|Home | About | Journals | Submit | Contact Us | Français|
Numerous candidate gene association studies of bipolar disorder (BP) have been carried out, but the results have been inconsistent. Individual studies are typically underpowered to detect associations with genes of small effect sizes. We conducted a meta-analysis of published candidate gene studies to evaluate the cumulative evidence. We systematically searched for all published candidate gene association studies of BP. We then carried out a random-effects meta-analysis on all polymorphisms that were reported on by three or more case–control studies. The results from meta-analyses of these genes were compared with the findings from a recent mega-analysis of eleven genome-wide association studies (GWAS) in BP performed by the Psychiatric GWAS Consortium (PGC). A total of 487 articles were included in our review. Among these,33 polymorphismsin 18genes were reported on by three or more case–control studies and included in the random-effects meta-analysis. Polymorphisms in BDNF, DRD4, DAOA, and TPH1, were found to be nominally significant with a P-value < 0.05. However, none of the findings were significant after correction for multiple testing. Moreover, none of these polymorphisms were nominally significant in the PGC-BP GWAS. A number of plausible candidate genes have been previously associated with BP. However, the lack of robust findings in our review of these candidate genes highlights the need for more atheoretical approaches to study the genetics of BP afforded by GWAS. The results of this meta-analysis and from other on-going genomic experiments in BP are available online at Metamoodics (http://metamoodics.igm.jhmi.edu).
Previous research from family, twin, and adoption studies have shown that genetic factors play an important role in the etiology of bipolar disorder (BP) [Todd and Botteron, 2002; Merikangas and Low, 2004]. Since the completion of the human genome project [Venter et al., 2001] over a decade ago, there has been an explosion in genetic association studies aimed at identifying susceptibility genes for BP. Studies carried out before the era of genome-wide association studies (GWAS) typically focused on specific genes suggested by various hypotheses about the underlying biology of BP. Reports from these candidate gene studies have implicated a number of “plausible” susceptibility genes for BP, but the findings have been inconsistent, making it difficult to draw definitive conclusions. We set out to review the literature and perform a meta-analysis of existing candidate gene studies to provide a comprehensive summary of findings on the leading susceptibility genes for BP to have emerged from the field over the past decade. We compared the results from the published candidate genes studies against those from the most comprehensive GWAS of BP to date [Sklar et al., 2011] carried out by the Psychiatric GWAS Consortium (PGC) to determine if the previous findings were corroborated by more recent genome-wide effort.
A literature search was conducted of the PubMed database on all articles published up through September 1, 2009 using the following keyword algorithm: (bipolar depression OR bipolar disorder OR mood disorder OR affective disorder) and (single nucleotide polymorphism OR SNP OR gene OR polymorphism). A total of 5,384 articles were returned. These were manually reviewed by looking at their titles, abstracts, keywords, and full text as needed in order to identify those that were case–control or family-based candidate gene association studies of BP meeting our inclusion/exclusion criteria described below. We further searched the references of these articles to identify any other articles that were potentially missed by the initial PubMed search. The rationale for using the September 1, 2009 cutoff for the PubMed search was so we could compare the results of our candidate gene meta-analyses to the results from the PGC, which began its efforts to conduct mega-analyses of existing GWAS in 2009 [Sklar et al., 2011].
Studies were included in our review if they met the following criteria: (i) reported on candidate gene association studies (case–control or family-based) in adult humans; (ii) examined individual level genotype data (i.e., no pooled genotyping studies) with an appropriate set of controls; (iii) tested for association with a dichotomous phenotype of BP that included diagnoses of bipolar I disorder (BPI), bipolar II disorder (BPII), and/or schizoaffective disorder bipolar sub-type (SABP); (iv) no overlap with other studies (if there was overlap with another study, we included the study with the most inclusive data); and (v) published in a peer-reviewed journal in English. Studies were excluded from our review if they: (i) did not distinguish between types of mood disorders; (ii) examined drug response or other quantitative traits; (iii) examined specific clinical features and/or co-morbidities within mood disorders; (v) did not report on primary data; and (vi) did not provide any characterization of the study sample. A total of 487 articles reporting on associations with 362 unique genes met our inclusion/exclusion criteria. Figure 1 shows the distribution of publication of these 487 articles by year.
We extracted into evidence tables the following data from each of the published articles: (i) PubMed ID, author, journal, and year of publication; (ii) study design; (iii) sample information including country of origin, ethnicity, source of ascertainment, and subject counts; (iv) diagnoses, diagnostic instrument, and diagnostic criteria used; (v) polymorphism information including dbSNP [Sherry et al., 2001] reference SNP identification number (rsID), base pair (BP) location, major/minor allele, and genotype/allele counts for cases and controls or transmitted/untransmitted alleles for families; and (vi) sample meta-information including percentage of males for cases and controls, age at interview and age at onset. To assure the data for each polymorphism was recorded with a common orientation across all studies, we used the HapMap [International HapMap Consortium, 2003]—release 24, Phase I and II CEPH (Centre d’Etude du Polymorphisme Humain, Utah residents with ancestry from northern and western Europe) to designate the strand orientation and major/minor alleles. If a polymorphism was not genotyped in the HapMap CEPH samples, we used the SNP database of the National Center of Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov/SNP/) dbSNP [Sherry et al., 2001] version 130 with human genome assembly build 36 and the University of California Santa Cruz (UCSC) Genome Browser Database [Fujita et al., 2011] (http://genome.ucsc.edu/) to determine a common orientation. Many studies did not provide rsIDs, but instead reported codes based on nucleotide or amino acid position (e.g., T102C in HTR2A or Thr25Asn also in HTR2A). To retrieve rsIDs for such polymorphisms we again used the dbSNP database. If a polymorphism still could not be mapped to an rsID we used the UCSC’s genome browser In-Silico PCR primer-BLAT [Kent, 2002] to map the polymorphisms using the provided primer sequences. When only allele counts were provided, we calculated the corresponding genotype counts assuming Hardy–Weinberg equilibrium. If the polymorphism was not in Hardy–Weinberg equilibrium in the controls, we kept the genotype counts as missing.
We conducted a meta-analysis for all polymorphisms reported on by at least three case–control studies. We focused only on the case–control association studies in the meta-analysis in order to compare our findings with the larger case–control GWAS. Allelic odds ratios (OR), standard errors (SE), 95% confidence intervals (CI), P-values (P), and HWE for each polymorphism were calculated individually for each study included in the meta-analysis. The minor allele for the polymorphisms was used as the reference to calculate the ORs. For polymorphisms with multiple alleles (e.g. variable number of tandem repeats (VNTRs) and microsatellite markers) we used the most common allele as the reference allele and calculated separate OR’s for all other alleles relative to this. We excluded from the meta-analysis studies with genotype data that was not in HWE in controls. For each polymorphism, we calculated Woolf’s chi-squared test [WOOLF, 1955] to assess for between-study heterogeneity. Woolf’s statistic is distributed as approximately χ2 with k – 1 degrees of freedom (df), where k is the number of studies. Allelic summary OR and 95% CI were estimated under the DerSimonian–Laird random-effects model [DerSimonian and Laird, 1986] using inverse-variance weights in R statistical programming [Development Core Team, 2010]. A random-effects model allows for both between-study and within-study heterogeneity, while a fixed-effects model considers only within-study heterogeneity. When there is no evidence of heterogeneity between studies, the random-effects and fixed-effects models yield similar results. The significance of the summary OR was determined using an asymptotic Z-test. We corrected the nominal P-values by the number of candidate genes that were meta-analyzed to control for multiple testing.
To further explore the relevance of the BP candidate genes included in the meta-analysis, we compared our results against those from a recent mega-analysis of GWAS carried out by the PGC on BP (PGC-BP) [Sklar et al., 2011]. The PGC-BP performed a combined analysis of individual-level data (i.e., mega-analysis) from 11 case–control GWAS of BP including 16,730 subjects (7,481 BP cases and 9,249 controls). The primary analyses consisted of allelic tests of associations for ~2,427,090 directly genotyped or imputed SNPs across the genome. The untyped SNPs in the individual studies were imputed based on the HapMap—release 24, Phase I and II CEPH samples as a reference. The association tests were carried out using logistic regression with estimated allelic dosages and controlling for five principal components capturing ancestral background and indicator variables for the 11 studies.
We examined all SNPs in HWE and with imputation quality r2 > 0.30 in an area 10-kb upstream and downstream of the longest RefSeq transcript of each candidate gene We compared results for the same variant that was examined in the candidate gene meta-analysis. If the same variant was not genotyped or imputed in the PGC-BP mega-analysis, then we examined the SNP that was in highest linkage disequilibrium with it and/or nearest in physical distance.
Of the 362 unique genes reported on by at least one of the 487 candidate gene association studies of BP, a total of 50 were examined by three or more such studies (case–control or family-based). These genes and the number of studies reporting on each of them are shown in Figure 2. The most widely studied gene by far was the serotonin transporter gene (SLC6A4 gene) with 41 different genetic studies of BP. The second most widely studied gene, 5-hydroxytryptamine (serotonin) receptor 2A (HTR2A), was examined in half as many studies (n = 21).
Eighteen of the 50 candidate genes identified here had 33 different polymorphisms that were examined in three or more case–control studies and therefore included in the meta-analysis. The 33 polymorphisms were investigated by a total of 140 unique studies (see Supplementary Table I for a list of references) and included 25 SNPs, two insertion/deletions (indels), and six VNTRs including four microsatellites, one minisatellite and one short tandem repeat (STR). Table I lists the top findings (i.e., lowest meta-analysis P-value) in each of the 18 candidate genes from the random-effects meta-analysis. Single polymorphisms in four different genes, BDNF, DRD4, DAOA, and TPH1, were found to be nominally significant with a P-value < 0.05. There was no evidence of heterogeneity between studies of these four polymorphisms (P > 0.05). The most significant polymorphism study-wide was rs1800532 (G/T) in tryptophan hydroxylase 1 (TPH1) (OR for the minor T allele = 1.24, 95% CI 1.09–1.41, P = 0.001). The meta-analysis of rs1800532 included data on a total of 1,848 cases and 2,075 controls from eight individual case–control studies. However, none of the findings, including with rs1800532, were significant after correcting for the 18 different genes that were meta-analyzed. We re-did the meta-analyses for these top 18 polymorphisms after removing non-Caucasian samples, but the results did not change meaningfully.
We compared our meta-analysis results against the recent mega-analysis of GWAS from the PGC-BP. We inspected the results from the same SNP if it was present in the PGC-BP data, or if not the one in highest linkage disequilibrium and/or nearest to the target SNP. Of the 18 top polymorphisms in 18 genes, there were 12 SNPs of which 8 of the same SNPs were present in the PGC-BP data. Of the remaining four SNPs, one of these was on the X chromosome (data not available from the PGC-BP) and for the remaining three the nearest neighbors were extracted. None of the six non-SNP polymorphisms were genotyped in the PGC-BP data, thus only the closest/nearest neighbors were extracted. Of the four polymorphisms that were nominally significant in our gene meta-analysis, none were nominally significant in PGC-BP. In fact, only one SNP of the 18 candidate genes (nearest neighbor to rs334558) included in our meta-analysis was nominally significant in the PGC-BP data (rs4688059 in GSK3β, OR = 1.15, P = 0.025).
Although family, twin, and adoption studies strongly implicate the role of genetic factors in the etiology of BP, genetic association studies of the disorder have been inconclusive. To provide a comprehensive assessment of the existing evidence, we conducted a literature review and meta-analysis of published candidate gene association studies. We identified 18 candidate genes with a total of 33 polymorphisms that were reported by three or more case–control association studies. Of these, there were four polymorphisms in four different genes that were found to be nominally significant upon meta-analysis. However, these did not remain significant after correcting for the 18 genes included in the meta-analysis. It should be noted that the Bonferroni correction method used in this context may be considered too conservative, in that each gene (and in many cases individual variants) represent specific hypotheses. Nevertheless, none of these polymorphisms were even nominally significant in the recently reported GWAS from the PGC-BP.
The lack of significant findings with previously implicated candidate genes in BP is in some ways telling. These genes were largely selected for investigation based on various hypotheses about the etiology of BP, and they emerged from the existing literature as leading candidates and attracted considerable attention in the field. However, that none appear to be associated with BP upon closer inspection suggests our understanding of the etiology of BP was inaccurate. However, these candidate genes should not be ruled out and it is still possible these genes do contribute to BP through rare variants or other types of variants not previously studied or captured by GWAS, but the overall evidence is not encouraging. These findings highlight what most have already come to accept, that is, that more atheoretical genome-wide approaches that are not constrained by our limited understanding of the disorder will be needed to shed light on its complex etiology.
This review provides a comprehensive evaluation of the existing literature on candidate genes studies of BP. Studies reported in the literature may be subject to publication biases in which positive studies are more likely to be published and this may unduly influence the inferences drawn in summarizing the findings. However, this is not a particular concern here, because all the results were essentially null. This review also provides a comparison with more recent GWAS findings, which were consistent with the conclusions from the meta-analysis. One concern with this comparison is that subjects included in the candidate gene studies may also have been included in the later GWAS, especially since investigators have attempted to GWAS every available BP sample, leading to non-independence in the comparison. Again, because the results were null, this is less of a concern.
It is clear that the etiology of BP is very complex. An increasing number of genomic experiments are being carried out to make sense of the complexity. It will be important to continue to monitor and evaluate the accumulating evidence about the role of different genetic factors that are identified by these studies using systematic reviews similar to the one carried out here. We have developed an online resource available to the research community called Metamoodics (http://metamoodics.igm.jhmi.edu) which will continue to gather and systematically review the results from these studies and provide tools for their quantitative analyses.
The authors declare that there are no conflicts of interest.
Additional supporting information may be found in the online version of this article.