Migraine pathophysiology is incompletely understood. While migraine is viewed as a primary disorder of the brain, vascular mechanisms are also implicated. For example, endothelial dysfunction and hypercoagulability [3] as well as altered vascular reactivity [4] are found in migraine patients. In addition, the current evidence on the association between migraine and cardiovascular disease (CVD) was summarized in a recent meta-analysis [5], which found a two-fold increased risk for ischemic stroke among migraineurs, in particular those with migraine with aura. Some individual studies suggest that the link between migraine and CVD also extends to myocardial infarction (MI), CVD death, and hemorrhagic stroke [6], [7], [8], [9], [10], [11], [12], [13], [14].

Family and twin studies have shown that genetic factors importantly contribute to the pathophysiology of both migraine and CVD. In addition, with respect to the established association between migraine and CVD, the influence of gene variants has recently been investigated, showing for example a modulatory effect of the MTHFR 677C>T [15] and ACE D/I [16] polymorphisms among women.

Given the unique physiology among migraineurs, specifically migraineurs with aura, which suggests the potential for a genetic risk of CVD, we sought to investigate the genetic liability for CVD among migraineurs. Such a design eliminates a potential confounding effect of migraine on the association between genetic markers and CVD. This contrasts with a study design comparing migraineurs to non-migraineurs, which would also investigate an interaction between the phenotype migraine and genetic markers on CVD risk.

Among the 5122 women with any migraine 164 developed a major CVD event during follow-up (73 MIs, 68 ischemic strokes, 35 CVD deaths). The respective numbers of CVD cases for migraine with aura and migraine without aura are: migraine with aura (1177 women)—55 major CVDs, 26 MIs, 21 ischemic strokes, 12 CVD deaths; migraine without aura (1826 women)—38 major CVDs, 18 MIs, 17 ischemic strokes, 7 CVD deaths.

Baseline characteristics of the participating women according to major CVD are summarized in Table 1. Women with migraine experiencing major CVD during the study were older, were more likely to have a history of hypertension, and had higher systolic and diastolic blood pressure compared to migraineurs without major CVD. Women with major CVD also had higher LDL and lower HDL cholesterol levels, were more likely to be postmenopausal, and to be current smokers.

In the genome-wide analysis no SNP reached the conventional significance threshold of p<5×10⁻⁸ in age-adjusted logistic models assuming an additive relationship between the minor allele dose and log-odds of major CVD, ischemic stroke, MI or CVD death. Nevertheless, investigation of the quantile-quantile plots of probability values for the association between SNPs and CVD events among migraineurs indicated a slight excess of small p-values for some of the associations after controlling for modest inflation of the test statistic (λ_GC=1.03, Figure 1). We thus decided to further investigate SNPs suggestive of an association with p-values <5×10⁻⁶.

Five SNPs on different chromosomes showed associations with CVD events among women with migraine at the p<5×10⁻⁶ threshold suggesting increased risks for carriers of the minor alleles (Table 2). Four of the associations were seen among women with migraine with aura: two SNPs suggested an increased risk for ischemic stroke (rs7698623 [Figure S1], odds ratio [OR]=6.37, 95% confidence interval [CI] 3.15–12.90, p=2.7×10⁻⁷; rs4975709 [Figure S2], OR=5.06, 95% CI 2.66–9.62, p=7.7×10⁻⁷), one for major CVD (rs2143678 [Figure S3], OR=3.05, 95% CI 1.98–4.69, p=4.3×10⁻⁷), and one for CVD death (rs1406961 [Figure S4], OR=12.33, 95% CI 4.62–32.87, p=5.2×10⁻⁷). Further, one SNP suggested an increased risk for CVD death among women with any migraine (rs1047964 [Figure S5], OR=4.67, 95% CI 2.53–8.62, p=8.0×10⁻⁷). All ORs are large compared with effect estimates for CVD from studies in populations not selected for migraine, which for example report of a doubling of the relative risk for ischemic stroke among migraineurs, particularly migraineurs with aura [5]. Further, our observed ORs need to be interpreted in terms of the rather low minor allele frequencies (MAF) for some of the SNPs (Table 2). The significance and magnitude of the associations were very similar in the multivariable-adjusted additive models (Table 2).

In addition to our model comparing migraineurs with CVD to migraineurs without CVD, we also explored a model comparing migraineurs with CVD to controls free of migraine and CVD (n=17,247). The top hits were the same for the associations between migraine with aura and ischemic stroke, migraine with aura and major CVD, and any migraine and CVD death (data not shown). For the association between migraine with aura and CVD death the SNP on chromosome 20 (rs1406961) was only the second most significant one, while rs1047964 (see previous paragraph) was the top SNP.

Migraine, in particular migraine with aura, has been associated with CVD. The evidence was summarized in a recent meta-analysis indicating a two-fold increased risk for ischemic stroke, which only appears for migraine with aura [5]. The evidence is less clear for other CVD events; however, individual studies suggest that this pattern may also extend to MI, CVD death, and hemorrhagic stroke [6], [7], [8], [9], [10],[11],[12],[13],[14]. The association between migraine and ischemic vascular events was independent of many CVD risk factors in many studies [7], [8]. Furthermore, the increased risk of ischemic stroke is more apparent among migraineurs without CVD risk factors [29], [30], [31]. In addition functional [4], cellular [32], and genetic data [33] indicate shared biological vulnerability between vascular disorders and migraine.

Investigating the genetic liability for CVD among migraineurs has only recently begun. Two analyses reported on the influence of single gene variants on the migraine-CVD association and showed a modulatory effect of the MTHFR 677C>T [15] and ACE D/I [16] polymorphisms among women. The two-fold increased risk of CVD among migraineurs with aura was further raised among those migraineurs with aura also carrying the MTHFR 677TT genotype. This was only seen for ischemic stroke and not for MI. With regard to the ACE D/I polymorphism, only migraineurs with aura carrying the DD/DI genotype seemed to be at increased risk for ischemic stroke and MI, but not carriers of the II genotype.

The approach of our study is novel in two different ways. First, we compared migraineurs with CVD events to migraineurs without CVD events in contrast to previous studies comparing migraineurs to non-migraineurs. This design eliminates a potential confounding effect of migraine with gene variants on CVD events and specifically addresses the question which gene variants are associated with CVD events among migraineurs. Second, our study investigates the effect of genetic markers among migraineurs on their risk for CVD events from a genome-wide perspective.

None of the investigated SNPs was significantly associated with CVD events among migraineurs at a genome-wide level. However, our analyses provide some indication of an association of five genetic variants with CVD events at a suggestive significance threshold. In addition, our results—similar to previous non-genetic studies—support the concept that the risk for CVD among migraineurs is not strongly confounded by classical CVD risk factors. First, results from the multivariable-adjusted models are very similar to age-adjusted models (Table 2), although CVD risk factors are more prevalent among migraineurs with CVD events than among those without CVD events (Table 1). This is likely due to a co-linearity between age and classical CVD risk factors. Second, none of the five implicated genetic variants has previously been reported to be associated with CVD risk factors or CVD events.

The potential function for the five variants with regard to an increased CVD risk among migraineurs is unclear. A summary of the known functional characteristics of these variants is presented in Table 3. However, it is important to note, that four of the SNPs were identified for the association between migraine with aura and CVD events and two of those specifically for the association with ischemic stroke. This agrees with data from observational studies suggesting a particularly strong link between migraine with aura and ischemic stroke [5].

Our study has several strengths, including CVD events being verified by an endpoints committee of physicians, detailed information on many potential CVD risk factors, and genetic information from 339,596 SNPs covering the whole genome. In addition, the homogenous nature of the cohort, consisting only of white women aged ≥45 years, may reduce confounding. For example, while there is no reason to assume that migraine pathophysiology is different between women and men, the phenotypic expression among migraineurs may differ by gender and age. This is suggested, for example, by data showing that the migraine-ischemic stroke association is greater among younger women than older women [5], which may depend on a changing pattern in cardiovascular risk profile [30].

However, several limitations should be considered. First, and most importantly, data to directly replicate our results are not available to verify our results or perform a meta-analysis. However, we believe that this should not hold back research, since our results may stimulate new research ideas and accelerate ongoing research. Second, we cannot exclude the possibility of spurious associations. The number of women with migraine experiencing a CVD event was limited (n=164) and the odds ratios for the five SNPs are rather large with effect sizes between 3 and 12 compared to usually much smaller odds ratios found in association studies for complex and heterogeneous disorder. For rs7698623, rs1047964, and rs1406961 this may also in part be due to the low MAFs (all <10%; Table 2). Hence, our data should be interpreted with caution. Third, migraine and aura status were self-reported and were not classified according to strict IHS criteria [2]. Although previous reports from the WHS have shown good agreement of our classification with modified IHS criteria from 1988 [8], and despite excellent agreement between self-reported migraine and migraine classification based on IHS criteria from 2004 in the WHS [20], non-differential misclassification is possible, which may in part explain some of our null findings. However, the prevalence of migraine (18.3%) and migraine aura (39.5% of migraineurs) is similar to those reported in other population-based studies [34], [35]. Finally, we only considered an additive genetic model of transmission. However, the vast majority of gene-phenotype associations follow an additive model, which would also have sufficient power to capture dominant modes of transmission. While recessive modes of transmission may be missed, we consider this a minor disadvantage because they are rare and mostly relevant for monogenic disorders.

Figure S1

Figure S2

Figure S3

Figure S4

Figure S5

Table S1