Sage Journals: Discover world-class research

Abstract

Observational studies have linked migraine to a wide range of risk factors and diseases, although the causality of these associations is uncertain. Mendelian randomization (MR) is an analytic paradigm that exploits germline genetic variants as natural experiments, providing a framework for causal inference from observational data. In this narrative review, we summarize the core principles and assumptions of MR and highlight methodological considerations unique to its application to migraine research. We review key MR findings across several domains of migraine research, noting areas of consistent evidence, as well as conflicting results. We then discuss limitations of MR in the context of migraine research, emphasizing that MR is not a panacea for causal inference. We conclude by underscoring the need for interdisciplinary collaboration between methodologists and domain experts to prioritize patient-relevant questions, refine analytic approaches and translate genetic insights into more effective therapies.

Graphical abstract

This is a visual representation of the abstract.

Keywords

GWAS Mendelian randomization migraine

Introduction

Migraine is a prevalent and leading cause of disability worldwide (1). To mitigate this disease burden, epidemiologic research has sought to identify associations between risk factors and migraine, including lifestyle behaviors (2), circulating biomarkers (3) and other medical conditions (4). Conversely, migraine itself has been examined as a risk factor for other adverse clinical outcomes. Among the most prominent of these are associations with greater risk of cardiovascular disease (5) to the extent that migraine is now included in the QRISK3 cardiovascular risk prediction algorithm (6).

Although numerous risk factors for migraine have been identified in observational studies, their causal contributions to disease risk remain uncertain. This uncertainty stems from the fact that, although traditional observational methods can identify associations, they are subject to inherent limitations that limit causal inference. Consequently, therapeutic strategies informed by observational studies are often not validated by subsequent randomized controlled trials, which constitute the gold standard for establishing causality (7). Residual confounding and reverse causality are among the most important biases that impact reliable inference in observational studies (Figure 1). Residual confounding arises when an unmeasured or inaccurately measured variable influences both the exposure and the outcome, leading to biased effect estimates. Reverse causality occurs when an apparent link association of an exposure with an outcome is actually driven by an unrecognized effect of the outcome on the exposure. This is particularly problematic for investigations of migraine, where the disorder itself alters behaviors such as diet, sleep and physical activity. Prospective, controlled studies can partially mitigate biases arising from confounding and reverse causality, although they cannot eliminate them entirely. These challenges highlight the need for alternative study designs for causal inference that avoid these biases.

Figure 1.

Directed acyclic graph demonstrating the relationships investigated in a Mendelian randomization (MR) analysis. The β symbols refer to the quantitative measure of association: β_O is the association of exposure with outcome in observational epidemiology; β_GE is the association of the genetic instrument with the exposure; β_GO is the association of the genetic instrument with the outcome. Solid arrows denote direct relationships. Direct relationships between the instrument and the outcome are biologically possible, but are considered pleiotropic and violate the assumptions of MR. Instead, the validity of MR requires that the effect of the instrument on the outcome should be mediated solely by the effect of the instrument on the exposure, The dashed line indicates the typical lack of confounding and reverse causation in genetic association that accounts for population structure, as emphasized by the “X” symbol. GWAS = genome-wide association studies; SNP = single nucleotide polymorphism.

Natural experiments have been applied across multiple research domains to strengthen causal inference in observational settings (8). These designs leverage scenarios in which exposure to a risk factor or intervention is driven by an external factor that results in approximate random allocation. By mimicking certain features of randomized controlled trials, natural experiments can identify effects that are less susceptible to conventional biases of observational research(9).

Human genetic variation provides a naturally randomized source of biological variation that can be leveraged under the natural experiment framework (10). Germline genetic variants are randomly assorted at gametogenesis and hence, under certain assumptions, are immune to environmental confounding. Moreover, germline genetic variants cannot be modified by disease processes, and therefore remain nearly unaffected by reverse causation. Mendelian randomization (MR) is an analytic technique that capitalizes on these properties by using genetic variants as proxies for exposures to infer causal effects from observational data (10,11). MR has been increasingly applied to investigate causal relationships in neurological diseases, including migraine (12). The first such application in migraine, published in 2017, evaluated the effect of serum calcium levels on migraine risk (13). Since then, the field has expanded rapidly: the number of published manuscripts using MR increased from 7 in 2020 to 57 in 2024 alone. However, the MR methodology is often inappropriately applied or interpreted, resulting in confusing and inconsistent findings (14). This highlights the importance of rigorous methodology and cautious interpretation to avoid unwarranted causal claims from MR analyses.

In this review, we provide a primer for understanding the application of MR to migraine research. We first outline the core assumptions and analytical framework of MR, and then summarize key findings from studies examining the causes and consequences of migraine. Finally, we discuss methodological challenges unique to migraine research and highlight priorities for future work.

Mendelian randomization methodology

Overview

In an MR analysis, genetic variants, typically common (allele frequency >1%) single nucleotide polymorphisms (SNPs), are used as proxies for an exposure to estimate its effect on an outcome. In the language of instrumental variable analysis, these variants are described as instruments for the exposure. Unlike studies focused on disease risk prediction or discovery of risk loci, MR uses these variants solely as tools for causal inference to support or refute a causal effect.

For valid causal inference in MR, the genetic variants used as instruments must meet three core assumptions. First, they must be strongly associated with the exposure. Investigators typically select variants that associate with the exposure at genome-wide statistical significance (p < 5 × 10⁻⁸), a threshold which controls for hypothesis testing across millions of variants. More lenient thresholds are sometimes used to increase the number of variants in an MR analysis, although this comes at the cost of including invalid instruments and introducing bias (15). These genetic associations are drawn from the summary statistics of genome-wide association studies (GWAS), which are now publicly available for thousands of traits. Summary-level data can also be used to obtain estimates of genetic associations with the outcome from a non-overlapping sample, in what is known as a two-sample MR design. By contrast, one-sample MR, which uses individual-level data from a single cohort, is less commonly applied due to limitations in data availability and potential bias from sample overlap (16).

The second MR assumption is that the association of the genetic variants with the outcome must be independent of confounders. An important source of confounding in this context is population stratification, in which genetic ancestry or population substructure correlate with variant allele frequencies and are in turn associated with the outcome. This potential bias may be mitigated through several analytic strategies, including restriction of analyses to homogeneous ancestry groups and application of statistical methods that adjust for genetic ancestry.

The final, and most stringent, analytic assumption is that the genetic variants used in the MR analysis impact the outcome solely through their effects on the exposure. Variants that exert effects on the outcome via other pathways are considered pleiotropic, and their inclusion in an MR analysis can bias results. The development of statistical methods to detect and mitigate this bias has been a central priority in methodological development for MR (17,18). Nevertheless, violations of the second and third MR assumptions are impossible to rule out entirely and represent inherent limitations of the approach.

Calculation of MR estimates

Statistical methods for MR are implemented in multiple publicly available software packages and are only briefly reviewed here (19,20). At the population level, a genetic instrument is interpreted as inducing differences in the mean level or risk of the exposure according to which alleles are inherited. Under a causal framework, this change in the exposure is expected to induce a consequent difference in the mean level or risk of the outcome. The Wald ratio quantifies this effect by dividing the variant's association with the outcome by its association with the exposure, providing an estimate of how much the outcome changes per unit difference in the exposure (Figure 1). In MR analyses involving multiple variants, individual Wald ratios may be combined using meta-analysis, with standard errors estimated using the delta method. Implicit in the interpretation of MR associations is the assumption that an exposure affects the outcome through a similar mechanism regardless of whether its variation arises from genetic or environmental factors (21).

GWAS data sources for migraine

Table 1 summarizes some GWAS data sources that have been used in two-sample MR analyses of migraine. The first migraine GWAS, published in 2010 by the International Headache Genetics Consortium (IHGC), included 5933 cases and identified a single genome-wide significant locus, although this association has not been replicated in subsequent studies (22). The following year, a second migraine GWAS reported three loci that have since been consistently replicated (23). Successive meta-analyses have rapidly increased case counts to 102,084 and identified 123 loci, including variants near genes that encode drug targets for migraine (e.g. HTR1F and CALCA/CALCB) (24,25).The most recent GWAS includes the largest sample sizes to date for the subtypes of migraine with aura (MA) and migraine without aura (MO), though statistical power remains limited for these outcomes. In parallel, the public release of summary statistics for a wide range of metabolic, lifestyle, cardiovascular and molecular traits has enabled efficient application of two-sample MR to investigate potential causal risk factors for migraine.

Table 1.

Genome-wide association study (GWAS) data sources for migraine research.

Phenotype	Cohort	Sample size (cases / controls)	Web link	Notes
Migraine	IHGC (24)	102,084 / 771,257	https://www.headachegenetics.org/datasets-cohorts	Summary statistics not available for migraine subtypes
Migraine	Meta-analysis of 7 cohorts (25)	79,495 / 1,259,808	https://www.decode.com/summarydata/	Summary statistics also available for migraine subtypes
Coronary artery disease	CARDIoGRAMplusC4D Consortium (26)	181,522 / 1,165,690	https://www.ebi.ac.uk/gwas/studies/GCST90132314
Ischemic stroke	GIGASTROKE (27)	110,182 / 1,503,898	https://www.ebi.ac.uk/gwas/publications/36180795	Includes data on ischemic stroke subtypes: small vessel, large-artery atherosclerotic and cardioembolic stroke
Wide range of binary and continuous variables	UK Biobank (analyzed by Yang lab) (28,29)	Up to 456,422 participants	https://yanglab.westlake.edu.cn/data/ukb_fastgwa/imp/	Includes anthropometric measures, disease risk factors and disease outcomes
Disease outcomes	FinnGen(29)	Up to 500,348 participants	https://www.finngen.fi/en/access_results	Includes disease outcomes

Only the two largest migraine GWAS were included in the table. GWAS = genome-wide association studies; IHGC = International Headache Genetics Consortium.

Mendelian randomization studies of the causes and consequences of migraine

We now provide a non-exhaustive overview of MR analyses that have leveraged these GWAS data sources to investigate the causes and consequences of migraine. In view of the large number of tested hypotheses, we focus on findings that surpassed study-specific statistical significance thresholds. Throughout, we highlight consistencies in findings at the same time as noting limitations that merit further investigation.

Vascular disease

For several decades, migraine was thought to be primarily initiated by vasodilation of meningeal and cerebral arteries, although this “vascular theory” of migraine has since been challenged by experimental physiological evidence (30,31). An alternative approach to investigating a vascular contribution to migraine pathophysiology would be to demonstrate evidence for associations between migraine and cardiovascular disease risk. Indeed, several large epidemiological studies have demonstrated independent associations between migraine and increased risk of incident stroke, coronary artery disease (CAD) and atrial fibrillation (6,32,33). These associations have been particularly pronounced for the subtype of migraine with aura. Whether these relationships are causal remains uncertain.

We have performed several MR analyses testing the associations between migraine and vascular diseases (Figure 2) (^34–37). In one of the earliest MR analyses applied to migraine, we investigated the association of genetic liability to migraine with risk of CAD and atrial fibrillation (36). In contrast to findings from observational studies, these analyses found a protective association between genetic liability to migraine and CAD risk, as well as no association with atrial fibrillation. These results raise the possibility that the observational association between migraine and increased CAD risk may be biased. Moreover, recent epidemiological studies have shown that individuals at high cardiovascular risk are less likely to have active migraine or develop incident migraine (38,39). The mechanisms underlying the protective effect warrant further investigation, and may involve developmental differences in the vasculature of patients with migraine, or differences in protective lifestyle behaviors in migraine patients such as avoidance of alcohol (36).

Figure 2.

Flowchart for conduct of Mendelian randomization (MR) analysis. GWAS = genome-wide association studies.

In a subsequent MR analysis, we investigated the association of genetic liability to migraine with ischemic stroke subtypes and cervical artery dissection (35). We identified an inverse association between genetic liability to migraine and large-artery atherosclerotic stroke, a stroke subtype that shares pathophysiological features with CAD. This finding was replicated by an independent research group analyzing the same datasets (40). There was no evidence for association with cardioembolic stroke, nor was there an association with small vessel stroke. However, genetic liability to migraine was associated with increased risk of cervical artery dissection, which is consistent with findings from observational research (41).

As another form of cerebrovascular disease, we investigated genetic liability to migraine in relation to intracranial aneurysm risk and found an association with increased risk of intracranial aneurysm and subarachnoid hemorrhage (34). Parallel analyses of abdominal and thoracic aortic aneurysms yielded similar associations, further supporting an influence of migraine liability on both intracranial and extracranial vasculature (34). Reverse MR analyses showed no association of genetically proxied intracranial aneurysm risk with migraine. This suggests that, on average, intracranial aneurysms do not cause headache, although we cannot exclude effects of aneurysm subtypes that are poorly represented in the studied populations (e.g. giant or rapidly growing aneurysms). These findings collectively suggest that genetic liability to migraine protects against atherosclerotic disease while predisposing to diseases of arterial wall elasticity and structural integrity (Figure 2).

MR has also been applied to investigate causal relationships between cardiovascular risk factors and migraine. We found evidence for a causal effect of higher diastolic blood pressure (DBP) adjusted for systolic blood pressure (SBP) on increased risk of migraine (37). This association was also reported in two independent MR analyses (42). Conversely, we found that genetically proxied higher pulse pressure and higher SBP adjusted for DBP were associated with reduced risk of migraine. In the reverse causal direction, genetic liability to migraine was associated with lower SBP and lower pulse pressure, but not with DBP. These findings provided further evidence supporting a vascular contribution to migraine risk.

In contrast, several cardiovascular risk factors have not been found to relate to migraine risk in MR analyses. Our group found no evidence from MR analyses for causal associations between migraine and lipid subfractions (43). In unidirectional MR analyses, Siewert et al. (42) also found no evidence for an association between genetically proxied low-density lipoprotein cholesterol or type 2 diabetes and risk of migraine. In a subsequent MR analysis that leveraged a larger migraine GWAS, Islam et al. (44) also detected no significant causal effects of glycemic traits on migraine risk, nor any reverse effects.

Taken together, these genetic findings support a vascular contribution to migraine pathophysiology. An emerging hypothesis is that migraine may serve as an indicator of arterial health, reflecting a tradeoff between increased susceptibility to disorders of vessel wall elasticity and structural integrity and reduced risk of atherosclerotic disease. Further research aimed at elucidating these mechanisms may reveal new therapeutic opportunities for both migraine and vascular disease.

Gastrointestinal diseases

Gastrointestinal symptoms are common in migraine, with emesis and nausea included in the International Classification of Headache Disorders criteria for migraine (45). Additionally, gastrointestinal diseases have long been recognized as comorbid with migraine (46). Several MR analyses have been conducted to investigate whether these associations may be causal. Welander et al. (47) investigated bidirectional relationships between genetic liability to migraine and risks of inflammatory bowel disease and celiac disease. They found no convincing evidence for causal associations in either direction. There was weak, nominal evidence suggesting an effect of genetic liability to celiac disease on migraine subtypes, although these findings did not survive correction for multiple testing (Figure 3).

Figure 3.

Divergent vascular effects of genetic liability to migraine. Red, pointed arrows indicate pathways that confer increased risk. Black arrows with a flathead indicate pathways that confer reduced risk. The “X” indicates the absence of evidence for a causal effect in either direction. Generated using the BioRender software ( https://www.biorender.com).

In a subsequent analysis using distinct analytic approaches, Chasman et al. (48) examined the bidirectional associations between genetic liability to migraine and gastrointestinal diseases. This study replicated the null association between genetic liability to migraine and inflammatory bowel disease. Although migraine and irritable bowel syndrome frequently co-occur, there was no evidence to support a causal relationship in either direction, suggesting that their comorbidity reflects shared etiological factors. By contrast, there was evidence for associations between genetic liability to several non-inflammatory gastrointestinal disorders (peptic ulcer disease, gastroesophageal reflux disease and diverticular disease) and migraine. However, in a random-effects inverse-variance weighted sensitivity analysis aiming to account for heterogeneity in SNP-specific MR estimates, only the association of diverticular disease with migraine remained statistically significant. Although brain–gut pathways may provide a broad conceptual framework for this finding, there is no strong a priori mechanistic explanation, particularly given the distinct tissue types involved in the pathogenesis of these disorders.

Cognition, dementia and neuroimaging phenotypes

Epidemiologic studies have reported associations between migraine and incident dementia, raising interest in a potential causal link (49). To evaluate this relationship, we conducted MR analyses examining the association of genetic liability to migraine with Alzheimer's disease (AD) and cognitive performance (50). We found no evidence in these analyses for a causal association. By contrast, subsequent MR analyses reported positive associations between genetic liability to migraine and AD risk (^51–53). However, methodological limitations complicate the interpretation of these findings. Several of these studies relied on migraine GWAS summary statistics from a single cohort (e.g. UK Biobank) to identify genetic proxies for migraine, capturing as few as 9% of the genetic instruments available in more recent, large-scale GWAS (52,53). Furthermore, these three studies utilized a GWAS of AD that included proxy-defined cases obtained using family history. This approach has been shown to be biased and can lead to misleading MR estimates (54). These limitations underscore the need for updated MR studies leveraging comprehensive genetic instruments and rigorously phenotyped AD cases.

The first MR investigation of magnetic resonance imaging (MRI)-derived brain volume phenotypes found no evidence for association (50). Three subsequent MR studies have investigated this relationship and yielded mixed results (Table 2). One of these studies reported no significant associations after applying Bonferroni correction for multiple comparisons (Table 2) (55). Zhang et al. (56) identified a protective association between thickness of the right superior parietal gyrus and reduced risk of migraine. Mitchell et al. (57) reported that increased size of the ventral diencephalon was associated with reduced migraine risk, and that genetic liability to migraine was linked to increased amygdala volume. Drawing definitive conclusions from these studies is challenging due to the lack of consistent or replicated findings, and methodological limitations in the application of MR to investigations of MRI-derived phenotypes (see Discussion). These results should therefore only be regarded as hypothesis-generating.

Table 2.

Comparison of Mendelian randomization studies examining the association of genetic liability to migraine with MRI-derived brain volume metrics.

Study	Migraine GWAS	Brain MRI GWAS	Brain MRI phenotypes	Identified associations
Daghlas et al. (50)	IHGC (2016) (58)	Enigma Consortium (2015) (59)	Intracranial volume and 9 subcortical brain volumes (including hippocampal volume)	Migraine → Brain volume: None Brain volume → Migraine: Not tested
Guo et al. (55)	IHGC (2022) (24)	Multiple GWAS (60 –63)	Hippocampal volume, cortical surface area and thickness, gray matter volume, white matter hyperintensities	Migraine → Brain volume: None Brain volume → Migraine: None
Zhang et al. (56)	FinnGen (2025) (29)	UK Biobank (63)	587 structural brain phenotypes	Migraine → Brain volume: None Brain volume → Migraine: Increased size of right superior parietal gyrus thickness associates with reduced risk of migraine
Mitchell et al. (57)	IHGC (2016) (58)	Meta-analysis of multiple cohorts including UK Biobank (57)	Intracranial volume and 9 subcortical brain volumes (including hippocampal volume)	Migraine → Brain volume: Association with smaller amygdala volume Brain volume → Migraine: Increased size of ventral diencephalon associates with reduced risk of migraine

The table only includes MR associations that were statistically significant after correction for multiple comparisons. GWAS = genome-wide association studies; IHGC = International Headache Genetics Consortium; MRI = magnetic resonance imaging.

Health-related behaviors

MR has been applied to investigate bidirectional relationships between migraine and health-related behaviors. Yuan et al. (64) investigated the bidirectional associations of genetically proxied alcohol consumption, smoking and coffee consumption on risk of migraine in the UK Biobank and FinnGen cohorts. Genetic liability to smoking initiation was positively associated with migraine risk, whereas coffee consumption showed a protective association. We extended this approach to sleep traits, finding significant evidence for an association between genetically proxied difficulty awakening (as a proxy for sleep inertia) with increased risk of migraine, and nominal evidence for an association between genetic liability to insomnia (65). This observation aligns with emerging clinical evidence that treatment of insomnia may reduce migraine frequency (66).

Molecular factors

Several targeted MR analyses have investigated the association of molecular factors with migraine risk. The first published MR study of migraine investigated the association of serum calcium levels with migraine risk, finding evidence that increased serum calcium increases migraine risk (13). However, this association was non-significant when examined in a subsequent, larger iteration of the IHGC migraine GWAS (42,67). Several circulating proteins have been investigated in isolation for their association with migraine, including insulin-like growth factor 1 (protective) (68), hypoxia-inducible factor 1-α (null) (69) and soluble Trem1 (nominal but non-significant after multiple comparisons adjustment) (70).

Several hypothesis-free MR investigations of molecular factors and migraine have been undertaken. In an analysis of 316 unique metabolites, Tanha et al. (71) identified potential causal effects between lower levels of X-11315, higher levels of LPE(20:4) and shorter fatty acid length on risk of migraine. They triangulated this finding with preliminary evidence from the migraine clinical literature suggesting benefit from supplementation with fish oil rich in very long-chain omega-3 fatty acids (71). Tanha et al. (72) also examined the association of 10 common laboratory-based tests, including lipids, iron status and liver enzyme tests, with migraine. The only association surpassing the Bonferroni-adjusted threshold was gamma-glutamyl transferase, although sensitivity analyses indicated the association was likely biased by horizontal pleiotropy.

Guo et al. (73) investigated twelve coagulation cascade-related measures and found that higher levels of coagulation factor VIII, von Willebrand factor levels and phosphorylated fibrinopeptide A levels were associated with increased risk of migraine. Stratified analyses suggested more pronounced effects for MA. Sensitivity analyses showed that the association of coagulation factor VIII with migraine was driven by genetic instruments near the pleiotropic ABO locus, and that independent associations between factor VIII and von Willebrand factor could not be distinguished (73). Although these findings hint at a connection between coagulation function and migraine, they rely on trans-acting variants located far from the genes encoding the relevant proteins, which makes pleiotropy more likely than with cis-acting variants situated near those genes (74,75). Of the proteins significantly associated with migraine in the primary analysis, only von Willebrand factor could be tested in sensitivity analyses restricted to cis-acting variants, which yielded effect estimates similar to the primary analysis, although with attenuated significance potentially due to reduced power (73).

Challenges and future directions

Conventional limitations of MR

All MR analyses are susceptible to bias from pleiotropic effects of variants that influence the outcome through pathways independent of the exposure (10). Although sensitivity analyses can provide some reassurance, they cannot fully exclude this potential bias. Additionally, genetic variants typically explain a modest proportion of variation in the exposure, which can limit statistical power. This limitation may be evaluated through power calculations or by examining the width of confidence intervals around MR estimates (76).

Interpreting effect estimates from MR analyses also requires attention to several additional assumptions and limitations (10). MR analyses estimate the impact of lifelong changes in an exposure on the outcome. In the case of the DBP analysis (37), the association represents the effect of a lifelong increase in DBP on migraine. This contrasts with randomized controlled trials which typically evaluate short-term effects of interventions. For exposures that exert a cumulative or time-dependent effect on an outcome, the MR estimate may be larger than the effect observed in interventional studies, although the effect directionality should be preserved. This may also impact the interpretation of biological effects that are age-dependent or emerge only during specific developmental windows, such as for MRI-derived brain volume measures. Additionally, conventional MR analyses assume linearity in the exposure-outcome relationship (77), which precludes assessment of non-linear effects and inference about extreme exposure levels. For example, findings from the MR analysis of DBP and migraine may not generalize to severe diastolic hypertension or hypotension.

In addition, the interpretation of MR analyses of binary exposures (e.g. migraine) is not straightforward. Such analyses do not estimate the effect of transitioning from the unexposed (e.g. migraine-free) to the exposed state (e.g. with migraine) but instead estimate the effect of genetic liability: a latent, continuous predisposition that influences the likelihood of developing the exposure. As an example of how this is expressed, in the MR analysis of migraine and subarachnoid hemorrhage, the MR estimate is reported as the odds ratio of subarachnoid hemorrhage per doubling in migraine risk (34). Point estimates from these analyses should not be directly compared with corresponding observational epidemiological estimates that use migraine status as an exposure in epidemiological studies. For these reasons, it has been proposed that MR estimates be interpreted primarily as a test of the causal null hypothesis, providing evidence about directionality rather than reliable estimates of effect size (21).

Challenges with MR analyses of neuroimaging features

MR studies of migraine and radiological outcomes have substantial analytic flexibility, including in the choice of GWAS sources for both migraine and brain volume traits. Because this flexibility allows many different analytic combinations and implicit hypothesis tests, it increases the risk of reporting false positive findings. Additionally, some volumetric measures encompass anatomically and functionally distinct subregions (e.g. thalamic volume), making it difficult to determine which components are specifically affected by migraine. Moreover, GWAS of neuroimaging features are often measured at a single time point per participant, which precludes determination of when during brain development a certain region may be affected by liability to migraine. Systematic approaches that rigorously control for false-positive rates and include independent replication would enhance the reliability of these findings.

Phenotyping challenges

GWAS of migraine subtypes have had limited sample sizes. MR analyses applied to these data are consequently underpowered, precluding assessment of subtype-specific causal effects. This limitation is particularly salient for MR analyses of vascular outcomes because MA has been selectively associated with cardiovascular disease in several epidemiological studies (78 –80). There may also be differences in causal effects of migraine on vascular disease according to age and sex, and delineating these differential effects requires stratified, individual-level analyses. However, the recent expansion of migraine GWAS sample sizes may soon permit well-powered MR investigations of subtype, age and sex-stratified effects (24).

Addressing questions that inform therapeutic development and clinical decision-making

To date, MR investigations of migraine have not yielded insights that have meaningfully improved patient care. Although this may reflect the relative nascency of the field, several study design-related factors may also contribute to the limited clinical impact. First, all studies referenced in this review relied on GWAS of migraine case–control status, and therefore primarily inform the biology of migraine susceptibility. While the pathophysiology underlying susceptibility is likely relevant to therapeutic strategies for individuals with established migraine, the genetic determinants of migraine severity may differ from those that drive disease onset (81 –84). MR analyses leveraging such outcomes may more directly generate insights that translate into therapeutic advances (85). Second, therapeutic insights may be accelerated through targeted application of the MR drug target paradigm to migraine. In this framework, the exposure is defined as perturbation of a protein drug target, allowing inferences that more directly inform drug development. This approach can be applied both to identify opportunities for drug repurposing and to evaluate the likely efficacy and safety of emerging targets (e.g. through phenome-wide MR analyses). Finally, confidence in MR findings can be strengthened through independent replication, which is becoming increasingly feasible with the widespread availability of GWAS data, particularly across diverse ancestries. Confidence can be further increased through triangulation using orthogonal forms of genetic analyses (e.g. rare variant analyses), as well as omics-based methodologies (e.g. proteomics).

Peer review

A thorough evaluation of findings from an MR analysis of migraine requires expertise in the disease process, genetic epidemiology, and the clinical context needed to prioritize relevant research questions. Identifying reviewers with this breadth of expertise is challenging, particularly with the increasing demands on the time of referees. While published recommendations and guidelines can assist in manuscript assessment, close collaboration between journals and experts in the field is also necessary to ensure the continued publication of high-quality studies (14). Taken together, these limitations highlight the importance of triangulating MR results with complementary methodologies, distinct data sources and the published literature (86).

Conclusions

MR has provided insights into the causes and consequences of migraine, although cautious interpretation of findings is essential. Cross-domain collaboration will be key to advancing the field and translating findings into tangible improvements in patient care.

Article highlights

Observational studies have linked migraine to a wide range of risk factors and diseases, though the causality of these associations is uncertain. MR is an analytic paradigm which exploits germline genetic variants as natural experiments, providing a framework for causal inference from observational data, and has been increasingly applied to investigate migraine and related disorders.

In this narrative review, we first summarize the core principles of MR and highlight methodological considerations unique to migraine research. We then synthesize key MR findings across several domains, noting areas of consistent evidence, as well as conflicting results. Finally, we discuss key limitations of MR in the context of migraine research, emphasizing that MR is not a panacea for causal inference.

We conclude by calling for closer collaboration between methodologists and disease domain experts to prioritize research questions, refine analytic approaches, and translate genetic insights into improved therapeutic strategies.

Footnotes

Author contributions

Iyas Daghlas: conceptualization, Investigation, Writing – original draft preparation.

Daniel I. Chasman: conceptualization, investigation, supervision, writing – original draft preparation, writing – review & editing.

Data availability

No new data were generated or analyzed in support of this research.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

Some of the research described in this review was supported by 1R21NS104398 from National Institute of Neurological Disorders and Stroke (NINDS)/NIH to DIC.

ORCID iD

Iyas Daghlas

References

Stovner

. Global, regional, and national burden of migraine and tension-type headache, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurol 2018; 17: 954–976.

Seng

Martin

Houle

. Lifestyle factors and migraine. Lancet Neurol 2022; 21: 911–921.

Tietjen

Khubchandani

. Vascular biomarkers in migraine. Cephalalgia 2015; 35: 95–117.

Scher

Bigal

Lipton

. Comorbidity of migraine. Curr Opin Neurol 2005; 18: 305–310.

Kalkman

Couturier

EGM

El Bouziani

, et al. Migraine and cardiovascular disease: what cardiologists should know. Eur Heart J 2023; 44: 2815–2828.

Hippisley-Cox

Coupland

Brindle

. Development and validation of QRISK3 risk prediction algorithms to estimate future risk of cardiovascular disease: prospective cohort study. Br Med J 2017; 357: j2099.

Smith

Phillips

. Correlation without a cause: an epidemiological odyssey. Int J Epidemiol 2020; 49: 4–14.

Craig

Cooper

Gunnell

, et al. Using natural experiments to evaluate population health interventions: new medical research council guidance. J Epidemiol Community Health (1978) 2012; 66: 1182–1186.

Baicker

Taubman

Allen

, et al. The Oregon experiment — effects of medicaid on clinical outcomes. N Engl J Med 2013; 368: 1713–1722.

10.

Davies

Holmes

Davey Smith

. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. Br Med J 2018; 362: k601.

11.

Smith

Ebrahim

. Mendelian randomization: prospects, potentials, and limitations. Int J Epidemiol 2004; 33: 30–42.

12.

Gagnon

Daghlas

Zagkos

, et al. Mendelian randomization applied to neurology. Neurology 2024; 102: e209128.

13.

Yin

Anttila

Siewert

, et al. Serum calcium and risk of migraine: a Mendelian randomization study. Hum Mol Genet 2016; 26: 820–828.

14.

Burgess

Woolf

Mason

, et al. Addressing the credibility crisis in Mendelian randomization. BMC Med 2024; 22: 74.

15.

Burgess

Thompson

. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol 2011; 40: 755–764.

16.

Burgess

Davies

Thompson

. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol 2016; 40: 597–608.

17.

Bowden

Smith

Burgess

. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol 2015; 44: 512–525.

18.

Bowden

Smith

Haycock

, , et al. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol 2016; 40: 304–314.

19.

Hemani

Zheng

Elsworth

, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife 2018; 7: e34408.

20.

Yavorska

Burgess

. Mendelianrandomization: an R package for performing Mendelian randomization analyses using summarized data. Int J Epidemiol 2017 1; 46: 1734–1739.

21.

Burgess

Labrecque

. Mendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates. Eur J Epidemiol 2018; 33: 947–952.

22.

Anttila

Stefansson

Kallela

, et al. Genome-wide association study of migraine implicates a common susceptibility variant on 8q22.1. Nat Genet 2010; 42: 869–873.

23.

Chasman

Schürks

Anttila

, et al. Genome-wide association study reveals three susceptibility loci for common migraine in the general population. Nat Genet 2011; 43: 695–698.

24.

Hautakangas

Winsvold

Ruotsalainen

, et al. Genome-wide analysis of 102,084 migraine cases identifies 123 risk loci and subtype-specific risk alleles. Nat Genet 2022; 54: 152–160.

25.

Bjornsdottir

Chalmer

Stefansdottir

, et al. Rare variants with large effects provide functional insights into the pathology of migraine subtypes, with and without aura. Nat Genet 2023; 55: 1843–1853.

26.

Aragam

Jiang

Goel

, et al. Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants. Nat Genet 2022; 54: 1803–1815.

27.

Mishra

Malik

Hachiya

, et al. Stroke genetics informs drug discovery and risk prediction across ancestries. Nature 2022; 611: 115–123.

28.

Jiang

Zheng

Fang

, et al. A generalized linear mixed model association tool for biobank-scale data. Nat Genet 2021; 53: 1616–1621.

29.

Kurki

M I

Karjalainen

Palta

, et al. FinnGen provides genetic insights from a well-phenotyped isolated population. Nature 2023; 613: 508–518.

30.

Jacobs

Dussor

. Neurovascular contributions to migraine: moving beyond vasodilation. Neuroscience 2016 3; 338: 130–144.

31.

Goadsby

. The vascular theory of migraine - A great story wrecked by the facts. Brain 2009; 132: –7.

32.

Adelborg

Szépligeti

Holland-Bill

, et al. Migraine and risk of cardiovascular diseases: danish population based matched cohort study. Br Med J 2018 31; 360: 96.

33.

Bigal

Kurth

Santanello

, et al. Migraine and cardiovascular disease. Neurology 2010; 74: 628–635.

34.

Daghlas

Rist

Chasman

. Genetically proxied liability to migraine and risk of intracranial aneurysm and subarachnoid hemorrhage. Headache 2025; 65: 391–398.

35.

Daghlas

Sargurupremraj

Danning

, et al. Migraine, Stroke, and Cervical Arterial Dissection: Shared Genetics for a Triad of Brain Disorders with Vascular Involvement. Neurol Genet 2022; 8: e653. DOI: 10.1212/NXG.0000000000000653

36.

Daghlas

Guo

Chasman

. Effect of genetic liability to migraine on coronary artery disease and atrial fibrillation: a Mendelian randomization study. Eur J Neurol 2020; 27: 550–556.

37.

Guo

Rist

Daghlas

, et al. A genome-wide cross-phenotype meta-analysis of the association of blood pressure with migraine. Nat Commun 2020; 11: 3368.

38.

Al-Hassany

MaassenVanDenBrink

Kurth

. Cardiovascular Risk Scores and Migraine Status. JAMA Netw Open 2024; 7: e2440577.

39.

Ibrahimi

Rist

Carpenet

, et al. Vascular risk score and associations with past, current, or future migraine in women: cohort study. Neurology 2022; 99: e1694–e1701.

40.

Lee

Bae

, et al. Exploring the causal inference of migraine on stroke: a Mendelian randomization study. Eur J Neurol 2022; 29: 335–338.

41.

Rist

Diener

Kurth

, et al. Migraine, migraine aura, and cervical artery dissection: a systematic review and meta-analysis. Cephalalgia 2011; 31: 886–896.

42.

Siewert

Klarin

Damrauer

, et al. Cross-trait analyses with migraine reveal widespread pleiotropy and suggest a vascular component to migraine headache. Int J Epidemiol 2020; 49: 1759.

43.

Guo

Daghlas

Gormley

, et al. Phenotypic and genotypic associations between migraine and lipoprotein subfractions. Neurology 2021; 97: e2223–e2235.

44.

Islam

Nyholt

. Cross-trait analyses identify shared genetics between migraine, headache, and glycemic traits, and a causal relationship with fasting proinsulin. Hum Genet 2023; 142: 1149–1172.

45.

The international classification of headache disorders, 3rd edition (beta version). Cephalalgia 2013; 33: 629–808.

46.

Van Hemert

Breedveld

Rovers

JMP

, et al. Migraine associated with gastrointestinal disorders: Review of the literature and clinical implications. Front Neurol 2014; 5: 241. DOI: 10.3389/fneur.2014.00241

47.

Welander

Rukh

Rask-Andersen

, et al. Migraine, inflammatory bowel disease and celiac disease: a Mendelian randomization study. Headache 2023; 63: 642–651.

48.

Chasman

Guo

Chan

, et al. Shared genetics of migraine and gastrointestinal disorders implicates underlying neurologic mechanisms yet heterogeneous etiologies. Neurol Genet 2024; 10: e200201. DOI: 10.1212/NXG.0000000000200201

49.

Wang

, et al. Meta-analysis of association between migraine and risk of dementia. Acta Neurol Scand 2022; 145: 87–93.

50.

Daghlas

Rist

Chasman

. Effect of genetic liability to migraine on cognition and brain volume: a Mendelian randomization study. Cephalalgia 2020; 40: 998–1002.

51.

Zhao

Tang

, et al. Genetic evidence for the causal relationships between migraine, dementia, and longitudinal brain atrophy. J Headache Pain 2024; 25: 93. DOI: 10.1186/s10194-024-01801-7

52.

Chen

Zhang

, et al. Genetic evidence for causal association between migraine and dementia: a Mendelian randomization study. BMC Med Genomics 2024; 17: 180. DOI: 10.1186/s12920-024-01956-x

53.

Fan

, et al. Independent causal effect of migraines on Alzheimer’s disease risk: a multivariate Mendelian randomization study. Front Neurol 2024; 15: 1401880. DOI: 10.3389/fneur.2024.1401880

54.

Sun

Zheng

, et al. Pervasive biases in proxy genome-wide association studies based on parental history of Alzheimer’s disease. Nat Genet 2024; 56: 2696–2703. DOI: 10.1038/s41588-024-01963-9

55.

Guo

Wang

Ying

, et al. Association between brain structures and migraine: A bidirectional Mendelian randomization study. Front Neurosci 2023; 17: 1148458. DOI: 10.3389/fnins.2023.1148458

56.

Zhang

Wang

Dong

, et al. Genetic influence of the brain imaging phenotypes, brain and cerebrospinal fluid metabolites and brain genes on migraine subtypes: a Mendelian randomization and multi-omics study. J Headache Pain 2025; 26: 124. DOI: 10.1186/s10194-025-02063-7

57.

Mitchell

Diaz-Torres

Bivol

, et al. Elucidating the relationship between migraine risk and brain structure using genetic data. Brain 2022; 145: 3214–3224.

58.

Gormley

Anttila

Winsvold

, et al. Meta-analysis of 375,000 individuals identifies 38 susceptibility loci for migraine. Nat Genet 2016; 48: 856–866.

59.

Hibar

Stein

Renteria

, et al. Common genetic variants influence human subcortical brain structures. Nature 2015; 520: 224–229.

60.

Hibar

Adams

HHH

Jahanshad

, et al. Novel genetic loci associated with hippocampal volume. Nat Commun 2017; 8: 13624.

61.

Grasby

Jahanshad

Painter

, et al. The genetic architecture of the human cerebral cortex. Science 2020; 367. DOI:https://doi.org/10.1126/science.aay6690

62.

Sargurupremraj

Suzuki

Jian

, et al. Cerebral small vessel disease genomics and its implications across the lifespan. Nat Commun 11. Epub ahead of print 1 December 2020. DOI: 10.1038/s41467-020-19111-2.

63.

Smith

Douaud

Chen

, et al. An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. Nat Neurosci 2021; 24: 737–745.

64.

Yuan

Daghlas

Larsson

. Alcohol, coffee consumption, and smoking in relation to migraine: a bidirectional Mendelian randomization study. Pain 2022; 163: e342–e348.

65.

Daghlas

Vgontzas

Guo

, et al. Habitual sleep disturbances and migraine: a Mendelian randomization study. Ann Clin Transl Neurol 2020; 7: 2370–2380.

66.

Smitherman

Kuka

Calhoun

, et al. Cognitive-behavioral therapy for insomnia to reduce chronic migraine: a sequential Bayesian analysis. Headache 2018; 58: 1052–1059.

67.

Yang

Duan

, et al. Insights into modifiable risk factors of migraine: a Mendelian randomization analysis. Neurol Res 2025; 47: 844–863.

68.

Abuduxukuer

Niu

Guo

, et al. Circulating insulin-like growth factor 1 levels and migraine risk: a Mendelian randomization study. Neurol Ther 2022; 11: 1677–1689.

69.

Liang

Wang

, et al. Association between hypoxia-inducible factor-1α and neurological diseases: a bidirectional two-sample mendelian randomization analysis. Brain Behav 2025; 15: e70398. DOI: 10.1002/brb3.70398

70.

Shi

Wei

, et al. The associations between plasma soluble Trem1 and neurological diseases: a Mendelian randomization study. J Neuroinflammation 2022; 19: 18.

71.

Tanha

Sathyanarayanan

, International Headache Genetics Consortium , et al. Genetic overlap and causality between blood metabolites and migraine. Am J Hum Genet 2021; 108: 2086–2098.

72.

Tanha

Martin

Whitfield

, et al. Association and genetic overlap between clinical chemistry tests and migraine. Cephalalgia 2021; 41: 1208–1221.

73.

Guo

Rist

Sabater-Lleal

, et al. Association between hemostatic profile and migraine: a Mendelian randomization analysis. Neurology 2021; 96: e2481–e2487.

74.

Daghlas

Gill

. Mendelian randomization as a tool to inform drug development using human genetics. Camb Prism Precis Med 2023; 1: 16.

75.

Daghlas

Karhunen

Kim

, et al. Application of human genetics to prioritize coagulation cascade protein targets for ischemic stroke prevention. Stroke 2025; 56: 1542–1553.

76.

Brion

M-JA

Shakhbazov

Visscher

. Calculating statistical power in Mendelian randomization studies. Int J Epidemiol 2013; 42: 1497–1501.

77.

Tian

Mason

Liu

, et al. Relaxing parametric assumptions for non-linear Mendelian randomization using a doubly-ranked stratification method. PLoS Genet 2023; 19: e1010823.

78.

Androulakis

Kodumuri

Giamberardino

, et al. Ischemic stroke subtypes and migraine with visual aura in the ARIC study. Neurology 2016; 87: 2527–2532.

79.

Kurth

Gaziano

Cook

, et al. Migraine and risk of cardiovascular disease in women. JAMA 2006; 296: 283–291.

80.

Sen

Androulakis

Duda

, et al. Migraine with visual aura is a risk factor for incident atrial fibrillation. Neurology 2018; 91: e2202–e2210. DOI: https://doi.org/10.1212/WNL.0000000000006650

81.

Kogelman

LJA

Esserlind

A-L

Francke Christensen

, et al. Migraine polygenic risk score associates with efficacy of migraine-specific drugs. Neurol Genet 2019 24; 5: e364.

82.

Gormley

Kurki

Hiekkala

, et al. Common variant burden contributes to the familial aggregation of migraine in 1,589 families. Neuron 2018; 98: 743–753.e4.

83.

Kossowsky

Schuler

Giulianini

, et al. Association of genetic variants with migraine subclassified by clinical symptoms in adult females. Front Neurol 2021; 11. Epub ahead of print 12 February 2021. DOI: 10.3389/fneur.2020.617472

84.

Chasman

Anttila

Buring

, et al. Selectivity in genetic association with sub-classified migraine in women. PLoS Genet 2014; 10: e1004366.

85.

Paternoster

Tilling

Davey Smith

. Genetic epidemiology and Mendelian randomization for informing disease therapeutics: conceptual and methodological challenges. PLoS Genet 2017; 13: e1006944.

86.

Lawlor

Tilling

Smith

. Triangulation in aetiological epidemiology. Int J Epidemiol 2016; 45: 1866–1886.

Mendelian randomization applied to migraine research: Opportunities and challenges

Abstract

Keywords

Introduction

Mendelian randomization methodology

Overview

Calculation of MR estimates

GWAS data sources for migraine

Mendelian randomization studies of the causes and consequences of migraine

Vascular disease

Gastrointestinal diseases

Cognition, dementia and neuroimaging phenotypes

Health-related behaviors

Molecular factors

Challenges and future directions

Conventional limitations of MR

Challenges with MR analyses of neuroimaging features

Phenotyping challenges

Addressing questions that inform therapeutic development and clinical decision-making

Peer review

Conclusions

Article highlights

Footnotes

Author contributions

Data availability

Declaration of conflicting interests

Funding

ORCID iD

References