Channeling Mental Illness: GWAS Links Ion Channels, Bipolar Disorder
19 August 2008. A collaborative study of bipolar disorder that combines data from two previously published genomewide association studies (GWASs) with a new GWAS associates two genes that encode components of voltage-gated ion channels with the illness. These findings raise the possibility that bipolar disorder may partly result from channelopathies, disruptions in ion channel subunits or other channel-related proteins. Such molecular lesions underlie a diverse group of disorders, including central nervous system disorders such as epilepsy, migraine, and ataxia.
The results of the study will be viewed with interest by researchers in schizophrenia genetics, as the bright diagnostic line between schizophrenia and bipolar disorder, first drawn by Emil Kraepelin in the late nineteenth century, is increasingly called into question. As has been pointed out by Cardiff, Wales-based Nick Craddock and Michael Owen, both authors on the new study, the emergence of overlapping candidate genes (e.g., DISC1, DTNBP1, NRG1) and the similarity between the psychotic phenotype seen in the manic phase of bipolar disorder and in schizophrenia suggest that there will be a growing cross-fertilization in genetic research in these two arenas (see Craddock and Owen, 2005, and SRF live discussion).
In the new multicenter study, bipolar cases and controls from the Wellcome Trust Case Control Consortium (the WTCCC study; Wellcome Trust Case Control Consortium, 2007; see SRF related news story) were added to those from the recent GWAS by Pamela Sklar and colleagues (the STEP-UCL study; Sklar et al., 2008, first reported in an SRF meeting report from the 2007 WPCG meeting), as well as a new sample of 1,098 cases and 1,267 controls, including some from the University of Edinburgh and Trinity College Dublin (the new data set was hence dubbed ED-DUB-STEP2), for a combined sample of 4,387 cases and 6,209 controls. In the case sample, 81 percent had been diagnosed with bipolar 1, and 16 percent with bipolar 2.
This large sample was directly genotyped on more than 325,000 overlapping SNPs, a number that was greatly increased, to about 1.8 million SNPs, when imputed HapMap SNPs were added using PLINK, a GWAS tool developed by study author Shaun Purcell of the Broad Institute of MIT and Harvard and colleagues (Purcell et al., 2007). “Applying a leave-out-one procedure for every genotyped SNP,” the authors write, “we estimated concordance between imputed and true genotypes as 0.987.”
Although there had been no overlap in the “top hits” identified in the WTCCC and the STEP-UCL studies—the former identified a gene-rich locus on chromosome 16, while the latter’s strongest SNPs were in MYO5B, TSPAN8, and EGFR—a post hoc comparison of the two data sets by the Sklar group found concordant SNP signals in CACNA1C, a gene on chromosome 12 that encodes the alpha 1C subunit (Cav 1.2) of the L-type voltage-dependent calcium channel.
In an initial analysis of the ED-DUB-STEP2 dataset alone, none of 14 chromosomal regions showing associations exceeded the researchers’ genomewide significance threshold of 5 x 10-8. However, one of these regions spanned the CACNA1C association found in the Sklar team’s analysis of the WTCCC and STEP-UCL samples, providing further support for the hypothesis that mutations in CACNA1C may contribute to bipolar disorder.
In the combined WTCCC/STEP-UCL/ED-DUB-STEP2 sample, the strongest association (P = 9.1 x 10-9) was found in ANK3 (Ankyrin-G) on chromosome 10q21, a gene that is required for the clustering of voltage-gated sodium channels at axon initial segments (Zhou et al., 1998) and nodes of Ranvier (Poliak and Peles, 2003), a configuration that underlies the rapid and efficient propagation of action potentials along myelinated axons. The second-strongest association was at rs1006737, in the third intron of CACNA1C (P = 7.0 x 10-8). A third association was found near C15orf53, a gene on 15q14 of unknown significance. No differential associations for these three regions were found across bipolar disorder subtypes, presence of psychosis, age of onset, sex, or response to treatment.
The ANK3 and CACNA1C associations reported in the new study are particularly intriguing in light of other recent work by study coauthor Hugh Gurling and colleagues (McQuillin et al., 2007), in which lithium carbonate, the gold standard for treatment of bipolar disorder, was shown to downregulate both ANK3 and subunits of the calcium channel. In addition, many antiepileptic drugs also used to manage symptoms in bipolar disorder and schizophrenia are known to affect voltage-gated sodium or calcium channels (Johannessen Landmark, 2008).
Though channelopathies have long been known to contribute to cystic fibrosis as well as several heritable cardiac (e.g., Brugada syndrome, Long QT syndrome) and motor (e.g., myasthenia gravis, periodic paralysis) disorders, the exploration of ion channel mutations in psychiatric disorders is a fairly recent development. However, an association between a missense mutation in CACNA1C and Timothy syndrome, a complex disorder that includes cardiac abnormalities, webbing of the fingers and toes, and autism, was recently reported (Splawski et al., 2004). KCNQ5, which encodes a potassium channel crucial for the M-current, an important regulator of neuronal excitability, maps to risk loci for attention-deficit hyperactivity disorder, bipolar disorder, and schizophrenia. Another channel gene, SK3, which encodes a small-conductance calcium-activated potassium channel, has also been implicated in schizophrenia (see Gargus, 2006, for a review of ion channel candidate genes in psychiatric disease).—Peter Farley.
Ferreira MAR and 62 others. Collaborative genome-wide association analysis supports a role for ANK3 and CACNA1C in bipolar disorder. Nat. Genet., published online 17 August, 2008.
Q&A With Nick Craddock. Questions by Hakon Heimer.
Q: This a rather large sample compared to previous ones, yet only three regions were found to have genomewide significance. Is that a disappointing result, and does it argue against either the multigenic theory of mental illness causation, on the one hand, or the use of 10-8 as a cut-off for determining genomewide significance on the other?
A: The findings of our analysis are entirely consistent with the existence of many genes each having a small effect on susceptibility to bipolar disorder. It is important to think beyond the top few “hits.” The pattern that we see in bipolar disorder is very much the same sort of pattern that we see in studies of non-psychiatric disorders. When we have access to even larger samples, more loci will achieve genomewide significance.
Q: Do you think there is validity to the argument that large-scale studies such as this may “wash out” signals of genes that contribute to disease in certain populations but not others?
A: Large-scale studies like ours provide optimal power to detect genetic variants that influence susceptibility broadly across the bipolar phenotype and across populations. It is, however, true that they may not be the optimal approach for detecting susceptibility variants that are either specific to certain populations or confer risk specifically to certain aspects of the clinical phenotype. In such situations, specific signals could be washed out.
Q: Is there any evidence from other sources to suggest that channelopathies might play a role in schizophrenia as well (or any evidence that they probably do not)?
A: This is an issue that has not previously received much attention—but obviously now warrants specific investigation.
Comments on News and Primary Papers
Comment by: Melvin G. McInnis
Submitted 19 August 2008
Posted 19 August 2008
The work by Ferreira et al. exemplifies the growing enthusiasm for collaborative work among investigators and marks the new era of collaborative genetic research in complex disorders. The LD data found in the extant HapMap SNPs allow investigators to use sophisticated computational approaches to impute genotypes based on these HapMap data sets and the data generated from the experimental sample, thereby maximizing the utility of the actual genotyping itself. Nothing short of brilliant. Correlates between imputed and true genotypes were estimated to be 0.987, which is quite good. The significance estimates of the combined data analyses of the three data sets identifies two genes (ANK3 and CACNA1C) in the genomewide significance range with a p value of 10-8, which is most reassuring and even more so considering that the CACNA1C gene was identified previously. The humbling fact in the mix is that the odds ratios are modest, ranging from 1.2 to 1.4, which is nonetheless in a similar arena as other complex genetic disorders such as diabetes. It is further humbling (and consistent with the modest ORs) to consider that the frequency of the risk allele for the CACNA1C gene is 7.5 percent in the BP cases and 5.6 percent in the unaffected control individuals. Finally, there was no effect of the sub-diagnostic categories, age of onset, presence of psychosis, or sex. The highly encouraging point is that these genes appear to be in pathways that are affected by lithium, the gold standard of care for BP disorder. The anchorage of a genetic finding within a mechanism of an established treatment for BP disorder (lithium) lends substantial credibility to overall results. The next questions of research will relate to the efficacy of lithium relative to genotypes of these genes and others within their pathways. These findings raise several clinical questions, and integration of clinical outcome patterns with genetic data can be expected to shed further light on the etiology of the disease and the genetics of treatment response. Long live lithium.
View all comments by Melvin G. McInnisComment by: John I. Nurnberger, Jr.
Submitted 19 August 2008
Posted 19 August 2008
Ferreira et al. propose two specific genes to be related to bipolar disorder, ANK3, which is indirectly related to sodium channels, and CACNA1C, which is a calcium channel subunit. They hypothesize that bipolar disorder is, at least in part, a channelopathy. This hypothesis is consistent with a number of physiological observations made over the past several decades, as reviewed elsewhere.
The genetic data these authors present is certainly suggestive. They have analyzed three independent data sets, STEP-UCL (Sklar et al., 2008), Wellcome Trust (Wellcome Trust Case Control Consortium, 2007), and a third set called ED-DUB-STEP2 (not yet published). Their total sample exceeds 4,000 cases and 6,000 controls. They have direct genotype data on >300,000 SNPs and have imputed nearly 1.5 million additional. Their highest significance values (10-7 to 10-9) include a combination of genotyped and imputed SNPs. For each of these, the combined p value is a product of modest but consistent associations in the three independent data sets.
ANK3 features rs10994336 at 9x10e-9 and rs1938526 at 1x10e-8. By my reading, these two polymorphisms are both slightly distal to the gene but the second is within 10-20 kB. The first of these is imputed, and thus the p value should probably be judged as more imprecise. Both of these polymorphisms are associated with an odds ratio of ~1.4 and a minor allele frequency of ~5 percent in controls.
The CACNA1C data is based on more common polymorphisms (~30 percent in controls) and an OR~1.2. Again two SNPs are featured (rs1006737 at 7x10e-8, genotyped, and rs1024582 at 2x10-7, imputed). A third region near an uncharacterized gene (on 15q14) is also featured.
Examination of available published data from STEP-UCL and WTCCC on ANK3 and CACNA1C does not show obvious evidence of association among SNPs across each of the named genes, but reasonably consistent signals of modest significance, which is what one might expect, and this does suggest that the featured SNPs are not completely anomalous, but may represent a pattern of genotype deviation across the two genes.
Needless to say, our investigators in the GAIN (Genetic Analysis Information Network) bipolar group are extremely interested in this report and are avidly following the lead provided by Ferreira to attempt to confirm these signals in our own data. I am also pleased to say that GAIN has provided the stimulus for an international consortium that includes representatives from the Ferreira group as well as many other investigators, dedicated to assembling yet larger samples of bipolar cases and controls to elucidate the genetics of this condition through genomewide methods.
This is an important report, and it may represent a breakthrough in bipolar genetic studies. The signals for ANK3 and CACNA1C appear very promising, and we hope that they prove to be consistently observed in other data sets as well. We anticipate that additional confirmed single genes will emerge soon as well, and that the genetic structure of these disorders will be elucidated using similar methods in large data sets in the coming years.
View all comments by John I. Nurnberger, Jr.Comment by: Peter P. Zandi
Submitted 21 August 2008
Posted 21 August 2008
Are we there yet? Have we in the field of bipolar genetics finally been delivered to the promised land by GWAS? For the past year or so since GWAS burst on the scene, we have had to watch with envy as an impressive list of genes were convincingly implicated in a range of other complex diseases like type 2 diabetes, the apparent poster child for GWAS. Now, is it our turn?
The first attempts at individual-level GWAS of bipolar disorder by WTCCC and STEP-UCL were exciting because of their novelty, but the results were not particularly overwhelming. None of the findings withstood correction for the massive multiple testing inherent in GWAS, and those at the top were of ambiguous relevance to bipolar disorder. Confronted with such uninspiring findings, one could not be faulted for experiencing pangs of doubt that maybe for psychiatric disorders, GWAS would prove no better than its dusty old predecessor, the genomewide linkage study, in illuminating the underlying genetic architecture.
Nevertheless, encouraged by the lessons learned from GWAS of type 2 diabetes that the road to the promised land is not paved in individual glory but in collaborations and consortiums, the investigators of WTCCC and STEP-UCL combined samples with a third previously unstudied collection (dubbed ED-DUB-STEP2) to assemble one of the largest samples in bipolar disorder yet to be analyzed by GWAS. The combined sample included 4,387 cases and 6,209 controls genotyped at 325,690 overlapping SNPs, which after imputation yielded data on 1.8 million variants. The results from this effort were recently reported in a manuscript by Ferreira and colleagues published in the latest edition of Nature Genetics.
Despite potential concerns about the genetic and/or clinical heterogeneity of the combined sample (e.g., the genomic inflation factor was estimated to be 1.11, even after controlling for two quantitative indices of population ancestry, which might suggest residual stratification or other unaccounted biases), the results from this effort are encouraging and provide some hope to those who may have been losing their faith. The most notable findings were in ANK3 on chromosome 10q21 and CACNA1C on chromosome 12p13. Multiple SNPs were associated across a 195-kb region of ANK3, and the top SNPs had p-values <5 x 10-8 which is often invoked as an appropriate threshold for genomewide significance in GWAS. Multiple nearby SNPs were also reassuringly associated in CACNA1C, although the top SNP just missed the threshold for genomewide significance. In both ANK3 and CACNA1C the top SNPs were consistently associated in the same direction across all three individual samples, lending credence to the claim that these associations are real. Lending further credence is the fact that ANK3 and CACNA1C are biologically plausible candidates for bipolar disorder, and indeed highlight the possibility that this disorder is an ion channelopathy. Interestingly, the associations with ANK3 and CACNA1C in each of the individual samples were relatively modest and became remarkable only in the combined sample, thus providing support for the rationale to combine samples in order to increase the power to detect those more modest signals that are presumably real but buried amidst the noise of a single study.
Although the evidence is promising, more samples will be needed to confirm the findings before we can say with confidence that we have in hand our first real bipolar susceptibility genes. Fortunately, several such samples have been or will soon be GWASed, including one from GAIN, and it will be of great interest to see whether the current findings are sustained in these new samples. Moreover, there are plans to combine all existing GWAS of bipolar disorder, as part of the initiative referred to as the Psychiatric GWAS Consortium, which should provide an even more definitive picture of the role of ANK3 and CACNA1C, as well as reveal other genes with more modest relative risks whose identities up until now have been obscured.
So, are we there yet? Maybe not just yet, but we are headed in the right direction and I think I can see the promised land.
View all comments by Peter P. Zandi
Comments on Related News
Related News: WCPG 2007—Schizophrenia, Bipolar GWA Results Prompt Calls for Bigger SamplesComment by: William Carpenter, SRF Advisor
Submitted 7 November 2007
Posted 8 November 2007
Terrific update and summary for those of us not attending the meeting.
View all comments by William Carpenter
Related News: Bigger Schizophrenia GWAS Yields More Hits
Comment by: Sven Cichon
Submitted 30 August 2013
Posted 30 August 2013
This paper is an important addition to the psychiatric genetics literature. One important message of it is that increasing the GWAS sample size in a complex neuropsychiatric phenotype such as schizophrenia identifies more common risk loci.
In the first-wave schizophrenia mega-analysis two years ago (Schizophrenia Psychiatric Genome-Wide Association Study Consortium, 2011), five risk loci at genomewide significance were detected, and it was speculated that more would follow with larger sample sizes. In fact, by performing a meta-analysis of a new Swedish schizophrenia sample (about 5,000 patients and 6,000 controls) and the first-wave PGC sample (about 9,000 patients and 12,000 controls) plus follow-up/replication in large, independent samples, the authors now find 22 genomic loci at genomewide significance. This is another important step forward in schizophrenia genetics.
It is reassuring (and important for the scientific community to know) that there is support for some of the previously reported loci. As expected, the findings are getting much more consistent now with the increasing power of the samples. Interestingly, some loci pop up now at genomewide significance that were known from bipolar disorder or combined phenotype analyses (schizophrenia + bipolar), such as CACNA1C and CACNB2. Together with the finding that there is an enrichment of smaller p values in genes encoding calcium channel subunits, there is now growing evidence that calcium signaling is involved in both bipolar disorder and schizophrenia. Calcium signaling is a crucial neuronal process and relevant in a number of human diseases, as the authors nicely review. Importantly, calcium channel complexes may be useful for clinical translation (e.g., as drug targets).
Variation in another gene that was implicated in bipolar disorder before, NCAN, was found at genomewide significance in schizophrenia in the study by Ripke et al. (in fact, an association with schizophrenia was already reported earlier by Mühleisen et al., 2012). It seems that another "theme" of disease-relevant genes may be neurodevelopmental effects in both bipolar disorder and schizophrenia. The different lincRNAs implicated now among the 22 genomewide significant findings may also point in this direction.
What I find particularly interesting are the considerations regarding the much debated genetic architecture of schizophrenia. The authors’ simulations, although certainly still not perfectly exact, substantiate previous calculations that common SNPs make substantial contributions to the risk for schizophrenia.
To my knowledge, for the first time there are estimates regarding the absolute number of common SNPs that contribute to the etiology of schizophrenia: The authors estimate "6,300 to 10,200 independent and mostly common SNPs." While it is difficult to deduce the number of genes or functional units covered by these SNPs, several thousand are well possible. Many of these loci/genes will fall into the same biological pathways, and the identification of a subset of these will probably suffice to identify the most important biological processes. The authors estimate that the top 2,000 loci might be sufficient, and maybe it is even fewer. At the same time, they estimate that such a number of identified loci is not completely out of reach. Sixty thousand patients and 60,000 controls should have enough statistical power to identify between 400 and 1,100 common loci at genomewide significance. Psychiatrists will agree that a great collaborative effort will be required to recruit such a large number of patients. But it is not impossible. The next wave of the PGC schizophrenia group is already moving in this direction.
Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium (2011) Genome-wide association study identifies five new schizophrenia loci. Nat Genet. 43:969-76. Abstract
Mühleisen TW, Mattheisen M, Strohmaier J, et al. (2012) Association between schizophrenia and common variation in neurocan (NCAN), a genetic risk factor for bipolar disorder. Schizophr Res. 138:69-73. Abstract
View all comments by Sven Cichon
Related News: Bigger Schizophrenia GWAS Yields More Hits
Comment by: Ole A. Andreassen, Martin Tesli
Submitted 6 October 2013
Posted 7 October 2013
I recommend the Primary Papers
The recent genomewide association study (GWAS) by Ripke and co-workers is a very important contribution to our knowledge of the genetic underpinnings of schizophrenia. By adding ~5,000 schizophrenia cases and ~6,000 healthy controls from Sweden to the Psychiatric Genomics Consortium (PGC) results from 2011 (PGC, 2011), the authors identified 22 gene loci (including 13 novel) at genomewide significance. These findings confirm that previous schizophrenia GWAS have been statistically underpowered, and that increasing the sample size is a successful approach to bridge the gap between the high heritability estimates in schizophrenia and low variance explained by currently identified genetic variants.
With increasing sample size, consistency is also enhanced. Reassuringly, of the 100 most significant SNPs in the Sweden/PGC meta-analysis, 90 percent had the same sign. Moreover, previously reported signals from immune-related genes in the MHC region on chromosome 6 were confirmed, as well as genes encoding calcium channel subunits, miR-137, and targets of miR-137. Additionally, the authors found enrichment in an extended set of genes with predicted miR-137 target sites. The results also pointed to long intergenic noncoding RNAs (lincRNAs), as 13 out of 22 identified regions contain lincRNAs. Interestingly, there is evidence that lincRNAs have functions related to epigenetic regulation.
Using two newly developed methods—genomewide complex trait analysis (GCTA) and applied Bayesian polygenic analysis (ABPA)—the authors estimated that common variants account for 52 percent and 78 percent, respectively, of the phenotypic variation in schizophrenia. These numbers are, although imprecisely, in accordance with the high heritability estimates from meta-analyses on twin studies (81 percent) (Sullivan et al., 2003) and large epidemiological studies (64 percent) (Lichtenstein et al., 2009), and indicate that a substantial proportion of the genetic risk for schizophrenia can be explained by common variants identified in GWAS. With the ABPA method, the authors also estimated that between 6,300 and 10,200 SNPs explain 50 percent of the variance in schizophrenia susceptibility. This clearly shows the high polygenicity in schizophrenia.
The authors suggest that 60,000 schizophrenia cases and 60,000 controls are warranted to identify ~800 markers at genomewide significance level. With the combined effort from the PGC, this might be achievable, and a larger proportion of the hidden heritability will undoubtedly be revealed. However, when using the PGC schizophrenia sample as the discovery set and the Swedish sample as the test set, explained variance (Nagelkerke pseudo R2) was only ~0.06 (contrasted by an impressive significance level of 2 x 10-114). This is slightly disappointing, as the explained variance with similar methodology was found to be ~0.03 in a study by Purcell and co-workers in 2009 (Purcell et al., 2009). By increasing the sample size five times (~6,000 vs. ~30,000 individuals), explained variance has only increased from 3 to 6 percent. Although a further increase in the PGC sample will provide important information on risk variants, a refinement of the method is probably also needed to discover larger parts of the hidden heritability for the highly polygenic disorder schizophrenia. Such methods might include Bayesian models for weighing SNPs based on prior evidence, for example, related to pleiotropic effects (Andreassen et al., 2013) and genic annotation (Schork et al., 2013). A combination of large numbers and methodological refinement might prove particularly potent, both in terms of explaining more of the variance and identifying underlying molecular pathways.
PGC. Genome-wide association study identifies five new schizophrenia loci. Nat Genet . 2011 Oct ; 43(10):969-76. Abstract
Andreassen OA, Thompson WK, Schork AJ, Ripke S, Mattingsdal M, Kelsoe JR, Kendler KS, O'Donovan MC, Rujescu D, Werge T, Sklar P, , , Roddey JC, Chen CH, McEvoy L, Desikan RS, Djurovic S, Dale AM. Improved detection of common variants associated with schizophrenia and bipolar disorder using pleiotropy-informed conditional false discovery rate. PLoS Genet . 2013 Apr ; 9(4):e1003455. Abstract
Lichtenstein P, Yip BH, Björk C, Pawitan Y, Cannon TD, Sullivan PF, Hultman CM. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet . 2009 Jan 17 ; 373(9659):234-9. Abstract
Purcell SM, Wray NR, Stone JL, Visscher PM, O'Donovan MC, Sullivan PF, Sklar P. Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature . 2009 Aug 6 ; 460(7256):748-52. Abstract
Schork AJ, Thompson WK, Pham P, Torkamani A, Roddey JC, Sullivan PF, Kelsoe JR, O'Donovan MC, Furberg H, Schork NJ, Andreassen OA, Dale AM. All SNPs are not created equal: genome-wide association studies reveal a consistent pattern of enrichment among functionally annotated SNPs. PLoS Genet . 2013 Apr ; 9(4):e1003449. Abstract
Sullivan PF, Kendler KS, Neale MC. Schizophrenia as a complex trait: evidence from a meta-analysis of twin studies. Arch Gen Psychiatry . 2003 Dec ; 60(12):1187-92. Abstract
View all comments by Ole A. Andreassen
View all comments by Martin Tesli