Significant associations at 38 independent genomic loci

The primary meta-analysis was performed on all migraine samples available through the IHGC, regardless of ascertainment. These case samples included both individuals diagnosed by a doctor as well as individuals with self-reported migraine via questionnaires. Study design and sample ascertainment for each individual study is outlined in theSupplementary Note (and summarized inSupplementary Table 1). The final combined sample consisted of 59,674 cases and 316,078 controls in 22 non-overlapping case-control datasets (Table 1). All samples were of European ancestry. Before including the largest study from 23andMe, we confirmed that it did not contribute any additional heterogeneity compared to the other population and clinic-based studies (Supplementary Table 2).

The 22 individual GWA studies completed standard quality control protocols (Online Methods) summarized inSupplementary Table 3. Missing genotypes were then imputed into each sample using a common 1000 Genomes Project reference panel¹⁶. Association analyses were performed within each study using logistic regression on the imputed marker dosages while adjusting for sex and other covariates where necessary (Online Methods andSupplementary Table 4). The association results were combined using an inverse-variance weighted fixed-effects meta-analysis. Markers were filtered for imputation quality and other metrics (Online Methods) leaving 8,094,889 variants for consideration in our primary analysis.

Among these variants in the primary analysis, we identified 44 genome-wide significant SNP associations (P < 5 × 10⁻⁸,Supplementary Figure 1) that are independent (r² < 0.1) with regards to linkage disequilibrium (LD). We validated the 44 SNPs by comparing genotypes in a subset of the sample to those obtained from whole-genome sequencing (Supplementary Table 5). To help identify candidate risk genes from these, we defined an associated locus as the genomic region bounded by all markers in LD (r² > 0.6 in 1000 Genomes, Phase I, EUR individuals) with each of the 44 index SNPs and in addition, all such regions in close proximity (< 250 kb) were merged. From these defined regions we implicate 38 genomic loci for the prevalent forms of migraine, 28 of which have not previously been reported (Figure 1).

These 38 loci replicate 10 of the 13 previously reported genome-wide associations to migraine and six loci contain a secondary genome-wide significant SNP not in LD (r² < 0.1) with the top SNP in the locus (Table 2). Five of these secondary signals were found in known loci (atLRP1/STAT6/SDR9C7,PRDM16,FHL5/UFL1,TRPM8/HJURP, and nearTSPAN2/NGF), while the sixth was found within one of the 28 new loci (PLCE1). Therefore, out of the 44 independent SNPs reported here, 34 represent new associations to migraine. Three previously reported loci that were associated to subtypes of migraine (rs1835740 nearMTDH for migraine with aura, rs10915437 nearAJAP1 for migraine clinical-samples, and rs10504861 nearMMP16 for migraine without aura)^8,11 show only nominal significance in the current meta-analysis (P = 5 × 10⁻³ for rs1835740,P = 4.4 × 10⁻⁵ for rs10915437, andP = 4.9 × 10⁻⁵ for rs10504861,Supplementary Table 6), however, these loci have since been shown to be associated to specific phenotypic features of migraine¹⁷ and therefore may require a more phenotypically homogeneous sample to be accurately assessed for association. Four out of 44 SNPs (atTRPM8/HJURP, nearZCCHC14,MRVI1, and nearCCM2L/HCK) exhibited moderate heterogeneity across the individual GWA studies (Cochran’s QP-value < 0.05,Supplementary Table 7) therefore at these markers we applied a random effects model¹⁸.

Characterization of the associated loci

In total, 32 of 38 (84%) loci overlap with transcripts from protein-coding genes, and 17 (45%) of these regions contain just a single gene (seeSupplementary Figure 2 for regional association plots andSupplementary Table 8 for additional locus information). Among the 38 loci, only two contain ion channel genes (KCNK5¹⁹ andTRPM8²⁰). Hence, despite previous hypotheses of migraine as a potential channelopathy^5,21, the loci identified to date do not support common variants in ion channel genes as strong susceptibility components in prevalent forms of migraine. However, three other loci do contain genes involved more generally in ion homeostasis (SLC24A3²²,ITPK1²³, andGJA1²⁴,Supplementary Table 9).

Several of the genes have previous associations to vascular disease (PHACTR1,^25,26TGFBR2,²⁷LRP1,²⁸PRDM16,²⁹RNF213,³⁰JAG1,³¹HEY2,³²GJA1³³,ARMS2³⁴), or are involved in smooth muscle contractility and regulation of vascular tone (MRVI1,³⁵GJA1,³⁶SLC24A3,³⁷NRP1³⁸). Three of the 44 migraine index SNPs have previously reported associations in the National Human Genome Research Institute (NHGRI) GWAS catalog at exactly the same SNP (rs9349379 atPHACTR1 with coronary heart disease^39–41, coronary artery calcification⁴², and cervical artery dissection²⁶; rs11624776 nearITPK1 with thyroid hormone levels⁴³; and rs11172113 atLRP1/STAT6/SDR9C7 with pulmonary function⁴⁴;Supplementary Table 10). Six of the loci harbor genes that are involved in nitric oxide signaling and oxidative stress (REST⁴⁵,GJA1⁴⁶,YAP1⁴⁷,PRDM16⁴⁸,LRP1⁴⁹, andMRVI1⁵⁰).

From each locus we chose the nearest gene to the index SNP to assess gene expression activity in tissues from the GTEx consortium (Supplementary Figure 3). While we found that most of the putative migraine loci genes were expressed in many different tissue types, we could detect tissue specificity in certain instances whereby some genes showed significantly higher expression in a particular tissue group relative to the others. For instance four genes were more actively expressed in brain (GPR149,CFDP1,DOCK4, andMPPED2) compared to other tissues, whereas eight genes were specifically active in vascular tissues (PRDM16,MEF2D,FHL5,C7orf10,YAP1,LRP1,ZCCHC14, andJAG1). Many other putative migraine loci genes were actively expressed in more than one tissue group.

Genomic inflation and LD-score regression analysis

To assess whether the 38 loci harbor true associations with migraine rather than reflecting systematic differences between cases and controls (such as population stratification) we analyzed the genome-wide inflation of test statistics in our primary meta-analysis. As expected for a complex polygenic trait, the distribution of test statistics deviates from the null (genomic inflation factor λ_GC = 1.24,Supplementary Figure 4) which is in line with other large GWA study meta-analyses^51–54. Since much of the inflation in a polygenic trait arises from LD between the causal SNPs and many other neighboring SNPs in the local region, we LD-pruned the data to create a set of LD-independent markers (i.e. in PLINK⁵⁵ with a 250-kb sliding window andr² > 0.2). The resulting genomic inflation was reduced (λ_GC = 1.15,Supplementary Figure 5) and likely reflects the inflation remaining due to the polygenic signal at many independent loci, including those not yet significantly associated.

To confirm that the observed inflation is primarily coming from true polygenic signal, we analyzed the data from all imputed markers using LD-score regression⁵⁶. This method tests for a linear relationship between marker test statistics and LD score, defined as the sum ofr² values between a marker and all other markers within a 1-Mb window. The primary analysis results show a linear relationship between association test statistics and LD-score (Supplementary Figure 6) and estimate that the majority (88.2%) of the inflation in test statistics can be ascribed to true polygenic signal rather than population stratification or other confounders. These results are consistent with the theory of polygenic disease architecture shown previously by both simulation and real data for GWAS samples of similar size⁵⁷.

Migraine subtype analyses

To elucidate pathophysiological mechanisms underpinning the migraine aura, we performed a secondary analysis by creating two subsets that included only samples with the subtypes; migraine with aura and migraine without aura. These subsets only included those studies where sufficient information was available to assign a diagnosis of either subtype according to classification criteria standardized by the International Headache Society (IHS)⁶. For the population-based studies this involved questionnaires, whereas for the clinic-based studies the diagnosis was assigned on the basis of a structured interview by telephone or in person. A stricter diagnosis is required for the subtypes as migraine aura is often challenging to distinguish from other neurological features that can present as symptoms from unrelated conditions.

As a result, the migraine subtype analyses consisted of considerably smaller sample sizes compared to the main analysis (6,332 casesvs. 144,883 controls for migraine with aura and 8,348 casesvs. 139,622 controls for migraine without aura,Table 1). As with the primary analysis, the test statistics for migraine with aura or migraine without aura were consistent with underlying polygenic architecture rather than other potential sources of inflation (Supplementary Figure 7–Supplementary Figure 8). For the migraine without aura subset analysis we found seven significantly associated genomic loci (nearTSPAN2,TRPM8,PHACTR1,FHL5,ASTN2, nearFGF6, andLRP1,Supplementary Table 11 and Supplementary Figure 9). All seven of these loci were already identified in the primary analysis, possibly reflecting the fact that migraine without aura is the most common form of migraine (around 2 in 3 cases) and likely drives these association signals in the primary analysis. Notably, no loci were associated to migraine with aura in the other subset analysis (Supplementary Figure 10).

To investigate whether excess heterogeneity could be contributing to the lack of associations in migraine with aura, we performed a heterogeneity analysis between the two subgroups (Online Methods andSupplementary Table 12). We selected the 44 LD-independent SNPs associated from the primary analysis and used a random-effects model to combine the migraine with aura and migraine without aura samples in a meta-analysis that allows for heterogeneity between the two migraine groups⁵⁸. We found little heterogeneity with only seven of the 44 loci (atMEF2D, PHACTR1, nearREST/SPINK2, ASTN2,PLCE1,MPPED2, and nearMED14/USP9X) exhibiting signs of heterogeneity across subtype groups (Supplementary Table 13).

Credible sets of markers within each locus

For each of the 38 migraine-associated loci, we defined a credible set of markers that could plausibly be considered as causal using a Bayesian-likelihood based approach⁵⁹. This method incorporates evidence from association test statistics and the LD structure between SNPs in a locus (Online Methods). A list of the credible set SNPs obtained for each locus is provided inSupplementary Table 14. We found three instances (inRNF213,PLCE1, andMRVI1) where the association signal could be credibly attributed to exonic missense polymorphisms (Supplementary Table 15). However, most of the credible markers at each locus were either intronic or intergenic, which is consistent with the theory that most variants detected by GWA studies involve regulatory effects on gene expression rather than disrupting protein structure^60,61.

Overlap with eQTLs in specific tissues

To identify migraine loci that might influence gene expression, we used previously published datasets that catalog expression quantitative trait loci (eQTLs) in either of two microarray-based studies from peripheral venous blood (N₁ = 3,754) or from human brain cortex (N₂ = 550). Additionally, we used a third study based on RNAseq data from a collection of 42 tissues and three cell lines (N₃ = 1,641) from the Genotype-Tissue Expression (GTEx) consortium⁶². While this data has the advantage of a diverse tissue catalog, the number of samples per tissue is relatively small (Supplementary Table 16) compared to the two microarray datasets, possibly resulting in reduced power to detect significant eQTLs in some tissues. Using these datasets we applied a method based on the overlap of migraine and eQTL credible sets to identify eQTLs that could explain associations at the 38 migraine loci (Online Methods). This approach merged the migraine credible sets defined above with credible sets fromcis-eQTL signals within a 1-Mb window and tested if the association signals between the migraine and eQTL credible sets were correlated. After adjusting for multiple testing we found no plausible eQTL associations in the peripheral blood or brain cortex data (Supplementary Tables 17–18 and Supplementary Figure 11). In GTEx, however, we found evidence for overlap from eQTLs in three tissues (Lung, Tibial Artery, and Aorta) at theHPSE2 locus and in one tissue (Thyroid) at theHEY2/NCOA7 locus (Supplementary Table 19 andSupplementary Figure 12).

In summary, from three datasets we implicate eQTL signals at only two loci (HPSE2 and HEY2). This low number (two out of 38) is consistent with previous studies which have observed that available eQTL catalogues currently lack sufficient tissue specificity and developmental diversity to provide enough power to provide meaningful biological insight⁵³. No plausibly causal eQTLs were observed in expression data from brain.

Gene expression enrichment in specific tissues

To understand if the 38 migraine loci as a group are enriched for expression in certain tissue types, we again used the GTEx pilot data⁶² (seeOnline Methods). We found four tissues that were significantly enriched (after Bonferroni correction) for expression of the migraine genes (Figure 2). The two most strongly enriched tissues were part of the cardiovascular system; theaorta andtibial artery. Two other significant tissues were from the digestive system;esophagus muscularis andesophageal mucosa. We replicated these enrichment results using the DEPICT⁶³ tool and an independent microarray-based gene expression dataset (Online Methods). DEPICT highlighted four tissues (Figure 3 andSupplementary Table 20) with significant enrichment of genes within the migraine loci; arteries (P = 1.58 × 10⁻⁵), the upper gastrointestinal tract (P = 2.97 × 10⁻³), myometrium (P = 3.03 × 10⁻³), and stomach (P = 3.38 × 10⁻³).

Taken together, the expression analyses implicate arterial and gastrointestinal (GI) tissues. To discover if this enrichment signature could be attributed to a more specific type of smooth muscle, we examined the expression of the nearest genes at migraine loci in a panel of 60 types of human smooth muscle tissue⁶⁴. Overall, migraine loci genes were not significantly enriched in a particular class of smooth muscle (Supplementary Figures 13–15). This suggests that the enrichment of migraine risk variants in genes expressed in tissues with a smooth muscle component is not specific to blood vessels, the stomach or GI tract, but rather appears to be generalizable across vascular and visceral smooth muscle types.

Combined, these enrichment results suggest that some of the genes affected by migraine-associated variants are highly expressed in vascular tissues and their dysfunction could play a role in migraine. Furthermore, the results suggest that other tissue types (e.g. smooth muscle) could also play a role and this may become evident once more migraine loci are discovered.

Enrichment in tissue-specific enhancers

To further assess the hypothesis that migraine variants might operate via effects on gene-regulation, we investigated the degree of overlap with histone modifications. Using candidate causal variants from the migraine loci, we examined their enrichment within cell-type specific enhancers from 56 primary human tissues and cell types from the Roadmap Epigenomics⁶⁵ and ENCODE projects⁶⁶ (Online Methods andSupplementary Table 21). These variants showed highest enrichment in tissues from the mid-frontal lobe and duodenum smooth muscle but were not significant after adjusting for multiple testing (Figure 4).

Gene set enrichment analyses

To implicate underlying biological pathways involved in migraine, we applied a Gene Ontology (GO) over-representation analysis of the 38 migraine loci (Online Methods). We found nine vascular-related biological function categories that were significantly enriched after correction for multiple testing (Supplementary Table 22). Interestingly, we found little statistical support from the identified loci for some molecular processes that have been previously linked to migraine, e.g. ion homeostasis, glutamate signaling, serotonin signaling, nitric oxide signaling, and oxidative stress. However, it is possible that the lack of enrichment for these functions may be explained by recognizing that current annotations for many genes and pathways are far from comprehensive, or that larger numbers of migraine loci need to be identified before we have sensitivity to detect enrichment in these mechanisms.

For a more comprehensive pathway analysis we used DEPICT, which incorporates gene co-expression information from microarray data to implicate additional, functionally less well-characterized genes in known biological pathways, protein-protein complexes and mouse phenotypes⁶³ (by forming so-called ‘reconstituted gene sets’). From DEPICT we identified 67 reconstituted gene sets that are significantly enriched (FDR < 5%) for genes found among the migraine associated loci (Supplementary Table 23). Because the reconstituted gene sets had genes in common, we clustered them into 10 distinct groups (Figure 5 andOnline Methods). Several gene sets, including the most significantly enriched reconstituted gene set (Abnormal Vascular Wound Healing;P = 1.86 × 10⁻⁶), were grouped into clusters related to cell-cell interactions (ITGB1 PPI, Adherens Junction, and Integrin Complex). Several of the other gene set clusters were also related to vascular-biology (Figure 5 andSupplementary Table 23). We still did not observe any support for molecular processes with hypothesized links to migraine (Supplementary Table 24), however, this could again be due to the reasons outlined above.

Study design and phenotyping

A description of the study design, ascertainment and phenotyping for each GWA study is provided in theSupplementary Note.

Quality Control

The 22 individual GWA studies were subjected to pre-established quality control (QC) protocols as recommended elsewhere^76,77. Differences in genotyping chips, DNA quality and calling pipelines necessitated that QC parameters were tuned separately for each study. At a minimum, we excluded markers with high missingness rates (>5%), low minor allele frequency (<1%), and failing a test of Hardy-Weinberg equilibrium. We also excluded individuals with a high proportion of missing genotypes (>5%) and used identity-by-descent (IBD) estimates to remove related individuals (IBD > 0.185). A summary of the genotyping platforms, QC, and software used in each study is provided inSupplementary Table 3. To control for population stratification within each study, we merged the genotypes passing QC filters with HapMap III data from three populations; European (CEU), Asian (CHB+JPT) and African (YRI). We then performed a principal components analysis on the merged dataset and excluded any (non-European) population outliers. To control for any sub-European population structure, we performed a second principal components analysis within each study to ensure that cases and controls were clustering together. Any principal components that were significantly associated with the phenotype were included as covariates in the model when calculating test statistics for the meta-analysis (Supplementary Table 4).

Imputation

Following study-level QC, estimated haplotypes were phased for each individual using (in most instances) the program SHAPEIT⁷⁸. Missing genotypes were then imputed into these haplotypes using the program IMPUTE2⁷⁹ and a mixed-population 1000 Genomes Project¹⁶ reference panel (March 2012, phase I, v3 release or later). A minority of contributing studies used alternative programs for phasing and imputation; BEAGLE⁸⁰, MACH⁸¹, MINIMAC⁸², or in-house custom software. A summary of software and procedures used is provided inSupplementary Table 3.

Statistical analysis

Individual study association analyses were implemented using logistic regression with an additive model on the imputed dosage of the effect allele. All models were adjusted for sex and other relevant covariates when appropriate (Supplementary Table 4). As age information was not available for individuals from all studies we were not able to adjust for it in our models. However, we note that all of the GWA studies were comprised of adults past the typical age of onset, hence, age is at most a non-confounding factor and false positive rates would not be affected by its inclusion/exclusion. The programs used for performing the within-study association analyses were either SNPTEST, PLINK or R (URLs). The program GWAMA was then used to perform a fixed-effects meta-analysis weighted by the inverse variances to obtain a combined effect size, standard error andP-value at each marker. We excluded markers in any study that had low imputation quality scores (IMPUTE2INFO < 0.6 or MACHr² < 0.6) or low minor allele frequency (MAF < 0.01). Additionally, we filtered out markers that were missing from more than half of all studies (12 or more) or exhibited high heterogeneity (heterogeneity indexi² > 0.75). After filtering, 8,045,569 total markers were tested in the meta-analysis.

Chromosome X meta-analysis

Due to the different ploidy of males and females on chromosome X, we implemented a model of X-chromosome inactivation (XCI) that assumes an equal effect of alleles in both males and females. This was achieved by scaling male dosages to the range 0–2 to match that of females. In total, 57,756 cases and 299,109 controls were available for the X-chromosome analysis (Supplementary Table 1). The reduced sample size compared to the autosomal data occurred because some of the individual GWA studies (EGCUT, Rotterdam III, Twins UK, and 846 controls from GSK for the ‘German MO’ study) did not contribute chromosome X data.

LD score regression analysis

We conducted a univariate heritability analysis based on summary statistics using LD score regression (LDSC) v1.0.0⁵⁶. For this analysis, high-quality common SNPs were extracted from the summary statistics by filtering the data on the following criteria: presence among the HapMap Project Phase 3 SNPs⁸³, allele matching to 1000 Genomes data, no strand ambiguity, INFO score > 0.9, MAF >= 1%, and missingness less than two thirds of the 90^th percentile of the total sample size. The HLA region (chromosome 6, 25–35 Mb) was excluded from the analysis. From this data, we used LDSC to quantify the proportion of the total inflation in chi-square statistics that can be ascribed to polygenic heritability by calculating the ratio of the LDSC intercept estimate and the chi-square mean using the formula described in the original publication⁵⁶.

Heterogeneity analysis of migraine subtypes

To determine if heterogeneity between the migraine subtypes might have affected our ability to identify new loci, we performed an additional meta-analysis using a subtype-differentiated approach that allows for different allelic effects between two groups⁵⁸. Since a large proportion of the controls were shared in the original migraine with aura and migraine without aura datasets (Table 1), for this analysis we created two additional subsets of the migraine subtype data that contained no overlapping controls between the two new subsets (Supplementary Table 12). The new migraine with aura subset consisted of 4,837 cases and 49,174 controls and the new migraine without aura subset consisted of 4,833 cases and 106,834 controls. To assess the heterogeneity observed, we chose the 44 index SNPs from the primary meta-analysis and applied the subtype-differentiated meta-analysis method to these. We observed that only seven out of the 44 SNPs exhibited heterogeneity in the subtype-differentiated test (HeterogeneityP-value < 0.05,Supplementary Table 13) suggesting that most loci likely affect risk for both subtypes.

Defining credible sets

Within each migraine-associated locus, we defined a credible set of variants that could be considered 99% likely to contain a causal variant. The method has been introduced in detail elsewhere^53,59 and is outlined again briefly here. AssumeD is the data including the genotype matrixX for all of theP variants (genotype for variantj is denoted asx_j) and disease statusY (forN individuals), and β is the model parameters. We define the ‘model’, denotedA, as the causal status for all of theP variants in the locus:A = {a_j}, in whicha_j is the causal status for variantj.a_j = 1 if the variantj is causal, whereasa_j = 0 if it is not. We assume that there is one and only one genuine signal for each locus, therefore, one and only one of theP variants is causal: Σ_ja_j = 1. For convenience, we defineA_j as the model that only variantj is causal, andA₀ as the model that no variant is causal (null model). The probability for modelA_j(where variantj is the only causal variant in this locus) given the data can be calculated using Bayes’ rule:

We estimateEquation (1) using the steepest descent approach⁸⁴. Making the assumption of a flat prior on the model parameters, we approximate the integral over the model parameters using their maximum likelihood estimator (β̂_j):

where the sample size is denoted byN and the number of fitted parameters for modelA_j is denoted by |β_j|· |β_j| is a constant because modelA_j has the same number of parameters across all variants. In the framework of a generalized linear model, the deviance for two nested models follows an approximate chi-square distribution. We therefore defineχ_j² as the deviance comparing the null model and the model in which variantj is causal

We further show thatχ_j² can be calculated as the chi-square statistic of fitting a binomial model with the disease status (Y) as the dependent variable and the genotype of variantj as the explanatory variable:

Pr(A_j|D) inEquation (2) is then a function of theχ_j²:

wherel₀ = Pr (D|A₀, β̂₀). We make the assumption that the prior causal probability for all variants is equal,i.e., Pr(A_j) is the same across all variantsj.Equation (5) can then be simplified with a constant for the term$l_0·N^{-\frac{|{\beta }_j|}{2}}·\frac{\text{Pr}(A_j)}{\text{Pr}(D)}$ and the probability that variantj is causal can be calculated using

which can be normalized across all variants as

Finally, the 99% credible set of variants is defined as the smallest set of models, with each model designating one causal variant, S = {A_j}, such that

This credible set of variants has 99% probability of containing the causal variant, given the assumption that there is a true association and that all possible causal variants have been genotyped (both assumptions are likely to be valid in genome-wide significant regions of data that have been imputed to 1000 Genomes). We have made the R-script for implementing the method freely available online (URLs).

eQTL credible set overlap analysis

To assess if the association statistics in the 38 migraine loci could be explained by credible overlapping eQTL signals, we used two eQTL microarray datasets. The first consisted of 3,754 samples from peripheral venous blood⁸⁵ and the second was from a meta-analysis of human brain cortex studies of 550 samples⁸⁶. From both studies we obtained summary statistics from an association test of putative cis-eQTLs between all SNP-transcript pairs within a 1-Mb window of each other. Then for the most significant eQTLs (P < 1 × 10⁻⁴) found for genes within a 1Mb window of migraine credible set variants (seeDefining credible sets), we created an additional credible set of markers for each eQTL. We then tested (using Spearman’s rank correlation) whether there was a significant correlation between the association test-statistics in each migraine credible set compared to the expression test-statistics in each overlapping eQTL credible set. Significant correlation between a migraine credible set and an eQTL credible set was taken as evidence of the migraine locus tagging a real eQTL. An appropriate significance threshold for multiple testing was determined by Bonferroni correction.

GTEx tissue enrichment analysis

Gene sets for each locus were obtained by taking all genes within 50kb of credible set SNPs. Identified genes were then analyzed for tissue enrichment using publicly available expression data from the pilot phase of the Genotype-Tissue Expression project (GTEx)⁶², version 3. In this dataset,postmortem samples from 42 human tissues and three cell lines across 1,641 samples (Supplementary Table 16) were used for bulk RNA sequencing according to a unified protocol. All samples were sequenced using Illumina 76 base-pair paired-end reads. Collapsed reads per kilobase per million mapped reads (RPKM) values for 52,577 transcripts were filtered for those with unique HGNC IDs (n = 20,932). We also excluded transcripts from any non-coding RNAs. All transcripts were ranked by mean RPKM across all samples and 100,000 permutations of each credible set gene list were generated by 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 converted into ranks for that transcript, and sums of ranks within each tissue were computed for each gene. EnrichmentP-values for each tissue were calculated by taking the total number of instances where the gene list of interest had a lower sum of ranks than the permuted sum of ranks (divided by the total number of permutations). We estimated the number of independent tissues via the matSpD tool⁸⁷ and then used Bonferroni correction to adjust for 27 independent tests (P < 1.90 × 10⁻³).

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 expression activity of each of these genes. As the number of samples for some tissues was small, we grouped individual tissues into four categories; brain, vascular, gastrointestinal, and other tissues (Supplementary Table 16). For each selected gene, we then tested whether the average expression (mean RPKM) was significantly higher in a particular tissue group compared to the ‘other tissues’ category. We assessed significance using a one-tailed t-test and used Bonferroni correction to adjust for 114 tests (38 genes × 3 tissue groups). While some genes were observed to be significantly expressed in multiple tissue groups, we determined that a gene was tissue-specific if it was only expressed highly in one tissue group (i.e. brain, vascular, or gastrointestinal,Supplementary Table 25).

eQTL credible set analysis in GTEx tissues

For all tissues and transcripts (filtered as above), we identified genome-wide significant (P < 2 × 10⁻¹³)cis-eQTLs within a 1Mb window of each transcript and created credible sets (seeDefining credible sets) for each eQTL locus identified in each tissue. We found a total of 35 of these significant eQTL credible sets within a 1Mb window of the migraine loci, however, only seven out of 35 contained variants that overlapped with a migraine credible set. For these seven eQTL credible sets, we then tested (Spearman’s rank correlation) if the test statistics between the two overlapping credible sets were significantly correlated. Significant correlation between a migraine credible set and an eQTL credible set was taken as evidence of the migraine locus tagging a real eQTL. Multiple testing was controlled for using Bonferroni correction (i.e. for seven tests atP < 7.1 × 10⁻³).

Enhancer enrichment analysis

Markers of gene regulation were defined using ChIP-seq datasets from ENCODE⁶⁶ and the NIH Roadmap Epigenome⁶⁵ projects. Based on the histone H3K27ac signal, which identifies active enhancers, we processed data from 56 cell lines and tissue samples to identify cell/tissue-specific enhancers, which we define as the 10% of enhancers with the highest ratio of reads in that cell/tissue type divided by the total reads⁸⁸. The raw data is publicly available (URLs) and a description of the 56 tissues/cell types is provided inSupplementary Table 21. We mapped the credible set variants at each migraine locus to these enhancer sites and compared the overlap observed with tissue-specific enhancers relative to a background of 10,000 randomly selected sets of SNPs of equal size. We restricted the background selection to 1000 Genomes project variants (MAF > 1%) that also passed QC filters in the meta-analysis (i.e. to only allow the selection of SNPs that had ana priori chance of being associated). The selection procedure then involved randomly selecting genomic regions that were of equivalent length and density of enhancers as found in the original locus. Once an appropriate region was found, a set of SNPs was randomly selected to match the number of SNPs in the credible set for that locus. If the selected SNPs mapped to an equal number of enhancer sites (of any tissue type) as credible SNPs from the original locus, then these were added to the background set of SNPs for comparison. If the selected SNPs did not map to the correct number of enhancers, the selection procedure was repeated until an appropriate set was found. This procedure was repeated 10,000 times for each locus to obtain an empirical null distribution. The enrichment significance was then estimated empirically by calculating the proportion of replicates that were greater than the observed value. Finally, we used Bonferroni correction to adjust for multiple testing of 56 tissue/cell types (P < 8.9 × 10⁻⁴).

Gene Ontology enrichment analysis

The set of 38 genes that are nearest to the index SNP in each migraine locus was chosen and tested for over-representation in Gene Ontology (GO) annotations. The PANTHER⁸⁹ tool (URLs) was used to perform the analysis implementing a binomial test to determine if the number of genes from the migraine test set found in each GO Pathway is likely to have occurred by chance alone. The associationP-values were adjusted for the number of pathways tested by Bonferroni correction.

DEPICT reconstituted gene set enrichment analysis

DEPICT⁶³ (Data-driven Expression Prioritized Integration for Complex Traits) is a computational tool, which, given a set of GWA study summary statistics, allows prioritization of genes in associated loci, enrichment analysis of reconstituted gene sets, and tissue enrichment analysis. DEPICT was run using 124 independent genome-wide significant SNPs as input (PLINK clumping parameters: --clump-p1 5e-8 --clump-p2 1e-5 --clump-r2 0.6 --clump-kb 250. Note, rs12845494 and rs140002913 could not be mapped). LD distance (r² > 0.5) was used to define locus boundaries (note that this locus definition is different than used elsewhere in the text) yielding 37 autosomal loci comprising 78 genes. DEPICT was run using default settings, that is, 500 permutations for bias adjustment, 20 replications for false discovery rate estimation, normalized expression data from 77,840 Affymetrix microarrays for gene set reconstitution (see reference⁹⁰), 14,461 reconstituted gene sets for gene set enrichment analysis, and testing 209 tissue/cell types assembled from 37,427 Affymetrix U133 Plus 2.0 Array samples for enrichment in tissue/cell type expression.

Post-analysis, we omitted reconstituted gene sets in which genes in the original gene set were not nominally enriched (Wilcoxon rank-sum test) because, by design, genes in the original gene set are expected to be enriched in the reconstituted gene set. Therefore, lack of enrichment complicates interpretation because the label of the reconstituted gene set may be inaccurate. Hence, the eight reconstituted gene sets were removed from the results: MP:0002089, MP:0002190, ENSG00000151577, ENSG00000168615, ENSG00000143322, ENSG00000112531, ENSG00000161021, and ENSG00000100320. We also removed an association identified for another gene set (ENSG00000056345 PPI,P = 1.7×10⁻⁴,FDR = 0.04) because it is no longer part of the Ensembl database. The Affinity Propagation tool⁹¹ was finally used to cluster related reconstituted gene sets into 10 groups (URLs).

DEPICT tissue enrichment analysis

DEPICT used data from 37,427 human microarray samples captured on the Affymetrix HGU133a2.0 platform to test if genes in the 38 migraine loci are highly expressed in 209 tissues/cell types with Medical Subject Heading (MeSH) annotations. The annotation procedure and method for normalizing expression profiles across annotations is outlined in the original publication⁶³. The tissue/cell type enrichment analysis algorithm was conceptually identical to the gene set enrichment analysis whereby enrichmentP-values were calculated empirically using 500 permutations for bias adjustment and 20 replications for false discovery rate estimation.

• 1.Vos T, et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990–2010: A systematic analysis for the Global Burden of Disease Study 2010. Lancet. 2012;380:2163–2196. doi: 10.1016/S0140-6736(12)61729-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 2.Vos T, et al. Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990–2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet. 2015;386:743–800. doi: 10.1016/S0140-6736(15)60692-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 3.Gustavsson A, et al. Cost of disorders of the brain in Europe 2010. Eur. Neuropsychopharmacol. 2011;21:718–779. doi: 10.1016/j.euroneuro.2011.08.008. [DOI] [PubMed] [Google Scholar]
• 4.Pietrobon D, Striessnig J. Neurological diseases: Neurobiology of migraine. Nature Reviews Neuroscience. 2003;4:386–398. doi: 10.1038/nrn1102. [DOI] [PubMed] [Google Scholar]
• 5.Tfelt-Hansen PC, Koehler PJ. One hundred years of migraine research: Major clinical and scientific observations from 1910 to 2010. Headache. 2011;51:752–778. doi: 10.1111/j.1526-4610.2011.01892.x. [DOI] [PubMed] [Google Scholar]
• 6.Headache Classification Committee of the International Headache Society (IHS) The International Classification of Headache Disorders, 3rd edition (beta version) Cephalalgia. 2013;33:629–808. doi: 10.1177/0333102413485658. [DOI] [PubMed] [Google Scholar]
• 7.Polderman TJC, et al. Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nat. Genet. 2015;47:702–709. doi: 10.1038/ng.3285. [DOI] [PubMed] [Google Scholar]
• 8.Anttila V, et al. Genome-wide association study of migraine implicates a common susceptibility variant on 8q22.1. Nat. Genet. 2010;42:869–873. doi: 10.1038/ng.652. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 9.Chasman DI, et al. Genome-wide association study reveals three susceptibility loci for common migraine in the general population. Nat Genet. 2011;43:695–698. doi: 10.1038/ng.856. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 10.Freilinger T, et al. Genome-wide association analysis identifies susceptibility loci for migraine without aura. Nat. Genet. 2012;44:777–782. doi: 10.1038/ng.2307. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 11.Anttila V, et al. Genome-wide meta-analysis identifies new susceptibility loci for migraine. Nat. Genet. 2013;45:912–917. doi: 10.1038/ng.2676. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 12.Ophoff RA, et al. Familial hemiplegic migraine and episodic ataxia type-2 are caused by mutations in the Ca2+ channel gene CACNL1A4. Cell. 1996;87:543–552. doi: 10.1016/s0092-8674(00)81373-2. [DOI] [PubMed] [Google Scholar]
• 13.De Fusco M, et al. Haploinsufficiency of ATP1A2 encoding the Na+/K+ pump alpha2 subunit associated with familial hemiplegic migraine type 2. Nat. Genet. 2003;33:192–196. doi: 10.1038/ng1081. [DOI] [PubMed] [Google Scholar]
• 14.Dichgans M, et al. Mutation in the neuronal voltage-gated sodium channel SCN1A in familial hemiplegic migraine. Lancet. 2005;366:371–377. doi: 10.1016/S0140-6736(05)66786-4. [DOI] [PubMed] [Google Scholar]
• 15.Nyholt DR, et al. A high-density association screen of 155 ion transport genes for involvement with common migraine. Hum. Mol. Genet. 2008;17:3318–3331. doi: 10.1093/hmg/ddn227. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 16.Altshuler DM, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491:56–65. doi: 10.1038/nature11632. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 17.Chasman DI, et al. Selectivity in Genetic Association with Sub-classified Migraine in Women. PLoS Genet. 2014;10:e1004366. doi: 10.1371/journal.pgen.1004366. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 18.Han B, Eskin E. Random-effects model aimed at discovering associations in meta-analysis of genome-wide association studies. Am. J. Hum. Genet. 2011;88:586–598. doi: 10.1016/j.ajhg.2011.04.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 19.Morton MJ, Abohamed A, Sivaprasadarao A, Hunter M. pH sensing in the two-pore domain K+ channel, TASK2. Proc. Natl. Acad. Sci. U. S. A. 2005;102:16102–16106. doi: 10.1073/pnas.0506870102. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 20.Ramachandran R, et al. TRPM8 activation attenuates inflammatory responses in mouse models of colitis. Proc. Natl. Acad. Sci. U. S. A. 2013;110:7476–7481. doi: 10.1073/pnas.1217431110. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 21.Hanna MG. Genetic neurological channelopathies. Nat. Clin. Pract. Neurol. 2006;2:252–263. doi: 10.1038/ncpneuro0178. [DOI] [PubMed] [Google Scholar]
• 22.Kraev A, et al. Molecular cloning of a third member of the potassium-dependent sodium-calcium exchanger gene family, NCKX3. J. Biol. Chem. 2001;276:23161–23172. doi: 10.1074/jbc.M102314200. [DOI] [PubMed] [Google Scholar]
• 23.Ismailov II, et al. A biologic function for an ‘orphan’ messenger: D-myo-inositol 3,4,5,6-tetrakisphosphate selectively blocks epithelial calcium-activated chloride channels. Proc. Natl. Acad. Sci. U. S. A. 1996;93:10505–10509. doi: 10.1073/pnas.93.19.10505. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 24.De Bock M, et al. Connexin channels provide a target to manipulate brain endothelial calcium dynamics and blood-brain barrier permeability. J. Cereb. Blood Flow Metab. 2011;31:1942–1957. doi: 10.1038/jcbfm.2011.86. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 25.Kathiresan S, et al. Genome-wide association of early-onset myocardial infarction with single nucleotide polymorphisms and copy number variants. Nat. Genet. 2009;41:334–341. doi: 10.1038/ng.327. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 26.Debette S, et al. Common variation in PHACTR1 is associated with susceptibility to cervical artery dissection. Nat. Genet. 2015;47:78–83. doi: 10.1038/ng.3154. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 27.Law C, et al. Clinical features in a family with an R460H mutation in transforming growth factor beta receptor 2 gene. J Med Genet. 2006;43:908–916. doi: 10.1136/jmg.2006.042176. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 28.Bown MJ, et al. Abdominal aortic aneurysm is associated with a variant in low-density lipoprotein receptor-related protein 1. Am. J. Hum. Genet. 2011;89:619–627. doi: 10.1016/j.ajhg.2011.10.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 29.Arndt AK, et al. Fine mapping of the 1p36 deletion syndrome identifies mutation of PRDM16 as a cause of cardiomyopathy. Am. J. Hum. Genet. 2013;93:67–77. doi: 10.1016/j.ajhg.2013.05.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 30.Fujimura M, et al. Genetics and Biomarkers of Moyamoya Disease: Significance of RNF213 as a Susceptibility Gene. J. stroke. 2014;16:65–72. doi: 10.5853/jos.2014.16.2.65. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 31.McElhinney DB, et al. Analysis of cardiovascular phenotype and genotype-phenotype correlation in individuals with a JAG1 mutation and/or Alagille syndrome. Circulation. 2002;106:2567–2574. doi: 10.1161/01.cir.0000037221.45902.69. [DOI] [PubMed] [Google Scholar]
• 32.Bezzina CR, et al. Common variants at SCN5A–SCN10A and HEY2 are associated with Brugada syndrome, a rare disease with high risk of sudden cardiac death. Nat. Genet. 2013;45:1044–1049. doi: 10.1038/ng.2712. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 33.Sinner MF, et al. Integrating genetic, transcriptional, and functional analyses to identify five novel genes for atrial fibrillation. Circulation. 2014;130:1225–1235. doi: 10.1161/CIRCULATIONAHA.114.009892. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 34.Neale BM, et al. Genome-wide association study of advanced age-related macular degeneration identifies a role of the hepatic lipase gene (LIPC) Proc. Natl. Acad. Sci. U. S. A. 2010;107:7395–7400. doi: 10.1073/pnas.0912019107. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 35.Desch M, et al. IRAG determines nitric oxide- and atrial natriuretic peptide-mediated smooth muscle relaxation. Cardiovasc. Res. 2010;86:496–505. doi: 10.1093/cvr/cvq008. [DOI] [PubMed] [Google Scholar]
• 36.Lang NN, Luksha L, Newby DE, Kublickiene K. Connexin 43 mediates endothelium-derived hyperpolarizing factor-induced vasodilatation in subcutaneous resistance arteries from healthy pregnant women. Am. J. Physiol. Heart Circ. Physiol. 2007;292:H1026–H1032. doi: 10.1152/ajpheart.00797.2006. [DOI] [PubMed] [Google Scholar]
• 37.Dong H, Jiang Y, Triggle CR, Li X, Lytton J. Novel role for K+-dependent Na+/Ca2+ exchangers in regulation of cytoplasmic free Ca2+ and contractility in arterial smooth muscle. Am. J. Physiol. Heart Circ. Physiol. 2006;291:H1226–H1235. doi: 10.1152/ajpheart.00196.2006. [DOI] [PubMed] [Google Scholar]
• 38.Yamaji M, Mahmoud M, Evans IM, Zachary IC. Neuropilin 1 is essential for gastrointestinal smooth muscle contractility and motility in aged mice. PLoS One. 2015;10:e0115563. doi: 10.1371/journal.pone.0115563. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 39.Lu X, et al. Genome-wide association study in Han Chinese identifies four new susceptibility loci for coronary artery disease. Nature Genetics. 2012;44:890–894. doi: 10.1038/ng.2337. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 40.Hager J, et al. Genome-wide association study in a Lebanese cohort confirms PHACTR1 as a major determinant of coronary artery stenosis. PLoS One. 2012;7:e38663. doi: 10.1371/journal.pone.0038663. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 41.The Coronary Artery Disease (C4D) Genetics Consortium. A genome-wide association study in Europeans and South Asians identifies five new loci for coronary artery disease. Nat. Genet. 2011;43:339–344. doi: 10.1038/ng.782. [DOI] [PubMed] [Google Scholar]
• 42.O’Donnell CJ, et al. Genome-wide association study for coronary artery calcification with follow-up in myocardial infarction. Circulation. 2011;124:2855–2864. doi: 10.1161/CIRCULATIONAHA.110.974899. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 43.Porcu E, et al. A meta-analysis of thyroid-related traits reveals novel loci and gender-specific differences in the regulation of thyroid function. PLoS Genet. 2013;9:e1003266. doi: 10.1371/journal.pgen.1003266. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 44.Soler Artigas M, et al. Genome-wide association and large-scale follow up identifies 16 new loci influencing lung function. Nat. Genet. 2011;43:1082–1090. doi: 10.1038/ng.941. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 45.Lu T, et al. REST and stress resistance in ageing and Alzheimer disease. Nature. 2014;507:448–454. doi: 10.1038/nature13163. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 46.Kar R, Riquelme MA, Werner S, Jiang JX. Connexin 43 channels protect osteocytes against oxidative stress-induced cell death. J. Bone Miner. Res. 2013;28:1611–1621. doi: 10.1002/jbmr.1917. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 47.Dixit D, Ghildiyal R, Anto NP, Sen E. Chaetocin-induced ROS-mediated apoptosis involves ATM-YAP1 axis and JNK-dependent inhibition of glucose metabolism. Cell Death Dis. 2014;5:e1212. doi: 10.1038/cddis.2014.179. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 48.Chuikov S, Levi BP, Smith ML, Morrison SJ. Prdm16 promotes stem cell maintenance in multiple tissues, partly by regulating oxidative stress. Nat. Cell Biol. 2010;12:999–1006. doi: 10.1038/ncb2101. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 49.Castellano J, et al. Hypoxia stimulates low-density lipoprotein receptor-related protein-1 expression through hypoxia-inducible factor-1α in human vascular smooth muscle cells. Arterioscler. Thromb. Vasc. Biol. 2011;31:1411–1420. doi: 10.1161/ATVBAHA.111.225490. [DOI] [PubMed] [Google Scholar]
• 50.Schlossmann J, et al. Regulation of intracellular calcium by a signalling complex of IRAG, IP3 receptor and cGMP kinase Ibeta. Nature. 2000;404:197–201. doi: 10.1038/35004606. [DOI] [PubMed] [Google Scholar]
• 51.Nalls MA, et al. Large-scale meta-analysis of genome-wide association data identifies six new risk loci for Parkinson’s disease. Nat. Genet. 2014;46:7989–7993. doi: 10.1038/ng.3043. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 52.Lambert JC, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat. Genet. 2013;45:1452–1458. doi: 10.1038/ng.2802. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 53.Ripke S, et al. Biological insights from 108 schizophrenia-associated genetic loci. Nature. 2014;511:421–427. doi: 10.1038/nature13595. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 54.Wood AR, et al. Defining the role of common variation in the genomic and biological architecture of adult human height. Nat. Genet. 2014;46:1173–1186. doi: 10.1038/ng.3097. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 55.Purcell S, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 2007;81:559–575. doi: 10.1086/519795. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 56.Bulik-Sullivan BK, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat. Genet. 2015;47:291–295. doi: 10.1038/ng.3211. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 57.Yang J, et al. Genomic inflation factors under polygenic inheritance. Eur. J. Hum. Genet. 2011;19:807–812. doi: 10.1038/ejhg.2011.39. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 58.Magi R, Lindgren CM, Morris AP. Meta-analysis of sex-specific genome-wide association studies. Genet. Epidemiol. 2010;34:846–853. doi: 10.1002/gepi.20540. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 59.Maller JB, et al. Bayesian refinement of association signals for 14 loci in 3 common diseases. Nat. Genet. 2012;44:1294–1301. doi: 10.1038/ng.2435. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 60.Nicolae DL, et al. Trait-associated SNPs are more likely to be eQTLs: Annotation to enhance discovery from GWAS. PLoS Genet. 2010;6:e1000888. doi: 10.1371/journal.pgen.1000888. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 61.Maurano MT, et al. Systematic Localization of Common Disease-Associated Variation in Regulatory DNA. Science. 2012;337:1190–1195. doi: 10.1126/science.1222794. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 62.The GTEx Consortium. The Genotype-Tissue Expression (GTEx) project. Nat. Genet. 2013;45:580–585. doi: 10.1038/ng.2653. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 63.Pers TH, et al. Biological interpretation of genome-wide association studies using predicted gene functions. Nat. Commun. 2015;6:5890. doi: 10.1038/ncomms6890. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 64.Chi JT, et al. Gene expression programs of human smooth muscle cells: Tissue-specific differentiation and prognostic significance in breast cancers. PLoS Genet. 2007;3:1770–1784. doi: 10.1371/journal.pgen.0030164. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 65.Bernstein BE, et al. The NIH Roadmap Epigenomics Mapping Consortium. Nat. Biotechnol. 2010;28:1045–1048. doi: 10.1038/nbt1010-1045. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 66.The ENCODE Project Consortium. An integrated encyclopedia of DNA elements in the human genome. Nature. 2012;489:57–74. doi: 10.1038/nature11247. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 67.Winsvold BS, et al. Genetic analysis for a shared biological basis between migraine and coronary artery disease. Neurol. Genet. 2015;1:e10. doi: 10.1212/NXG.0000000000000010. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 68.Malik R, et al. Shared genetic basis for migraine and ischemic stroke: A genome-wide analysis of common variants. Neurology. 2015;84:2132–2145. doi: 10.1212/WNL.0000000000001606. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 69.Ferrari MD, Klever RR, Terwindt GM, Ayata C, van den Maagdenberg AMJM. Migraine pathophysiology: lessons from mouse models and human genetics. Lancet. Neurol. 2015;14:65–80. doi: 10.1016/S1474-4422(14)70220-0. [DOI] [PubMed] [Google Scholar]
• 70.Olesen J, Burstein R, Ashina M, Tfelt-Hansen P. Origin of pain in migraine: evidence for peripheral sensitisation. Lancet Neurol. 2009;8:679–690. doi: 10.1016/S1474-4422(09)70090-0. [DOI] [PubMed] [Google Scholar]
• 71.Hadjikhani N, et al. Mechanisms of migraine aura revealed by functional MRI in human visual cortex. Proc. Natl. Acad. Sci. 2001;98:4687–4692. doi: 10.1073/pnas.071582498. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 72.Lauritzen M. Pathophysiology of the migraine aura. The spreading depression theory. Brain. 1994;117:199–210. doi: 10.1093/brain/117.1.199. [DOI] [PubMed] [Google Scholar]
• 73.Olesen J. The role of nitric oxide (NO) in migraine, tension-type headache and cluster headache. Pharmacol Ther. 2008;120:157–171. doi: 10.1016/j.pharmthera.2008.08.003. [DOI] [PubMed] [Google Scholar]
• 74.Ashina M, Hansen JM, Olesen J. Pearls and pitfalls in human pharmacological models of migraine: 30 years’ experience. Cephalalgia. 2013;33:540–553. doi: 10.1177/0333102412475234. [DOI] [PubMed] [Google Scholar]
• 75.Read SJ, Parsons AA. Sumatriptan modifies cortical free radical release during cortical spreading depression: A novel antimigraine action for sumatriptan? Brain Res. 2000;870:44–53. doi: 10.1016/s0006-8993(00)02400-8. [DOI] [PubMed] [Google Scholar]
• 76.Anderson CA, et al. Data quality control in genetic case-control association studies. Nat. Protoc. 2010;5:1564–1573. doi: 10.1038/nprot.2010.116. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 77.Winkler TW, et al. Quality control and conduct of genome-wide association meta-analyses. Nat. Protoc. 2014;9:1192–1212. doi: 10.1038/nprot.2014.071. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 78.Delaneau O, Marchini J, Zagury J-F. A linear complexity phasing method for thousands of genomes. Nature Methods. 2011;9:179–181. doi: 10.1038/nmeth.1785. [DOI] [PubMed] [Google Scholar]
• 79.Howie B, Fuchsberger C, Stephens M, Marchini J, Abecasis GR. Fast and accurate genotype imputation in genome-wide association studies through pre-phasing. Nature Genetics. 2012;44:955–959. doi: 10.1038/ng.2354. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 80.Browning SR, Browning BL. Rapid and accurate haplotype phasing and missing-data inference for whole-genome association studies by use of localized haplotype clustering. Am. J. Hum. Genet. 2007;81:1084–1097. doi: 10.1086/521987. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 81.Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: Using sequence and genotype data to estimate haplotypes and unobserved genotypes. Genet. Epidemiol. 2010;34:816–834. doi: 10.1002/gepi.20533. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 82.Fuchsberger C, Abecasis GR, Hinds DA. minimac2: faster genotype imputation. Bioinformatics. 2015;31:782–784. doi: 10.1093/bioinformatics/btu704. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 83.The International HapMap 3 Consortium. Integrating common and rare genetic variation in diverse human populations. Nature. 2010;467:52–58. doi: 10.1038/nature09298. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 84.Schwarz G. Estimating the Dimension of a Model. The Annals of Statistics. 1978;6:461–464. [Google Scholar]
• 85.Wright FA, et al. Heritability and genomics of gene expression in peripheral blood. Nat. Genet. 2014;46:430–437. doi: 10.1038/ng.2951. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 86.Richards AL, et al. Schizophrenia susceptibility alleles are enriched for alleles that affect gene expression in adult human brain. Mol. Psychiatry. 2012;17:193–201. doi: 10.1038/mp.2011.11. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 87.Nyholt DR. A simple correction for multiple testing for single-nucleotide polymorphisms in linkage disequilibrium with each other. Am. J. Hum. Genet. 2004;74:765–769. doi: 10.1086/383251. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 88.Farh KK-H, et al. Genetic and epigenetic fine mapping of causal autoimmune disease variants. Nature. 2014;518:337–343. doi: 10.1038/nature13835. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 89.Mi H, Muruganujan A, Casagrande JT, Thomas PD. Large-scale gene function analysis with the PANTHER classification system. Nat. Protoc. 2013;8:1551–1566. doi: 10.1038/nprot.2013.092. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 90.Fehrmann RSN, et al. Gene expression analysis identifies global gene dosage sensitivity in cancer. Nat. Genet. 2015;47:115–125. doi: 10.1038/ng.3173. [DOI] [PubMed] [Google Scholar]
• 91.Frey BJ, Dueck D. Clustering by Passing Messages Between Data Points. Science. 2007;315:972–976. doi: 10.1126/science.1136800. [DOI] [PubMed] [Google Scholar]

Supplementary Materials