l e t t e r s
Genome-wide association study reveals three susceptibility loci for common migraine in the general population
Daniel I Chasman1,2,23, Markus Schürks1,3,23, Verneri Anttila4,5,22, Boukje de Vries6, Ulf Schminke7, Lenore J Launer8, Gisela M Terwindt9, Arn M J M van den Maagdenberg6,9, Konstanze Fendrich10, Henry Völzke11, Florian Ernst12, Lyn R Griffiths13, Julie E Buring1, Mikko Kallela14,22, Tobias Freilinger15,22, Christian Kubisch16,22, Paul M Ridker1,2, Aarno Palotie4,5,17–19,22, Michel D Ferrari9, Wolfgang Hoffmann10, Robert Y L Zee1,24 & Tobias Kurth1,20,21,24
Migraine is a common, heterogeneous and heritable attacks of headache associated with gastrointestinal and autonomic neurological disorder. Its pathophysiology is incompletely nervous system symptoms2. Up to one third of affected individuals understood, and its genetic influences at the population level may also experience transient focal neurological symptoms known as are unknown. In a population-based genome-wide analysis aura. Current concepts view migraine primarily as a multi-factorial including 5,122 migraineurs and 18,108 non-migraineurs, brain disorder with heritability estimates as high as 50% (ref. 3). rs2651899 (1p36.32, PRDM16), rs10166942 (2q37.1, TRPM8) Yet, progress in genetic analysis has been largely restricted to rare and rs11172113 (12q13.3, LRP1) were among the top monogenic subtypes of migraine, and very little is known about the seven associations (P < 5 × 10−6) with migraine. These SNPs underlying genetic variants for common forms of migraine, including were significant in a meta-analysis among three replication migraine with and without aura4. A recent genome-wide association cohorts and met genome-wide significance in a meta-analysis study (GWAS) identified a genetic variant on chromosome 8q22.1 combining the discovery and replication cohorts (rs2651899, associated with migraine in a large clinic-based sample of European odds ratio (OR) = 1.11, P = 3.8 × 10−9; rs10166942, OR = 0.85, individuals with migraine5. P = 5.5 × 10−12; and rs11172113, OR = 0.90, P = 4.3 × 10−9). In order to identify common genetic variants for migraine at the The associations at rs2651899 and rs10166942 were specific population level, we performed a GWAS among 23,230 women with for migraine compared with non-migraine headache. None complete genotype and migraine information and verified European Nature America, Inc. All rights reserved. All rights Inc. America, 1 Nature of the three SNP associations was preferential for migraine ancestry from the Women’s Genome Health Study (WGHS)6, a large
© 20 with aura or without aura, nor were any associations population-based cohort (Supplementary Note). Migraine was reported specific for migraine features. TRPM8 has been the focus by 5,122 women compared to 18,108 women not reporting migraine. of neuropathic pain models, whereas LRP1 modulates The clinical profile between migraineurs and non-migraineurs differed neuronal glutamate signaling, plausibly linking both genes most strongly for age, postmenopausal hormone use, physical activity to migraine pathophysiology. and alcohol consumption (Supplementary Table 1). In the discovery stage genome-wide scan among genotyped SNPs, Migraine is a common and often debilitating disorder affecting up to no SNPs reached the conventional threshold association of P < 5 × 10−8 20% of the population, with women being affected three to four times in age-adjusted logistic models assuming an additive relationship more often than men1. Clinically, migraine manifests with recurrent between the minor allele dose and the log-odds of migraine.
1Division of Preventive Medicine, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts, USA. 2Donald W. Reynolds Center for Cardiovascular Disease Prevention, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts, USA. 3Department of Neurology, University Hospital Essen, Essen, Germany. 4Wellcome Trust Sanger Institute, Hinxton, Cambridge, UK. 5Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Helsinki, Finland. 6Department of Human Genetics, Leiden University Medical Centre, Leiden, The Netherlands. 7Department of Neurology, Ernst-Moritz- Arndt University, Greifswald, Germany. 8National Institute of Aging, Laboratory for Epidemiology, Demography, and Biometry, Bethesda, Maryland, USA. 9Department of Neurology, Leiden University Medical Centre, Leiden, The Netherlands. 10Institute for Community Medicine, Section Epidemiology of Health Care and Community Health, Ernst-Moritz-Arndt University, Greifswald, Germany. 11Institute for Community Medicine, Section Clinical Epidemiological Research, Ernst-Moritz-Arndt University, Greifswald, Germany. 12Interfaculty Institute for Genetics and Functional Genomics, Ernst-Moritz-Arndt University, Greifswald, Germany. 13Genomics Research Centre, Griffith Health Institute, Griffith University, Gold Coast, Queensland, Australia. 14Department of Neurology, Helsinki University Central Hospital, Helsinki, Finland. 15Department of Neurology, Klinikum Großhadern, Ludwig-Maximilians-Universität and Institute for Stroke and Dementia Research, Klinikum der Universität München, Munich, Germany. 16Institute of Human Genetics, University of Ulm, Ulm, Germany. 17Department of Medical Genetics, University of Helsinki, Helsinki, Finland. 18Department of Medical Genetics, Helsinki University Central Hospital, Helsinki, Finland. 19The Broad Institute of MIT and Harvard, Boston, Massachusetts, USA. 20INSERM Unit 708—Neuroepidemiology, Paris, France. 21UPMC Univ Paris 06, F-75005, Paris, France. 22On behalf of the International Headache Genetics Consortium (IHGC) (full list of consortium members appears in the Supplementary Note). 23These authors contributed equally to this work. 24These authors jointly directed this work. Correspondence should be addressed to M.S. ([email protected]) or T.K. ([email protected]).
Received 7 March; accepted 16 May; published online 12 June 2011; doi:10.1038/ng.856
Nature Genetics VOLUME 43 | NUMBER 7 | JULY 2011 695 l e t t e r s
Table 1 Discovery and replication statistics for association of three genome-wide significant SNPs with migraine in meta-analysis Locus position a b b 2 SNP (candidate) A1/A2 Stage: cohort (Nmig/Nnon-mig) MAF OR (95% CI) P I % (Phet) rs2651899 1p36.32 C/T Discovery: WGHS (5,122/18,108) 0.43 1.13 (1.08–1.18) 1.4 × 10−7 3,073,572 Replication: GEM (774/942) 0.41 1.07 (0.93–1.23) 0.37 (PRDM16) Replication: SHIP (306/2,260) 0.47 1.25 (1.04–1.49) 0.02 Replication: IHGC (2,748/10,747) 0.42 1.07 (1.00–1.14) 0.04 All replicationc: (3,828/13,949) 1.08 (1.03–1.14) 4.2 × 10−3 21.8 (0.28) Discovery and replicationc: (8,950/32,057) 1.11 (1.07–1.15) 3.8 × 10−9 18.9 (0.30) rs10166942 2q37.1 C/T Discovery: WGHS 0.19 0.86 (0.81–0.91) 2.3 × 10−7 234,489,832 Replication: GEM 0.22 0.80 (0.67–0.96) 0.01 (TRPM8) Replication: SHIP 0.18 0.86 (0.67–1.09) 0.21 Replication: IHGC 0.18 0.85 (0.78–0.93) 0.0002 All replicationc 0.84 (0.79–0.91) 5.0 × 10−6 0.0 (0.80) Discovery and replicationc 0.85 (0.82–0.89) 5.5 × 10−12 0.0 (0.90) rs11172113 12q13.3 C/T Discovery: WGHS 0.41 0.90 (0.86–0.94) 3.7 × 10−6 55,813,550 Replication: GEM 0.41 0.96 (0.84–1.11) 0.62 (LRP1) Replication: SHIP 0.42 0.81 (0.67–0.98) 0.03 Replication: IHGC 0.41 0.90 (0.85–0.96) 0.0014 All replicationc 0.90 (0.85–0.95) 3.0 × 10−4 0.3 (0.37) Discovery and replicationc 0.90 (0.87–0.93) 4.3 × 10−9 0.0 (0.57) aNumber of migraineurs/non-migraineurs. bSingle cohort analysis association statistics from an age-adjusted (WGHS), age- and sex-adjusted (GEM and SHIP) or sex-adjusted and country-of- origin–adjusted (IHGC) logistic regression model. Meta-analysis association statistics from an inverse-variance weighted fixed effects model.c Meta-analysis combining estimates from individual cohorts. A1, minor allele ‘+’ strand; A2, major allele ‘+’ strand; MAF, minor allele frequency; OR, odds ratio; CI, confidence interval. All genomic information is from human genome build hg18. 2 I , heterogeneity as estimated proportion of total variance; Phet, P value from test of heterogeneity. WGHS, Women’s Genome Health Study; GEM, Genetic Epidemiology of Migraine study; SHIP, Study of Health in Pomerania; IHGC, International Headache Genetics Consortium.
Nevertheless, the top SNPs were more significant than expected for the IHGC, the latter of which is the largest replication cohort (P = the 95% confidence interval for the ordered test statistic under the 0.02 (0.57 × 3 replication cohorts)). In IHGC, there were nominally null hypothesis and, except for one SNP, remained more significant significant associations for rs2078371 and the three SNPs that had after controlling for modest inflation of the test statistic (λGC = 1.03; concordant effects in all three replication cohorts. rs10166942 in GEM Supplementary Fig. 1). Seven independent loci had at least one SNP and both rs2651899 and rs11172113 in SHIP were also nominally with P < 5 × 10−6. We selected the most significant SNP from each significant (Table 1 and Supplementary Table 2). locus for further analysis (Table 1 lists three SNPs which ultimately We then performed a meta-analysis of the results for the primary reached genome-wide significance; Supplementary Table 2 lists all SNPs using a fixed-effects model with inverse-variance weight- SNPs). None of the seven SNPs was associated with non-migraine ing. Combining only the replication cohorts, the meta-analysis headache in the WGHS (all P > 0.05, N = 3,001 compared with 14,959 supported association (P < 0.008 (0.05/6)) for the three fully con- controls). The significance and magnitude of the associations of the cordant SNPs: rs2651899 (OR = 1.08, 95% CI 1.03–1.14, P = 4.2 × Nature America, Inc. All rights reserved. All rights Inc. America, 1 Nature seven SNPs with migrane were essentially unchanged in sensitiv- 10−3), rs10166942 (OR = 0.84, 95% CI 0.79–0.91, P = 5.0 × 10−6) and ity analyses using (i) an allele frequency test, (ii) logistic regression rs11172113 (OR = 0.90, 95% CI 0.85–0.95, P = 3.0 × 10−4) (see Table 1 © 20 without adjustment or (iii) logistic regression adjusted for clinical and Supplementary Table 4). There was no strong evidence of hetero- 2 characteristics, eigenvector parameters for sub-European popula- geneity among the studies (all I < 22%, Phet > 0.28). Furthermore, all tion structure or both (Supplementary Table 3). There was no evi- three SNPs met the genome-wide standards of significance in a meta- dence of non-additive modes of association with migraine for any of analysis combining the discovery cohort with the replication cohorts the top SNPs, and no SNPs in the entire genome-wide scan reached (Table 1; all P ≤ 4.3 × 10−9). None of the remaining SNPs reached genome-wide significance in non-additive models (D.I.C. & M.S., either nominal significance in the meta-analysis of the replication data not shown). Finally, analysis performed with genotypes imputed cohorts or genome-wide significance in the meta-analysis combining for approximately 2.6 million SNPs in HapMap (release 22)7 revealed all cohorts (Supplementary Table 4). the same top seven loci found with the genotyped SNPs. All three replicating SNPs map within or near transcribed regions We evaluated the seven SNPs in two additional population-based of known genes. rs2651899 at 1p36.32 is within the first intron of cohorts: the Dutch Genetic Epidemiology of Migraine study (GEM8; PRDM16 in a block of moderate linkage disequilibrium (LD) extend- 774 migraineurs and 942 non-migraineurs) and the German Study ing about 22 kb in the 5′ direction and about 25 kb in the 3′ direction of Health in Pomerania (SHIP9; 306 migraineurs and 2,260 non- (Fig. 1a). The transcript for the second closest gene, ACTRT2, termi- migraineurs), as well as in the previously reported clinic-based nates 144 kb from rs2651899. rs10166942 at 2q37.1 is 950 bp 5′ to the case-control samples from the International Headache Genetics transcription start site for TRPM8 in a block of moderate LD spanning Consortium (IHGC5; 2,748 migraineurs and 10,747 population- approximately 168 kb. This LD block begins about 117 kb 5′ from based controls), all with European ancestry (Supplementary Note). the SNP and extends through TRPM8 (Fig. 1b), a gene previously Genotyping failed for rs17172526 in both GEM and SHIP. The effect identified as a potential candidate for association with migraine in estimates for rs2651899, rs10166942 and rs11172113 were concordant the IHGC cohort on the basis of sub–genome-wide significance5. The in direction and magnitude among all replication cohorts with the second closest gene, HJURP, is situated 61.9 kb from rs10166942. effects in the WGHS (Table 1 and Supplementary Table 2; P = 0.03 rs11172113 at 12q13.3 maps to the first intron of LRP1 in a gene-rich for consistency of direction). Moreover, among the three replication region that also includes the second closest gene, STAT6, 22.1 kb away cohorts, all seven SNPs were concordant between the WGHS and (Fig. 1c). At each of the three loci, only the primary SNP was retained
696 VOLUME 43 | NUMBER 7 | JULY 2011 Nature Genetics l e t t e r s
Migraine associations at 12q13.3 a b c near rs11172113 12 2 20 kb Migraine associations at 2q37.1 r Migraine associations at 1p36.32 near rs10166942 10 0.8 rs11172113
near rs2651899 12 2 P 8 r 20 kb P = 3.7 × 10–6 12 r 2 20 kb 10 10 6 0.5
rs10166942 Recombination (cM/Mb) 0.8 –7 10 –log 4 P 8 P = 2.3 × 10 rs2651899 0.8 –7 10 P 8 P = 1.4 × 10 6 0.5 2 Recombination (cM/Mb) 10 6 0.5 –log 4 0
Recombination (cM/Mb) 13
–log 4 2 5 2 0 0 0 46 RDH16 ZBTB39 TMEM194A LRP1 R3HDM2 25 0 0 GPR182 NAB2 NXPH4 DNAJB3 HJURP SPP2 ACTRT2 PRDM16 TRPM8 TAC3 STAT6 SHMT2 Genes Genes Genes UGT1A1–10 2.9 3.0 3.1 3.2 MYO1A NDUFA4L2 234.3 234.4 234.5 234.6 234.7 Chromosome 1 position (Mb) STAC3 Chromosome 2 position (Mb) 55.6 55.7 55.8 55.9 56.0 Figure 1 Association P values and genomic context for candidate SNPs. (a–c) Shown are candidate SNPs Chromosome 12 position (Mb) at 1p36.32 (PRDM16) (a), 2q37.1 (TRPM8) (b) and 12q13.3 (LRP1) (c). P values for experimentally determined genotypes are indicated by diamond shapes, including the lead SNPs (Table 1), which are indicated by the large diamond shape. P values for SNPs with imputed genotype are indicated with circle shapes. Linkage disequilibrium relationships (as r2) are from the HapMap (r22)7.
in stepwise model selection for association with migraine, suggesting hypotheses testing, although some met nominal significance the absence of additional, non-redundant or conditional associations (Supplementary Table 8). In contrast, rs2651899 and rs10166942 (Online Methods). (but not rs11172113) were significantly associated with migraine As migraine is more prevalent in women, we examined the effects compared to non-migraine headache (rs2651899, OR = 1.12, 95% CI of the three replicating SNPs on migraine in the replication cohorts, 1.05–1.20, P = 5 × 10−4; and rs10166942, OR = 0.90, 95% CI 0.82–0.98, which include both women and men, in contrast to the WGHS. In the P = 0.01) with effects similar to the association of migraine compared meta-analysis of the six sex-specific strata among the three replica- to non-migraine controls, reinforcing their specificity for migraine tion cohorts, effect estimates were comparable to the main analysis, (Supplementary Table 9, compare with Table 1). which was not stratified by sex (compare Supplementary Table 5, left TRPM8 encodes a sensor for cold and cold-induced burning pain10, and Table 1). Notably, the heterogeneity estimate (I2) for rs10166942 which is primarily expressed in sensory neurons and dorsal root increased from 0% to 45.8%, although neither the heterogeneity ganglion neurons11. Members of the mammalian TRP superfamily P value nor a potential sex effect at this SNP in the meta-regression are channels activated by stimuli of chemo- and somato-sensation. was significant (P = 0.10 and P = 0.23, respectively; Supplementary TRPM8 particularly is a target in animal models of neuropathic Table 5, right). However, a potential sex interaction for rs10166942 pain12. As migraine shares some characteristics with neuropathic pain may have been confounded by differences in study design and disorders13, TRPM8 could be a pathophysiological link between both migraine ascertainment between IHGC (clinic based) and GEM and pain syndromes. Nature America, Inc. All rights reserved. All rights Inc. America, 1 Nature SHIP (both population based). A meta-analysis by sex strata restricted LRP1 is expressed in many tissues including brain and vascula- to the population-based studies revealed even greater heterogeneity for ture14. As a member of the lipoprotein receptor family, it serves as a © 20 rs10166942 (I2 = 67.2%, heterogeneity P value = 0.03; Supplementary sensor of the extracellular environment and also modulates synaptic Table 6, left). In the meta-regression of this SNP in these two studies transmission. LRP1 protein and glutamate (NMDA) receptors are alone, the association with migraine suggested a significant sex inter- co-localized on neurons and interact. This integrates well with find- action (P = 0.004) and essentially no evidence of association in men ings from the previous GWAS in migraine reporting a genetic variant (OR = 1.08, 95% CI 0.85–1.38, P = 0.52; Supplementary Table 6, implicated in glutamate homeostasis5 as well as with recent pharmaco- right). The potential differential association according to sex did logical approaches to migraine targeting glutamate receptors15. not appear to be related to estrogen receptor 1 (encoded by ESR1), A potential role of PRDM16 in migraine is unclear. PRDM16 was as there was no interaction between rs10166942 and ESR1 SNPs in originally identified near a chromosomal breakpoint associated with WGHS, including SNPs associated with other clinical traits in other myelodysplastic syndrome and acute myeloid leukemia16, but sub- studies (Online Methods and data not shown). Similarly, none of the sequent research has focused on its transcriptional role in brown fat other primary SNPs showed an interaction with ESR1 SNPs (data development17. Structurally, PRDM16 contains two arrays of C2H2 not shown). zinc-finger domain repeats, which are often linked to transcriptional We investigated whether there was evidence for a disproportion- activity; PRDM16 also contains a putative SET domain, which is a ate association of the three genome-wide significant SNPs from the conserved region among histone lysine methyltransferases. meta-analysis with migraine aura status and features recognized Our findings may be compared with the recent GWAS reporting an by International Headache Society diagnostic criteria (unilateral association of rs1835740 at 8q22.1 with migraine, especially migraine pain location, pulsating pain quality, sensitivity to light or sound, with aura5. The nearby candidate genes PGCP and MTDH are attack duration, nausea and/or vomiting, aggravation by physical involved in glutamate homeostasis, consistent with current concepts activity and inhibition of physical activity; Online Methods). None of migraine pathophysiology. However, in the WGHS, rs1835740 was of these associations was more significant than for overall migraine neither associated with overall migraine (P = 0.22) nor migraine with (Supplementary Table 7). Moreover, the allele frequency differences or without aura separately (data not shown). Similarly, across the entire of the three SNPs between migraineurs with or without each of the fea- region, no SNP reached locus-wide significant thresholds for associa- tures were not significant enough to overcome correction for multiple tion with migraine, and there was no evidence of stronger associations
Nature Genetics VOLUME 43 | NUMBER 7 | JULY 2011 697 l e t t e r s
with migraine with aura than without aura (Supplementary Fig. 2). The IHGC study was supported, among others, by the Academy of Finland The most significant SNP for association with migraine in the WGHS (200923 to A.P.), the Wellcome Trust (grant number 089062), the European was rs1864729 (OR = 1.15, 95% CI 1.06–1.25, P = 0.001). The differ- Community’s Seventh Framework Programme (FP7/2007-2013) (through the 5 SYNSYS Consortium (grant agreement no. 242167) and the ENGAGE Consortium ential findings between the current and the previous study may (grant agreement no. 201413)), the Helsinki University Central Hospital (to M.K.) partly relate to the differences in migraine ascertainment (see also and the Finnish Culture Foundation (to V.A.). Funding by the German Federal the analysis of sex interaction above). Ministry of Education and Research (BMBF) within the National Genome There are three major implications of our study. First, we have Research Network (NGFNplus, EMINet-01GS08120 for C.K., 01GS08121 to M. Dichgans), the Deutsche Forschungsgemeinschaft (to C.K.) and the Center for identified three SNPs with genome-wide significant association for Molecular Medicine Cologne (to C.K.). For a full list of acknowledgements, common migraine at the population level. Two of the SNPs signi- please see reference 5. ficantly distinguish migraine from non-migraine headache. In addi- tion, the association of rs10166942 may be stronger among women, AUTHOR CONTRIBUTIONS which may be related to but would not explain the higher prevalence Obtained funding: J.E.B., M.D.F., W.H., T.K., A.M.J.M.v.d.M., A.P., P.M.R., U.S., H.V., R.Y.L.Z. Overall study design: D.I.C., T.K., M.S. Cohort supervision and of migraine in women. Second, although one new locus (LRP1) sup- phenotyping: V.A., J.E.B., K.F., M.D.F., T.F., W.H., M.K., C.K., T.K., L.J.L., A.P., ports prevailing (glutamatergic) concepts of neurotransmitter path- P.M.R., U.S., M.S., G.M.T., H.V. Analysis and genotyping: V.A., D.I.C., K.F., ways in migraine4, we also identified a second new locus (TRPM8) L.R.G., W.H., A.M.J.M.v.d.M., P.M.R., U.S., M.S., F.E., B.d.V., R.Y.L.Z. Manuscript explicitly implicated in a pain related pathway. The functional role writing: D.I.C., M.S. All authors participated in critical review of the manuscript for intellectual content. of the third locus (PRDM16) pathway is still unknown. Last, whereas some studies have focused on differences between migraine with and COMPETING FINANCIAL INTERESTS 4 without aura , our results suggest a shared pathophysiology among The authors declare no competing financial interests. common types of migraine (Supplementary Tables 7,8). Ongoing large GWAS will continue to identify additional genetic Published online at http://www.nature.com/naturegenetics/. risk variants for migraine and further delineate the pathophysiologi- Reprints and permissions information is available online at http://www.nature.com/ reprints/index.html cal basis of migraine. Meanwhile, the three new loci identified in the present work provide hypotheses for immediate further exploration. 1. Haut, S.R., Bigal, M.E. & Lipton, R.B. Chronic disorders with episodic manifestations: focus on epilepsy and migraine. Lancet Neurol. 5, 148–157 (2006). URLs. R, http://www.r-project.org/; Metal, http://www.sph.umich. 2. Headache Classification Subcommittee of the International Headache Society. The edu/csg/abecasis/Metal/index.html. International Classification of Headache Disorders: 2nd edition. Cephalalgia 24 (Suppl 1), 9–160 (2004). 3. Mulder, E.J. et al. Genetic and environmental influences on migraine: a twin study Methods across six countries. Twin Res. 6, 422–431 (2003). Methods and any associated references are available in the online 4. de Vries, B., Frants, R.R., Ferrari, M.D. & van den Maagdenberg, A.M. Molecular genetics of migraine. Hum. Genet. 126, 115–132 (2009). version of the paper at http://www.nature.com/naturegenetics/. 5. Anttila, V. et al. Genome-wide association study of migraine implicates a common susceptibility variant on 8q22.1. Nat. Genet. 42, 869–873 (2010). Note: Supplementary information is available on the Nature Genetics website. 6. Ridker, P.M. et al. Rationale, design, and methodology of the Women’s Genome Health Study: a genome-wide association study of more than 25,000 initially healthy Acknowledgments American women. Clin. Chem. 54, 249–255 (2008). This study is supported by a grant from the National Institute of Neurological 7. Frazer, K.A. et al. A second generation human haplotype map of over 3.1 million Disorders and Stroke (NS-061836). The Women’s Health Study and the Women’s SNPs. Nature 449, 851–861 (2007). 8. Launer, L.J., Terwindt, G.M. & Ferrari, M.D. The prevalence and characteristics of
Nature America, Inc. All rights reserved. All rights Inc. America, 1 Nature Genome Health Study are supported by grants from the National Heart, Lung, and migraine in a population-based cohort: the GEM study. Neurology 53, 537–542 Blood Institute (HL-043851, HL-080467 and HL-099355) and the National Cancer (1999). Institute (CA-47988). Part of the research for this work was supported by grants © 20 9. Völzke, H. et al. Cohort profile: the study of health in pomerania. Int. J. Epidemiol. 40, from the Donald W. Reynolds Foundation and the Leducq Foundation. Genome- 294–307 (2011). wide genotyping and collaborative scientific support was provided by Amgen. 10. Proudfoot, C.J. et al. Analgesia mediated by the TRPM8 cold receptor in chronic Genotyping in the Genetic Epidemiology of Migraine Study was supported by neuropathic pain. Curr. Biol. 16, 1591–1605 (2006). the Netherlands Organisation for Scientific Research (NWO) VICI (918.56.602) 11. Peier, A.M. et al. A TRP channel that senses cold stimuli and menthol. Cell 108, and Spinoza (2009) grants and the Center for Medical Systems Biology (CMSB) 705–715 (2002). 12. Dray, A. Neuropathic pain: emerging treatments. 101, 48–58 established by the Netherlands Genomics Initiative/Netherlands Organisation for Br. J. Anaesth. (2008). Scientific Research (NGI/NWO), project no. 050-060-409. The GEM study was 13. Biondi, D.M. Is migraine a neuropathic pain syndrome? Curr. Pain Headache Rep. 10, supported by the Ministry of Health, Welfare and Sport and the National Institute 167–178 (2006). of Public Health and the Environment, The Netherlands. 14. Lillis, A.P., Van Duyn, L.B., Murphy-Ullrich, J.E. & Strickland, D.K. LDL receptor- SHIP is part of the Community Medicine Research net of the University of related protein 1: unique tissue-specific functions revealed by selective gene Greifswald, Germany, which is funded by the Federal Ministry of Education and knockout studies. Physiol. Rev. 88, 887–918 (2008). Research (grants no. 01ZZ9603, 01ZZ0103 and 01ZZ0403), the Ministry of Cultural 15. Andreou, A.P. & Goadsby, P.J. Therapeutic potential of novel glutamate Affairs as well as the Social Ministry of the Federal State of Mecklenburg-West receptor antagonists in migraine. Expert Opin. Investig. Drugs 18, 789–803 (2009). Pomerania. Genome-wide data have been supported by the Federal Ministry of 16. Secker-Walker, L.M., Mehta, A. & Bain, B. Abnormalities of 3q21 and 3q26 Education and Research (grant no. 03ZIK012) and a joint grant from Siemens in myeloid malignancy: a United Kingdom Cancer Cytogenetic Group study. Healthcare, Erlangen, Germany and the Federal State of Mecklenburg-West Pomerania. Br. J. Haematol. 91, 490–501 (1995). The SHIP authors are grateful to the contribution of A. Teumer, A. Hoffmann and 17. Seale, P., Kajimura, S. & Spiegelman, B.M. Transcriptional control of brown adipocyte A. Petersmann in generating the SNP data. The University of Greifswald is a member development and physiological function–of mice and men. Genes Dev. 23, 788–797 of the ‘Center of Knowledge Interchange’ program of Siemens AG. (2009).
698 VOLUME 43 | NUMBER 7 | JULY 2011 Nature Genetics ONLINE METHODS covariates: age (continuous), body mass index (continuous), history of diabetes Genotyping. Genotyping in WGHS was performed using the HumanHap300 (yes, no), history of hypertension (yes, no), postmenopausal hormone use Duo ‘+’ chips or the combination of the HumanHuman300 Duo and iSelect (yes, no), physical activity (rarely or never, <1/week, 1–3 times/week or ≥4 chips (Illumina) with the Infinium II protocol and has been described else- times/week), alcohol consumption (rarely or never, 1–3 drinks/month, 1–6 where6. Among individuals with successful genome-wide genotyping, 23,294 drinks/week or ≥1 drink/day), smoking (never, past or current) and family were identified as having self-reported European ancestry that could be veri- history of myocardial infarction before age 60 (yes, no). Statistical analysis in fied on the basis of multidimensional scaling analysis of identity by state using the GEM replication cohort was performed in PLINK using logistic regression 1,443 ancestry informative markers in PLINK v. 1.07 (ref. 18). In the final data- models assuming additive models with optional adjustment for age or sex. set, we retained SNPs with minor allele frequency >1%, successful genotyp- Association analysis in SHIP was performed using R 2.4.1 (see URLs). In the ing in 90% of subjects and deviations from Hardy-Weinberg equilibrium not IHGC cohort, the analysis for directly genotyped SNPs was performed using exceeding P = 10−6 in significance. Within this sample, genotypes for 2,608,509 PLINK with logistic regression models assuming additive effects, with covari- SNPs were imputed from the experimental genotypes and LD relationships ate adjustments for sex, population identity within IHGC and an additional implicit in the HapMap (r22) CEU samples using MACH 1.0.16 (ref. 19). interaction component between sex and genotype in the interaction analysis. SNPs selected from WGHS for replication were genotyped among partici- For the imputed SNPs, dosage data from the individual population cohorts pants from GEM with the iPLEX (Sequenom) method by means of matrix- was analyzed using SNPTEST v2.2.0 (ref. 20) and the results combined using assisted laser desorption ionization time-of-flight mass spectrometry method GWAMA23. To identify a non-redundant set of SNPs influencing migraine at (MALDI-TOF MS, Mass Array, Sequenom) according to the manufacturer’s each of the candidate loci, we performed forward-backward, stepwise selection instructions. Genotype information for all SNPs was available in 98.3% of in age-adjusted logistic models to optimize the Bayesian information criterion the samples with the exception of rs2203834, which was available for 84% among SNPs within 100 kb of the locus lead SNP. of samples. The SHIP samples were genotyped using the Affymetrix Human SNP Array 6.0. Meta-analysis of discovery and replication cohorts. Finally, in the absence of Hybridization of genomic DNA was done in accordance with the manufacturer’s heterogeneity among the studies, a fixed-effects meta-analysis was performed standard recommendations. The genetic data analysis workflow was created for SNPs with P < 5 × 10−6 in the WGHS discovery cohort among the replica- using the software InforSense. Genetic data were stored using the database tion cohorts using the software METAL and the metafor package in R24. Caché (InterSystems). Genotypes were determined using the Birdseed2 clustering algorithm. For quality control purposes, several control samples Analysis of gene-gene interaction. To investigate for potential gene-gene were added. On the chip level, only subjects with a genotyping rate on quality interaction of our top three SNPs with gene variants in ESR1 (the estrogen control probesets (QC callrate) of at least 86% were included. The overall mean receptor 1 gene), we considered SNPs represented on the WGHS genotyp- genotyping efficiency of the GWAS was 98.55% with a minimum sample call ing platform supplemented by SNPs with imputed genotype that have been rate of 92%. Imputation of genotypes in SHIP was performed with the program reported for association with other clinical traits: rs1038304, rs1999805, IMPUTE v0.5.0 based on HapMap II CEU dataset20. Except for rs2203834, rs2504063, rs2941740, rs4870044, rs6929137, rs2228480, rs1801132, rs2234693, imputed genotypes were used for the replication analysis of the candidate rs6557170, rs2347867, rs6557171 and rs4870062 (ref. 25). SNPs and all had quality measure observed to expected variance ratio >0.91. Additional programming was performed in R. All annotations derive from Experimental genotypes for rs2203834 were 99% complete. human genome reference sequence hg18 (NCBI build 36.1), the UCSC RefSeq Genotyping of the IHGC cases was performed using Illumina arrays as as of October 27, 2008 and the dbSNP database (build 129) as represented by described previously5. Control samples were genotyped with Illumina 660K the UCSC database. Locus region plots follow the conventions of SNAP26. or 610K platforms for the Finnish sample and the Illumina 550K platform for both the German and Dutch samples. Experimental genotype information was Power calculation. There was an estimated 80% power at the genome-wide available for all replicated SNPs except rs17172526 and was successful in at significance (P < 5 × 10−8) level to detect associations with odds ratios of 1.16 least 99.9% of the samples for each. For rs17172526, genotypes were imputed increased risk (or 0.86 reduced risk) and a minor allele frequency of 0.5 or 1.21 Nature America, Inc. All rights reserved. All rights Inc. America, 1 Nature by strata of nationality and case status with the program IMPUTE (v2.1.2) (or 0.83) and a minor allele frequency of 0.2. Because of the relatively large based on the HapMap III and 1000 Genomes data with IMPUTE information number of cases in the WGHS compared with the other two cohorts, the power © 20 measure IA > 0.945 (ref. 20). was comparable for the meta-analysis including all three studies.
Statistical analysis. Age-adjusted logistic regression in PLINK18 (experimental 21 18. Purcell, S. et al. PLINK: a tool set for whole-genome association and population- genotype information) and ProbAbel (imputed genotype information) were based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007). used to investigate the association between gene variants and migraine in 19. Li, Y., Willer, C., Sanna, S. & Abecasis, G. Genotype imputation. Annu. Rev. the WGHS discovery cohort, and effect estimates were calculated as ORs Genomics Hum. Genet. 10, 387–406 (2009). and 95% CIs. In the primary genome-wide analyses, we assumed an additive 20. Marchini, J. & Howie, B. Genotype imputation for genome-wide association studies. Nat. Rev. Genet. 11, 499–511 (2010). relationship between the number of copies of the minor allele of each SNP 21. Aulchenko, Y.S., Struchalin, M.V. & van Duijn, C.M. ProbABEL package for genome- and the age-adjusted, log-odds of migraine (additive model) and considered wide association analysis of imputed data. BMC Bioinformatics 11, 134 (2010). a threshold of P < 5 × 10−8 for genome-wide significance22. The significance 22. Hirschhorn, J.N. & Daly, M.J. Genome-wide association studies for common diseases of the consistency of the direction of effect between the WGHS and the three and complex traits. Nat. Rev. Genet. 6, 95–108 (2005). 23. Mägi, R. & Morris, A.P. GWAMA: software for genome-wide association meta- Σ nSNPs replication cohorts was computed as (choose(6,nSNPs) × 0.125 × analysis. BMC Bioinformatics 11, 288 (2010). (6-nSNPs) 3 0.875 ), with the sum over nSNPs between 3 and 6, where 0.125 (0.5 ) 24. Viechtbauer, W. Conducting meta-analyses in R with the metafor package. J. Stat. is the probability of having all three replication cohorts with the same direc- Softw. 36, 1–48 (2010). tion of effect as the WGHS. For sensitivity analysis, the genome-wide scan was 25. Hindorff, L.A., Junkins, H.A., Hall, P.N., Mehta, J.P. & Manolio, T.A. A catalog of 2 published genome-wide association studies.
doi:10.1038/ng.856 Nature Genetics