been implicated in a number of linkage studies of BPAD and schizophrenia, though these genes have not previously been implicated in either disorder. Psychotic features strengthened the association signals, consistent with the linkage findings and our prior hypothesis.

These results should, however, be viewed in the context of a number of limitations. First, nominally significant results in the HMG2L1/TOM1 region were not obtained in the case-control analysis. This might be due to lesser power in this analysis as there were only 51% as many affecteds in the case sample as in the family sample. Based on analyses using the Genetic Power Calculator [Purcell et al., 2003], our case-control sample had 17–76% power to detect an effect of the magnitude derived from our family-based association results, given the markers studied, across a range of disease allele frequencies between 0.15 and 0.50. Second, because only SNPs in and near genes were assayed, substantial gaps in the region remain unstudied. Previously undetected genes as well as regulatory sequences may exist in the unstudied gaps. Third, our study was not designed to detect or rule out rare variants, even those with a large contribution to disease risk. Fourth, the magnitude of risk suggested for the putative disease genotype is modest. A modest effect would, however, be consistent with the failure to detect linkage in a number of BPAD and schizophrenia samples. Fifth, we presented a correction of our strongest finding based on having tested 98 LD regions. This might be considered anti-conservative because several tests—for the standard phenotpye and psychotic features, allelle-wise and genotype-wise analyses, and a recessive and a dominant model—were performed on the markers. However, the correction is conservative insofar as it does not take into account the substantial LD that exists between the 98 LD regions. Finally, we could not test additional sub-phenotypes that might bear on the relationship between BPAD and schizophrenia, such as first-rank symptoms, negative symptoms, or thought disorder, as we did not have complete data on these.

Psychosis Subtype

If either HMG2L1 or TOM1 does play a role in susceptibility to the psychotic form of BPAD, then it might also be a factor in schizophrenia vulnerability. The existence of shared vulnerability genes has been predicted based on family and twin studies, which show overlapping liability that is particularly strong between psychotic mood disorder and schizophrenia (reviewed in Potash et al. [2003b]). These findings, along with evidence of BPAD/schizophrenia linkage overlap, have led several investigators to search for shared susceptibility loci using linkage approaches [Pulver et al., 2000; Maziade et al., 2001; Sklar et al., 2004]. The strongest evidence to date for a BPAD/schizophrenia overlap gene(s), comes from the G72/G30 gene complex on chromosome 13q33 with reports of association in five BPAD and five schizophrenia samples (see Craddock et al. [2005] for references). One prior association study of BPAD employed psychotic features to define a phenotypic subtype, as we did; modest association of the subtype with a serotonin transporter polymorphism was reported [Ho et al., 2000]. The reciprocal approach, using mood symptoms to subtype schizophrenia subjects, was employed in an analysis of five functional candidate genes where it yielded a suggestive association with the serotonin 2A receptor gene [Fanous et al., 2004].

Comparison With Prior Studies on Chromosome 22

Other genes on chromosome 22 may influence BPAD risk. The region identified in this report is 16 Mb telomeric to the velocardiofacial syndrome (VCFS) region of chromosome 22q11.2 that has been previously implicated in schizophrenia particularly, and to a lesser extent in BPAD [Carlson et al., 1997; Murphy et al., 1999]. The VCFS region includes one gene, COMT, that could be implicated in both BPAD [Shifman et al., 2004] and schizophrenia [Shifman et al., 2002], though there have been numerous negative studies for each illness (see Glatt et al. [2003] and references in [Shifman et al., 2004]). Two additional VCFS region genes may be implicated in schizophrenia, PRODH [Liu et al., 2002], for which there have been non-replications [Williams et al., 2003], and ZDHHC8 [Mukai et al., 2004]. Other chromosome 22 genes have also been studied in both BPAD and schizophrenia. GRK3, 9.7 Mb centromeric to the region in this report, was modestly associated with BPAD in two samples [Barrett et al., 2003] though not in a schizophrenia sample [Yu et al., 2004]. XBP1, 6.5 Mb centromeric to the region described here, was implicated in two BPAD samples [Kakiuchi et al., 2003] and one schizophrenia [Chen et al., 2004] sample, but was not associated in several large BPAD samples [Cichon et al., 2004]. The APOL1, APOL2, and APOL4 genes, reported to be upregulated in schizophrenia [Mimmack et al., 2002], are located within the region we studied. We found no evidence for association of these genes with BPAD or with psychotic BPAD.

Signaling Pathways

An intriguing aspect of our finding is the evidence that the protein products of both HMG2L1 and TOM1 interact with signaling pathways potentially relevant to the pathogenesis of BPAD and schizophrenia. One report implicates HMG2L1 as a negative regulator of Wnt signaling [Yamada et al., 2003], while another implicates TOM1 as a negative regulator of signaling pathways induced by tumor necrosis factor-α and interleukin-1β [Yamakami and Yokosawa, 2004]. The prior data implicating Wnt signaling is particularly compelling. The mood stabilizer medications lithium [Klein and Melton, 1996] and valproic acid [Chen et al., 1999], and the anti-psychotic medications haloperidol [Emamian et al., 2004], and clozapine [Kang et al., 2004], all activate the pathway, suggesting a role for Wnt activation in the mechanism of action of these drugs. In a recent report, dopamine, the excess of which is the best-established mechanism in psychosis, increased activity of glycogen synthase kinase 3β, a key inhibitory enzyme in the Wnt signaling pathway, an effect that was reversed by dopamine D2 receptor blockade and by lithium [Beaulieu et al., 2004]. This result suggests that reduced Wnt signaling might be causally related to illness. Only two association studies of Wnt signaling genes in BPAD have been published, with each reporting modestly positive findings [Toyota et al., 2003; Benedetti et al., 2004]. Evidence for the involvement of tumor necrosis factor-α and interleukin-1β induced signaling pathways in BPAD and schizophrenia is more ambiguous, as increased serum levels of each cytokine have been associated with schizophrenia in some samples, but not others [Erbagci et al., 2001; Theodoropoulou et al., 2001], while a major downstream effector of both pathways, nuclear factor-κB, was increased in frontal cortex brain samples from BPAD subjects [Sun et al., 2001].

CONCLUSIONS

We have demonstrated evidence for association of a chromosomal region containing HMG2L1 and part of TOM1 with psychotic BPAD. Further work is needed to confirm these results and uncover the functional variation underlying the association signal. Replication of the current finding should be sought in additional BPAD samples and in schizophrenia samples. Within BPAD samples, further focus on the psychotic subtype may be productive. Within schizophrenia samples, the presence of a mood syndrome might define a phenotypic subtype most closely associated with the putative disease locus, or alternatively, simply the presence of hallucinations and/or delusions might define the associated subtype.

Acknowledgements

This study was supported by grants from the NIMH and the Stanley Medical Research Institute, and by the NIMH Intramural Research Program. Data and biomaterial were partly collected in four projects as part of the NIMH Bipolar Disorder Genetics Initiative. The principal investigators and co-investigators were as follows: Indiana University, Indianapolis—John Nurnberger, M.D., Ph.D., Marvin Miller, M.D., and Elizabeth Bowman, M.D. (supported by grant U01 H46282); Washington University, St. Louis—Theodore Reich, M.D., Allison Goate, Ph.D., and John Rice, Ph.D. (supported by grant U01 MH46280); Johns Hopkins University, Baltimore—J. Raymond DePaulo, Jr., M.D., Sylvia Simpson, M.D., M.P.H., and Colin Stine, Ph.D. (supported by U01 H46274); NIMH Intramural Research Program, Clinical Neurogenetics Branch, Bethesda—Elliot Gershon, M.D., Diane Kazuba, B.A., and Elizabeth Maxwell, M.S.W. We thank Luana Galver for managing our project at Illumina, Jay Tischfield of the Rutgers Cell Repository for providing DNA samples from the NIMH Genetics Initiative families, Dani Fallin for statistical advice, many research assistants, interviewers and best estimate diagnosticians for helping generate the samples, and the patients and families for generously giving their time for this study.

Ancillary