|Home | About | Journals | Submit | Contact Us | Français|
Straub et al. (2002b) located a susceptibility region for schizophrenia at the DTNBP1 locus. At least 40 studies (including one study in US populations) attempted to replicate this original finding, but the reported findings are highly diverse and at least five pathways by which dysbindin protein might be involved in schizophrenia have been proposed. The present study aimed to test the association in two common US populations by using powerful analytic methods.
Six markers at DTNBP1 were genotyped by mass spectroscopy (“MassARRAY” technique) in a sample of 663 subjects, including 346 healthy subjects [298 European-Americans (EAs) and 48 African-Americans (AAs)] and 317 subjects with schizophrenia (235 EAs and 82 AAs). Thirty-eight ancestry-informative markers (AIMs) were genotyped in this sample to infer the ancestry proportions. Diplotype, haplotype, genotype, and allele frequency distributions were compared between cases and controls, controlling for possible population stratification, admixture, and sex-specific effects, and taking interaction effects into account, using a logistic regression analysis (an extended structured association (SA) method).
Conventional case-control comparisons showed that genotypes of the markers P1578 (rs1018381) and P1583 (rs909706) were nominally associated with schizophrenia in EAs and in AAs, respectively. These associations became less or non-significant after controlling for population stratification and admixture effects (using SA or regression analysis), and became non-significant after correction for multiple testing. However, regression analysis demonstrated that the common diplotypes (ACCCTT/GCCGCC or GCCGCC/GCCGCC) and the interaction effects of haplotypes GCCGCC × GCCGCC significantly affected risk for schizophrenia in EAs, effects that were modified by sex. Fine-mapping using δ or J statistics located the specific markers (δ: P1328; J: P1333) closest to the putative risk sites in EAs.
The present study shows that DTNBP1 is a risk gene for schizophrenia in EAs. Variation at DTNBP1 may modify risk for schizophrenia in this population.
At least twenty-three complete or nearly complete genome scans for schizophrenia in 27 samples have been published, which have localized risk regions for schizophrenia to numerous different chromosomes (reviewed by Sullivan 2005). Since Straub et al. (1995) and Kendler et al. (1996) initially reported the linkage of markers mapped to chromosome 6p24-21 to schizophrenia spectrum disorders, there have been at least 15 additional linkage studies; of these, at least 7 provided supportive evidence for susceptibility loci on chromosome 6p (Schwab et al. 1995; Levinson et al. 1996; Maziade et al. 1997; Lindholm et al. 1999; Turecki et al. 1997; Straub et al. 2002a; Lewis et al. 2003). These susceptibility loci span a broad region of 25Mb between D6S296 and D6S291, including four possibly distinct subregions: 6p25-24, 6p24, 6p23-22, and 6p21 (reviewed by Straub et al. 2002b). Association studies using linkage disequilibrium mapping methods have served to fine-map the risk alleles within these subregions. Using a family-based association method, Straub et al. (2002b) initially identified the dystrobrevin-binding protein 1 gene, i.e., the dysbindin gene (DTNBP1) at 6p22.3, as a susceptibility gene for schizophrenia based on a set of 270 Irish high-density pedigrees. They found several polymorphisms within this gene that associated with schizophrenia (see Table 2). At least 40 family-based or population-based association studies have attempted to replicate this initial finding (summarized in Table 1), although not necessarily in the strict sense of repeating the design and methods of the initial study. One linkage study in an Israeli isolate directly located a risk region for major psychiatric disorders at the DTNBP1 locus (Kohn et al. 2004). At least twenty-two association studies supported the associations between DTNBP1 and schizophrenia in different populations, but seven did not. However, the positive findings from these studies were variable: (1) Some putative risk alleles (even at the same marker locus) are minor alleles in some populations [e.g., P1635^G in Irish (Straub et al. 2002b)] but common alleles in other populations [e.g., P1635^A in Bulgarian (Kirov et al. 2004) and in German-Israeli (Schwab et al. 2003)]; some risk haplotypes are rare in some populations [e.g., in Irish (Straub et al. 2002b; van den Oord et al. 2003)] but common in other populations [e.g., in German (Schwab et al. 2003), in Chinese (Tang et al. 2003) and in Japanese (Numakawa et al. 2004)], and some common or rare haplotypes protect against disease (Williams et al. 2004); (2) Some markers or haplotypes, even in the same population (e.g., Irish), are associated with schizophrenia in some studies (e.g., Straub et al. 2002b and Williams et al. 2004), but not in other studies (e.g., Morris et al. 2003); (3) The most significantly-associated risk markers are different across different studies (e.g., P1635 in Straub et al. 2002b, Kirov et al. 2004 and Numakawa et al. 2004; but P1320 in Schwab et al. 2003); (4) The risk or protective haplotypes have different block boundaries in different populations and the numbers of these haplotypes differ among studies.
Most researchers (e.g., Straub et al. 2002b; Funke et al. 2004; Bray et al. 2005) have attributed the diversity of findings to allelic heterogeneity or haplotypic heterogeneity per se. However, other potential issues need to be considered, particularly, sampling bias and sampling variance. The sampling in all of these studies was non-random. Serious sampling bias may lead to “surprising” (i.e., unexpected) findings that might be false. Mild sampling bias or sampling variance may lead to inconsistent findings within a single gene, because of various stratification effects from variables such as population, familiality, age, sex, etc., which are not random in the sampling. (1) Population stratification effects. Different populations have different evolutionary histories with different numbers of generations. The difference in generation number (i.e., the age of the population) leads to a difference in recombination that leads to LD decay and thus results in different haplotype block sizes between populations. Population-specific gene frequency distributions and haplotype block sizes often lead to population-specificity of associations between genes and diseases. To guard against false positives, and because of unfeasibility of random sampling from all populations, many researchers limit sampling to one specific population (even for family-based studies) or conduct analyses only within single populations. Thus, the findings are also limited to that specific population, and replication is required in other populations if the findings are to be generalized. Although more than 30 studies have attempted to replicate the original linkage findings, the results were highly diverse; only two studies were performed in US populations. Additionally, ten studies were performed in at least seven European subpopulations, with diverse findings (see Tables 1, ,2),2), which suggests that the European-Americans that originate from different geographic regions should be taken as potentially admixed. African-Americans (AAs) and Hispanics have high admixture with EAs (Parra et al, 1998; Hoggart et al, 2003), and thus should also be taken as admixed populations. Within a single population, especially those admixed populations, population-based studies are also vulnerable to admixture effects, but this has not been considered in previous studies (see Table 1). Although the family-based design is thought to be immune to population stratification effect, it may be not immune to other stratification effects. (2) Familiality stratification effects. Familial patients represent a specific subgroup, different from sporadic patients. The sporadic/familial distinction might lead to the different findings between the family-based studies and the population-based studies, even in the same population. Familiality could confound the association analysis. (3) Age stratification effects. Schizophrenia is an age-dependent phenotype. Age could be a stratification factor confounding the association analysis. (4) Sex stratification effects. Sex-specificity of schizophrenia has been reported by several studies (e.g., Franzek and Beckmann, 1992; Hafner et al. 1993; Kitamura et al. 1993; Sham et al. 1994; Leung and Chue, 2000; Aleman et al. 2003). In the present study, we stratified the sample by sex or took sex as a covariate in the regression analysis. (5) Other known or unknown factors might generate stratification effects that contribute to the diversity of findings. Because completely random sampling so as to randomly distribute these factors usually is unfeasible, replication is very important. As for previous studies, important factors were not randomized in our case-control sample, but the stratification effects of the main confounding factors, including population and sex, were controlled for in our analysis. Additionally, our sample is reasonably representative of the general population, based on similar allele and haplotype frequency distributions of different markers to those from other independent studies within the same populations, e.g., the OPRM1 haplotype frequency distribution in our AA controls (Luo et al, 2003) is similar to that in the study by Crowley et al. (2003) (comment in Luo et al, 2005a; other comparisons are not shown here), thus, our sample is apparently reasonable for a replication study.
The diversity of findings may also result from design variance and variation in methods among studies. (1) The replication studies used family-based and population-based association designs, which differ in power, e.g., several studies claimed that case-control studies can be more powerful than family-based studies in identifying disease genes, both for qualitative traits (Risch et al. 1996; 1998; 2000) and for quantitative traits (Van den Oord. 1999). (2) Multiple genes with minor effects might contribute to the risk for complex diseases. Detection of these minor effects is sensitive to study power. Differing power may be due to different designs, different sample sizes, different marker sets, and different analytic methods, which may lead to different results. For example, most studies do not exactly “repeat” the initial design and methods, but involve further exploratory analyses and aim to generally replicate the findings in the sense of identifying some relationship between markers or haplotypes at the locus, and the phenotype.
In summary, the present study aimed to replicate the study by Straub et al. (2002b), but adopted design features intended to overcome some limitations present in some other replication studies. These include (1) controlling for population stratification, admixture effects, and sex stratification effects; (2) preserving the haplotypes with unknown phases in the analysis; (3) taking marker-marker epistasis into account; (4) waiving the requirement of the HWE assumption on haplotype reconstruction; (5) avoiding multiple tests due to involving multiple populations and multiple markers; and (6) fine-mapping the risk sites.
Six hundred sixty-three subjects were included in the study: 346 healthy controls [298 European-Americans (EAs) and 48 African-Americans (AAs)] and 317 subjects with schizophrenia (235 EAs and 82 AAs). Four hundred twenty-seven subjects were male and 208 were female. Males constituted 98.2% of the cases and 41.3% of the controls. Cases and controls showed a roughly matched age distribution. The population groups for individual subjects were classified by ancestry proportions rather than self-report (see below). The diagnosis of schizophrenia was according to DSM-III-R criteria (American Psychiatric Association, 1987) as determined by the Structured Clinical Interview for DSM-III-R (SCID) (Spitzer et al, 1992). The control subjects were screened using the SCID, the Computerized Diagnostic Interview Schedule for DSM-III-R (Blouin et al, 1988), or the Schedule for Affective Disorders and Schizophrenia (Spitzer and Endicott 1975) to exclude major Axis I disorders, including substance dependence, psychotic disorders (including schizophrenia or schizophrenia-like disorders), mood disorders, and major anxiety disorders.
Subjects were recruited at the VA Connecticut Healthcare System, West Haven Campus, the University of Connecticut Health Center, or 14 other Veterans Affairs medical centers (described in Rosenheck et al, 1997). The study was approved by the Institutional Review Boards (IRB) at Yale University School of Medicine, University of Connecticut Health Center, VA Connecticut Healthcare Center, and in some cases additional IRBs at sample collection sites. All subjects signed informed consent, with the exception of a subsample collected at Highland Drive VA (Pittsburgh), which was determined by the Yale IRB to be exempt from review because the research involved use of existing anonymous samples.
Six markers within DTNBP1 were genotyped in the present study, including two markers (P1583: rs909706 and P1578: rs1018381) at intron 1, one marker (P1320: rs760761) at intron 3, one marker (P1655: rs2619539) at intron 5, one marker (P1333: rs742105) at intron 7, and one marker (P1328: rs742106) at intron 9 (see Table 3). These markers were selected from the original 12 markers in the study by Straub et al. (2002b), because they could be genotyped by multiplex PCR in the MassARRAY system, and we could validate their allele frequencies in a small sample prior to the high-throughput genotyping. All six markers have also been examined in many other studies and most of them were found to be associated with schizophrenia (see Table 2). The six markers span a total of 136Kb, with an average intermarker distance of 22kb. Most of them are tagSNPs in the HapMap database (www.hapmap.org), and cover most of the information content of DTNBP1.
Thirty-eight ancestry-informative markers (AIMs) unlinked to DTNBP1, including 37 STRs and one Duffy antigen gene (FY) marker (rs2814778) that is highly ancestry-informative, were also genotyped, to examine the population structure of our sample. These markers were employed in the studies by Stein et al. (2004), Kaufman et al (2004), and Luo et al. (2005b;c); their characteristics were described in the study by Yang et al. (2005), and the genotyping methods have been described in these studies.
Genomic DNA was extracted from peripheral blood by standard methods. The six SNPs were genotyped by Matrix-Assisted Laser Desorption Ionization - Time of Flight (MALDI-TOF) Mass Spectrometry via the Sequenom MassARRAY system (SEQUENOM, Inc., San Diego, CA, USA) in three 2-plex PCRs, using six pairs of primers. These multiplexes and primers were designed using the MassARRAY™ Assay Design Software and all primers were extended by a 5′ “cap” sequence “ACGTTGGATG” to increase the molecular weight of these primers to > 9000 daltons, so that any residual PCR primer would not interfere with the SNP genotyping software, that is, the PCR primers would not be in the mass range of 5000–9000 Daltons that is used in the genotyping process. PCR was performed in a final volume of 5 μl for each system, which included 2.5–5.0 ng genomic DNA, 200 nM each PCR primer for uniplex reactions or 50 nM each PCR primer for multiplex reactions, 200 μM each dNTP, 1 × HotStar buffer, HotStar Taq polymerase (Qiagen, Inc., Valencia, CA, USA), and 2.5 mM MgCl2. A strict validation experiment was performed prior to high-throughput genotyping: (1) PCR conditions for multiplex PCR were optimized based on the manufacturer’s recommendation until the genotypes completely agreed with those by the uniplex PCR; (2) One large CEPH family pedigree including 27 individuals, from whom DNAs were available through Coriell (http://locus.umdnj.edu/nigms/ceph/ceph.html), were genotyped by the optimized PCR to assure that the genotypes completely agreed with the Mendelian rule; (3) The success rate for each multiplex PCR was higher than 90%. Both positive controls (CEPH DNA sample) and negative controls (water and blank) were included in the high-throughput genotyping.
Pairwise LD between any two DTNBP1 markers was analyzed separately by population, i.e., EAs and AAs. The value of the standardized disequilibrium coefficient, D′, for each LD pair was calculated and the statistical significance for D′ was tested. HWE of the genotype frequency distribution for each marker was tested within different populations, and separately in cases and controls.
The allele and genotype frequencies of the DTNBP1 markers in different phenotype groups are shown in Table 4. Associations between either the alleles or the genotypes and the phenotypes were analyzed by comparing the allele and genotype frequency distributions between cases and controls (within EAs and AAs, respectively) with exact tests. All of the above tests were performed via PowerMarker software (Liu & Muse, 2004).
EAs and AAs can be taken as admixed populations with different degrees of admixture (Parra et al. 1998; Hoggart et al. 2003; Shriver et al. 2003; Collins-Schramm et al. 2004). The extent of admixture (i.e., ancestry proportions) can be estimated using the program STRUCTURE (Pritchard et al, 2000a) to analyze the 38 AIMs (Yang et al, 2005; Luo et al, 2005c). The case-control design is vulnerable to admixture effects, but the admixture effects on case-control association analysis can be controlled for using the program STRAT (Pritchard et al, 2000b), which adjusts for ancestry proportions to yield a so-called structured association (SA) analysis. The SA method is limited to genotypewise and allelewise analyses. Therefore, the ancestry proportions were also entered into the regression models described below for an extended analysis, which included diplotypewise, haplotypewise, genotypewise, and allelewise analyses and tested for the population-specificity of associations.
In the present study, in order to increase statistical power by expanding the sample, EAs and AAs were combined as a single admixed sample for association study. Then, EAs and AAs were analyzed separately to identify the sources of the observed associations.
The program PHASE was used to reconstruct haplotypes and to estimate the probabilities of all likely pairs of haplotypes (i.e., diplotypes) for every individual in this study. This program was developed by Stephens et al. (2001; 2003), based on a Bayesian approach and the Partition Ligation algorithm. These algorithms may be more accurate in reconstructing haplotypes than the Expectation-Maximum (EM) algorithm, especially when the HWE does not hold among some markers, as is the case for our data (see Table 4) (Stephens et al. 2001; Stephens and Donnely 2003; Niu et al. 2002). In spite of its advantages, PHASE still has its limitations, and thus the below regression analysis on the diplotype and haplotype probabilities estimated by PHASE has to be considered as being exploratory. The haplotypes were reconstructed within two separate subgroups, that is, the genetically-inferred EAs (European ancestry proportion>0.5) and the genetically-inferred AAs (African ancestry proportion>0.5).
A backward stepwise logistic regression analysis was used to test associations between gene and disease. We modeled the analysis with the following equation: ln[p/(1−p)]= β0 + ΣβiXi + ΣβijXiXj, where p is probability of disease; XiXj is the interaction between Xi and Xj; β is regression coefficient; βi can be interpreted as the magnitude of main effect of Xi, when all other predictor items are equal to 0; Σβ=βi +βij can be interpreted as the magnitude of total effect of Xi, when Xj=1 and all other Xs are equal to 0; other Xs can be interpreted similarly to this. Four kinds of regression models were employed in the present study: Xi includes African ancestry proportions predicted by the program STRUCTURE, sex of individuals, and diplotype probabilities (model 1), haplotype probabilities (model 2), genotypes (model 3), or alleles (model 4). In models 1 and 2, only diplotypes or haplotypes with frequencies >0.01 (see Table 5) were included; the interaction effects between haplotypes were also considered (diplotypes and haplotypes per se have incorporated the interaction information between SNPs). In models 3 and 4, only two genotypes and one allele from each SNP were included, respectively; and the two-way interaction effects between alleles or between genotypes from different SNPs were included as well. In all four models, the interaction effects between sex and diplotypes, haplotypes, genotypes, or alleles were also considered.
Regression analysis using the haplotype probabilities as predictors (model 2) is called haplotype trend regression (HTR). The probabilities, instead of the categories, of haplotypes being included in HTR makes HTR more powerful, because the probabilities preserve more information than does the direct use of categorical variables. The rationale of HTR was first described by Zaykin et al. (2002) and HTR has been widely applied. Regression analysis using the diplotype probabilities as predictors (model 1) is called diplotype trend regression (DTR). DTR has been successfully applied in many previous studies (e.g., Luo et al., 2005b; c; 2006) and its advantages have been demonstrated. DTR increases effective sample size by combining different populations in a single model, avoids multiple testing that would accrue due to the inclusion of multiple populations and markers, controls for population stratification and admixture effects and the potential confounding by sex, allows uncertainty for haplotype inference, obviates the HWE assumption, and takes marker–marker interactions into account.
At a single locus, two alleles could be incorporated into a genotype. Similarly, at the multiple loci, the haplotype information content could be incorporated into the diplotype. The information content of alleles and genotypes from multiple loci could be incorporated into multi-locus haplotypes and diplotypes, respectively. Therefore, the above four regression models are not independent of each other; they actually are equivalent to a single regression model that does not require for correction for multiple testing. Among the four regression models, the diplotype trend regression model (model 1) is most powerful.
Within each regression analysis, multiple predictor variables are tested. These kinds of multiple testing are corrected by the degree of freedom. Thus, p-values derived from the regression analysis do not require for further correction for multiple testing and the significant level (α) is set at 0.05.
Many measures for LD in case-control samples, e.g., the population attributable risk δ (Levin and Bertell, 1978; Devlin and Risch 1995), have been advanced as tools to fine-map risk loci. Many measures for HWD in case-only samples have also been advanced to fine-map the risk loci, including F, F′, J and J′ (Feder et al. 1996; Jiang et al. 2001). These statistics were used for fine-mapping the risk locus in the present study. Because there are no methods available to test the statistical significance for δ or J statistics, we used diplotype trend regression analysis or haplotype trend regression analysis (at the “whole gene” level) to test the statistical significance of gene-disease association first, and then we used a δ or J statistic to fine-map the risk site (at a “single-point” level) within this gene. The marker with the highest δ or J value is thought to be closest to the putative disease locus.
The present study demonstrated that the diplotypes and haplotypes at DTNBP1 locus affected risk for schizophrenia in EAs. We conclude that DTNBP1 is a risk gene for schizophrenia, and it may harbor a risk locus for the disorder. The present study has also provided additional map information regarding the probable location of functional variants within the locus.
Conventional case-control comparisons on allele and genotype frequency distributions showed that two polymorphisms (P1578: p=0.015 and P1583: p=0.052) were nominally associated with schizophrenia in EAs and in AAs, respectively. The associations became less significant after controlling for population stratification and admixture effects using the SA method and were no longer significant after the correction for multiple tests. These findings suggest that the methods were not powerful enough to identify the gene as a susceptibility gene, possibly due to a small effect size. Moreover, these two methods have other limitations. For example, the conventional case-control comparison method is vulnerable to population stratification and admixture effects; the SA method cannot handle unphased haplotype data; both methods are limited by multiple testing; potential confounders such as sex and age cannot be controlled for by either method; and neither method is capable of considering marker-marker interaction effects. These limitations reduce the statistical power, accuracy, and robustness of both methods, so that the results are considered to be exploratory.
Regression analysis overcomes these limitations and thus increases the statistical power and leads to more accurate and robust findings. Cases and controls, and EAs and AAs, were combined in one regression model to increase sample size; different markers were entered in one regression model to avoid multiple tests; ancestry proportions were entered as a covariate in the regression model to control for population stratification and admixture effects on association analysis; data on sex were entered in the regression model as a covariate to take into account the sex-specificity of the prevalence of schizophrenia and correct for asymmetric sampling of cases and controls, thereby controlling for its stratification effects and potential confounding effects on the association analysis; the phased and unphased diplotype and haplotype data, which are thought to contain more information than single markers in many cases, were included in the analysis; finally, marker-marker interaction effects and marker-covariate interaction effects were considered to avoid erroneously interpreting the main effect of each marker in the presence of a significant interaction.
P1328 was found to be in HWD in AA cases but in HWE in AA controls, which may be an indication of association between P1328 and schizophrenia (Feder et al. 1996; Nielsen et al. 1999; Jiang et al. 2001; Hoh et al. 2001; Lee 2003; Hao et al. 2004; Wittke-Thompson et al. 2005; Luo et al. 2005b). P1320 and P1578 were in HWD in EA controls, and P1333 and P1655 were in HWD in AA controls, which most likely resulted from sampling bias, or unrecognized copy number variation in this genomic region [such variation in other regions has been related to schizophrenia risk (Walsh et al, 2008)]; it is unlikely to have resulted from genotyping errors [the genotyping missing rates for these four markers in those groups were 2.01%, 2.68%, 2.08% and 0%, respectively]. The presence of HWD led us to use a Bayesian approach and the Partition Ligation algorithm instead of the Expectation-Maximum (EM) algorithm to reconstruct diplotypes and haplotypes. When the diplotype and haplotype data were analyzed in our regression models, the regression method was independent of the HWE assumption. Additionally, the predicted diplotype and haplotype probabilities that can be analyzed by regression methods are continuous variables, which usually are more informative than diplotypes or haplotypes (i.e., categorical variables).
In view of the advantages of the regression method and the limitations of conventional association analysis methods (including the HWD test, case-control comparison and the SA method), we believe that the results from the regression analysis are more accurate and robust, and the different results may reflect the greater accuracy and robustness of the regression method. Using regression analysis, we found that: (1) Males predominate in cases both in EAs and AAs in our sample. Although the incidence of schizophrenia differs by sex (McGrath et al, 2004), the imbalance on sex in our sample is mostly due to a sampling bias (males constituted 98.2% of the cases and 41.3% of the controls). Thus, sex data were taken as a key confounder for gene-disease association analysis and the interaction effects between sex and gene were also considered. (2) African ancestry was more common in cases (82/317 (25.9%) were AA) than controls (48/346 (13.9%) were AA). However, when we analyzed the data separately for EAs and AAs, we noted that African ancestry was also more common in EA cases than controls (2.9% vs. 1.0%; p=0.001), and European ancestry was more common in AA cases than controls (5.8% vs. 2.8%; p=0.052), suggesting that the degree of admixture per se was higher in patients with schizophrenia than in controls. This asymmetry in the degree of admixture between cases and controls is probably also attributable to sampling bias. It is difficult to avoid biased sampling in relation to admixture, since it is not feasible to measure the degree of admixture clinically in order to match cases and controls during the sampling process. Instead, a genetic experiment makes it possible to measure the extent of admixture, so that its potential confounding effects on association analysis can be controlled for. An alternative explanation for the association between the degree of admixture and schizophrenia is that admixture per se may increase risk for schizophrenia. Further studies are warranted to test this hypothesis. (3) In the combined sample or in EAs, the most common diplotype and haplotype were ACCCTT/GCCGCC (f=0.161) and GCCGCC (f=0.370), respectively; the second most common diplotype is GCCGCC/GCCGCC (f=0.153). In AAs, the most common diplotype and haplotype were GCCGCC/GCCGCC (f=0.157) and GCCGCC (f=0.291). Generally, for the majority of individuals in a population, the common haplotypes and diplotypes protect against a disease that is present at low frequency in the population, as observed in the present study. However, these gene effects can be modified by sex. For example, in the combined sample, the common diplotype ACCCTT/GCCGCC and the interaction of the common haplotypes GCCGCC × GCCGCC protected against (β<0) schizophrenia only in males, which constituted the majority of our sample. This is basically consistent with the results from straightforward case-control comparison on diplotype frequency distributions, i.e., the frequencies of diplotype GCCGCC/GCCGCC were lower in male cases than male controls both in EAs and in combined sample. In females, the common diplotype and haplotype increased risk for schizophrenia (β>0), both in EAs and in the combined sample. This is also basically consistent with the results from straightforward case-control comparison on diplotype frequency distributions, i.e., the frequencies of diplotype GCCGCC/GCCGCC were higher in female cases than female controls both in EAs and in combined sample. (The findings in females might be chance findings given that only 2% of cases were women). The associations of diplotypes and haplotypes with schizophrenia suggest that DTNBP1 may harbor a disease locus for schizophrenia, which is in LD with these risk or protective diplotypes or haplotypes. The magnitude of the interaction effect of GCCGCC × GCCGCC in EAs (β=3.016) did not increase when combined with AAs (Σβ=1.967), which suggests that the gene effects were significant mainly in EAs and that the addition of AA subjects did not increase information. Fine-mapping using δ or J located the specific markers (δ: P1328; J: P1333) closest to the putative risk sites in EAs. These fine-mapping methods have limitations; for example, δ is subject to the assumption of HWE and J ignores the information from controls, which may explain the different localization provided by the methods.
All of the risk markers were located in introns spanning DTNBP1. They may affect risk for schizophrenia in three possible ways. First, these markers may be in LD with nearby functional variation. However, despite intensive resequencing efforts (e.g., Liao and Chen, 2004; Williams et al, 2004), no DTNBP1 coding variants have yet been identified. Second, DTNBP1 has at least 12 different known mRNA transcripts resulting from alternative splicing (Williams et al. 2004), and these markers may be involved in the post-transcriptional alternative RNA splicing process, so that genetic variation in pre-mRNA may yield distinct mature mRNAs, which can be translated to distinct dysbindin proteins with differing, or even opposing activities. Third, these intronic variants and their haplotypes per se might directly affect the expression of dysbindin protein in the brain and thus directly affect the susceptibility to schizophrenia. Recently, Bray et al. (2003, 2005) detected strong allele-specific and haplotype-specific expression of DTNBP1 in the brain, and several other studies have reported significant reduction of DTNBP1 expression in the brains of patients with schizophrenia (Weickert et al, 2004; Numakawa et al, 2004; McClintock et al, 2003; Talbot et al, 2004), raising the possibility that cis-acting variation may contribute to the role of DTNBP1 in the etiology of schizophrenia. Because the present study does not repeat the design and methods of the initial study, but involve novel approaches, replication of our findings is warranted in the future.
How dysbindin protein affects risk for schizophrenia is still under investigation. Because dysbindin protein, binding with β-dystrobrevin, is likely a component of the brain dystrophin protein complex (DPC) (Benson et al. 2001), Straub et al. (2002b) speculated that dysbindin protein’s involvement in the development of schizophrenia may be mediated by DPC via three possible pathways (reviewed by Straub et al. (2002b)). In addition, Talbot et al. (2004) identified a new presynaptic signaling pathway that is not mediated by DPC. Specifically, the dysbindin protein is also located in presynaptic glutamatergic neurons, independent of DPC. Presynaptic dysbindin reductions are frequent in schizophrenia and are related to glutamatergic alterations in intrinsic hippocampal formation connections. Such changes may contribute to the cognitive deficits common in schizophrenia. Finally, there has also been recent speculation that the mechanism involves a phosphatidylinositol 3-kinase -Akt (PI3-kinase-Akt) signaling pathway (Weickert et al. 2004; Numakawa et al. 2004; McClintock et al. 2003; Talbot et al. 2004; Emamian et al. 2004; Numakawa et al. 2004). Further research is needed to specify the mechanism(s) by which DTNBP1 contributes to the risk of schizophrenia.
This work was supported in part NIH grants R01-DA12849, R01-DA12690, K24-DA15105, R01-AA11330, P50-AA12870, K08-AA13732, K24-AA13736, R01-AA016015, K02-MH01387, and M01-RR06192 (University of Connecticut General Clinical Research Center), by funds from the U.S. Department of Veterans Affairs (the VA Medical Research Program, and the VA Connecticut Massachusetts Mental Illness Research, Education and Clinical Center [MIRECC], and the VA Research Enhancement Award Program [REAP] research center), and Alcoholic Beverage Medical Research Foundation (ABMRF) grant award R06932 (X Luo). Dr. Karl Hager provided valuable assistance with generating genotypes. The helpful comments of the two anonymous reviewers are highly appreciated.