New and sex-specific migraine susceptibility loci identified from a multiethnic genome-wide meta-analysis.

Migraine is a common disabling primary headache disorder that is ranked as the most common neurological cause of disability worldwide. Women present with migraine much more frequently than men, but the reasons for this difference are unknown. Migraine heritability is estimated to up to 57%, yet much of the genetic risk remains unaccounted for, especially in non-European ancestry populations. To elucidate the etiology of this common disorder, we conduct a multiethnic genome-wide association meta-analysis of migraine, combining results from the GERA and UK Biobank cohorts, followed by a European-ancestry meta-analysis using public summary statistics. We report 79 loci associated with migraine, of which 45 were novel. Sex-stratified analyses identify three additional novel loci (CPS1, PBRM1, and SLC25A21) specific to women. This large multiethnic migraine study provides important information that may substantially improve our understanding of the etiology of migraine susceptibility.

Introduction

Migraine is a common disabling disorder that can be accompanied by a wide range of symptoms of varying intensity, including headache pain that is often one-sided, and accompanied by nausea, sound and light sensitivity, and disturbed vision[1-3]. Epidemiological studies have repeatedly found that women have a substantially higher prevalence of migraine compared to men; however, the reasons for this difference are poorly understood[1,4-8].

Migraine has a moderate genetic component, with heritability estimates ranging between 0.10 and 0.57[9-16], depending on the type of estimate (i.e., based on twins studies vs. single-nucleotide polymorphism (SNP)-based heritability). In the past decade, genome-wide association studies (GWASs) have reported 42 genetic loci associated with migraine at genome-wide significance[17-22]. The most recent large genetic study of migraine was conducted by the International Headache Genetics Consortium (IHGC), by combining 22 European ancestry GWASs in a meta-analysis, they identified 38 loci, including 28 novel loci awaiting independent replication[21]. To our knowledge, no studies have yet conducted a genetic analysis of migraine in large and ethnically diverse cohorts. Therefore, there is a clear need for research to illuminate the genetic underpinnings of migraine.

Here, we present a large and ethnically diverse human genetic study of migraine, including, for the first time to our knowledge, East Asian, African American, and Hispanic/Latino adult individuals. Our study utilizes data from 554,569 individuals (28,852 migraine cases) from the Genetic Epidemiology Research in Adult Health and Aging (GERA) cohort and the UK Biobank (UKB) cohort (Table 1). We also used GWAS summary statistics data from the study of Gormley et al.[21], consisting of 375,752 participants (including 59,674 migraine cases) from the IHGC, which were publicly accessible. We conduct multiethnic GWAS meta-analyses, identifying several novel loci, including sex-specific ones. The associated loci provide potential causal variants, candidate genes, and relevant pathways underlying migraine susceptibility.

Table 1

Characteristics of the migraine cases and controls from GERA and UKB cohorts.

		GERA		UKB
		Controls	Cases	Controls	Cases
Total, N (proportion that are cases)		60,282	11,320 (15.8%)	465,435	17,532 (3.6%)
Age, mean ± SD		63.6 ± 13.5	58.9 ± 13.5	57.12 ± 8.10	55.80 ± 7.87
Sex, N (proportion that are cases)	Female	31,578	8982 (22.1%)	248,184	13,518 (5.1%)
	Male	28,704	2338 (7.5%)	217,251	4014 (1.8%)
Ethnicity, N (proportion that are cases)	Non-Hispanic white/European	49,036	9343 (16.0%)	438,178	16,709 (3.6%)
	Hispanic/Latino	4695	1142 (19.6%)	NA	NA
	East Asian	4804	539 (10.1%)	1815	30 (1.6%)
	African American/African	1747	296 (14.5%)	8357	208 (2.4%)
	South Asian	NA	NA	9480	278 (2.8%)
	Mixed	NA	NA	7605	307 (3.8%)

Results

Multiethnic meta-analysis of GERA and UKB

We first undertook GWAS analysis of migraine in the GERA cohort and UKB cohort, stratified by ethnic group, followed by a meta-analysis combining results from GERA and UKB. This combined meta-analysis identified 22 loci associated with migraine (P < 5 × 10−8), of which 10 were novel (Fig. 1, Table 2, Supplementary Fig. 1, and Supplementary Data 1). The effect estimates of the ten lead SNPs at novel loci were consistent across the two studies (Supplementary Fig. 2), and no significant heterogeneity was observed between them (Table 2).

Fig. 1

Manhattan plot of the multiethnic combined (GERA + UKB) GWAS meta-analysis of migraine.

The y-axis represents the –log10(P value); all P values derived from logistic regression model are two-sided. The red dotted line represents the threshold of P = 5 × 10−8, which is the commonly accepted threshold of adjustments for multiple comparisons in GWAS. Locus names in blue are for the novel loci and the ones in dark are for the previously reported ones.

Table 2

Migraine loci identified in the combined (GERA + UKB) GWAS multiethnic meta-analysis.

SNP	Chr	Pos	Locus	EA /OA	OR (SE)	P	DE (GERA-UKB)	I2 (Q)
rs1393064	1	3099138	PRDM16	C/G	1.15 (0.013)	2.38 × 10−27	++	83.41 (0.014)
rs10399665	1	8407287	SLC45A1-RERE	T/C	0.95 (0.009)	3.67 × 10−10	−−	0 (0.32)
rs10927732	1	15543202	TMEM51	C/A	1.07 (0.012)	3.89 × 10−8	++	0 (0.54)
rs12134493	1	115677946	TSPAN2-NGF	A/C	1.10 (0.013)	1.20 × 10−13	++	88.35 (0.0034)
rs2274319	1	156450873	MEF2D	C/T	0.94 (0.0096)	7.40 × 10−11	−−	40.65 (0.19)
rs4663983	2	234815005	MSL3P1-TRPM8	G/A	0.92 (0.011)	3.44 × 10−13	−−	56.04 (0.13)
rs13087932	3	19740933	MIR4791-EFHB	T/G	0.92 (0.016)	3.38 × 10−8	−−	0 (0.61)
rs9349379	6	12903957	PHACTR1	G/A	0.92 (0.0092)	2.59 × 10−18	−−	93.14 (1.0 × 10−4)
rs829470	6	72437128	LINC00472-RIMS1	C/T	1.05 (0.009)	2.71 × 10−8	++	0 (0.72)
rs9486715	6	97059769	FHL5	C/A	1.10 (0.0093)	2.62 × 10−26	++	78.36 (0.032)
rs35136812	6	121815335	GJA1-HSF2	GA/G	1.09 (0.012)	7.05 × 10−13	++	27.1 (0.24)
rs549745067	7	40417816	C7of10	T/TCTC	1.10 (0.014)	7.32 × 10−12	++	79.41 (0.028)
rs7855905	9	71700035	FXN	G/C	0.94 (0.0089)	9.95 × 10−13	−−	0 (0.74)
rs7858153	9	119245085	ASTN2-LOC100128505	A/G	1.09 (0.010)	1.54 × 10−16	++	90.15 (0.0014)
rs10795669	10	8720639	GATA3-SFTA1P	T/G	0.95 (0.010)	3.81 × 10−8	−−	0 (0.44)
rs10833535	11	3259478	MRGPRE-ZNF195	A/G	1.06 (0.0089)	9.88 × 10−10	++	0 (0.40)
rs4909945	11	10673739	MRVI1	C/T	1.06 (0.0098)	9.42 × 10−9	++	0 (0.88)
rs11023404	11	15004340	CALCB	T/C	1.06 (0.0092)	7.68 × 10−10	++	0 (0.77)
rs2160875	12	4527322	FGF23-FGF6	T/C	0.93 (0.009)	4.70 × 10−15	−−	18.49 (0.27)
rs11172113	12	57527283	LRP1-MIR1228	C/T	0.90 (0.0092)	1.12 × 10−28	−−	87.29 (0.005)
rs11655891	17	81015295	B3GNTL1-METRNL	G/A	1.07 (0.011)	1.43 × 10−8	++	0 (0.84)
rs79545715	21	35589328	LINC00310-KCNE2	C/T	0.90 (0.017)	3.80 × 10−9	−−	0 (0.41)

Loci in bold are novel.

SNP single-nucleotide polymorphism, Chr chromosome, Pos position, EA effect allele, OA other allele, SE standard error, DE direction of effect, GERA Genetic Epidemiology Research on Adult Health and Aging, UKB UK Biobank, I2 heterogeneity index (0–100%), Q P value for Cochrane’s Q statistic.

Replication in the IHGC data

We then tested the ten lead SNPs representing each of the ten novel loci for replication in the most recent large genetic study of migraine conducted by the IHGC[21]. However, as the GWAS summary statistics data, publicly accessible, reported only the SNPs from the Gormley et al. study with a P value of less than 1.0 × 10−5, some of the strongest SNPs reported by Gormley et al. were different than ours. Six loci, including TMEM51, MIR4791-EFHB, LINC00472-RIMS1, FXN, GATA3-SFTA1P, and LINC00310-KCNE2, replicated at Bonferroni significance (P < 0.05/10 novel loci = 5.0 × 10−3) (Supplementary Data 2). Our lead SNPs within the remaining four novel loci (i.e., SLC45A1/RERE, MRGPRE-ZNF195, CALCB, and B3GNTL1-METRNL) were not reported in the publicly accessible GWAS summary statistic from the Gormley et al. study; however, those may replicate at a Bonferroni significance threshold in IHGC (1.0 × 10−5 ≤ P < 5.0 × 10−3) but were not publicly accessible.

Replication of previous migraine GWAS results

We also investigated the lead SNPs within 38 loci associated with migraine at a genome-wide significance level from the most recent and exhaustive GWAS of migraine conducted to date[21] (Supplementary Data 3). Ten lead SNPs of the 36 available replicated at a genome-wide level of significance in our combined (GERA + UKB) multiethnic meta-analysis (including rs10218452 at PRDM16, rs2078371 near TSPAN2/NGF, rs1925950 at MEF2D, rs10166942 at TRPM8/HJURP, rs9349379 at PHACTR1, rs28455731 near GJA1, rs186166891 at C7orf10, rs6478241 at ASTN2, rs1024905 near FGF6, and rs11172113 at LRP1) (Supplementary Data 3). Furthermore, 14 additional SNPs replicated at Bonferroni significance (P < 0.05/36 = 1.39 × 10−3), and 4 showed nominal evidence of association (P < 0.05). In contrast, eight SNPs (including rs140002913 near NOTCH4, rs10155855 near DOCK4/IMMP2L, rs2506142 at NRP1, rs561561 at IGSF9B, rs75213074 near WSCD1/NLRP1, rs17857135 at RNF213, rs144017103 near CCM2L/HCK, and rs12845494 near MED14/USP9X) were not validated in the current combined (GERA + UKB) multiethnic meta-analysis (P > 0.05).

Ethnic-specific and conditional analyses

For ethnic groups represented in each cohort, we conducted ethnic-specific meta-analyses of each group. In the European ancestry (GERA non-Hispanic whites + UKB Europeans + IHGC Europeans only; 85,726 migraine cases and 803,292 controls) meta-analysis, we identified 73 loci, of which 35 were additional novel (Supplementary Data 4). To identify independent signals within the 73 genomic regions identified in the European-specific meta-analysis, we performed a multi-SNP-based conditional and joint association analysis (COJO)[23], which revealed two additional independent SNPs within the known loci TSPAN2-NGF (rs2207237) and ADAMTSL4-ECM1 (rs7524797). Conducting a GWAS meta-analysis of East Asian-specific cohorts (GERA + UKB East Asian ancestry individuals only; 569 migraine cases and 6619 controls) and a GWAS meta-analysis of African-specific cohorts (GERA + UKB African ancestry individuals only; 504 migraine cases and 10,104 controls) did not result in the identification of genome-wide significant findings. We may have been underpowered to detect effects with statistical significance in those non-European ancestry meta-analyses.

Sex-specific analyses identified additional loci

Next, we conducted meta-analyses (GERA + UKB) for migraine by sex. The women-specific (GERA + UKB) meta-analysis (22,500 migraine cases and 279,762 controls) identified 19 loci (P < 5.0 × 10−8), of which three, CPS1 on chromosome 2, PBRM1 on chromosome 3, and SLC25A21 on chromosome 14, were additional novel loci (Fig. 2 and Supplementary Data 5). CPS1 rs1047891, PBRM1 rs11718509, and SLC25A21 rs10150336 were significantly associated with migraine in women (P < 5.0 × 10−8) but not in men (P > 0.05) (Supplementary Fig. 3a–c). Furthermore, among the loci also identified in the multiethnic meta-analysis (GERA + UKB), we observed that ASTN2 locus was significantly differently associated in women and men (ASTN2 rs7858153: OR = 1.11 and P = 1.83 × 10−19 in women; OR = 1.02, P = 0.40 in men; Z = 3.59, P = 1.63 × 10−4) (Supplementary Fig. 3d). Regional association plots illustrate the sex-specific association signals (Supplementary Fig. 3). For each of the four women-specific lead SNPs, we compared the effect allele frequencies (EAF) between cases and controls in women and men separately, as well as between women and men cases and women and men controls (Supplementary Data 6). We found that the EAF of each SNP were near identical when comparing women cases to men cases, or when comparing women controls to men controls. This shows that the differences in association signals in the women and men meta-analyses were not due to the difference in case and/or control EAF in the women and men samples. The men-specific (GERA + UKB) meta-analysis (6352 migraine cases and 245,955 controls) did not result in the identification of additional novel genome-wide significant findings (Fig. 2 and Supplementary Data 7). To further evaluate the shared genetic basis of migraine between women and men, we compared the GWAS results from the two sex-specific meta-analyses by performing a linkage disequilibrium (LD) score regression (LDSC). We observed a high genetic correlation (rg) between women and men for migraine (rg = 0.76, P = 2.39 × 10−15).

Fig. 2

Chicago plot of the sex-stratified multiethnic GWAS meta-analyses of migraine.

Results from the meta-analysis combining women from GERA and UKB are presented on upper panel, while results from the meta-analysis combining men from GERA and UKB are presented on the lower panel. The y-axis represents the –log10(P value); all P values derived from logistic regression model are two-sided. The red dotted line represents the threshold of P = 5 × 10−8, which is the commonly accepted threshold of adjustments for multiple comparisons in GWAS. Locus names in black are for those previously reported. Locus names in bold (i.e., CPS1, PBRM1, and SLC25A21) are for the additional novel loci specific to women (compared to the multiethnic meta-analysis (GERA + UKB)). Novel loci significantly associated (P < 5 × 10−8) with migraine in women are highlighted in blue.

SNP prioritization and annotations

To prioritize variants within the 22 loci identified in the combined multiethnic (GERA + UKB) meta-analysis and within the 73 loci identified in the combined (GERA + UKB + IHGC) European-specific meta-analysis, we applied a Bayesian approach (CAVIARBF)[24]. For each of the associated signals, we computed each variant’s capacity to explain the identified signal within a 2 Mb window (±1.0 Mb with respect to the original top variant) and derived the smallest set of variants that included the causal variant with 95% probability (95% credible set). For the 22 loci identified in the combined multiethnic (GERA + UKB) meta-analysis, two sets included a unique variant (Supplementary Data 8). These include the previously reported intronic variant rs9349379 at PHACTR1[21], and the newly identified intergenic variant rs13087932 at MIR4791-EFHB with 100.0% and 97.2% posterior probability of being the causal variants, respectively, suggesting that these variants are more likely to be the true causal variants. For the 73 loci identified in the European-specific meta-analysis, four sets included a unique variant (Supplementary Data 9). In addition to rs9349379 at PHACTR1, we found that the intronic variants rs5763529 at ASCC2 and rs11172113 at LRP1, and the intergenic variant rs28451064 at LINC00310-KCNE2 were more likely to be the true causal variants with 100.0%, 99.9%, and 97.3% posterior probability, respectively.

Gene-based association analysis and gene prioritization

To identify additional genes associated with migraine at a gene level, we conducted a gene-based association analysis using the functional mapping and annotation of genetic associations (FUMA)[25] integrative tool using the combined multiethnic GWAS meta-analysis (GERA + UKB) results. FUMA implements Multi-marker Analysis of GenoMic Annotation (MAGMA)[26] gene-based analysis, which employs a multiple linear regression approach to properly incorporate LD between genetic variants and to detect multi-variant effects. As 19,933 genes were tested, the P value adjusted for Bonferroni correction was set as P < 2.51 × 10−6 (0.05/19,933). We found significant associations with migraine for 47 genes, with the strongest association for STAT6 (P = 1.24 × 10−23), followed by UFL1 (P = 7.26 × 10−19), and FHL5 (P = 1.25 × 10−18) (Supplementary Data 10). Out of the 47 genes, 9 were located outside the loci identified in the current study, including PRKCE, RCHY1, THAP6, MAPK9, RP11-508N12.4, LRCH1, PNKP, AKT1S1, and TBC1D17.

To prioritize genes within the 73 loci identified in the combined (GERA + UKB + IHGC) European-specific meta-analysis, we used the DEPICT[27] integrative tool. DEPICT gene prioritization analysis detected 15 genes, of which 9 were within novel migraine-associated loci, to prioritize after false-discovery rate (FDR) correction (Supplementary Data 11). These included: LEPR on chromosome 1, TJP2 on chromosome 9, AMBRA1 on chromosome 11, HOXB2, HOXB3, HOXB6, and POLR2A on chromosome 17, TGFB1 on chromosome 19, and JAG1 on chromosome 20.

Biological pathway annotations and prioritization

While DEPICT gene-set enrichment analysis using independent genome-wide significant genetic variants from the combined (GERA + UKB + IHGC) European-specific meta-analysis did not detect pathways to prioritize after FDR correction (Supplementary Data 12), FUMA[25] gene-set enrichment analysis based on the GWAS multiethnic meta-analysis (GERA + UKB) results highlighted many gene-sets involved in the flavonoid glucuronide biosynthetic process, the ascorbate and aldarate metabolic pathways, the response to endogenous stimulus or wounding, and the regulation of muscle contraction or system process (Supplementary Data 13).

FUMA tissue expression quantitative trait loci (eQTL) specificity analysis highlighted the sigmoid colon as the main tissue for which expression was affected by migraine-associated variants, followed by the esophagus muscularis and gastroesophageal junction (P < 9.43 × 10−4; Bonferroni significance after correcting for 53 GTEx tissues tested—Supplementary Fig. 4 and Supplementary Data 14). In contrast, DEPICT tissue-enrichment analysis using independent genome-wide significant genetic variants from the combined (GERA + UKB + IHGC) European-specific meta-analysis identified two tissues or cell type annotations to prioritize after FDR correction: the arteries (cardiovascular system), consistent with the previously reported Gormley et al. study[21], and the serous membrane (Supplementary Data 15).

Genetic correlation between migraine and other phenotypes

Genome-wide genetic correlations of migraine were calculated with a total of 772 complex traits and diseases by comparing allelic effects using a LDSC with the European-specific migraine meta-analysis (GERA + UKB) summary statistics (Methods). A total of 75 significant genetic correlations were observed (P < 6.48 × 10−5 which corresponds to 0.05/772 phenotypes tested; Supplementary Data 16). Among those 75 genetic correlations, 38 reached genome-wide level of significance (Supplementary Fig. 5). The strongest positive correlations were observed with medications for pain relief, constipation, or heartburn; neck, shoulder or back pain; alcohol intake frequency; mood swings, anxiety, depression or neuroticism; and sleeplessness or insomnia. There was also a strong negative correlation with physical activity.

Discussion

In this study, a multiethnic meta-analysis combining the GERA and UKB cohorts and a meta-analysis of migraine across European ancestry individuals identified 79 loci, of which 45 were novel. Our multiethnic meta-analysis also validated 78% of the migraine-associated loci identified in the most recent and exhaustive GWAS of migraine conducted to date[21]. Furthermore, our sex-stratified analyses identified three additional novel loci (CPS1, PBRM1, and SLC25A21) specific to women.

The identified loci give new insight and additional evidence about the genes and pathways/systems underlying migraine susceptibility. For instance, we identified a new region associated with migraine at 17q21 and our DEPICT gene analysis prioritized three members of the Antp homeobox family genes (i.e., HOXB2, HOXB3, and HOXB6) at this region that encode proteins with a homeobox DNA-binding domain. Those three genes have been involved in the early development[28-32] (i.e., hindbrain, nervous system, or epidermal development) and common variants in HOXB3 have been shown to be associated with motion sickness, which is a condition that shares underlying genetic factors with migraine[33]. Our DEPICT gene analysis also prioritized TGFB1 at the novel 19q13 migraine-associated locus. TGFB1 encodes the transforming growth factor beta 1 protein that is a multifunctional proinflammatory cytokine that regulates cell proliferation, differentiation, and growth. Early works suggested that TGFB1 could play a role in migraine susceptibility. Plasma level of TGFB1 has been shown to increase in patients with migraine during headache-free periods compared to healthy subjects without headache[34]. Another study investigated the TGFB1 genotype in pediatric migraine patients and reported significant differences between control and migraine patients[35]. Our DEPICT gene analysis also prioritized JAG1 at the novel 20p12 migraine-associated locus. JAG1 encodes that the jagged 1 protein is the ligand for the receptor notch 1, which is involved in signaling processes. JAG1 plays a role in the formation of blood cellular components[36-38] and has been involved in the pathogenesis of patent foramen ovale, which is an atrial septal deformity associated with major causes of morbidity, including stroke and migraine[39,40]. Future investigations may provide insights into how these genes influence migraine susceptibility.

Our study also reported, for the first time to our knowledge, sex-specific loci associated with migraine susceptibility in women but not in men. Previous studies[18,41] evaluated the concordance of genetic risk for migraine between women and men but did not reveal any heterogeneity in the effect sizes of the genome-wide significant loci. Among the four sex-specific loci identified in the current study, two, PBRM1 and ASTN2, have been involved in susceptibility to bipolar disorder and other psychiatric phenotypes. PBRM1 encodes polybromo 1, which is a subunit of ATP-dependent chromatin-remodeling complexes. Common genetic polymorphisms at PBRM1 (also named 3p21.1 locus) have been implicated in susceptibility to bipolar disorder, as well as major depression and schizophrenia[42-46]. ASTN2 encodes astrotactin 2 that is expressed in the brain and is involved in neuronal migration. Deletions of ASTN2 have been associated with schizophrenia and other psychiatric and neurodevelopmental disorders[47-49]. Similarly, our genetic correlation results indicate that migraine was significantly correlated with mood swings, anxiety, depression, or neuroticism, consistent with a previous genetic correlation analysis study[50]. Thus, these findings suggest that migraine shares common variation risk with psychiatric disorders, such as schizophrenia, for which genome-wide significant loci involved in brain function have been reported[51,52].

Among the sex-specific loci associated with migraine susceptibility in women but not in men, less is known about the role of the identified CPS1 and SLC25A21 in regard to the biologic pathways underlying migraine. CPS1 encodes a mitochondrial enzyme named carbamoyl-phosphate synthase 1 that catalyzes synthesis of carbamoyl phosphate from ammonia and bicarbonate. Mutations in this gene have been associated with metabolic deficiencies such as urea cycle disorder and hyperammonemia[53,54]. Individuals with a CPS1 deficiency can present with a wide range of clinical manifestations, including headache, behavioral or psychiatric problems, learning disabilities, sleep disorder, periodic vomiting, seizures, coma, and even death[55-57]. SLC25A21 (also known as ODC1) encodes the solute carrier family 25 member 21, which is essential for ammonium fixation and lysine biosynthesis[58]. While rare mutations in this gene cause a syndromic neurometabolic disorder associated with macrocephaly, developmental delay, alopecia, and dysmorphic features[59,60], common polymorphisms have been associated with smoking behaviors, and could contribute to Alzheimer’s disease[61,62]. Thus, our findings will help to better understand the biological mechanisms underlying migraine susceptibility, particularly in women, even if follow-up experimental studies are needed to validate the role of the identified genes in migraine susceptibility.

Our results also illustrate the association between disorders of the large bowel and migraine susceptibility, as our eQTL analysis highlighted the sigmoid colon tissue for which expression was affected by migraine-associated variants. Similarly, our multiethnic meta-analysis results identified variants in CALCB associated with migraine susceptibility. CALCB encodes the calcitonin-related polypeptide beta (CGRP), which has been shown to contribute to migraine[63-65]. Several monoclonal antibodies targeting CGRP or its receptor have been proven to be effective therapeutics for the preventive treatment of migraine[66,67] and have been recently approved by the U.S. Food and Drug Administration[68,69]. In parallel, common genetic polymorphisms at CALCB have been reported to contribute to diverticular disease that is a common of intestinal neuromuscular function[70]. Of note, several epidemiologic studies reported that gastrointestinal disorders such as the “irritable bowel syndrome” are the most commonly reported comorbidities associated with migraine[71,72]. Our results suggest that the relationship between the intestine and migraine, also referred to “gut-brain axis,“ could have some genetic origin.

Our genetic correlation results identified the strongest positive correlations between migraine and neck, shoulder or back pain, suggesting shared genetic factors underlying those conditions. This result is consistent with early work demonstrating that migraine patients have higher pain responses in the splenius and temporalis muscles after a cognitive stress test compared to controls, and their muscle pain responses are regional (i.e., more pain in the neck and trapezius)[73]. Thus, our results support the concept that sensitization of pain pathways and the muscular system seem to be important in migraine susceptibility.

We recognize several potential limitations of our study. First, it is important to note phenotypic differences for migraine between the two study cohorts. Although migraine cases in GERA were identified in the Kaiser Permanente Northern California (KPNC) electronic health record system using our previously described and validated migraine probability algorithm (MPA)[74], which is based on migraine-specific prescriptions and International Classification of Disease, Ninth (ICD9) or Tenth Revision (ICD10) diagnosis codes, most of the migraine cases in UKB were based on self-reported data. This may lead to phenotype misclassification that may have affected, for instance, the high positive genetic correlation between migraine and neck, shoulder, or back pain. However, our meta-analysis combining GERA and UKB results showed consistency of the SNPs effect estimates between cohorts. Furthermore, the previously reported associations[21] were well validated in our multiethnic meta-analysis and our novel loci were well replicated in the most recent large genetic study of migraine conducted by the IHGC[21]. Second, because of the limited sample of men cases (compared to women), we may have been underpowered to detect effects with statistical significance in the men-specific analysis.

In conclusion, our results identified additional loci that contribute to migraine susceptibility and represent potential candidates for the development of new therapeutic targets for this common neurological cause of disability. Although these findings will help to better understand the etiology of migraine susceptibility, additional genetic studies are needed to validate these associations in more large cohorts.

Methods

GERA

The GERA cohort contains genome-wide genotype, clinical, and demographic data of over 110,000 adult members from mainly four ethnic groups (non-Hispanic white, Hispanic/Latino, East Asian, and African American) of the KPNC integrated healthcare system[75,76]. The Institutional Review Board of the Kaiser Foundation Research Institute approved all study procedures. Patients with migraine were identified in the KPNC electronic health record system using the previously validated MPA[74] (Supplementary Data 17), which is based on migraine-specific prescriptions and ICD9 diagnosis code: 346.XX and ICD10: G43.XX. We defined migraine cases as patients with a score >10 on the MPA (any evidence of migraine). After excluding individuals with any evidence of headache without a migraine diagnosis, as well as individuals with a MPA score = 10, our control group included all the non-cases. In total, 11,320 migraine cases and 60,282 controls from GERA were included in this study.

Protocols for participant genotyping, data collection and quality control (QC) have been described in detail[76]. Briefly, GERA participants’ DNA samples were extracted from Oragene kits (DNA Genotek Inc., Ottawa, ON, Canada) at KPNC and genotyped at the Genomics Core Facility of the University of California, San Francisco. DNA samples were genotyped at over 665,000 genetic markers on four ethnic-specific Affymetrix Axiom arrays (Affymetrix, Santa Clara, CA, USA) optimized for European, Latino, East Asian, and African American individuals[77,78]. Genotype QC procedures and imputation were conducted on an array-wise basis[76], after an updated genotyping algorithm with an advanced normalization step specifically for SNPs in batches not recommended or flagged by the outlier plate detector than has previously been done. Subsequently, variants were excluded if: >3 clusters were identified; the number of batches was <38/42 (EUR array), <3/5 (AFR), <3/6 (EAS), or <7/9 (LAT); and the ratio of expected allele frequency variance across packages was <100 (EUR), <50 (AFR), <100 (EAS), <200 (LAT). On the EUR array, variants were additionally excluded if heterozygosity >0.52 or <0.02, and if an association test between Reagent kit v1.0 and v2.0 had P < 10−4. Imputation was done by array, and we additionally removed variants with call rates <90%. Genotypes were then pre-phased with Eagle[79] v2.3.2, and then imputed with Minimac3[80] v2.0.1, using two reference panels. Variants were preferred if present in the EGA release of the Haplotype Reference Consortium (HRC; n = 27,165) reference panel[81], and from the 1000 Genomes Project Phase III release if not (n = 2504; e.g., indels)[82].

UK Biobank

The UKB is a large prospective study following the health of approximately 500,000 participants from five ethnic groups (European, East Asian, South Asian, African British, and mixed ancestries) resident in the UK aged between 40 and 69 years at the baseline recruitment visit[83,84]. Demographic information and medical history were ascertained through touch-screen questionnaires. Participants also underwent a wide range of physical and cognitive assessments, including blood sampling. Migraine cases (N = 17,532) were defined as participants with a self-reported migraine (data field 20002 code 1265) and/or a diagnosis code for migraine (ICD10: G43). After excluding participants who self-reported headaches (data field 20002, code 1436) and/or who had a diagnosis code for headaches (ICD10: G44), the control group included 465,435 participants who were not cases. Phenotyping, genotyping, and imputation were carried out by members of the UKB team. Imputation to the Haplotype Reference Consortium reference panel has been described (www.ukbiobank.ac.uk). Following QC, over 10 million variants in 487,622 individuals were tested for migraine adjusting for age, sex, and genetic ancestry principal components (PCs). The analyses presented in this paper were carried out under UKB Resource project #14105.

The International Headache Genetics Consortium (IHGC)

GWAS summary statistics data (SNPs with P < 1.0 × 10−5) from the study of Gormley et al.[21], consisting of 375,000 participants from the IHGC, were publicly accessible (http://eagle-i.itmat.upenn.edu/i/00000155-e1db-73aa-c956-e86e80000000).

GWAS and adjustment in GERA

We first analyzed each ethnic group (non-Hispanic white, Hispanic/Latino, East Asian, and African American) separately. We ran a logistic regression of migraine and each SNP using PLINK[85] v1.9 (www.cog-genomics.org/plink/1.9/) adjusting for age, sex, and ancestry PCs, which were previously[75] assessed within each ethnic group using Eigenstrat[86] v4.2. We included as covariates the top ten ancestry PCs for the non-Hispanic whites, whereas we included the top six ancestry PCs for the three other ethnic groups. To adjust for genetic ancestry, we also included the percentage of Ashkenazi (ASHK) ancestry as a covariate for the non-Hispanic white sample analyses[75]. For comparison, the GWAS analyses were also conducted using a new approach accounting for relatedness that fits a whole genome regression model, implemented in REGENIEv2.0.2[87] (https://rgcgithub.github.io/regenie/).

GWAS meta-analyses

First, a meta-analysis of migraine was conducted in GERA to combine the results of the 4 ethnic groups using the R[88] (https://www.R-project.org) package “meta.” Similarly, a meta-analysis was conducted in UKB to combine the results of the five ethnic groups. The meta-analysis GWAS results generated using REGENIE were similar compared to the results generated using PLINK (Supplementary Data 18 and Supplementary Fig. 6). Three ethnic-specific meta-analyses were also performed: (1) combining European-specific samples (i.e., GERA non-Hispanic whites + UKB Europeans + IHGC Europeans); (2) combining East Asian-specific samples (i.e., GERA and UKB East Asians); and (3) combining African-specific samples (i.e., GERA African Americans and UKB Africans). A meta-analysis was then conducted to combine the results from GERA and UKB. The overall (GERA + UKB) meta-analysis results generated using REGENIE were similar compared to the results generated using PLINK (Supplementary Fig. 6). Two sex-specific meta-analyses were also performed: (1) combining women from GERA and UKB; and (2) combining men from GERA and UKB. Fixed-effects summary estimates were calculated for an additive model. We also estimated heterogeneity index, I (0–100%) and P value for Cochrane’s Q statistic among different groups, and studies. For each locus, we defined the top SNP as the most significant variant within a 2 Mb window. Novel loci were defined as those that were located over 1 Mb apart from any previously reported locus[17,18,20,21].

Conditional and joint (COJO) analysis

A multi-SNP-based COJO[23] was performed on the combined European-specific (GERA non-Hispanic whites + UKB Europeans + IHGC Europeans) meta-analysis results to potentially identify independent signals within the 73 identified genomic regions. To calculate LD patterns, we used 10,000 randomly selected samples from GERA non-Hispanic white ethnic group as a reference panel. A P value less than 5.0 × 10−8 was considered as the significance threshold for this COJO analysis.

Variants prioritization

To prioritize variants within the identified genomic regions for follow-up functional evaluation, a Bayesian approach (CAVIARBF)[24] was used, which is available publicly at https://bitbucket.org/Wenan/caviarbf. Each variant’s capacity to explain the identified signal within a 2 Mb window (±1.0 Mb with respect to the original top variant) was computed for each identified genomic region. Then the smallest set of variants that included the causal variant with 95% probability (95% credible set) was derived.

FUMA for gene-based association analysis and gene-set analysis

To prioritize genes and biological pathways, and highlight gene-set and tissue/cell enrichments within the 22 migraine-associated loci identified in the combined multiethnic (GERA + UKB) meta-analysis, we used the FUMA[25] integrative tool. FUMA uses input GWAS summary statistics to compute gene-based P values (gene analysis) and gene-set P value (gene-set analysis) using the MAGMA[26] tool, which employs multiple regression to obtain gene-based and gene-set P values. GWAS summary statistics from the combined multiethnic (GERA + UKB) meta-analysis served as input for the “SNP2GENE” function. We used 10,000 random samples from GERA non-Hispanic whites as a reference panel. This “SNP2GENE” function provides extensive functional annotation for all SNPs in genomic areas identified and prioritizes gene-set and genes (using the MAGMA gene-set association analysis and the MAGMA gene analysis, respectively) for the next FUMA function, which is “GENE2FUNC.” Then, this “GENE2FUNC” function annotates the prioritized genes in biological context. For gene analysis, the gene-based P value is computed for protein-coding genes by mapping SNPs to genes if SNPs are located within the genes. For gene-set analysis, the gene-set P value is computed using the gene-based P value for 4728 curated gene-sets (including canonical pathways) and 6166 GO terms obtained from MsigDB v5.2. For both analyses, the default MAGMA setting (SNP-wise model for gene analysis and competitive model for gene-set analysis) is used, and the Bonferroni correction (gene) or FDR (gene-set) was used to correct for multiple testing. For the MAGMA gene-based association analysis conducted on the combined (GERA + UKB) meta-analysis results, the P value adjusted for Bonferroni correction was set as P < 2.51 × 10−6 (0.05/19,933 genes tested).

FUMA tissue eQTL specificity

To highlight and visualize tissue eQTL enrichments within the 22 migraine-associated genomic regions identified in the combined multiethnic (GERA + UKB) meta-analysis, we used the FUMA[25] integrative tool. FUMA is an integrative web-based platform that accommodates eQTL, and provides tissue enrichment results for each of 53 tissue types based on the genotype-tissue expression (GTEx) v6 RNA-seq data[89].

DEPICT

To prioritize genes and biological pathways, and highlight gene-set and tissue/cell enrichments within the 73 migraine-associated loci identified in the combined (GERA + UKB + IHGC) European meta-analysis, we used the following integrative tool: DEPICT[27]. All independent genome-wide significant genetic variants (P < 5.0 × 10−8) served as input, and as the reference panel, we used 10,000 random samples from GERA non-Hispanic white ethnic group. Genes, gene-sets, and tissue/cell annotations that achieved a nominal significance level of 0.05 after FDR correction were subsequently prioritized.

Genetic correlations

To estimate the genetic correlation of migraine with more than 700 diseases/traits from different publicly available resources/consortia, we used the LD Hub web interface[90], which performs automated LDSC. In the LDSCs, we included only HapMap3 SNPs with MAF > 0.01. Genetic correlations were considered significant after Bonferroni adjustment for multiple testing (P < 6.48 × 10−5 which corresponds to 0.05/772 phenotypes tested).