Gormley 2016

Articles
Meta-analysis of 375,000 individuals identifies 38

susceptibility loci for migraine
Padhraig Gormley1–4,81, Verneri Anttila2,3,5,81, Bendik S Winsvold6–8, Priit Palta9, Tonu Esko2,10,11,
Tune H Pers2,11–13, Kai-How Farh2,5,14, Ester Cuenca-Leon1–3,15, Mikko Muona9,16–18, Nicholas A Furlotte19,
Tobias Kurth20,21, Andres Ingason22, George McMahon23, Lannie Ligthart24, Gisela M Terwindt25, Mikko Kallela26,
Tobias M Freilinger27,28, Caroline Ran29, Scott G Gordon30, Anine H Stam25, Stacy Steinberg22, Guntram Borck31,
Markku Koiranen32, Lydia Quaye33, Hieab H H Adams34,35, Terho Lehtimäki36, Antti-Pekka Sarin9, Juho Wedenoja37,
© 2016 Nature America, Inc. All rights reserved.
David A Hinds19, Julie E Buring21,38, Markus Schürks39, Paul M Ridker21,38, Maria Gudlaug Hrafnsdottir40,
Hreinn Stefansson22, Susan M Ring23, Jouke-Jan Hottenga24, Brenda W J H Penninx41, Markus Färkkilä26,
Ville Artto26, Mari Kaunisto9, Salli Vepsäläinen26, Rainer Malik28, Andrew C Heath42, Pamela A F Madden42,
Nicholas G Martin30, Grant W Montgomery30, Mitja I Kurki1–3,9,43, Mart Kals10, Reedik Mägi10, Kalle Pärn10,
Eija Hämäläinen9, Hailiang Huang2,3,5, Andrea E Byrnes2,3,5, Lude Franke44, Jie Huang4, Evie Stergiakouli23,
Phil H Lee1–3, Cynthia Sandor45, Caleb Webber45, Zameel Cader46,47, Bertram Muller-Myhsok48, Stefan Schreiber49,
Thomas Meitinger50,51, Johan G Eriksson52,53, Veikko Salomaa53, Kauko Heikkilä54, Elizabeth Loehrer34,55,
Andre G Uitterlinden56, Albert Hofman34, Cornelia M van Duijn34, Lynn Cherkas33, Linda M Pedersen6,
Audun Stubhaug57,58, Christopher S Nielsen57,59, Minna Männikkö32, Evelin Mihailov10, Lili Milani10,
Hartmut Göbel60, Ann-Louise Esserlind61, Anne Francke Christensen61, Thomas Folkmann Hansen62,
Thomas Werge63–65, International Headache Genetics Consortium66, Jaakko Kaprio9,37,67, Arpo J Aromaa53,
Olli Raitakari68,69, M Arfan Ikram34,35,70, Tim Spector33, Marjo-Riitta Järvelin32,71–73, Andres Metspalu10,
Christian Kubisch74, David P Strachan75, Michel D Ferrari25, Andrea C Belin29, Martin Dichgans28,76,
Maija Wessman9,16, Arn M J M van den Maagdenberg25,77, John-Anker Zwart6–8, Dorret I Boomsma24,
George Davey Smith23, Kari Stefansson22,78, Nicholas Eriksson19, Mark J Daly2,3,5, Benjamin M Neale2,3,5,82,
Jes Olesen61,82, Daniel I Chasman21,38,82, Dale R Nyholt79,82 & Aarno Palotie1–5,9,80,82
Migraine is a debilitating neurological disorder affecting around one in seven people worldwide, but its molecular mechanisms
remain poorly understood. There is some debate about whether migraine is a disease of vascular dysfunction or a result of
neuronal dysfunction with secondary vascular changes. Genome-wide association (GWA) studies have thus far identified
13 independent loci associated with migraine. To identify new susceptibility loci, we carried out a genetic study of migraine
on 59,674 affected subjects and 316,078 controls from 22 GWA studies. We identified 44 independent single-nucleotide
polymorphisms (SNPs) significantly associated with migraine risk (P < 5 × 10−8) that mapped to 38 distinct genomic loci,
including 28 loci not previously reported and a locus that to our knowledge is the first to be identified on chromosome X.
In subsequent computational analyses, the identified loci showed enrichment for genes expressed in vascular and smooth
muscle tissues, consistent with a predominant theory of migraine that highlights vascular etiologies.
Migraine is the third most common disease worldwide, with a life- vascular changes representing downstream effects that are not them-
time prevalence of 15–20%, affecting up to 1 billion people across the selves causative of migraine4,5. However, genetic evidence favoring
globe1,2. It ranks as the seventh most disabling disease worldwide (and one theory over the other is lacking. At the phenotype level, migraine
the most disabling neurological disease) in terms of years of life lost is defined by diagnostic criteria from the International Headache
to disability1, and it is the third most costly neurological disorder, Society6. There are two prevalent subforms: migraine without aura,
after dementia and stroke3. There is debate about whether migraine which is characterized by recurrent attacks of moderate or severe
is a disease of vascular dysfunction or of neuronal dysfunction with headache associated with nausea or hypersensitivity to light and
A full list of authors and affiliations appears at the end of the paper.
Received 27 October 2015; accepted 26 May 2016; published online 20 June 2016; doi:10.1038/ng.3598
Nature Genetics ADVANCE ONLINE PUBLICATION

Articles
sound, and migraine with aura, which is characterized by transient candidate risk genes, we defined an associated locus as the genomic
visual, sensory, or speech symptoms usually followed by a headache region bounded by all markers in LD (r2 > 0.6 in 1000 Genomes
phase similar to migraine without aura. Project, phase 1, EUR individuals) with each of the 44 index SNPs,
Family and twin studies estimate a heritability of 42% (95% con- and in addition, all such regions in close proximity (<250 kb) were
fidence interval = 36–47%) for migraine7, pointing to a genetic merged. On the basis of these defined regions, we implicated 38
component of the disease. Despite this, genetic association studies genomic loci for the prevalent forms of migraine, 28 of which had
have uncovered relatively little about the molecular mechanisms that not been reported previously (Fig. 1).
contribute to migraine’s pathophysiology. Understanding has been These 38 loci replicated 10 of the 13 previously reported genome-
limited partly because so far only 13 genome-wide significant risk wide associations with migraine, and 6 loci contained a secondary
loci have been identified for the prevalent forms of migraine8–11. genome-wide significant SNP not in LD (r2 < 0.1) with the top SNP
For familial hemiplegic migraine (FHM), a rare Mendelian form of in the locus (Table 2). Five of these secondary signals were found
the disease, three ion-transport-related genes (CACNA1A, ATP1A2, in known loci (at LRP1–STAT6–SDR9C7, PRDM16, FHL5–UFL1,
and SCN1A) have been implicated12–14. These findings suggest that TRPM8–HJURP, and near TSPAN2–NGF), whereas the sixth was
mechanisms that regulate neuronal ion homeostasis might also be found within one of the 28 new loci (PLCE1). Therefore, out of the
involved in migraine more generally; however, no genes related 44 independent SNPs reported here, 34 represent new associations
to ion transport have yet been identified for the more prevalent with migraine. Three previously reported loci that were associated
forms of migraine15. with subtypes of migraine (rs1835740 near MTDH for migraine with
We conducted a meta-analysis of 22 GWA studies, including aura, rs10915437 near AJAP1 for migraine clinical samples, and
data for a total of 59,674 affected subjects and 316,078 controls col- rs10504861 near MMP16 for migraine without aura)8,11 showed
lected from six tertiary headache clinics and 27 population-based only nominal significance in the current meta-analysis (P = 5 × 10−3
cohorts through our worldwide collaboration with the International for rs1835740, P = 4.4 × 10−5 for rs10915437, and P = 4.9 × 10−5
Headache Genetics Consortium (IHGC). This combined data set for rs10504861; Supplementary Table 6); however, these loci have
contained more than 35,000 new migraine cases not included in pre- since been shown to be associated with specific phenotypic features of
viously published GWA studies. Here we present the findings of this migraine17, and therefore a more phenotypically homogeneous sam-
meta-analysis, including 38 genomic loci harboring 44 independent ple may be required for an accurate assessment of association. Four
association signals identified at levels of genome-wide significance, out of 44 SNPs (MRVI1, at TRPM8–HJURP, near ZCCHC14, and near
which support current theories of migraine pathophysiology and also CCM2L–HCK) showed moderate heterogeneity across the individual
offer new insights into the disease. GWA studies (Cochran’s Q test, P < 0.05; Supplementary Table 7);
therefore, at these markers we applied a random-effects model18.
RESULTS
Significant associations at 38 independent genomic loci Characterization of the associated loci
The primary meta-analysis was carried out on all samples from In total, 32 of 38 (84%) loci overlapped with transcripts from protein-
subjects with migraine available through the IHGC, regardless of coding genes, and 17 (45%) of these regions contained just a single
ascertainment. These case samples came from both individuals gene (see Supplementary Fig. 2 for regional association plots and
diagnosed by a doctor and individuals with self-reported migraine Supplementary Table 8 for additional locus information). Among the
as stated on questionnaires. Study design and sample ascertainment 38 loci, only two contained ion channel genes (KCNK5 and TRPM8)19,20.
for each individual study are outlined in the Supplementary Note Thus, despite previous hypotheses that migraine is a potential chan-
(and summarized in Supplementary Table 1). The final combined nelopathy5,21, the loci identified to date do not support the idea of
sample consisted of 59,674 case samples and 316,078 controls in 22 common variants in ion channel genes being strong susceptibility
non-overlapping case–control data sets (Table 1). All subjects were components in prevalent forms of migraine. However, three other
of European ancestry (EUR). Before including the largest study from loci do contain genes involved more generally in ion homeostasis 22–24
23andMe, we confirmed that it did not contribute any additional (SLC24A3, ITPK1, and GJA1; Supplementary Table 9).
heterogeneity compared with the other population and clinic-based Several of the identified genes have previously been associated with
studies (Supplementary Table 2). vascular disease (PHACTR1, TGFBR2, LRP1, PRDM16, RNF213, JAG1,
The 22 individual GWA studies included standard quality control HEY2, GJA1, and ARMS2)25–34 or are involved in smooth muscle con-
protocols (Online Methods), summarized in Supplementary Table 3. tractility and regulation of vascular tone (MRVI1, GJA1, SLC24A3,
Missing genotypes were then imputed into each sample using a com- and NRP1)35–38. Three of the 44 migraine index SNPs had previ-
mon 1000 Genomes Project reference panel16. Association analyses ously reported associations in the National Human Genome Research
were carried out within each study using logistic regression on the Institute GWA study catalog at exactly the same SNP (rs9349379 at
imputed marker dosages, with adjustments made for sex and other PHACTR1 with coronary heart disease39–41, coronary artery calcifica-
covariates where necessary (Online Methods and Supplementary tion42, and cervical artery dissection26; rs11624776 near ITPK1 with
Table 4). The association results were combined in an inverse-vari- thyroid hormone levels43; and rs11172113 at LRP1–STAT6–SDR9C7
ance weighted fixed-effects meta-analysis. Markers were filtered for with pulmonary function44; Supplementary Table 10). Six of the loci
imputation quality and other metrics (Online Methods), leaving harbor genes that are involved in nitric oxide (NO) signaling and oxi-
8,094,889 variants for consideration in our primary analysis. dative stress (REST, GJA1, YAP1, PRDM16, LRP1, and MRVI1)45–50.
Among the variants in the primary analysis, we identified For each locus, we chose the gene nearest to the index SNP to
44 genome-wide significant SNP associations (P < 5 × 10−8; assess gene expression activity in tissues from the Genotype-Tissue
Supplementary Fig. 1) that were independent (r2 < 0.1) with regard Expression (GTEx) Consortium project (Supplementary Fig. 3).
to linkage disequilibrium (LD). We validated these 44 SNPs by com- Although we found that most of the putative migraine loci genes
paring genotypes in a subset of the sample to those obtained from were expressed in many different tissue types, we were able to detect
whole-genome sequencing (Supplementary Table 5). To help identify tissue specificity in certain instances in which some genes showed
aDVANCE ONLINE PUBLICATION Nature Genetics

Articles
Table 1 Individual IHGC GWA studies with numbers of case and control samples used in the primary analysis (all migraine) and in the
subtype analyses (migraine with aura and migraine without aura)
All migraine Migraine with aura Migraine without aura
GWA study ID Full name of GWA study Cases Controls Cases Controls Cases Controls
23andMe 23andMe Inc. 30,465 143,147 - - - -
ALSPAC Avon Longitudinal Study of Parents and Children 3,134 5,103 - - - -
ATM Australian Twin Migraine 1,683 2,383 - - - -
B58C 1958 British Birth Cohort 1,165 4,141 - - - -
Danish HC Danish Headache Center 1,771 1,000 775 1,000 996 1,000
DeCODE deCODE Genetics Inc. 3,135 95,585 366 95,585 608 95,585
Dutch MA Dutch migraine with aura 734 5,211 734 5,211 - -
Dutch MO Dutch migraine without aura 1,115 2,028 - - 1,115 2,028
EGCUT Estonian Genome Center, University of Tartu 813 9,850 76 9,850 94 9,850
Finnish MA Finnish migraine with aura 933 2,715 933 2,715 - -
German MA German migraine with aura 1,071 1,010 1,071 1,010 - -
German MO German migraine without aura 1,160 1,647 - - 1,160 1,647
Health 2000 Health 2000 136 1,764 - - - -
HUNT Nord-Trøndelag Health Study 1,395 1,011 290 1,011 980 1,011
NFBC Northern Finnish Birth Cohort 756 4,393 - - - -
NTR/NESDA Netherlands Twin Register and the Netherlands 1,636 3,819 544 3,819 615 3,819
Study of Depression and Anxiety

Rotterdam III Rotterdam Study III 487 2,175 106 2,175 381 2,175
Swedish Twins Swedish Twin Registry 1,307 4,182 - - - -
Tromsø The Tromsø Study 660 2,407 - - - -
Twins UK Twins UK 618 2,334 202 2,334 416 2,334
WGHS Women’s Genome Health Study 5,122 18,108 1,177 18,108 1,826 18,108
Young Finns Young Finns 378 2,065 58 2,065 157 2,065
Total 59,674 316,078 6,332 144,883 8,348 139,622
Note that chromosome X genotype data were unavailable from three of the individual GWA studies (EGCUT, Rotterdam III, and Twins UK) and were partially unavailable from
some of the control samples (specifically, the GSK controls) used for the German MO study, meaning that the samples analyzed for chromosome X represented 57,756 cases and
299,109 controls. Complete data were available on the autosomes for all samples.
significantly higher expression in a particular tissue group relative to and control samples (such as population stratification), we analyzed
the others. For instance, four genes were expressed more actively in the genome-wide inflation of test statistics in our primary meta-
brain (GPR149, CFDP1, DOCK4, and MPPED2) than in other tissues, analysis. As expected for a complex polygenic trait, the distribution
and eight genes were specifically active in vascular tissues (PRDM16, of test statistics deviated from the null (genomic inflation factor
MEF2D, FHL5, C7orf10, YAP1, LRP1, ZCCHC14, and JAG1). Many λGC = 1.24; Supplementary Fig. 4), which is in line with other large
other putative migraine loci genes were actively expressed in more GWA study meta-analyses51–54. Because much of the inflation for
than one tissue group. a polygenic trait arises from LD between the causal SNPs and
many other neighboring SNPs in the local region, we LD-pruned the
Genomic inflation and LD-score regression analysis data to create a set of LD-independent markers (in PLINK55 with
To assess whether the 38 loci harbored true associations with a 250-kb sliding window and r2 > 0.2). The resulting genomic
migraine, rather than reflecting systematic differences between case inflation was comparatively reduced (λGC = 1.15; Supplementary
50 LRP1/STAT6/SDR9C7
Known loci
New loci
40
PRDM16
–log10 P value
30
TRPM8/HJURP FHL5/UFL1
Near TSPAN2/NGF
MEF2D PHACTR1
20 ARMS2/HTRA1 SLC24A3
HEY2/NCOA7 Near FGF6
C7orf10 PLCE1
KCNK5 MRVI1
Near ADAMTSL4/ECM1 Near TGFBR2 Near WSCD1/NLRP1
Near GJA1 ASTN2 Near JAG1
Near GPR149 HPSE2 YAP1 Near ZCCHC14
IGSF9B Near CCM2L/HCK
10 1p31.1 Near REST/SPINK2
Near DOCK4/IMMP2L
NRP1
MPPED2 CFDP1 RNF213 Near MED14/USP9X
CARF Near NOTCH4 Near ITPK1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 X
Chromosome
Figure 1 Manhattan plot showing results of the primary meta-analysis of all migraine samples (59,674 case and 316,078 control). The horizontal
axis shows the chromosomal position, and the vertical axis shows the significance of tested markers combined in a fixed-effects meta-analysis.
Markers that reached genome-wide significance (P < 5 × 10−8, chi-square test, 1 d.f.) at previously known and/or newly identified loci are highlighted.

Articles
Fig. 5) and probably reflects the inflation remaining owing to the values between a marker and all other markers within a 1-Mb win-
polygenic signal at many independent loci, including those not yet dow. The primary analysis results show a linear relationship between
significantly associated. association test statistics and LD score (Supplementary Fig. 6) and
To confirm that the observed inflation came primarily from true suggest that the majority (88.2%) of the inflation in test statistics can
polygenic signal, we analyzed the data from all imputed markers using be ascribed to true polygenic signal rather than population stratifi-
LD-score regression56. This method tests for a linear relationship cation or other confounders. These results are consistent with the
between marker test statistics and LD score, defined as the sum of r2 theory of polygenic disease architecture demonstrated previously by
Table 2 Summary of the 38 genomic loci associated with the prevalent types of migraine
All migraine Secondary signal Migraine without aura Previous
Locus Minor OR Index Index publication
rank Locus Chr Index SNP allele MAF (95% CI) P SNP P SNP P PMID
1 LRP1–STAT6– 12 rs11172113 C 0.42 0.90 (0.89–0.91) 5.6 × 10−49 rs11172055 1.3 × 10−9 rs11172113 4.3 × 10−16 21666692
SDR9C7
2 PRDM16 1 rs10218452 G 0.22 1.11 (1.10–1.13) 5.3 × 10−38 rs12135062 3.7 × 10−10 - - 21666692
3 FHL5–UFL1 6 rs67338227 T 0.23 1.09 (1.08–1.11) 2.0 × 10−27 rs4839827 5.7 × 10−10 rs7775721 1.1 × 10−12 23793025
4 Near 1 rs2078371 C 0.12 1.11 (1.09–1.13) 4.1 × 10−24 rs7544256 8.7 × 10−9 rs2078371 7.4 × 10−9 23793025
TSPAN2–NGF
5 TRPM8–HJURP 2 rs10166942 C 0.20 0.94 (0.89–0.99) 1.0 × 10−23 rs566529 2.5 × 10−9 rs6724624 1.1 × 10−9 21666692
6 PHACTR1 6 rs9349379 G 0.41 0.93 (0.92–0.95) 5.8 × 10−22 - - rs9349379 2.1 × 10−9 22683712
7 MEF2D 1 rs1925950 G 0.35 1.07 (1.06–1.09) 9.1 × 10−22 - - - - 22683712
8 SLC24A3 20 rs4814864 C 0.26 1.07 (1.06–1.09) 2.2 × 10−19 - - - - -
9 Near FGF6 12 rs1024905 G 0.47 1.06 (1.04–1.08) 2.1 × 10−17 - - rs1024905 2.5 × 10−9 -
10 C7orf10 7 rs186166891 T 0.11 1.09 (1.07–1.12) 9.7 × 10−16 - - - - 23793025
11 PLCE1 10 rs10786156 G 0.45 0.95 (0.94–0.96) 2.0 × 10−14 rs75473620 5.8 × 10−9 - - -
12 KCNK5 6 rs10456100 T 0.28 1.06 (1.04–1.07) 6.9 × 10−13 - - - - -
13 ASTN2 9 rs6478241 A 0.36 1.05 (1.04–1.07) 1.2 × 10−12 - - rs6478241 1.2 × 10−10 22683712
14 MRVI1 11 rs4910165 C 0.33 0.94 (0.91–0.98) 2.9 × 10−11 - - - - -
15 HPSE2 10 rs12260159 A 0.07 0.92 (0.89–0.94) 3.2 × 10−10 - - - - -
16 CFDP1 16 rs77505915 T 0.45 1.05 (1.03–1.06) 3.3 × 10−10 - - - - -
17 RNF213 17 rs17857135 C 0.17 1.06 (1.04–1.08) 5.2 × 10−10 - - - - -
18 NRP1 10 rs2506142 G 0.17 1.06 (1.04–1.07) 1.5 × 10−9 - - - - -
19 Near GPR149 3 rs13078967 C 0.03 0.87 (0.83–0.91) 1.8 × 10−9 - - - - -
20 Near JAG1 20 rs111404218 G 0.34 1.05 (1.03–1.07) 2.0 × 10−9 - - - - -
21 Near 4 rs7684253 C 0.45 0.96 (0.94–0.97) 2.5 × 10−9 - - - - -
REST–SPINK2
22 Near ZCCHC14 16 rs4081947 G 0.34 1.03 (1.00–1.06) 2.5 × 10−9 - - - - -
23 HEY2–NCOA7 6 rs1268083 C 0.48 0.96 (0.95–0.97) 5.3 × 10−9 - - - - -
24 Near 17 rs75213074 T 0.03 0.89 (0.86–0.93) 7.1 × 10−9 - - - - -
WSCD1–NLRP1
25 Near GJA1 6 rs28455731 T 0.16 1.06 (1.04–1.08) 7.3 × 10−9 - - - - -
26 Near TGFBR2 3 rs6791480 T 0.31 1.04 (1.03–1.06) 7.8 × 10−9 - - - - 22683712
27 Near ITPK1 14 rs11624776 C 0.31 0.96 (0.94–0.97) 7.9 × 10−9 - - - - -
28 Near 1 rs6693567 C 0.27 1.05 (1.03–1.06) 1.2 × 10−8 - - - - -
ADAMTSL4–ECM1
29 Near 20 rs144017103 T 0.02 0.85 (0.76–0.96) 1.2 × 10−8 - - - - -
CCM2L–HCK
30 YAP1 11 rs10895275 A 0.33 1.04 (1.03–1.06) 1.6 × 10−8 - - - - -
31 Near X rs12845494 G 0.27 0.96 (0.95–0.97) 1.7 × 10−8 - - - - -
MED14–USP9X
32 Near 7 rs10155855 T 0.05 1.08 (1.05–1.12) 2.1 × 10−8 - - - - -
DOCK4–IMMP2L
33 1p31.1a 1 rs1572668 G 0.48 1.04 (1.02–1.05) 2.1 × 10−8 - - - - -
34 CARF 2 rs138556413 T 0.03 0.88 (0.84–0.92) 2.3 × 10−8 - - - - -
35 ARMS2–HTRA1 10 rs2223089 C 0.08 0.93 (0.91–0.95) 3.0 × 10−8 - - - - -
36 IGSF9B 11 rs561561 T 0.12 0.94 (0.92–0.96) 3.4 × 10−8 - - - - -
37 MPPED2 11 rs11031122 C 0.24 1.04 (1.03–1.06) 3.5 × 10−8 - - - - -
38 Near NOTCH4 6 rs140002913 A 0.06 0.91 (0.88–0.94) 3.8 × 10−8 - - - - -
aThe nearest coding gene (LRRIQ3) to this locus is 592 kb away.
Ten loci were previously reported (PubMed IDs (PMID) are listed for these loci), and 28 were newly found in this study. Each locus is labeled with protein-coding genes that overlap the
region. Intergenic loci are also labeled as “near” to highlight the additional uncertainty in identifying relevant genes. Effect sizes and P values for each SNP were calculated for each
study with an additive genetic model using logistic regression adjusted for sex and then combined in a fixed-effects meta-analysis. For loci that contain a secondary LD-independent
signal passing genome-wide significance, the secondary index SNP and P value are given. For the seven loci reaching genome-wide significance in the ‘migraine without aura’ subtype
analysis, the corresponding index SNP and P value are also given. Evidence for significant heterogeneity was found at four loci (TRPM8–HJURP, MRVI1, near ZCCHC14, and near
CCM2L–HCK), so for those we present the results of a random-effects model. Chr, chromosome; MAF, minor allele frequency; OR, odds ratio; CI, confidence interval.

Articles
Figure 2 Expression enrichment of genes from Aorta

the migraine loci in GTEx tissue samples. Tibial artery
Expression data from 1,641 samples were Coronary artery
obtained using RNA-seq for 42 tissues and Heart—left ventricle
three cell lines. Enrichment P values were Heart—atrial appendage
Esophagus muscularis
assessed empirically for each tissue using a
Esophageal mucosa
permutation procedure (100,000 replicates),
Stomach
and the red vertical line shows the significance Colon—transverse
threshold after adjustment for multiple Tissue category
Liver
Cardiovascular system
testing by Bonferroni correction (Online Pancreas
Digestive system
Methods). LCL, Epstein−Barr transformed Testis
Exo-/endocrine system
lymphoblastoid cell lines. Ovary
Hemic and immune
Prostate Integumentary system
Adipose—visceral (omentum) Musculoskeletal
GWA studies using samples of similar size Adipose—subcutaneous Nervous
Pituitary gland
with both simulated and real data57. Mammary tissue
Respiratory
Urogenital
Kidney cortex
Migraine subtype analyses Adrenal gland
Thyroid
To elucidate pathophysiological mechanisms
GTEx tissue
Whole blood
underpinning the migraine aura, we car- LCL
ried out a secondary analysis of two created Skin—not sun-exposed
subsets that included only samples with the Skin—sun-exposed
Skin cells—transformed fibroblasts

subtypes ‘migraine with aura’ and ‘migraine Skeletal muscle
without aura’. These subsets included samples Caudate (basal ganglia)
only from studies in which sufficient informa- Nucleus accumbens
tion was available to assign a diagnosis of one Putamen
Anterior cingulate cortex (BA24)
or the other subtype according to classification Hypothalamus
criteria standardized by the International Frontal cortex (BA9)
Headache Society6. For the population-based Amygdala
Substantia nigra
studies, this involved questionnaires, whereas
Nerve—tibial
for the clinic-based studies the diagnosis was Cortex
assigned on the basis of a structured interview Spinal cord (cervical C1)
by telephone or in person. A stricter diagno- Hippocampus
Cerebellum
sis was required for the migraine subtypes, as Cerebellar cortex
it is often challenging to distinguish migraine Lung
aura from other neurological features that Uterus
Vagina
can present as symptoms from unrelated
Fallopian tube
conditions. As a result, the migraine subtype
0 −1 −2 −3 −4 −5
analyses involved considerably smaller sam- 10 10 10 10 10 10
Enrichment P value
ple sizes than the main analysis (6,332 case
samples and 144,883 controls for migraine
with aura; 8,348 case samples and 139,622 controls for migraine 44 loci (at MEF2D, PHACTR1, near REST–SPINK2, ASTN2, PLCE1,
without aura; Table 1). MPPED2, and near MED14–USP9X) exhibiting signs of heterogeneity
As with the primary analysis, the test statistics for migraine with across subtype groups (Supplementary Table 13).
aura and for migraine without aura were consistent with underlying
polygenic architecture rather than other potential sources of inflation Credible sets of markers within each locus
(Supplementary Figs. 7 and 8). In our analysis for migraine without For each of the 38 migraine-associated loci, we defined a credible
aura, we found seven significantly (P < 5 × 10−8) associated genomic loci set of markers that could plausibly be considered as causal using a
(near TSPAN2, TRPM8, PHACTR1, FHL5, ASTN2, near FGF6, and LRP1; Bayesian-likelihood-based approach59. This method incorporated
Supplementary Table 11 and Supplementary Fig. 9). All seven of these evidence from association-test statistics and the LD structure between
loci had already been identified in the primary analysis, possibly reflecting SNPs in a locus (Online Methods). A list of the credible set of SNPs
the fact that migraine without aura is the most common form of migraine obtained for each locus is provided in Supplementary Table 14.
(around two in three cases) and probably drove these association signals in We found three loci (in RNF213, PLCE1, and MRVI1) for which the
the primary analysis. Notably, no loci were associated with migraine with association signal could be credibly attributed to exonic mis-sense
aura in the other subset analysis (Supplementary Fig. 10). polymorphisms (Supplementary Table 15). However, most of the
To investigate whether excess heterogeneity could be contributing credible markers at each locus were either intronic or intergenic,
to the lack of associations for migraine with aura, we carried out a which is consistent with the theory that most variants detected by
heterogeneity analysis of the two subgroups (Online Methods and GWA studies involve regulatory effects on gene expression rather
Supplementary Table 12). We selected the 44 LD-independent SNPs than disruption of protein structure60,61.
from the primary analysis and used a random-effects model to com-
bine the ‘migraine with aura’ and ‘migraine without aura’ samples Overlap with eQTLs in specific tissues
in a meta-analysis that allowed for heterogeneity between the two To identify migraine loci that might influence gene expression, we
migraine groups58. We found little heterogeneity, with only 7 of the used previously published data sets cataloging expression quantitative

Articles
Figure 3 Expression enrichment of genes from the migraine loci in 209 5 Arteries
tissue or cell type annotations by DEPICT. Expression data were obtained
from 37,427 human microarray samples, and then genes in the migraine 4
loci were assessed for high expression in each of the annotation categories.
We determined enrichment P values by comparing the expression patterns
−log10 P value
from the migraine loci to those from 500 randomly generated loci, and we 3
estimated the false discovery rate (horizontal dashed line denotes
FDR < 0.05) to control for multiple testing (Online Methods). Red bars
2
indicate categories that were significant after FDR correction. A full list of
these enrichment results is provided in Supplementary Table 20.
1
trait loci (eQTLs) in either of two microarray-based studies of samples

from peripheral venous blood (n = 3,754) or from human brain cor- 0
al
us
o g ry
ic
l
tex (n = 550). Additionally, we used a third study based on RNA-seq
ta
la
r
tiv
in
un
th
et
ta
at to
vo
ni
cu
cr
es
na
el
m
en
ge
er
as
do
om pir
sk
im
data from a collection of 42 tissues and three cell lines (n = 1,641)
ig
um
ro
N
ov
En
St es
D
lo
d
U
g
cu
di
an
R
te
ar
from the GTEx Consortium project62. Although these data offered the
us
In
ic
C
M
em
advantage of a diverse tissue catalog, the number of samples per tis-
H
Physiological system
sue was relatively small (Supplementary Table 16) compared with the
two microarray data sets, which may have resulted in reduced power that the enrichment of migraine risk variants in genes expressed in
to detect significant eQTLs in some tissues. Using these data sets, we tissues with a smooth muscle component is not specific to blood
applied a method based on the overlap of migraine and eQTL credible vessels, the stomach, or the gastrointestinal tract; rather, it seems to be
sets to identify eQTLs that could explain associations at the 38 migraine generalizable across vascular and visceral smooth muscle types.
loci (Online Methods). This approach merged the migraine credible Combined, these enrichment results suggest that some of the genes
sets defined above with credible sets from cis-eQTL signals within a affected by migraine-associated variants have high expression in vas-
1-Mb window and tested whether the association signals between the cular tissues, and their dysfunction could have a role in migraine.
migraine and eQTL credible sets were correlated. After adjusting for Furthermore, the results suggest that other tissue types (for example,
multiple testing, we found no plausible eQTL associations in the periph- smooth muscle) could also have a role, which may become evident
eral blood or brain cortex data (Supplementary Tables 17 and 18 and once more migraine loci are discovered.
Supplementary Fig. 11). In the GTEx data, however, we found evidence
for overlap from eQTLs in three tissues (lung, tibial artery, and aorta) at Enrichment in tissue-specific enhancers
the HPSE2 locus and in one tissue (thyroid) at the HEY2–NCOA7 locus To further assess the hypothesis that migraine variants might operate
(Supplementary Table 19 and Supplementary Fig. 12). via effects on gene regulation, we investigated the degree of overlap
In summary, from three data sets we found evidence implicating with histone modifications. Using candidate causal variants from the
eQTL signals at only two loci (HPSE2 and HEY2). This low number migraine loci, we examined their enrichment in cell-type-specific
(2 out of 38) is consistent with previous studies noting that available enhancers from 56 primary human tissues and cell types from the
eQTL catalogs currently lack sufficient tissue specificity and develop- Roadmap Epigenomics65 and ENCODE projects66 (Online Methods
mental diversity to provide enough power for meaningful biological and Supplementary Table 21). These variants showed the greatest
insight53. No plausibly causal eQTLs were observed in expression data enrichment in tissues from the mid-frontal lobe and duodenum
from brain tissue samples. smooth muscle, but their enrichment was not significant after adjust-
ment for multiple testing (Fig. 4).
Gene expression enrichment in specific tissues
To understand whether the 38 migraine loci as a group are enriched Gene set enrichment analyses
for expression in certain tissue types, we again used the GTEx pilot To implicate underlying biological pathways involved in migraine,
data62 (Online Methods). We found four tissues that were signifi- we applied a Gene Ontology over-representation analysis of the 38
cantly enriched (after Bonferroni correction) for expression of the migraine loci (Online Methods). We found nine vascular-related bio-
migraine-associated genes (Fig. 2). The two most strongly enriched logical function categories that were significantly enriched (adjusted
tissues were part of the cardiovascular system (aorta and tibial artery). P < 0.05) after correction for multiple testing (Supplementary Table
The two other significantly enriched tissues were from the digestive 22). Notably, data for the identified loci provided little statistical sup-
system (esophagus muscularis and esophageal mucosa). We replicated port for some molecular processes that have been previously linked
these enrichment results using the DEPICT63 tool and an independ- to migraine, including ion homeostasis, glutamate signaling, serot-
ent microarray-based gene expression data set (Online Methods). onin signaling, NO signaling, and oxidative stress. However, a pos-
DEPICT highlighted four tissues (Fig. 3 and Supplementary Table 20) sible explanation for the lack of enrichment for these functions is
with significant enrichment of genes within the migraine loci: arteries that current annotations for many genes and pathways are far from
(P = 1.58 × 10−5), the upper gastrointestinal tract (P = 2.97 × 10−3), comprehensive, or that larger numbers of migraine loci need to be
myometrium (P = 3.03 × 10−3), and stomach (P = 3.38 × 10−3). identified before sensitivity can be sufficient to detect enrichment
Taken together, the expression analyses implicated arterial and for these mechanisms.
gastrointestinal tissues. To discover whether this enrichment signa- For a more comprehensive pathway analysis, we used DEPICT,
ture could be attributed to a more specific type of smooth muscle, which incorporates gene coexpression information from microar-
we examined the expression of the nearest genes at migraine loci ray data to implicate additional, functionally less well-characterized
in a panel of 60 types of human smooth muscle tissue64. Overall, genes in known biological pathways, protein–protein complexes, and
migraine loci genes were not significantly enriched in a particular mouse phenotypes63 (by forming so-called reconstituted gene sets).
class of smooth muscle (Supplementary Figs. 13–15). This suggests From DEPICT, we identified 67 reconstituted gene sets that were

Articles
Figure 4 Enrichment of the migraine loci

in sets of tissue- and cell-type-specific
enhancers. Credible sets from the migraine
Hippocampus middle
loci were mapped to sets of enhancers under
Anterior caudate
active expression in 56 tissues and cell lines Substantia nigra
(identified by H3K27ac histone marks from Cingulate gyrus
the Roadmap Epigenomics65 and ENCODE66 Mid-frontal lobe
Angular gyrus
projects). We assessed enrichment P values Inferior temporal lobe
empirically by randomly generating a background Neurosphere
set of matched loci for comparison (10,000 HUES 6
replicates); the vertical dotted line indicates HUES 64
hiPS−20b
the significance threshold after adjustment iPS−18a
for 56 separate tests by Bonferroni correction H1
(Online Methods). BM MSC, bone marrow Kidney
mesenchymal stem cells; dif, differentiated; Pancreatic islets
Liver
NHDF, normal human dermal fibroblasts; Duodenum mucosa
NHLF, normal human lung fibroblasts; NH-A, Colonic mucosa
normal human astrocytes; MACS, magnetic- Rectal mucosa
Adipose nuclei
activated cell sorting; HSMM, human skeletal
Skeletal muscle
muscle myoblast; HUVEC, human umbilical Duodenum smooth muscle
vein endothelial cells; NHEK, normal human Rectal smooth muscle
epidermal keratinocyte; HMEC, human mammary Stomach smooth muscle
Colon smooth muscle
epithelial cells; TH, helper T cell; Tmem, memory
BM MSC
T cell; Treg, regulatory T cell; stim, stimulated. Chondrogenic dif cells
NH−osteoblast
NHDF
NHLF
significantly enriched (false discovery rate NH−A
HSMM−myotube
(FDR) < 5%) for genes found among the HSMM
migraine-associated loci (Supplementary SK−N−MC
A673
Table 23). Because the reconstituted gene K562
sets had genes in common, we clustered them HepG2
HUVEC
into ten distinct groups (Fig. 5 and Online NHEK
Methods). Several gene sets, including the HMEC
HeLaS3
most significantly enriched reconstituted DND−41
gene set (abnormal vascular wound healing; Mobilized CD34
P = 1.86 × 10−6), were grouped into clus- Adult CD14
GM12878
ters related to cell–cell interactions (ITGB1 Adult CD20
protein-protein interaction, adherens junction, CD19
TH2
and integrin complex). Several of the other TH1
gene set clusters were also related to vascular – –
TH0
CD25 IL17 TH stim MACS
biology (Fig. 5 and Supplementary Table 23). – +
CD25 IL17 TH17 stim
int +
We still did not observe any support for CD25 CD127 Tmem
+ –
CD25 CD127 Treg
molecular processes with hypothesized links –
CD25 CD45RA naive
+
to migraine (Supplementary Table 24); +

CD3 primary cells
however, this might again have been due to 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
the reasons outlined above. –log10 P value
DISCUSSION dysfunction and possibly also smooth muscle dysfunction are likely
In what is to our knowledge the largest genetic study of migraine so to have roles in migraine pathogenesis.
far, we identified 38 distinct genomic loci harboring 44 independent The support for vascular and smooth muscle enrichment of the
susceptibility markers for the prevalent forms of migraine. We provide loci is strong, with multiple lines of evidence from independent
evidence that migraine-associated genes are involved in both arterial methods and independent data sets. However, it remains likely that
and smooth muscle function. Two separate analyses, the DEPICT and neurogenic mechanisms are also involved in migraine. For example,
GTEx gene expression enrichment analyses, pointed to the involve- several lines of evidence from previous studies have pointed to
ment of vascular and smooth muscle tissues in common variant sus- such mechanisms5,69–72. We found some support for this in our
ceptibility to migraine. The vascular finding is consistent with known examination of the expression of individual genes at the 38 loci
comorbidities and previously reported shared polygenic risk among (Supplementary Fig. 3 and Supplementary Table 25), which
migraine, stroke, and cardiovascular diseases 67,68. Furthermore, a showed that several genes were specifically active in brain tissues.
recent GWA study of cervical artery dissection identified a genome- We did not observe statistically significant enrichment in brain
wide significant association at the same index SNP (rs9349379 in across all loci, but it may be that more associated loci are needed in
the PHACTR1 locus) as is associated with migraine, suggesting the order for this to be detected. Alternatively, the lack of significant
possibility of partially shared genetic components between migraine enrichment could be due to difficulties in collecting appropriate
and cervical artery dissection26. These results suggest that vascular brain tissue samples with enough specificity, or other technical

Articles
a Transcription factor b Abnormal vascular

binding
wound healing
Transforming growth Integrin binding Integrin cell surface
Partial postnatal interactions
factor β–activated
lethality
receptor activity
ITGB5 PPI NID2 PPI
Vascular endothelial
growth factor Blood vessel
receptor signaling development
ITGB7 PPI Collagen binding
pathway
Decreased cardiac
ITGB1 protein muscle contractility Gene set P values
ITGB4 PPI
complex ITGA5 PPI P < 10–3
P < 10–4
Protein-complex P < 10–5
Positive regulation Of Adherens junction GIPC1 PPI
binding
cell morphogenesis
Gene set overlap
involved in
Low
differentiation Integrin complex Abnormal cell Abnormal
ITGB1 PPI Medium
migration vasoconstriction
High
Figure 5 DEPICT network of the reconstituted gene sets that were significantly enriched (FDR < 0.05, determined empirically by permutation) for genes at
the migraine loci (Online Methods). Enriched gene sets are represented as nodes, with the extent of pairwise overlap denoted by the width of the connecting
lines, and empirical enrichment P values are indicated by color according to the key. (a) The 67 significantly enriched gene sets clustered by similarity
into ten group nodes, with each group node named for the most representative gene set in the group. (b) One example of gene sets that were clustered
within the now expanded ITGB1 protein-protein interaction (PPI) group. A full list of the 67 significantly enriched reconstituted gene sets can be found in
Supplementary Table 23.
challenges. Additionally, there is less clarity regarding the biologi- However, the heterogeneity analysis (Supplementary Tables 12
cal mechanisms for a neurological disease like migraine compared and 13) demonstrated that most of the identified loci were impli-
with some other common diseases, such as autoimmune or cardio- cated in both migraine subtypes. This suggests that the absence of
metabolic diseases for which intermediate risk factors and underly- significant loci in the analysis for migraine with aura was mainly due
ing mechanisms are better understood. to a lack of power owing to the smaller sample size. Additionally,
Interestingly, some of the analyses highlighted gastrointestinal tis- as shown by the LD-score analysis (Supplementary Figs. 6–8), the
sues. Although migraine attacks may include gastrointestinal symptoms amount of heritability captured by the data set for migraine with
(e.g., nausea, vomiting, diarrhea)6, it is likely that the signals observed aura was considerably lower than that for migraine without aura,
here broadly represent smooth muscle signals rather than gastrointes- such that in order for comparable power to be achieved, a sample
tinal specificity. Smooth muscle is a predominant tissue of the intestine, size two to three times larger would be required. This might reflect a
but specific smooth muscle subtypes were not available to test this higher degree of heterogeneity in the clinical capture, more complex
hypothesis in our primary enrichment analyses. Instead we examined underlying biology, or even a greater contribution to risk from low-
a range of 60 smooth muscle subtypes and found that the migraine loci frequency and rare variation for this form of the disease.
were expressed in many types of smooth muscle, including vascular In conclusion, the 38 genomic loci identified in this study support
(Supplementary Figs. 14 and 15). These results, although not conclu- the notion that factors in vascular and smooth muscle tissues con-
sive, suggest that the enrichment of the migraine loci in smooth muscle tribute to migraine pathophysiology and that the two major subtypes
is not specific to the stomach and gastrointestinal tract. of migraine—migraine with aura and migraine without aura—have a
Our results implicate cellular pathways and provide an opportunity partially shared underlying genetic susceptibility profile.
to determine whether the genomic data support previously presented
hypotheses of mechanisms linked to migraine. One prevailing hypoth- URLs. 1000 Genomes Project, http://www.1000genomes.org/; BEAGLE,
esis, stimulated by findings in FHM, has been that migraine is a chan- http://faculty.washington.edu/browning/beagle/beagle.html;
nelopathy5,21. Among the 38 migraine loci in the current study, only DEPICT, https://github.com/perslab/DEPICT; credible set fine-mapping
two harbor known ion channels (KCNK5 and TRPM8)19,20, and three script, https://github.com/hailianghuang/FM-summary; GTEx, http://
others (SLC24A3, near ITPK1, and near GJA1) can be linked to ion www.gtexportal.org; GWAMA, http://www.well.ox.ac.uk/gwama/;
homeostasis22–24. This further supports the findings of previous stud- IMPUTE2, https://mathgen.stats.ox.ac.uk/impute/impute_v2.html;
ies that, in common forms of migraine, ion channel dysfunction is not IHGC, http://www.headachegenetics.org/; MACH, http://www.sph.
the major pathophysiological mechanism15. However, more generally, umich.edu/csg/abecasis/MACH/tour/imputation.html; matSpD,
genes involved in ion homeostasis could be a component of genetic http://neurogenetics.qimrberghofer.edu.au/matSpD; MINIMAC,
susceptibility. Moreover, we cannot exclude that ion channels might still http://genome.sph.umich.edu/wiki/Minimac; PANTHER Gene
be important contributors in migraine with aura, the form most closely Ontology enrichment, http://geneontology.org/page/go-enrichment-
resembling FHM, as identifying loci for this subgroup is more challeng- analysis; PLINK, https://www.cog-genomics.org/plink2; ProbABEL,
ing. Another hypothesis relates to oxidative stress and NO signaling73–75. http://www.genabel.org/packages/ProbABEL; R, https://www.
Six genes with known links to oxidative stress and NO were identified r-project.org/; Roadmap Epigenomics data, http://www.epigenomeatlas.
in these 38 loci (REST, GJA1, YAP1, PRDM16, LRP1, and MRVI1)45–50. org; SHAPEIT, http://www.shapeit.fr; SNPTEST, https://mathgen.
This is in line with previous findings11; however, the DEPICT pathway stats.ox.ac.uk/genetics_software/snptest/snptest.html.
analysis did not show an association between NO-related reconstituted
gene sets and migraine (FDR > 0.54; Supplementary Table 24). Methods
Notably, in the migraine-subtype analyses, it was possible to identify Methods and any associated references are available in the online
specific loci for migraine without aura but not for migraine with aura. version of the paper.

Articles
Note: Any Supplementary Information and Source Data files are available in the 11. Anttila, V. et al. Genome-wide meta-analysis identifies new susceptibility loci for
online version of the paper. migraine. Nat. Genet. 45, 912–917 (2013).
12. Ophoff, R.A. et al. Familial hemiplegic migraine and episodic ataxia type-2 are caused
by mutations in the Ca2+ channel gene CACNL1A4. Cell 87, 543–552 (1996).
Acknowledgments
13. De Fusco, M. et al. Haploinsufficiency of ATP1A2 encoding the Na+/K+ pump α2 subunit
We thank the numerous individuals who contributed to sample collection, associated with familial hemiplegic migraine type 2. Nat. Genet. 33, 192–196 (2003).
storage, handling, phenotyping, and genotyping for each of the individual cohorts. 14. Dichgans, M. et al. Mutation in the neuronal voltage-gated sodium channel SCN1A
We also acknowledge the important contribution to research made by the study in familial hemiplegic migraine. Lancet 366, 371–377 (2005).
participants. We are grateful to H. Zhao (QIMR Berghofer Medical Research 15. Nyholt, D.R. et al. A high-density association screen of 155 ion transport genes for
Institute) for helpful correspondence on the pathway analyses. We acknowledge involvement with common migraine. Hum. Mol. Genet. 17, 3318–3331 (2008).
the GTEx Consortium for support and contribution of pilot data. A list of study- 16. 1000 Genomes Project Consortium et al. An integrated map of genetic variation
specific acknowledgments and funding information can be found in from 1,092 human genomes. Nature 491, 56–65 (2012).
17. Chasman, D.I. et al. Selectivity in genetic association with sub-classified migraine
the Supplementary Note.
in women. PLoS Genet. 10, e1004366 (2014).
18. Han, B. & Eskin, E. Random-effects model aimed at discovering associations
AUTHOR CONTRIBUTIONS in meta-analysis of genome-wide association studies. Am. J. Hum. Genet. 88,
P.G., V. Anttila, G.W.M., M.I.K., M. Kals, R. Mägi, K.P., E.H., E.L., A.G.U., L.C., 586–598 (2011).
E.M., L.M., A.-L.E., A.F.C., T.F.H., A.J.A., D.I.C., and D.R.N. performed the 19. Morton, M.J., Abohamed, A., Sivaprasadarao, A. & Hunter, M. pH sensing in
experiments. P.G., V. Anttila, B.S.W., P.P., T.E., T.H.P., K.-H.F., M. Muona, N.A.F., the two-pore domain K+ channel, TASK2. Proc. Natl. Acad. Sci. USA 102,
A.I., G.M., L.L., S.G.G., S. Steinberg, L.Q., H.H.H.A., D.A.H., J.-J.H., R. Malik, 16102–16106 (2005).
A.E.B., E.S., C.M.v.D., E.M., D.P.S., N.E., B.M.N., D.I.C., and D.R.N. performed 20. Ramachandran, R. et al. TRPM8 activation attenuates inflammatory responses in
mouse models of colitis. Proc. Natl. Acad. Sci. USA 110, 7476–7481 (2013).
the statistical analyses. P.G., V. Anttila, B.S.W., P.P., T.E., T.H.P., K.-H.F., E.C.-L.,
21. Hanna, M.G. Genetic neurological channelopathies. Nat. Clin. Pract. Neurol. 2,
N.A.F., A.I., G.M., L.L., M. Kallela, T.M.F., S.G.G., S. Steinberg, M. Koiranen, L.Q., 252–263 (2006).
H.H.H.A., T.L., J.W., D.A.H., S.M.R., M.F., V. Artto, M. Kaunisto, S.V., R. Malik, 22. Kraev, A. et al. Molecular cloning of a third member of the potassium-dependent
M.I.K., M. Kals, R. Mägi, K.P., H.H., A.E.B., J.H., E.S., C.S., C.W., Z.C., K.H., E.L., sodium-calcium exchanger gene family, NCKX3. J. Biol. Chem. 276, 23161–23172
L.M.P., A.-L.E., A.F.C., T.F.H., J.K., A.J.A., O.R., M.A.I., M.-R.J., D.P.S., M.W., (2001).
G.D.S., N.E., M.J.D., B.M.N., J.O., D.I.C., D.R.N., and A.P. participated in data 23. Ismailov, I.I. et al. A biologic function for an ‘orphan’ messenger: D-myo-inositol
analysis and/or interpretation. P.G., V. Anttila, B.S.W., T.H.P., K.-H.F., E.C.-L., 3,4,5,6-tetrakisphosphate selectively blocks epithelial calcium-activated chloride
T.K., G.M.T., M. Kallela, C.R., A.H.S., G.B., M. Koiranen, T.L., M.S., M.G.H., M.F., channels. Proc. Natl. Acad. Sci. USA 93, 10505–10509 (1996).
24. De Bock, M. et al. Connexin channels provide a target to manipulate brain
V. Artto, M. Kaunisto, S.V., R. Malik, A.C.H., P.A.F.M., N.G.M., G.W.M., H.H.,
endothelial calcium dynamics and blood-brain barrier permeability. J. Cereb. Blood
A.E.B., L.F., J.H., P.H.L., C.S., C.W., Z.C., B.M.-M., S. Schreiber, T.M., J.G.E., V.S., Flow Metab. 31, 1942–1957 (2011).
A.G.U., C.M.v.D., A.S., C.S.N., H.G., A.-L.E., A.F.C., T.F.H., T.W., A.J.A., O.R., 25. Kathiresan, S. et al. Genome-wide association of early-onset myocardial infarction
M.-R.J., C.K., M.D.F., A.C.B., M.D., M.W., J.-A.Z., B.M.N., J.O., D.I.C., D.R.N., with single nucleotide polymorphisms and copy number variants. Nat. Genet. 41,
A.-P.S., J.E.B., P.M.R., and A.P. contributed materials and/or analysis tools. T.E., 334–341 (2009).
T.K., T.L., H.S., B.W.J.H.P., A.C.H., P.A.F.M., N.G.M., G.W.M., L.F., A.H., A.S., 26. Debette, S. et al. Common variation in PHACTR1 is associated with susceptibility
C.S.N., M. Männikkö, T.W., J.K., O.R., M.A.I., T.S., M.-R.J., A.M., C.K., D.P.S., to cervical artery dissection. Nat. Genet. 47, 78–83 (2015).
M.D.F., A.M.J.M.v.d.M., J.-A.Z., D.I.B., G.D.S., K.S., N.E., B.M.N., J.O., D.I.C., 27. Law, C. et al. Clinical features in a family with an R460H mutation in transforming
growth factor β receptor 2 gene. J. Med. Genet. 43, 908–916 (2006).
D.R.N., and A.P. supervised the research. T.K., G.M.T., G.B., T.L., J.E.B., M.S.,
28. Bown, M.J. et al. Abdominal aortic aneurysm is associated with a variant in
P.M.R., H.S., B.W.J.H.P., A.C.H., P.A.F.M., N.G.M., G.W.M., L.F., V.S., A.H., L.C., low-density lipoprotein receptor-related protein 1. Am. J. Hum. Genet. 89, 619–627
A.S., C.S.N., H.G., J.K., A.J.A., O.R., M.A.I., M.-R.J., A.M., C.K., D.P.S., M.D., (2011).
A.M.J.M.v.d.M., D.I.B., G.D.S., N.E., M.J.D., B.M.N., D.I.C., D.R.N., and A.P. 29. Arndt, A.K. et al. Fine mapping of the 1p36 deletion syndrome identifies mutation of
conceived and designed the study. P.G., V. Anttila, B.S.W., P.P., T.E., T.H.P., E.C.-L., PRDM16 as a cause of cardiomyopathy. Am. J. Hum. Genet. 93, 67–77 (2013).
H.H., B.M.N., J.O., D.I.C., D.R.N., and A.P. wrote the paper. All authors contributed 30. Fujimura, M. et al. Genetics and biomarkers of Moyamoya disease: significance of
to the final version of the manuscript. RNF213 as a susceptibility gene. J. Stroke 16, 65–72 (2014).
31. McElhinney, D.B. et al. Analysis of cardiovascular phenotype and genotype-
COMPETING FINANCIAL INTERESTS phenotype correlation in individuals with a JAG1 mutation and/or Alagille syndrome.
Circulation 106, 2567–2574 (2002).
The authors declare competing financial interests: details are available in the online
32. Bezzina, C.R. et al. Common variants at SCN5A–SCN10A and HEY2 are associated
version of the paper. with Brugada syndrome, a rare disease with high risk of sudden cardiac death.
Nat. Genet. 45, 1044–1049 (2013).
Reprints and permissions information is available online at http://www.nature.com/ 33. Sinner, M.F. et al. Integrating genetic, transcriptional, and functional analyses to
reprints/index.html. identify five novel genes for atrial fibrillation. Circulation 130, 1225–1235 (2014).
34. Neale, B.M. et al. Genome-wide association study of advanced age-related macular
1. Vos, T. et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases degeneration identifies a role of the hepatic lipase gene (LIPC). Proc. Natl. Acad.
and injuries 1990–2010: a systematic analysis for the Global Burden of Disease Sci. USA 107, 7395–7400 (2010).
Study 2010. Lancet 380, 2163–2196 (2012). 35. Desch, M. et al. IRAG determines nitric oxide- and atrial natriuretic peptide-
2. Vos, T. et al. Global, regional, and national incidence, prevalence, and years lived mediated smooth muscle relaxation. Cardiovasc. Res. 86, 496–505 (2010).
with disability for 301 acute and chronic diseases and injuries in 188 countries, 36. Lang, N.N., Luksha, L., Newby, D.E. & Kublickiene, K. Connexin 43 mediates
1990–2013: a systematic analysis for the Global Burden of Disease Study 2013. endothelium-derived hyperpolarizing factor-induced vasodilatation in subcutaneous
Lancet 386, 743–800 (2015). resistance arteries from healthy pregnant women. Am. J. Physiol. Heart Circ. Physiol.
3. Gustavsson, A. et al. Cost of disorders of the brain in Europe 2010. Eur. 292, H1026–H1032 (2007).
Neuropsychopharmacol. 21, 718–779 (2011). 37. Dong, H., Jiang, Y., Triggle, C.R., Li, X. & Lytton, J. Novel role for K+-dependent
4. Pietrobon, D. & Striessnig, J. Neurological diseases: neurobiology of migraine. Nat. Na+/Ca2+ exchangers in regulation of cytoplasmic free Ca2+ and contractility in arterial
Rev. Neurosci. 4, 386–398 (2003). smooth muscle. Am. J. Physiol. Heart Circ. Physiol. 291, H1226–H1235 (2006).
5. Tfelt-Hansen, P.C. & Koehler, P.J. One hundred years of migraine research: major 38. Yamaji, M., Mahmoud, M., Evans, I.M. & Zachary, I.C. Neuropilin 1 is essential for
clinical and scientific observations from 1910 to 2010. Headache 51, 752–778 gastrointestinal smooth muscle contractility and motility in aged mice. PLoS One
(2011). 10, e0115563 (2015).
6. Headache Classification Committee of the International Headache Society (IHS) 39. Lu, X. et al. Genome-wide association study in Han Chinese identifies four new
The International Classification of Headache Disorders, 3rd edition (beta version). susceptibility loci for coronary artery disease. Nat. Genet. 44, 890–894 (2012).
Cephalalgia 33, 629–808 (2013). 40. Hager, J. et al. Genome-wide association study in a Lebanese cohort confirms PHACTR1
7. Polderman, T.J.C. et al. Meta-analysis of the heritability of human traits based on as a major determinant of coronary artery stenosis. PLoS One 7, e38663 (2012).
fifty years of twin studies. Nat. Genet. 47, 702–709 (2015). 41. The Coronary Artery Disease (C4D) Genetics Consortium. A genome-wide association
8. Anttila, V. et al. Genome-wide association study of migraine implicates a common study in Europeans and South Asians identifies five new loci for coronary artery
susceptibility variant on 8q22.1. Nat. Genet. 42, 869–873 (2010). disease. Nat. Genet. 43, 339–344 (2011).
9. Chasman, D.I. et al. Genome-wide association study reveals three susceptibility 42. O’Donnell, C.J. et al. Genome-wide association study for coronary artery calcification
loci for common migraine in the general population. Nat. Genet. 43, 695–698 with follow-up in myocardial infarction. Circulation 124, 2855–2864 (2011).
(2011). 43. Porcu, E. et al. A meta-analysis of thyroid-related traits reveals novel loci and
10. Freilinger, T. et al. Genome-wide association analysis identifies susceptibility loci gender-specific differences in the regulation of thyroid function. PLoS Genet. 9,
for migraine without aura. Nat. Genet. 44, 777–782 (2012). e1003266 (2013).

Articles
44. Soler Artigas, M. et al. Genome-wide association and large-scale follow up identifies 59. Maller, J.B. et al. Bayesian refinement of association signals for 14 loci in 3 common
16 new loci influencing lung function. Nat. Genet. 43, 1082–1090 (2011). diseases. Nat. Genet. 44, 1294–1301 (2012).
45. Lu, T. et al. REST and stress resistance in ageing and Alzheimer disease. Nature 60. Nicolae, D.L. et al. Trait-associated SNPs are more likely to be eQTLs: annotation
507, 448–454 (2014). to enhance discovery from GWAS. PLoS Genet. 6, e1000888 (2010).
46. Kar, R., Riquelme, M.A., Werner, S. & Jiang, J.X. Connexin 43 channels protect 61. Maurano, M.T. et al. Systematic localization of common disease-associated variation
osteocytes against oxidative stress-induced cell death. J. Bone Miner. Res. 28, in regulatory DNA. Science 337, 1190–1195 (2012).
1611–1621 (2013). 62. The GTEx Consortium. The Genotype-Tissue Expression (GTEx) project. Nat. Genet.
47. Dixit, D., Ghildiyal, R., Anto, N.P. & Sen, E. Chaetocin-induced ROS-mediated 45, 580–585 (2013).
apoptosis involves ATM-YAP1 axis and JNK-dependent inhibition of glucose 63. Pers, T.H. et al. Biological interpretation of genome-wide association studies using
metabolism. Cell Death Dis. 5, e1212 (2014). predicted gene functions. Nat. Commun. 6, 5890 (2015).
48. Chuikov, S., Levi, B.P., Smith, M.L. & Morrison, S.J. Prdm16 promotes stem cell 64. Chi, J.T. et al. Gene expression programs of human smooth muscle cells: tissue-
maintenance in multiple tissues, partly by regulating oxidative stress. Nat. Cell Biol. specific differentiation and prognostic significance in breast cancers. PLoS Genet.
12, 999–1006 (2010). 3, 1770–1784 (2007).
49. Castellano, J. et al. Hypoxia stimulates low-density lipoprotein receptor-related 65. Bernstein, B.E. et al. The NIH Roadmap Epigenomics Mapping Consortium. Nat.
protein-1 expression through hypoxia-inducible factor-1α in human vascular smooth Biotechnol. 28, 1045–1048 (2010).
muscle cells. Arterioscler. Thromb. Vasc. Biol. 31, 1411–1420 (2011). 66. The ENCODE Project Consortium. An integrated encyclopedia of DNA elements in
50. Schlossmann, J. et al. Regulation of intracellular calcium by a signalling complex the human genome. Nature 489, 57–74 (2012).
of IRAG, IP3 receptor and cGMP kinase Iβ. Nature 404, 197–201 (2000). 67. Winsvold, B.S. et al. Genetic analysis for a shared biological basis between migraine
51. Nalls, M.A. et al. Large-scale meta-analysis of genome-wide association data and coronary artery disease. Neurol. Genet. 1, e10 (2015).
identifies six new risk loci for Parkinson’s disease. Nat. Genet. 46, 989–993 68. Malik, R. et al. Shared genetic basis for migraine and ischemic stroke: a genome-
(2014). wide analysis of common variants. Neurology 84, 2132–2145 (2015).
52. Lambert, J.C. et al. Meta-analysis of 74,046 individuals identifies 11 new 69. Ferrari, M.D., Klever, R.R., Terwindt, G.M., Ayata, C. & van den Maagdenberg,
susceptibility loci for Alzheimer’s disease. Nat. Genet. 45, 1452–1458 (2013). A.M.J.M. Migraine pathophysiology: lessons from mouse models and human
53. Schizophrenia Working Group of the Psychiatric Genomics Consortium. Biological genetics. Lancet Neurol. 14, 65–80 (2015).
insights from 108 schizophrenia-associated genetic loci. Nature 511, 421–427 70. Olesen, J., Burstein, R., Ashina, M. & Tfelt-Hansen, P. Origin of pain in migraine:
(2014). evidence for peripheral sensitisation. Lancet Neurol. 8, 679–690 (2009).
54. Wood, A.R. et al. Defining the role of common variation in the genomic and 71. Hadjikhani, N. et al. Mechanisms of migraine aura revealed by functional MRI in
biological architecture of adult human height. Nat. Genet. 46, 1173–1186 human visual cortex. Proc. Natl. Acad. Sci. USA 98, 4687–4692 (2001).
(2014). 72. Lauritzen, M. Pathophysiology of the migraine aura. The spreading depression theory.
55. Purcell, S. et al. PLINK: a tool set for whole-genome association and population- Brain 117, 199–210 (1994).
based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007). 73. Olesen, J. The role of nitric oxide (NO) in migraine, tension-type headache and
56. Bulik-Sullivan, B.K. et al. LD score regression distinguishes confounding from cluster headache. Pharmacol. Ther. 120, 157–171 (2008).
polygenicity in genome-wide association studies. Nat. Genet. 47, 291–295 (2015). 74. Ashina, M., Hansen, J.M. & Olesen, J. Pearls and pitfalls in human pharmacological
57. Yang, J. et al. Genomic inflation factors under polygenic inheritance. Eur. J. Hum. models of migraine: 30 years’ experience. Cephalalgia 33, 540–553 (2013).
Genet. 19, 807–812 (2011). 75. Read, S.J. & Parsons, A.A. Sumatriptan modifies cortical free radical release during
58. Magi, R., Lindgren, C.M. & Morris, A.P. Meta-analysis of sex-specific genome-wide cortical spreading depression: a novel antimigraine action for sumatriptan? Brain
association studies. Genet. Epidemiol. 34, 846–853 (2010). Res. 870, 44–53 (2000).
1Psychiatric and Neurodevelopmental Genetics Unit, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA. 2Medical and
Population Genetics Program, Broad Institute of MIT and Harvard, Cambridge, Massachusetts, USA. 3Stanley Center for Psychiatric Research, Broad Institute of MIT
and Harvard, Cambridge, Massachusetts, USA. 4Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, UK. 5Analytic and Translational Genetics
Unit, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA. 6FORMI, Oslo University Hospital, Oslo, Norway. 7Department of
Neurology, Oslo University Hospital, Oslo, Norway. 8Institute of Clinical Medicine, University of Oslo, Oslo, Norway. 9Institute for Molecular Medicine Finland (FIMM),
University of Helsinki, Helsinki, Finland. 10Estonian Genome Center, University of Tartu, Tartu, Estonia. 11Division of Endocrinology, Boston Children’s Hospital,
Boston, Massachusetts, USA. 12Department of Epidemiology Research, Statens Serum Institut, Copenhagen, Denmark. 13Novo Nordisk Foundation Center for
Basic Metabolic Research, University of Copenhagen, Copenhagen, Denmark. 14Illumina, San Diego, California, USA. 15Pediatric Neurology, Vall d’Hebron
Research Institute, Barcelona, Spain. 16Folkhälsan Institute of Genetics, Helsinki, Finland. 17Neuroscience Center, University of Helsinki, Helsinki, Finland.
18Molecular Neurology Research Program, Research Programs Unit, University of Helsinki, Helsinki, Finland. 1923andMe, Inc., Mountain View, California, USA.
20Institute of Public Health, Charité–Universitätsmedizin Berlin, Berlin, Germany. 21Division of Preventive Medicine, Brigham and Women’s Hospital, Boston,
Massachusetts, USA. 22deCODE Genetics, Reykjavik, Iceland. 23Medical Research Council (MRC) Integrative Epidemiology Unit, University of Bristol, Bristol, UK.
24Department of Biological Psychology, Vrije Universiteit, Amsterdam, the Netherlands. 25Department of Neurology, Leiden University Medical Centre, Leiden,
the Netherlands. 26Department of Neurology, Helsinki University Central Hospital, Helsinki, Finland. 27Department of Neurology and Epileptology, Hertie-Institute
for Clinical Brain Research, University of Tuebingen, Tuebingen, Germany. 28Institute for Stroke and Dementia Research, Klinikum der Universität München,
Ludwig-Maximilians-Universität München, Munich, Germany. 29Department of Neuroscience, Karolinska Institutet, Stockholm, Sweden. 30Department of Genetics
and Computational Biology, QIMR Berghofer Medical Research Institute, Brisbane, Queensland, Australia. 31Institute of Human Genetics, Ulm University, Ulm,
Germany. 32Center for Life Course Epidemiology and Systems Medicine, University of Oulu, Oulu, Finland. 33Department of Twin Research and Genetic Epidemiology,
King’s College London, London, UK. 34Department of Epidemiology, Erasmus University Medical Center, Rotterdam, the Netherlands. 35Department of Radiology,
Erasmus University Medical Center, Rotterdam, the Netherlands. 36Department of Clinical Chemistry, Fimlab Laboratories, School of Medicine, University of
Tampere, Tampere, Finland. 37Department of Public Health, University of Helsinki, Helsinki, Finland. 38Harvard Medical School, Boston, Massachusetts, USA.
39Department of Neurology, University Duisburg–Essen, Essen, Germany. 40Landspitali University Hospital, Reykjavik, Iceland. 41Department of Psychiatry, VU
University Medical Centre, Amsterdam, the Netherlands. 42Department of Psychiatry, Washington University School of Medicine, St. Louis, Missouri, USA.
43Department of Neurosurgery, NeuroCenter, Kuopio University Hospital, Kuopio, Finland. 44Department of Genetics, University Medical Center Groningen, University
of Groningen, Groningen, the Netherlands. 45MRC Functional Genomics Unit, Department of Physiology, Anatomy & Genetics, Oxford University, Oxford, UK.
46Nuffield Department of Clinical Neuroscience, University of Oxford, Oxford, UK. 47Oxford Headache Centre, John Radcliffe Hospital, Oxford, UK. 48Max Planck
Institute of Psychiatry, Munich, Germany. 49Institute of Clinical Molecular Biology, Christian Albrechts University, Kiel, Germany. 50Institute of Human Genetics,
Helmholtz Zentrum München, Neuherberg, Germany. 51Institute of Human Genetics, Technische Universität München, Munich, Germany. 52Department of General
Practice and Primary Health Care, University of Helsinki and Helsinki University Hospital, Helsinki, Finland. 53National Institute for Health and Welfare, Helsinki,
Finland. 54Institute of Clinical Medicine, University of Helsinki, Helsinki, Finland. 55Department of Environmental Health, Harvard T.H. Chan School of Public Health,
Boston, Massachusetts, USA. 56Department of Internal Medicine, Erasmus University Medical Center, Rotterdam, the Netherlands. 57Department of Pain Management
and Research, Oslo University Hospital, Oslo, Norway. 58Medical Faculty, University of Oslo, Oslo, Norway. 59Department of Ageing and Health, Norwegian Institute of
Public Health, Oslo, Norway. 60Kiel Pain and Headache Center, Kiel, Germany. 61Danish Headache Center, Department of Neurology, Rigshospitalet, Glostrup
Hospital, University of Copenhagen, Copenhagen, Denmark. 62Institute of Biological Psychiatry, Mental Health Center Sct. Hans, University of Copenhagen, Roskilde,
Denmark. 63Institute of Biological Psychiatry, MHC Sct. Hans, Mental Health Services Copenhagen, Copenhagen, Denmark. 64Institute of Clinical Sciences, Faculty of
Medicine and Health Sciences, University of Copenhagen, Copenhagen, Denmark. 65iPSYCH—The Lundbeck Foundation Initiative for Integrative Psychiatric
Research, Copenhagen, Denmark. 66A full list of members and affiliations appears at the end of the paper and in the Supplementary Note. 67Department of Health,
National Institute for Health and Welfare, Helsinki, Finland. 68Research Centre of Applied and Preventive Cardiovascular Medicine, University of Turku, Turku,
10 aDVANCE ONLINE PUBLICATION Nature Genetics

Articles
Finland. 69Department of Clinical Physiology and Nuclear Medicine, Turku University Hospital, Turku, Finland. 70Department of Neurology, Erasmus University
Medical Center, Rotterdam, the Netherlands. 71Department of Epidemiology and Biostatistics, MRC Health Protection Agency (HPE) Centre for Environment and
Health, School of Public Health, Imperial College London, London, UK. 72Biocenter Oulu, University of Oulu, Oulu, Finland. 73Unit of Primary Care, Oulu University
Hospital, Oulu, Finland. 74Institute of Human Genetics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany. 75Population Health Research Institute,
St George’s, University of London, London, UK. 76Munich Cluster for Systems Neurology (SyNergy), Munich, Germany. 77Department of Human Genetics, Leiden
University Medical Centre, Leiden, the Netherlands. 78Faculty of Medicine, University of Iceland, Reykjavik, Iceland. 79Statistical and Genomic Epidemiology
Laboratory, Institute of Health and Biomedical Innovation, Queensland University of Technology, Kelvin Grove, Queensland, Australia. 80Department of Neurology,
Massachusetts General Hospital, Boston, Massachusetts, USA. 81These authors contributed equally to this work. 82These authors jointly supervised this work.
Correspondence should be addressed to A.P. ([email protected]).
The International Headache Genetics Consortium

Verneri Anttila2,3, Ville Artto26, Andrea Carmine Belin29, Dorret I Boomsma24, Sigrid Børte83, Daniel I Chasman38,
Lynn Cherkas33, Anne Francke Christensen61, Bru Cormand84, Ester Cuenca-Leon2,3, George Davey Smith23,
Martin Dichgans28, Cornelia van Duijn34, Else Eising77, Tonu Esko10, Ann-Louise Esserlind61, Michel Ferrari25,
Rune R Frants77, Tobias M Freilinger27, Nicholas A Furlotte19, Padhraig Gormley2,3, Lyn Griffiths85,
Eija Hamalainen9, Thomas Folkmann Hansen61, Marjo Hiekkala16, M Arfan Ikram34,35,70, Andres Ingason22,
Marjo-Riitta Järvelin32,72,73, Risto Kajanne9, Mikko Kallela26, Jaakko Kaprio9, Mari Kaunisto16,
Christian Kubisch74, Mitja Kurki2,3, Tobias Kurth38, Lenore Launer86, Terho Lehtimaki36, Davor Lessel74,
Lannie Ligthart24, Nadia Litterman19, Arn M J M van den Maagdenberg25,77, Alfons Macaya15, Rainer Malik28,
Massimo Mangino33, George McMahon23, Bertram Muller-Myhsok48, Benjamin M Neale2,3, Carrie Northover19,
Dale R Nyholt85, Jes Olesen61, Aarno Palotie2,3, Priit Palta9, Linda M Pedersen6, Nancy Pedersen87,
Danielle Posthuma88, Patricia Pozo-Rosich89, Alice Pressman90, Lydia Quaye33, Olli Raitakari91,
Markus Schürks38, Celia Sintas84, Kari Stefansson22, Hreinn Stefansson22, Stacy Steinberg22, David Strachan75,
Gisela M Terwindt25, Marta Vila-Pueyo15, Maija Wessman16, Bendik S Winsvold6–8, William Wrenthal92,
Huiying Zhao85 & John-Anker Zwart6–8
83Department of Neurology, Oslo University Hospital and University of Oslo, Oslo, Norway. 84Department of Genetics, University of Barcelona, Barcelona, Spain.
85Institute of Health and Biomedical Innovation, Queensland University of Technology, Brisbane, Queensland, Australia. 86National Institute on Aging, Bethesda,
Maryland, USA. 87Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden. 88Department of Clinical Genetics, Vrije
Universiteit, Amsterdam, the Netherlands. 89Department of Medicine, Universitat Autònoma de Barcelona, Barcelona, Spain. 90Sutter Health, Sacramento,
California, USA. 91Department of Medicine, University of Turku, Turku, Finland. 92Broad Institute of MIT and Harvard, Cambridge, USA.
Nature Genetics ADVANCE ONLINE PUBLICATION 11

ONLINE METHODS For this analysis, we extracted high-quality common SNPs from the summary
Study design and phenotyping. A description of the study design, statistics by filtering the data according to the following criteria: presence
ascertainment, and phenotyping for each GWA study is provided in the among the HapMap Project Phase 3 SNPs83, allele matching to 1000 Genomes
Supplementary Note. Project data, no strand ambiguity, INFO score > 0.9, MAF ≥ 1%, and missing-
ness less than two-thirds of the 90th percentile of the total sample size. The
Quality control. The 22 individual GWA studies were subjected to pre- HLA region (chromosome 6, 25–35 Mb) was excluded from the analysis. With
established quality control (QC) protocols as recommended elsewhere76,77. these data, we used LDSC to quantify the proportion of the total inflation in
Differences in genotyping chips, DNA quality, and calling pipelines neces- chi-square statistics that could be ascribed to polygenic heritability by calcu-
sitated that the QC parameters be tuned separately for each study. At a mini- lating the ratio of the LDSC intercept estimate and the chi-square mean using
mum, we excluded markers that had high ‘missingness’ rates (>5%), that had the formula described in the original publication56.
low minor allele frequency (<1%), and that failed a test of Hardy–Weinberg
equilibrium. We also excluded individuals with a high proportion of missing Heterogeneity analysis of migraine subtypes. To determine whether
genotypes (>5%) and used identity-by-descent estimates to remove related heterogeneity between the migraine subtypes might have affected our ability to
individuals (identity by descent > 0.185). A summary of the genotyping plat- identify new loci, we performed an additional meta-analysis using a subtype-
forms, QC, and software used in each study is provided in Supplementary differentiated approach that allows for different allelic effects between two
Table 3. To control for population stratification within each study, we merged groups58. Because a large proportion of the controls were shared between the
the genotypes passing QC filters with HapMap III data from three populations: original data sets for migraine with aura and migraine without aura (Table 1),
European (CEU), Asian (CHB+JPT), and African (YRI). We then performed for this analysis we created two additional subsets of the migraine subtype data
a principal-components analysis on the merged data set and excluded any that contained no overlapping controls (Supplementary Table 12). The new
(non-European) population outliers. To control for any sub-European popula- ‘migraine with aura’ subset consisted of 4,837 cases and 49,174 controls, and
tion structure, we performed a second principal-components analysis within the new ‘migraine without aura’ subset consisted of 4,833 cases and 106,834
each study to ensure that cases and controls were clustering together. Any controls. To assess the heterogeneity observed, we chose the 44 index SNPs
principal components that were significantly associated with the phenotype from the primary meta-analysis and applied the subtype-differentiated meta-
were included as covariates in the model when we calculated test statistics for analysis method to them. We observed that only 7 out of the 44 SNPs exhibited
the meta-analysis (Supplementary Table 4). heterogeneity in the subtype-differentiated test (heterogeneity P value < 0.05;
Supplementary Table 13), suggesting that most loci probably affect risk for
Imputation. After study-level QC, estimated haplotypes were phased for each both subtypes.
individual using (in most instances) the program SHAPEIT78. Missing geno-
types were then imputed into these haplotypes using the program IMPUTE2 Defining credible sets. Within each migraine-associated locus, we defined
(ref. 79) and a mixed-population 1000 Genomes Project16 reference panel a credible set of variants that could be considered 99% likely to contain
(March 2012, phase 1, v3 release or later). A minority of contributing studies a causal variant. The method has been described in detail elsewhere53,59
used alternative programs for phasing and imputation (BEAGLE 80, MACH81, and is outlined briefly here. Assume D represents the data including the
MINIMAC82, or in-house custom software). A summary of software and genotype matrix X for all of the P variants (the genotype for variant j is
procedures used is provided in Supplementary Table 3. denoted as xj) and disease status Y (for N individuals), and β represents the
model parameters. We define the model, denoted by A, as the causal status
Statistical analysis. Individual-study association analyses were implemented for all of the P variants in the locus: A ≡ {aj}, in which aj is the causal sta-
using logistic regression with an additive model on the imputed dosage of the tus for variant j. aj = 1 if the variant j is causal, and aj = 0 if it is not. We
effect allele. All models were adjusted for sex and other relevant covariates assume that there is one and only one genuine signal for each locus; therefore,
when appropriate (Supplementary Table 4). Because age information was not one and only one of the P variants is causal: Σjaj = 1. For convenience, we
available for all individuals from all studies, we were not able to adjust for it in define Aj as the model in which only variant j is causal, and A0 as the model
our models. However, we note that all of the GWA studies included adults past in which no variant is causal (null model). The probability of model Aj (where
the typical age of onset; thus age was at most a non-confounding factor, and variant j is the only causal variant in the locus) given the data can be calculated
false positive rates would therefore not be affected by its inclusion or exclusion. using Bayes’s rule:
For the within-study association analyses, we used SNPTEST, PLINK, or R.
The program GWAMA was then used to perform a fixed-effects meta-analysis Pr( A j )
(1)
weighted by the inverse variances to obtain a combined effect size, standard
Pr( A j | D) = ∫b Pr(D, b | A j ) ⋅ Pr(D) ⋅ db
error, and P value at each marker. We excluded markers in any study that had
low imputation quality scores (IMPUTE2 INFO < 0.6 or MACH r2 < 0.6) or We estimate equation (1) using the steepest-descent approach84. Making
low minor allele frequency (MAF < 0.01). Additionally, we filtered out mark- the assumption of a flat prior on the model parameters, we approximate
ers that were missing from more than half of all studies (12 or more) or that the integral over the model parameters using their maximum likelihood
exhibited high heterogeneity (heterogeneity index I2 > 0.75). After filtering, bˆ (bˆ j ) :
estimator
8,045,569 total markers were tested in the meta-analysis.
−|b |/ 2 Pr( A j )
bˆ Pr( A j | D) ≈ Pr(D | A j , bˆ j ) ⋅ N j ⋅ (2)
Chromosome X meta-analysis. Because of the different ploidy of males Pr(D)
and females on chromosome X, we implemented a model of X-chromosome
inactivation that assumes an equal effect of alleles in both males and females. where the sample size is denoted by N and the number of fitted parameters for
We achieved this by scaling male dosages to the range of 0–2 to match that model Aj is denoted by |βj|. |βj| is a constant because model Aj has the same
of females. In total, 57,756 cases and 299,109 controls were available for the number of parameters across all variants. In the framework of a generalized
X-chromosome analysis (Supplementary Table 1). The sample size was linear model, the deviance for two nested models follows an approximate
smaller than that for the autosomal data because some of the individual GWA chi-square distribution. We therefore define c 2j as the deviance comparing
studies (EGCUT, Rotterdam III, Twins UK, and 846 controls from GSK for the null model and the model in which variant j is causal:
the German MO study) did not contribute chromosome X data.
Pr(D | A0 , bˆ0 )
LD-score regression analysis. We conducted a univariate heritability analy- bˆ c 2j ≡ −2 log (3)
Pr(D | A j , bˆ j )
sis based on summary statistics using LD-score regression (LDSC)56 v1.0.0.
Nature Genetics doi:10.1038/ng.3598

We further show that c 2j can be calculated as the chi-square statistic of postmortem samples from 42 human tissues and three cell lines across 1,641
fitting a binomial model with the disease status (Y) as the dependent variable samples (Supplementary Table 16) were used for bulk RNA sequencing
and the genotype of variant j as the explanatory variable: according to a unified protocol. All samples were sequenced using Illumina
76-base-pair paired-end reads. Collapsed reads per kilobase per million
Pr({xi }, Y | {ai = 0},{bˆi , 0 }) mapped reads (RPKM) for 52,577 transcripts were filtered for those with
bˆ c 2j = − 2 log unique HGNC IDs (n = 20,932). We also excluded transcripts from any
Pr({xi }, Y | {a j = 1, a− j = 0}, {bˆ j , bˆ − j , 0 })
noncoding RNAs. We ranked all transcripts by mean RPKM across all samples,
= − 2 log
∏i Pr(x , Y | a = 0, bˆ )
i i i, 0
(4)
and we generated 100,000 permutations of each credible set gene list by
Pr(x j , Y | a j = 1, bˆ j )∏ i ≠ j Pr(xi , Y | ai = 0, bˆi , 0 ) selecting a random transcript for each entry in the credible set within ±100
ranks of the transcript for that gene. For each sample, the RPKM values were
Pr(x j , Y | a j = 0, bˆ j , 0 ) converted into ranks for that transcript, and sums of ranks within each tissue
= − 2 log were computed for each gene. We calculated enrichment P values for each
Pr(x , Y | a = 1, bˆ )
j j j tissue by taking the total number of instances when the gene list of interest
had a lower sum of ranks than the permuted sum of ranks (divided by the
2
Pr(Aj|D) in equation (2) is then a function of c j : total number of permutations). We estimated the number of independent
tissues using the matSpD tool87 and then used Bonferroni correction to
 c 2j  adjust for 27 independent tests (P < 1.90 × 10−3).
−|b |/ 2 Pr( A j )
Pr( A j | D) ≈ exp   ⋅ l0 ⋅ N j ⋅ (5)
 2  Pr(D) Specificity of individual genes in GTEx tissues. We selected the nearest gene
to the index SNP at each migraine locus and then investigated the individual
wherebˆ l0 = Pr(D | A0 , bˆ0 ). We make the assumption that the prior causal expression activity of each of the selected genes. Because the number of samples
probability for all variants is equal, that is, Pr(Aj) is the same across all for some tissues was small, we grouped individual tissues into four categories:
variants j. Equation (5) can then be simplified with a constant for the term brain, vascular, gastrointestinal, and other tissues (Supplementary Table 16).
−|b |/ 2 For each selected gene, we then tested whether the average expression (mean
l0 ⋅ N j ⋅ (Pr( A j )/ Pr(D)), and the probability that variant j is causal can
RPKM) was significantly higher in a particular tissue group compared with the
be calculated using
‘other tissues’ category. We assessed significance using a one-tailed t-test and
 c 2j  used Bonferroni correction to adjust for 114 tests (38 genes × 3 tissue groups).
Pr( A j | D) ∝ exp   (6)
 2  Although some genes were observed to be significantly expressed in multiple
tissue groups, we determined that a gene was tissue specific if it had high
which can be normalized across all variants as expression in only one tissue group (i.e., brain, vascular, or gastrointestinal;
Supplementary Table 25).
P( A j ) ≡ Pr( A j | D)/ Σ k Pr( Ak | D) (7)
eQTL credible set analysis in GTEx tissues. For all tissues and tran-
Finally, the 99% credible set of variants is defined as the smallest set of models, scripts (filtered as described above), we identified genome-wide significant
with each model designating one causal variant, S = {Aj}, such that (P < 2 × 10−13) cis-eQTLs within a 1-Mb window of each transcript and created
credible sets (see “Defining credible sets”) for each eQTL identified in each
∑ A j ∈S P(A j ) ≥ 99% (8) tissue. We found a total of 35 of these significant eQTL credible sets within a
1-Mb window of the migraine loci; however, only 7 out of the 35 contained
This credible set of variants has 99% probability of containing the causal vari- variants that overlapped with a migraine credible set. For these seven eQTL
ant, given the assumption that there is a true association and that all possible credible sets, we then tested (using Spearman’s rank correlation) whether the
causal variants have been genotyped (both assumptions are likely to be valid test statistics between the two overlapping credible sets were significantly
in genome-wide significant regions of data that have been imputed to the 1000 correlated. Significant correlation between a migraine credible set and an
Genomes Project). We have made the R script for implementing the method eQTL credible set was taken as evidence that the migraine locus tagged a real
freely available online (see “URLs”). eQTL. Multiple testing was controlled for the use of Bonferroni correction
(i.e., for seven tests at P < 7.1 × 10−3).
eQTL credible set overlap analysis. To assess whether the association
statistics in the 38 migraine loci could be explained by credible overlapping Enhancer enrichment analysis. Markers of gene regulation were defined using
eQTL signals, we used two eQTL microarray data sets. The first consisted ChIP-seq data sets from ENCODE66 and the NIH Roadmap Epigenome65
of 3,754 samples from peripheral venous blood 85, and the second was from projects. On the basis of the histone H3K27ac signal, which identifies active
a meta-analysis of human brain cortex studies of a total of 550 samples86. enhancers, we processed data from 56 cell lines and tissue samples to identify
From both studies, we obtained summary statistics from an association test of cell-type-specific and tissue-specific enhancers, which we define as the 10%
putative cis-eQTLs between all SNP–transcript pairs within a 1-Mb window. of enhancers with the highest ratio of reads in that cell or tissue type divided
Then, for the most significant eQTLs (P < 1 × 10−4) found for genes within by the total reads88. The raw data are publicly available (see “URLs”), and a
a 1-Mb window of migraine credible set variants (see “Defining credible description of the 56 tissues and cell types is provided in Supplementary
sets”), we created an additional credible set of markers for each eQTL. We Table 21. We mapped the credible set variants at each migraine locus to these
then tested (using Spearman’s rank correlation) whether there was a signifi- enhancer sites and compared the overlap observed with tissue-specific enhanc-
cant correlation between the association test statistics in each migraine cred- ers relative to a background of 10,000 randomly selected sets of SNPs of equal
ible set compared to the expression test statistics in each overlapping eQTL size. We restricted the background selection to 1000 Genomes Project vari-
credible set. Significant correlation between a migraine credible set and an ants (MAF > 1%) that also passed QC filters in the meta-analysis (to allow
eQTL credible set was taken as evidence that the migraine locus tagged a the selection of only SNPs that had an a priori chance of being associated).
real eQTL. An appropriate significance threshold for multiple testing was The selection procedure then involved randomly selecting genomic regions
determined by Bonferroni correction. with length and density of enhancers equivalent to those found in the original
locus. Once an appropriate region was found, a set of SNPs was randomly
GTEx tissue enrichment analysis. We obtained gene sets for each locus selected to match the number of SNPs in the credible set for that locus. If the
by taking all genes within 50 kb of credible-set SNPs. We then analyzed selected SNPs mapped to an equal number of enhancer sites (of any tissue
identified genes for tissue enrichment using publicly available expression type) as credible SNPs from the original locus, then they were added to the
data from the pilot phase of the GTEx project62, version 3. In this data set, background set of SNPs for comparison. If the selected SNPs did not map to
doi:10.1038/ng.3598 Nature Genetics

the correct number of enhancers, the selection procedure was repeated until an whether genes in the 38 migraine loci were highly expressed in 209 tissues and
appropriate set was found. This procedure was repeated 10,000 times for each cell types with Medical Subject Heading (MeSH) annotations. The annotation
locus to obtain an empirical null distribution. We then estimated the enrich- procedure and method for normalizing expression profiles across annotations
ment significance empirically by calculating the proportion of replicates that is outlined in the original publication63. The tissue/cell-type enrichment analy-
were greater than the observed value. Finally, we used Bonferroni correction sis algorithm was conceptually identical to the gene set enrichment analysis
to adjust for multiple testing of 56 tissue and cell types (P < 8.9 × 10−4). whereby enrichment P values were calculated empirically using 500 permuta-
tions for bias adjustment and 20 replications for FDR estimation.
GO enrichment analysis. The set of 38 genes nearest to the index SNP in
each migraine locus was chosen and tested for over-representation in Gene Data access. All genome-wide significant and suggestive SNP associations
Ontology (GO) annotations. We used the PANTHER89 tool (see “URLs”) to (P < 1 × 10−5) from the meta-analysis can be obtained directly from the IHGC
perform the analysis, implementing a binomial test to determine whether the website (http://www.headachegenetics.org/content/datasets-and-cohorts).
number of genes from the migraine test set found in each GO pathway was For access to deeper-level data, please contact the data access committee
likely to have occurred by chance alone. The association P values were adjusted ([email protected]).
for the number of pathways tested by Bonferroni correction.
DEPICT reconstituted gene set enrichment analysis. DEPICT63 (Data-driven

76. Anderson, C.A. et al. Data quality control in genetic case-control association studies.
Expression Prioritized Integration for Complex Traits) is a computational tool Nat. Protoc. 5, 1564–1573 (2010).
that, given a set of GWA study summary statistics, allows prioritization of 77. Winkler, T.W. et al. Quality control and conduct of genome-wide association meta-
genes in associated loci, enrichment analysis of reconstituted gene sets, and analyses. Nat. Protoc. 9, 1192–1212 (2014).
78. Delaneau, O., Marchini, J. & Zagury, J.-F. A linear complexity phasing method for
tissue enrichment analysis. DEPICT was run using 124 independent genome-
thousands of genomes. Nat. Methods 9, 179–181 (2011).
wide significant SNPs as input (PLINK clumping parameters: --clump-p1 79. Howie, B., Fuchsberger, C., Stephens, M., Marchini, J. & Abecasis, G.R. Fast and
5e-8 --clump-p2 1e-5 --clump-r2 0.6 --clump-kb 250; note that rs12845494
accurate genotype imputation in genome-wide association studies through pre-

and rs140002913 could not be mapped). LD distance (r2 > 0.5) was used to phasing. Nat. Genet. 44, 955–959 (2012).
80. Browning, S.R. & Browning, B.L. Rapid and accurate haplotype phasing and
define locus boundaries (note that this locus definition is different than that
missing-data inference for whole-genome association studies by use of localized
used elsewhere in the text), yielding 37 autosomal loci comprising 78 genes. haplotype clustering. Am. J. Hum. Genet. 81, 1084–1097 (2007).
DEPICT was run using default settings, that is, 500 permutations for bias 81. Li, Y., Willer, C.J., Ding, J., Scheet, P. & Abecasis, G.R. MaCH: using sequence
adjustment, 20 replications for FDR estimation, normalized expression data and genotype data to estimate haplotypes and unobserved genotypes. Genet.
Epidemiol. 34, 816–834 (2010).
from 77,840 Affymetrix microarrays for gene set reconstitution (see ref. 90),
82. Fuchsberger, C., Abecasis, G.R. & Hinds, D.A. minimac2: faster genotype imputation.
14,461 reconstituted gene sets for gene set enrichment analysis, and testing of Bioinformatics 31, 782–784 (2015).
209 tissue/cell types assembled from 37,427 Affymetrix U133 Plus 2.0 array 83. The International HapMap 3 Consortium. Integrating common and rare genetic
samples for enrichment in tissue/cell-type expression. variation in diverse human populations. Nature 467, 52–58 (2010).
84. Schwarz, G. Estimating the dimension of a model. Ann. Stat. 6, 461–464
After the analysis, we omitted reconstituted gene sets in which genes in
(1978).
the original gene set were not nominally enriched (Wilcoxon rank-sum test) 85. Wright, F.A. et al. Heritability and genomics of gene expression in peripheral blood.
because, by design, genes in the original gene set were expected to be enriched Nat. Genet. 46, 430–437 (2014).
in the reconstituted gene set. Therefore, a lack of enrichment complicated 86. Richards, A.L. et al. Schizophrenia susceptibility alleles are enriched for alleles
that affect gene expression in adult human brain. Mol. Psychiatry 17, 193–201
interpretation because the label of the reconstituted gene set may have been
(2012).
inaccurate. Hence eight reconstituted gene sets were removed from the 87. Nyholt, D.R. A simple correction for multiple testing for single-nucleotide
results: MP:0002089, MP:0002190, ENSG00000151577, ENSG00000168615, polymorphisms in linkage disequilibrium with each other. Am. J. Hum. Genet. 74,
ENSG00000143322, ENSG00000112531, ENSG00000161021, and 765–769 (2004).
88. Farh, K.K.-H. et al. Genetic and epigenetic fine mapping of causal autoimmune
ENSG00000100320. We also removed an association identified for another
disease variants. Nature 518, 337–343 (2015).
gene set (ENSG00000056345 PPI, P = 1.7 × 10−4, FDR = 0.04) because it was no 89. Mi, H., Muruganujan, A., Casagrande, J.T. & Thomas, P.D. Large-scale gene function
longer part of the Ensembl database. Finally, we used the Affinity Propagation analysis with the PANTHER classification system. Nat. Protoc. 8, 1551–1566
tool91 to cluster related reconstituted gene sets into ten groups (see “URLs”). (2013).
90. Fehrmann, R.S.N. et al. Gene expression analysis identifies global gene dosage
sensitivity in cancer. Nat. Genet. 47, 115–125 (2015).
DEPICT tissue enrichment analysis. DEPICT used data from 37,427 human 91. Frey, B.J. & Dueck, D. Clustering by passing messages between data points. Science
microarray samples captured on the Affymetrix HGU133a2.0 platform to test 315, 972–976 (2007).
Nature Genetics doi:10.1038/ng.3598

Gormley 2016

Uploaded by

Copyright:

Available Formats

Gormley 2016

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gormley 2016

Uploaded by

Copyright:

Available Formats

Articles

Meta-analysis of 375,000 individuals identifies 38

Nature Genetics ADVANCE ONLINE PUBLICATION 

 aDVANCE ONLINE PUBLICATION Nature Genetics

Study of Depression and Anxiety

Nature Genetics ADVANCE ONLINE PUBLICATION 

 aDVANCE ONLINE PUBLICATION Nature Genetics

Figure 2 Expression enrichment of genes from Aorta

Skin cells—transformed fibroblasts

Nature Genetics ADVANCE ONLINE PUBLICATION 

trait loci (eQTLs) in either of two microarray-based studies of samples

 aDVANCE ONLINE PUBLICATION Nature Genetics

Figure 4 Enrichment of the migraine loci

to migraine (Supplementary Table 24); +

Nature Genetics ADVANCE ONLINE PUBLICATION 

a Transcription factor b Abnormal vascular

Supplementary Table 23.

 aDVANCE ONLINE PUBLICATION Nature Genetics

Nature Genetics ADVANCE ONLINE PUBLICATION 

10 aDVANCE ONLINE PUBLICATION Nature Genetics

The International Headache Genetics Consortium

Nature Genetics ADVANCE ONLINE PUBLICATION 11

Nature Genetics doi:10.1038/ng.3598

doi:10.1038/ng.3598 Nature Genetics

DEPICT reconstituted gene set enrichment analysis. DEPICT63 (Data-driven

accurate genotype imputation in genome-wide association studies through pre-

Nature Genetics doi:10.1038/ng.3598

You might also like

Nature Genetics ADVANCE ONLINE PUBLICATION

aDVANCE ONLINE PUBLICATION Nature Genetics

Nature Genetics ADVANCE ONLINE PUBLICATION

aDVANCE ONLINE PUBLICATION Nature Genetics

Nature Genetics ADVANCE ONLINE PUBLICATION

aDVANCE ONLINE PUBLICATION Nature Genetics

Nature Genetics ADVANCE ONLINE PUBLICATION

aDVANCE ONLINE PUBLICATION Nature Genetics

Nature Genetics ADVANCE ONLINE PUBLICATION