What this is
- This research investigates the genetic and spatial differences between two migraine subtypes: () and ().
- Using a multi-omics approach, the study integrates genome-wide association studies () with single-cell transcriptomics to identify distinct genetic risk loci and tissue-specific mechanisms.
- The findings reveal that is associated with neuroimmune dysregulation, while is linked to vascular and metabolic pathways.
Essence
- The study identifies distinct genetic and spatial mechanisms underlying migraine heterogeneity, revealing that is driven by neuroimmune-epigenetic dysregulation and by vascular-metabolic perturbations.
Key takeaways
- Strong genetic correlations exist between and , with notable differences in their functional architectures. is enriched in conserved regulatory elements, while shows prominence in vascular pathways.
- Single-cell transcriptomics revealed spatially distinct niches for and . is enriched in neural crest-derived tissues and hypothalamic microglial areas, whereas shows a peripheral focus in vascular smooth muscle and gut-brain interfaces.
- High-confidence genes associated with migraine subtypes were identified, including LRP1 and PHACTR1, which may serve as targets for subtype-specific therapies.
Caveats
- The study's findings are based on data primarily from European ancestry populations, which may limit the generalizability of the results to other ethnic groups.
- The reliance on post-mortem expression profiles for transcriptomic data could introduce biases related to tissue degradation or confounding factors.
- Functional validation of the identified genes is necessary to confirm their roles in migraine pathogenesis and therapeutic potential.
Definitions
- Migraine with aura (MA): A subtype of migraine characterized by reversible neurological symptoms preceding headache, such as visual disturbances.
- Migraine without aura (MO): A subtype of migraine that manifests as unilateral pulsatile pain without preceding neurological symptoms.
- Genome-wide association study (GWAS): An observational study design that looks for associations between genetic variants and traits across the genome.
AI simplified
Introduction
Migraine is a primary neurological disorder characterized by recurrent episodes of moderate-to-severe throbbing headache, commonly accompanied by nausea, vomiting, photophobia, and phonophobia. Its pathophysiology involves dysregulation of the trigeminovascular system, neurogenic inflammation (e.g., upregulated CGRP signaling), and cortical spreading depression (CSD)-induced electrophysiological disturbances [1β3]. CSD, a wave of neuronal and glial depolarization, not only underlies migraine with aura (MA) but also exacerbates nociceptive transmission by releasing inflammatory mediators (e.g., ATP, potassium ions) that activate trigeminal nerve terminals [4, 5]. According to the International Classification of Headache Disorders, 3rd edition (ICHD-3), migraine is categorized into MA and MO. MA is defined by reversible neurological symptoms (e.g., visual scotomas, sensory paresthesia, or dysphasia) preceding headache, whereas MO manifests as unilateral pulsatile pain aggravated by physical activity [6β8]. Epidemiological studies indicate that migraine affects approximately 14.7% of the global population, with a significantly higher prevalence in females (18.9%) than males (9.8%), potentially linked to estrogen-mediated modulation of trigeminovascular reactivity [9, 10]. Notably, MA patients exhibit an age-dependent increase in stroke risk, possibly attributable to CSD-induced blood-brain barrier disruption and platelet hyperreactivity [11, 12]. Current therapies, including acute agents (e.g., NSAIDs, triptans, CGRP receptor antagonists) and preventive treatments (e.g., Ξ²-blockers, antiepileptics, CGRP monoclonal antibodies), fail to achieve adequate relief in 30%β40% of patients. Moreover, MA and MO demonstrate divergent therapeutic responses (e.g., triptans exhibit a 19% lower response rate in MA patients compared to MO) that might due to the gaps in subtype-specific pathomechanisms([13β15]. Despite the transformative impact of CGRP-targeted therapies, long-term use may lead to antibody resistance or cardiovascular adverse effects (e.g., blood pressure fluctuations), with limited efficacy against CSD-related pathways [16β18]. Therefore, elucidating the molecular heterogeneity between MA and MO through identification of subtype-specific genetic risk loci and tissue-enriched signatures is critical for overcoming current therapeutic limitations.
Recent advances in functional genomics have propelled the development of integrative analytical frameworks that bridge genomic variation with molecular phenotypes. Transcriptome-wide association studies (TWAS) exemplify this approach by synthesizing expression quantitative trait loci (eQTL) datasets with genome-wide association study (GWAS) summary statistics, enabling systematic prioritization of trait-associated genes and mechanistic insights into gene regulatory networks [19]. Computational tools such as Multi-marker Analysis of GenoMic Annotation (MAGMA), which evaluates gene-level associations through SNP aggregation [20], and Functional Summary-based Imputation (FUSION), leveraging transcriptome prediction models to infer causal genes [21], have become cornerstones in this field. Beyond single-omics investigations, contemporary research increasingly adopts multi-omics convergence strategies, harmonizing data from transcriptomics, proteomics, and epigenomics to establish cross-dimensional validation. For instance, joint analysis of methylation quantitative trait loci (mQTL) and eQTL with GWAS signals allows triangulation of disease-associated variants across epigenetic regulation, transcriptional activity, and post-translational modifications [22]. Emerging methodologies further incorporate single-cell resolution profiles and spatial transcriptomics to dissect cell-type-specific regulatory cascades and tissue-microenvironment interactions [23]. Such integrative paradigms not only mitigate false-positive discoveries inherent to single-layer analyses but also illuminate context-dependent biological pathways, offering a multidimensional roadmap for biomarker discovery and therapeutic innovation [23]. By transcending traditional reductionist frameworks, these approaches address the polygenic complexity of human diseases, fostering precision medicine through systems-level elucidation of genotype-phenotype architecture.
This study aims to elucidate the genetic underpinnings of migraine by identifying risk genes through systematic genetic analyses, with a focus on delineating the polygenic risk architecture distinguishing MA and MO. By integrating genome-wide association data, transcriptomic profiles, and tissue-specific regulatory annotations, we seek to uncover subtype-specific susceptibility loci and their enriched biological pathways. Furthermore, we will characterize the molecular divergence between MA and MO, including tissue-specific risk gene expression patterns. Through integrative analyses of multi-omics datasets, this work aims to establish a subtype-stratified genetic framework, advancing precision diagnostics and targeted interventions for heterogeneous migraine populations.
Methods
We obtained GWAS summary statistics for overall migraine, MA, and MO from the GWAS Catalog and FinnGen R11. The overall migraine dataset was derived from seven high-quality studies, harmonized via inverse-variance weighted meta-analysis to enhance statistical power and reliability, while subtype-specific data (MA and MO) were extracted from the FinnGen R11 cohort (detailed metadata in Supplementary Table S1). Genetic correlations between overall migraine and MA/MO, as well as between MA and MO, were assessed using Linkage Disequilibrium Score Regression (LDSC) and High-Definition Likelihood (HDL) to validate the genetic coherence of these traits. Partitioned heritability analysis via LDSC was employed to quantify the contribution of distinct genomic regions (e.g., coding, regulatory elements) to subtype-specific heritability, coupled with tissue-specific enrichment analysis to delineate divergent tissue architectures underlying overall migraine, MA, and MO. Multi-omics integration included Summary Mendelian Randomization (SMR) for colocalizing eQTL and mQTL, alongside TWAS using FUSION, MAGMA, and Joint-Tissue Imputation Enhanced PrediXcan Analysis (JTI-PrediXcan). Finally, the Polygenic Priority Score (PoPS) algorithm was applied to rank candidate genes based on functional genomic annotations and pleiotropic evidence, prioritizing targets with robust subtype-specific associations. To validate tissue-enriched specificity in MA and MO, we integrated single-cell spatial transcriptomics (sc-ST) and GWAS data, leveraging embryonic murine ST data from the E16.5 stage (spanning 25 organs) to systematically map MA and MO-associated cellular patterns at single-cell resolution.
All analyses utilized R (R-4.4.1) and Python, with detailed software and methodological specifications provided in Supplementary Table S2. The flow chart is shown in Fig. 1.
Flow chart of our study. GWAS: Genome-Wide Association Study; mQTL: Methylation Quantitative Trait Loci; eQTL: Expression Quantitative Trait Loci; TWAS: Transcriptome-Wide Association Study; BSGS: Brisbane Systems Genetics Study; LBC: Lothian Birth Cohort; GTEx: Genotype-Tissue Expression Project; LDSC: Linkage Disequilibrium Score Regression; HDL: High-Definition Likelihood; S-LDSC: Sparse Linkage Disequilibrium Score Regression; LDSC-SEG: Linkage Disequilibrium Score Regression-Stratified by Expression of Genes; gsMap: genetically informed spatial mapping of cells for complex traits; SMR: Summary Mendelian Randomization; FDR: False Discovery Rate; HEIDI: Heterogeneity in Dependent Instruments Test; MAGMA: Multi-marker Analysis of GenoMic Annotation; JTI: Joint-Tissue Imputation; PrediXcan: Predicting Gene Expression; UTMOST: Unified Test for Molecular Signature; PoPS: Polygenic Priority Score; gnomAD: Genome Aggregation Database; pLI: Probability of Loss-of-Function Intolerance; ENCODE: Encyclopedia of DNA Elements; MA: Migraine with Aura; MO: Migraine without Aura; CACNA1A: Calcium Voltage-Gated Channel Subunit Alpha1 A; KLHDC8B: Kelch Domain Containing 8B; ACO2: Aconitase 2; BCAR1 (Breast Cancer Anti-Estrogen Resistance 1
Sources of methylation, gene expression data
The methylation quantitative trait loci (mQTL) reference panel was constructed from epigenomic profiling of peripheral whole blood specimens in two independent population-based cohorts: the Brisbane Systems Genetics Study (BSGS; n = 614) and the Lothian Birth Cohort (LBC; n = 1,366) [24, 25]. Transcriptomic regulation patterns were interrogated using cis-expression QTL (cis-eQTL) summary statistics from the eQTLGen Consortiumβs meta-analysis of 31,684 blood samples [26]. Comprehensive metadata, including cohort characteristics, phenotype harmonization protocols, and quality control pipelines for each dataset, are cataloged in Supplementary Table S3 alongside primary literature references.
Statistic analysis
Genetic correlation analysis
To assess the genetic coherence between overall migraine and its subtypes (MA/MO), we evaluated pairwise genetic correlations using LDSC and HDL. LDSC estimated genetic correlation coefficients (rg) by modeling the relationship between GWAS summary statistics and linkage disequilibrium (LD) patterns across the genome, with robustness to sample overlap and inflation from genome-wide significant loci [27]. To enhance precision, we complemented LDSC with HDL, a full-likelihood approach that minimizes approximation bias through iterative restricted maximum likelihood (REML) optimization [28]. These analyses collectively validated the genetic architecture consistency across migraine phenotypes, ensuring subtype stratification aligned with polygenic risk profiles.
Partitioned heritability
To dissect the subtype-specific heritability architecture of migraine, we implemented partitioned heritability analysis via Sparse Linkage Disequilibrium Score Regression (S-LDSC) [29]. SNPs were stratified according to functional genomic annotations (e.g., regulatory elements, coding regions), followed by computation of LD score metrics for each annotation stratum. These metrics were subsequently modeled to quantify annotation-specific heritability enrichment and their proportional contributions to the aggregate genetic correlations between overall migraine and MA/MO, as well as between MA and MO. This framework enabled systematic decomposition of shared versus subtype-restricted polygenic effects across functional genomic compartments, circumventing confounding from pleiotropic loci through annotation-specific variance partitioning.
Tissue specific enrichment of heritability
To delineate tissue-enriched signatures underlying migraine and its subtype heterogeneity, we performed tissue-specific heritability enrichment analyses leveraging LDSC-Stratified by Expression of Genes (LDSC-SEG) [30] and transcriptomic profiles from the Genotype-Tissue Expression (GTEx) project (v8). This framework quantified the tissue-level regulatory contributions of migraine risk loci by integrating stratified LD scores with eQTL data spanning 53 human tissues, including brain subregions and vascular systems. Comparative analyses across migraine subtypes (overall migraine, MA, MO) revealed divergent tissue architectures, particularly between MA and MO, through metrics of heritability enrichment and cross-tissue specificity indices. GTEx annotations enabled systematic prioritization of tissues where subtype-associated genes exhibit heightened regulatory activity, thereby mapping molecular heterogeneity to anatomically distinct pathophysiological pathways.
Integrated analysis of methylation, gene expression levels, TWAS validation, and pops score
To resolve the molecular heterogeneity underlying migraine and its subtypes (MA/MO), we implemented a convergent multi-omics framework integrating methylomic, transcriptomic evidence. Cross-dimensional validation was achieved through SMR applied to cis-acting methylation and expression quantitative trait loci, with variant selection restricted to 2-Mb cis-windows flanking gene bodies and genome-wide significant loci (p < 5 Γ 10β»βΈ). Population stratification was mitigated by excluding SNPs with minor allele frequency (MAF) discrepancies exceeding 0.3 across ancestral cohorts [31]. Pleiotropy was quantified via the HEIDI heterogeneity test, where associations exhibiting HEIDI heterogeneity p-values < 0.05 were flagged as non-causal and excluded, thereby strengthening causal inference [31]. To address multiple testing inflation while preserving sensitivity, we adopted false discovery rate (FDR) control (Benjamini-Hochberg procedure) rather than Bonferroni correction, maintaining an expected FDR threshold of 5% (FP/(FP + TP) β€ 0.05) [32, 33].
Complementary insights were derived through a tripartite TWAS strategy: [1] FUSION delineated spatially resolved gene-trait associations by synthesizing GWAS signals with multi-tissue expression predictors (GTEx v8) [21]; [2] JTI-PrediXcan enhanced causal inference robustness through cross-tissue regulatory modeling, improving expression prediction accuracy by integrating shared genetic architectures across tissues [34]; and [3] MAGMA identified enriched biological pathways (Gene Ontology, KEGG, Reactome) and cell-type-specific expression signatures via competitive gene-set analysis [20]. This integrative paradigm reconciled mechanistic hypotheses with statistical associations, leveraging FUSIONβs spatial precision, JTI-PrediXcanβs augmented power for weak cis-regulatory signals, and MAGMAβs systems-level annotation breadth to validate SMR candidates and nominate novel risk loci([20, 21]. A FDR-adjusted P-value threshold of < 0.05 was applied across all three methods to define statistical significance.
To functionally contextualize prioritized genes within migraine and its subtypes (MA/MO) pathways, we employed the Polygenic Priority Scoring (PoPS) algorithm. PoPS synthesizes multi-modal evidenceβincluding GWAS association strengths, cis-regulatory QTL effects, protein interactome topology (STRING v11), evolutionary constraint metrics (gnomAD pLI scores), and regulatory potential (ENCODE chromatin accessibility)βvia a gradient-boosted decision tree model [35]. Genes were ranked by composite scores reflecting convergence across orthogonal data layers, with subtype-specific weights assigned to MA/MO risk architectures (detailed in Supplementary Table S2) [35].
Tissue enrichment specificity of MA and MO in sc-ST
To investigate tissue-enriched specificity of MA and MO, we integrated sc-ST data with GWAS statistics, enabling systematic mapping of MA and MO-associated cellular patterns at single-cell resolution. Using a novel algorithm termed genetically informed spatial mapping of cells for complex traits (gsMap) [23], which combines cross-species analyses of mouse embryonic/brain tissues, macaque cerebral cortex, and human GWAS datasets, we successfully identified spatial distribution patterns of disease-associated cell populations. The key principle involves leveraging GWAS-derived trait-related genes to map their expression patterns onto spatially resolved cells, thereby evaluating cellular-level associations between specific anatomical regions and complex traits. We mapped MA and MO tissue-enriched specificity using ST profiles of mouse embryonic tissues at E16.5 (25 organs), establishing a single-cell-resolution atlas for MA and MO spatial pathogenesis. The methodology is described in Supplementary Table S2.
Results
Genetic Correlation Analysis
Our analyses revealed robust genetic correlations between migraine subtypes and the overarching migraine phenotype, as quantified by both LDSC and HDL methods (Table 1). Significant positive genetic correlations were observed between overall migraine and MA (LDSC: rg=0.571 Β± 0.061, P = 4.36 Γ 10β21; HDL: rg=0.669 Β± 0.095, P = 2.17 Γ 10β12), as well as between overall migraine and MO (LDSC: rg=0.687 Β± 0.059, P = 1.27 Γ 10β31; HDL: rg=0.898 Β± 0.084, P = 7.20 Γ 10β27). Notably, the genetic correlation between MA and MO was exceptionally strong (LDSC: rg=0.806 Β± 0.078, P = 3.32 Γ 10β25; HDL: rg=0.932 Β± 0.137, P = 1.14 Γ 10β11), suggesting substantial shared genetic architecture across subtypes.
| Trait pairs | LDSC | HDL | ||
|---|---|---|---|---|
| P | (SE)rg | P | (SE)rg | |
| Overall migraine & Migraine with aura | 4.36e-21 | 0.5707 (0.0606) | 2.17e-12 | 0.6694 (0.0953) |
| Overall migraine & Migraine without aura | 1.27e-31 | 0.6872 (0.0587) | 7.2e-27 | 0.8978 (0.0837) |
| Migraine with aura & Migraine without aura | 3.32e-25 | 0.8057 (0.0777) | 1.14e-11 | 0.9321 (0.1373) |
Partitioned heritability
Our S-LDSC analyses identified distinct functional genomic architectures across migraine subtypes. For overall migraine, SNPs demonstrated significant enrichment in 18 of 38 functional categories, with the strongest signals in MAF_Adj_LLD_AFRL2_0 (Enrichment=β150.87, p = 1.41 Γ 10β7), Backgrd_Selection_StatL2_0 (Enrichment = 1.24, p = 2.28 Γ 10β7), and Nucleotide_Diversity_10kbL2_0 (Enrichment = 0.86, p = 1.79 Γ 10β6). For MA, the top enriched categories were Backgrd_Selection_StatL2_0 (Enrichment = 1.38, p = 5.47 Γ 10β6), MAF_Adj_LLD_AFRL2_0 (Enrichment = β162.47, p = 2.19 Γ 10β4), and GERP.NSL2_0 (Enrichment = 1.74, p = 1.17 Γ 10β3). In contrast, MO showed prominent enrichment in GERP.NSL2_0 (Enrichment = 2.12, p = 1.04 Γ 10β6), FetalDHS_TrynkaL2_0 (Enrichment = 9.73, p = 3.56 Γ 10β5), and Nucleotide_Diversity_10kbL2_0 (Enrichment = 0.78, p = 8.26 Γ 10β5). As demonstrated in Fig. 2, despite originating from the same population, MA and MO exhibited divergent contributions of genomic functional annotations to their genetic correlations. When compared to the overall migraine phenotype, MA showed the strongest enrichment in Backgrd_Selection_StatL2_0, while MO was predominantly enriched in GERP.NSL2_0. Direct comparison between MA and MO revealed a more pronounced divergence: MA-specific signals peaked in MAF_Adj_LLD_AFRL2_0, whereas MO remained strongly associated with GERP.NSL2_0, underscoring subtype-specific regulatory architectures. The specific results are shown in Fig. 2 and Supplementary Table S4.
Partitioned Heritability Analysis of Genetic Correlations Between Overall Migraine and Subtypes Using LDSC. This figure illustrates the partitioned heritability contributions of distinct genomic components to the aggregate genetic correlations between overall migraine, migraine with aura (MA), and migraine without aura (MO), as quantified via Linkage Disequilibrium Score Regression (LDSC). The analysis delineates the proportion of heritability attributable to specific functional annotations (e.g., regulatory elements, conserved regions) for each subtype, highlighting subtype-specific and shared genetic architectures
Tissue specific enrichment of heritability
LDSC-SEG tissue enrichment analysis revealed distinct tissue-specific patterns across migraine subtypes. The specific results are shown in Fig. 3 and Supplementary Table S5. Overall migraine demonstrated significant enrichment (P < 0.05) in vascular and smooth muscle-related tissues, including Artery Coronary (Coefficient: 2.998 Γ 10β10, P = 0.0058), Artery Tibial (2.645 Γ 10β10, P = 0.0137), Artery Aorta (2.351 Γ 10β10, P = 0.0168), Uterus (2.238 Γ 10β10, P = 0.0146), and Fallopian Tube (1.913 Γ 10β10, P = 0.0401), underscoring vascular and reproductive tissue involvement. MA exhibited no statistically significant tissue enrichment at the predefined threshold (P < 0.05), though modest associations were observed in Artery Tibial (β3.525 Γ 10β10, P = 0.7255) and Cells Transformed Fibroblasts (7.401 Γ 10β10, P = 0.1569), suggesting potential neuroimmune interplay requiring further validation. In contrast, Migraine without aura (MO) showed significant vascular specificity in Artery Tibial (1.451 Γ 10β9, P = 0.0124) and Artery Coronary (1.315 Γ 10β9, P = 0.0366), alongside non-significant trends in peripheral tissues such as Skin Sun Exposed Lower Leg (6.988 Γ 10β10, P = 0.1258) and Colon Sigmoid (5.741 Γ 10β10P = 0.2017). Comparative analysis highlighted stark divergence: while Overall migraine and MO shared vascular enrichment (e.g., Artery Tibial), MA lacked coherent tissue signals. MO uniquely implicated peripheral vascular and epithelial tissues, whereas MA showed no significant neural or immune associations. These results emphasize MOβs peripheral vascular dysregulation and MAβs etiological complexity, underscoring the necessity of subtype-specific investigations to delineate their distinct biological mechanisms.
Bidirectional bar plot of tissue enrichment results (LDSC-SEG) across 53 tissues. The dashed line indicates the significance threshold (=β0.05). Positive bars represent tissues with enrichment for migraine risk, while negative bars reflect depletion. Tissues surpassing the threshold (<β0.05) are highlighted in bold. LDSC-SE: Linkage Disequilibrium Score Correction-Stratified Expression Gene P P
Integrated Analysis of Methylation, Gene Expression Levels, TWAS Validation, and PoPS Score
Our multi-omics findings are summarized as follows: Significant mQTL and eQTL SMR results are reported in Supplementary Tables S6 and S7, respectively. Transcriptome-wide associations from FUSION, MAGMA, and JTI-PrediXcan analyses are detailed in Supplementary Tables S8, S9 (with visualization in Fig. 4), and Supplementary Table S10. PoPS results for Overall Migraine, MA, and MO are provided in Supplementary Tables S11, S12, and S13, respectively. Integrating findings from LDSC-SEG and MAGMA tissue enrichment analyses revealed both convergent and subtype-specific biological pathways in migraine. Overall Migraine consistently implicated vascular tissues (e.g., coronary artery, tibial artery) across both methods, with MAGMA further highlighting reproductive tissues (uterus, fallopian tube), aligning with LDSC-SEGβs emphasis on vascular-smooth muscle involvement. MA showed limited coherence between methods: while LDSC-SEG identified no significant tissue enrichment, MAGMA suggested tentative associations with neural regions (basal ganglia, hypothalamus) and immune niches (cultured fibroblasts), hinting at unresolved neuroimmune mechanisms. In contrast, MO demonstrated robust consistency across methods, with both LDSC-SEG and MAGMA prioritizing vascular tissues (tibial artery) and gastrointestinal systems (transverse colon, gastroesophageal junction). Critically, MA and MO diverged in tissue specificity. MAβs lack of significant LDSC-SEG signals contrasted with MAGMAβs weak neural trends, suggesting central pathways may involve regulatory mechanisms not captured by genetic enrichment alone. Conversely, MOβs strong convergence across methods reinforced its association with peripheral vascular and epithelial dysregulation. These results underscore methodological complementarity: LDSC-SEG emphasizes broad genetic architecture, while MAGMA refines tissue-specific functional context. MAβs etiological complexity and MOβs peripheral focus highlight the necessity of multi-method approaches to dissect subtype heterogeneity.
Table 2 shows the results of comprehensive multi-omics screening of significant genes. The reference panel utilized in this study comprises genes that demonstrated positive associations with overall migraine susceptibility through comprehensive screening procedures. Integrative transcriptome-wide association analyses across migraine subtypes revealed a complex genetic architecture characterized by both subtype-specific and shared associations, with 68 genes achieving significance in at least two analytical methods. Overall migraine identified 52 genes, demonstrating the broadest genetic landscape, while MA and MO implicated 18 and 22 genes, respectively, reflecting their distinct pathobiological underpinnings. PHACTR1 (4 methods; PoPS Score Overall: 2.65, MA: 0.33, MO: 1.28) emerged as the most robust pleiotropic gene, implicated in vascular regulation across all subtypes. STAT6 (3 methods; PoPS Score Overall: 3.00, MA: 0.71, MO: 2.29) demonstrated consistent associations with immune modulation in both Overall migraine and MO, while also showing weaker links to MA. PRDM16 (3 methods; PoPS Score Overall: 4.76, MO: 1.44) was strongly prioritized in Overall migraine and MO, with high PoPS Scores reflecting its role in adipocyte differentiation and neuroprotection. LRP1 (3 methods; PoPS Score Overall: 3.67, MO: 0.80), a lipoprotein receptor critical for blood-brain barrier integrity, showed significant overlap between Overall migraine and MO. CFDP1 (3 methods; PoPS Score Overall: 0.33, MO: 0.42) and ITPKB (3 methods; PoPS Score Overall: 0.35, MO: 1.32) were recurrently linked to MO, with ITPKB further implicating inositol phosphate metabolism in synaptic signaling. Genes shared between Overall migraine and MA included RERG (3 methods; PoPS Score Overall: 0.50, MA: 0.94), associated with neuroinflammatory pathways, and CELSR3 (3 methods; PoPS Score Overall: 0.12, MA: 0.09), involved in planar cell polarity. CALCA (2 methods; PoPS Score Overall: 2.37, MO: 1.41), encoding calcitonin gene-related peptide (CGRP), was notable for its high PoPS Score in Overall migraine and MO, despite fewer methodological supports. REEP3 (3 methods; PoPS Score MA: 0.23, MO: 0.57) and ZBED4 (3 methods; PoPS Score Overall: 0.03, MA: 0.31, MO: 0.27) exhibited cross-subtype associations, though with modest effect sizes. Additionally, SLC26A6 (3 methods; PoPS Score Overall: 0.25, MO: β0.08) and CELSR3 highlighted roles in ion transport and cell adhesion, respectively. Genes with multi-trait and multi-method support also included RASGRF2 (3 methods; PoPS Score Overall: 0.57, MO: 0.54), a synaptic plasticity regulator, and UFL1 (3 methods; PoPS Score Overall: 0.14, MO: 0.87), involved in protein ubiquitination. Notably, TRIM26 (3 methods; PoPS Score Overall: β0.15, MA: 0.02, MO: 0.15) and TRIM40 (3 methods; PoPS Score Overall: β0.14, MA: 0.01, MO: β0.04) displayed subtype-specific regulatory effects despite lower PoPS Scores. CFDP1 and ITPKB further underscored MO-specific epigenetic and metabolic dysregulation. Genes with the highest methodological and phenotypic concordance included LRP1, PHACTR1, RDH16, STAT6, TTC24, ZBTB39, FHL5, MEF2D, NAB2, and UFL1, all supported by β₯ 2 methods and associated with multiple subtypes. PHACTR1, STAT6, and LRP1 merit particular attention, as they were implicated in Overall Migraine, MA, and MO, with methodological support from β₯ 2 approaches (up to 4 for PHACTR1). Despite lacking MA associations, PRDM16 demonstrated robust methodological concordance (β₯ 3 methods) in Overall Migraine. Notably, REEP3, though supported by only one method in Overall Migraine and MA, showed cross-subtype relevance with two methods in MO. Collectively, LRP1, PHACTR1, RDH16, STAT6, TTC24, ZBTB39, FHL5, MEF2D, NAB2, UFL1, and REEP3 emerged as high-confidence candidates due to their multi-method and multi-subtype consistency. Our analysis also identified distinct subtype-specific genetic profiles: MA-specific genes included CACNA1A and KLHDC8B, while MO-specific genes comprised ACO2, BCAR1, CCDC134, CFDP1, CHADL, GLCE, ITPKB, L3MBTL2, MED19, PAM, PPP2R5A, RHOF, SREBF2, TEF, TMEM170A, TNFRSF13C, and ZC3H7B. Methodologically, FUSION demonstrated the highest yield, identifying 62 genes across subtypes, with 48 in Overall migraine, 14 in MA, and 18 in MO, supported by robust multi-method validation (e.g., PHACTR1, STAT6). JTI-PrediXcan exhibited superior cross-tissue specificity, prioritizing genes like REEP3 (PoPS Score MA: 0.23, MO: 0.57) and STAT6 (PoPS Score MO: 2.29), validated across multiple methods. MAGMA, while primarily contributing to pathway enrichment, reinforced key loci such as CALCA (PoPS Score Overall: 2.37). Genes identified by β₯ 3 methods (e.g., PHACTR1, STAT6, LRP1) showed the highest reliability, with 85% associated with multiple traits, emphasizing their biological centrality. For instance, LRP1 (PoPS Score Overall: 3.67, MO: 0.80), a lipoprotein receptor involved in blood-brain barrier integrity, and CFDP1 (PoPS Score Overall: 0.33, MO: 0.42), a chromatin modifier, were recurrently linked to both Overall migraine and MO, suggesting convergent mechanisms in neurovascular and epigenetic regulation. These multi-method, multi-trait genes represent high-priority candidates for functional studies and therapeutic targeting, emphasizing the need to integrate cross-subtype analyses in migraine research.
MAGMA analysis results for overall migraine and its subtypes. From left to right: Manhattan plots of significant genes, bar plots of enriched pathways for significant genes, and bar plots of tissue enrichment results for significant genes. From top to bottom: analysis results for overall migraine, migraine with aura (MA), and migraine without aura (MO). MAGMA: Multi-marker Analysis of GenoMic Annotation
| Gene | Methods(Overall migraine) | MA | Methods(MA) | MO | Methods(MO) | PoPS Score (Overall migraine) | PoPS Score (MA) | PoPS Score (MO) |
|---|---|---|---|---|---|---|---|---|
| ADAMTSL4 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.2403722909175 | 0.0225216051005641 | 0.538046589464348 |
| C12orf4 | FUSION | Γ | NA | β | FUSION, JTI-PrediXcan | 0.290385658560395 | 0.182271555039095 | β0.095572564 |
| RDH16 | FUSION, JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.359786028524236 | 0.15239712350494 | 0.125319051952641 |
| TTC24 | FUSION, JTI-PrediXcan | β | JTI-PrediXcan | Γ | NA | β0.112578051 | 0.0903664465752124 | β0.036658648 |
| ZBTB39 | FUSION, JTI-PrediXcan | β | JTI-PrediXcan | Γ | NA | β0.023474569 | β0.101597819 | β0.185836871 |
| B9D2 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.365151042906335 | β0.022523441 | 0.0693930801751631 |
| CALCA | FUSION, MAGMA | Γ | NA | Γ | NA | 2.36973513232638 | 0.0864374882312646 | 1.41224311053889 |
| CSPG5 | FUSION, MAGMA | Γ | NA | Γ | NA | 1.04631306588356 | 0.586294556766031 | 0.147118031776585 |
| DHX30 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.224381852042872 | 0.159239728549263 | β0.233115176 |
| FHL5 | FUSION, MAGMA | Γ | NA | β | FUSION, MAGMA | 1.64468664258626 | 0.0327313463894807 | 0.376216783304532 |
| HHIPL1 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.314520642014664 | β0.269973021 | 0.117091828856953 |
| HJURP | FUSION, MAGMA | Γ | NA | Γ | NA | 0.801295049083255 | β0.177180613 | β0.101596149 |
| KTN1 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.541569481289006 | 0.0696303706283601 | β0.223006165 |
| LRP1 | FUSION, MAGMA | β | FUSION | β | FUSION, MAGMA, JTI-PrediXcan | 3.67314339988961 | 0.134803950405592 | 0.803977371715713 |
| MAP4 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.845551776264265 | β0.061472362 | β0.041466686 |
| NAB2 | FUSION, MAGMA | Γ | NA | β | FUSION | 1.20077407828595 | β0.078670971 | 0.943103770824176 |
| SFXN2 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.420981586948363 | 0.082919704924982 | 0.392160387675544 |
| SLC24A3 | FUSION, MAGMA | Γ | NA | Γ | NA | 0.837878120862327 | 0.100893017728755 | 0.665708322191937 |
| SUGCT | FUSION, MAGMA | Γ | NA | Γ | NA | 0.878587017841562 | 0.131659853992558 | 0.216142057032876 |
| UFL1 | FUSION, MAGMA | Γ | NA | β | FUSION, MAGMA, JTI-PrediXcan | 0.137318524376632 | β0.075747007 | 0.871789162394911 |
| MRGPRE | FUSION, MAGMA, eQTL | Γ | NA | Γ | NA | 0.623454352941175 | 0.105909178755406 | 0.193315828231861 |
| STAT6 | FUSION, MAGMA, JTI-PrediXcan | β | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | 3.00490427420246 | 0.707922348716691 | 2.29100240771294 |
| CALCB | FUSION, MAGMA, mQTL | Γ | NA | Γ | NA | 0.836950570707407 | 0.111723399679157 | 0.782220816061467 |
| MEF2D | FUSION, MAGMA, mQTL | Γ | NA | β | FUSION, JTI-PrediXcan | 1.8446832215119 | 0.103947682824509 | 1.24795266915079 |
| PRDM16 | FUSION, MAGMA, mQTL | Γ | NA | β | FUSION, MAGMA, mQTL | 4.76496934758302 | 0.229525850990023 | 1.43549514400552 |
| PHACTR1 | FUSION, MAGMA, mQTL, JTI-PrediXcan | β | FUSION, JTI-PrediXcan | β | FUSION, JTI-PrediXcan | 2.64863880801219 | 0.329033379461558 | 1.28491453226306 |
| TMEM51 | FUSION, mQTL | Γ | NA | Γ | NA | 0.324723002719683 | 0.0175358582031461 | β0.114450971 |
| ALG12 | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | β0.232168941 | 0.00221505422599431 | β0.138432492 |
| ANKK1 | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | β0.009960156 | β0.163077601 | 0.0184594591191178 |
| ARIH2OS | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.290381289112369 | 0.0407333716497951 | β0.012036367 |
| ARL6IP4 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.150787885327833 | β0.041809937 | 0.227140200882244 |
| C3orf62 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.200587778633348 | 0.146091349019112 | 0.00741614359066453 |
| CCDC36 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | NA | NA | NA |
| CCDC71 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.256585706910546 | 0.166309349327275 | β0.078357276 |
| CELSR3 | JTI-PrediXcan | β | FUSION, MAGMA, eQTL, JTI-PrediXcan | Γ | NA | 0.119999597557478 | 0.09429551597812 | 0.178045792245991 |
| CRELD2 | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | 0.381953248424051 | 0.140496210618467 | 0.0676003643226458 |
| DRC3 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.175710876496949 | 0.22497425544632 | 0.0145874102023276 |
| HIST1H2BD | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | NA | NA | NA |
| IP6K2 | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | 0.00707926381654044 | 0.325263806565343 | β0.308951157 |
| LINC02166 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | NA | NA | NA |
| MAEA | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | β0.296921421 | 0.991778979546547 | 0.00711687416957938 |
| NCKIPSD | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.170083332653109 | 0.163753771059098 | β0.085890517 |
| OGFOD2 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.233177314686005 | 0.107483481869043 | 0.213635996087082 |
| PIM3 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.275509582578445 | β0.01763925 | β0.249641656 |
| RASGRF2 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.571025712599019 | 0.612050252046555 | 0.539191146561724 |
| RERG | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | 0.50096582432266 | 0.94236482303448 | 0.897008803529776 |
| RPP21 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.124184224357975 | β0.023418647 | β0.476777546 |
| RRBP1 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | β0.265711307 | 0.437368887395571 | 0.243605543130937 |
| SLC26A6 | JTI-PrediXcan | β | FUSION, MAGMA, eQTL, JTI-PrediXcan | Γ | NA | 0.248037034603348 | 0.0326101601997732 | β0.075148727 |
| SMCR5 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | NA | NA | NA |
| SOX7 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | β0.235742387 | 0.161934197469218 | β0.171281885 |
| SPATA33 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | β0.020486223 | 0.0474121846803396 | β0.032791392 |
| SREBF1 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | β0.047492627 | 0.198028017303947 | β0.240782539 |
| TMEM89 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.0973901895964615 | β0.099338175 | β0.03698422 |
| TRIM10 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | β0.12645781 | β0.104819613 | β0.103203709 |
| TRIM26 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | β0.151431133 | 0.0192352055323646 | 0.149795816481255 |
| TRIM31 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.0208054653841703 | 0.143075880887615 | 0.117138106328655 |
| TRIM40 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | β0.13590979 | 0.00663920590144585 | β0.041325751 |
| UQCRC1 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.322730330789477 | 0.132632615498281 | β0.021382396 |
| WDR6 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | β0.082409051 | 0.232064088655754 | 0.145731123558279 |
| XKR6 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.218985261309357 | 0.121925434029205 | 0.0898575636645161 |
| ZBED4 | JTI-PrediXcan | β | FUSION, MAGMA, JTI-PrediXcan | Γ | NA | 0.028836782724248 | 0.308754958148029 | 0.265081694171281 |
| ZFAND5 | JTI-PrediXcan | β | MAGMA, JTI-PrediXcan | Γ | NA | 0.279721844161871 | 0.588195700698123 | 0.0947171576814631 |
| ZNF697 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | β0.364594847 | 0.00442179768563452 | 0.142940889982599 |
| ZNF808 | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | Γ | NA | 0.0981477489738863 | 0.189390038746002 | β0.057942593 |
| REEP3 | JTI-PrediXcan | β | JTI-PrediXcan | β | FUSION, JTI-PrediXcan | 0.118243164040487 | 0.23020461305631 | 0.572021449697311 |
| TJP2 | MAGMA, eQTL | Γ | NA | Γ | NA | 0.705879412025416 | β0.629753827 | 0.0881011087319735 |
| TRPM8 | MAGMA, mQTL | Γ | NA | Γ | NA | 0.657785153855803 | β0.057995162 | 0.502279618117342 |
| CACNA1A | NA | NA | FUSION, MAGMA | Γ | NA | 0.753372975465257 | 2.53671237848025 | β0.255234053 |
| KLHDC8B | NA | NA | FUSION, MAGMA | Γ | NA | 0.404411695783461 | 0.335351528678147 | 0.0987640739918263 |
| ACO2 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.0888046564808445 | 0.0269877764599751 | 0.465084717303068 |
| BCAR1 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | β0.036898536 | β0.210686811 | β0.012173616 |
| CCDC134 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.0741073757871275 | 0.075451494580366 | 0.212591586718194 |
| CFDP1 | NA | NA | NA | NA | FUSION, MAGMA, JTI-PrediXcan | 0.325398088367892 | 0.13320628367269 | 0.416450599434954 |
| CHADL | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.18369554796958 | 0.167523392886105 | 0.165344455281645 |
| GLCE | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.262511443361146 | β0.113172068 | 0.551782310544156 |
| ITPKB | NA | NA | NA | NA | FUSION, MAGMA, JTI-PrediXcan | 0.349148413502103 | β0.225519525 | 1.31939005847472 |
| L3MBTL2 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.287926091959072 | 0.486445143650934 | 0.251500321836007 |
| MED19 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.0353095246676063 | β0.409493228 | β0.008130699 |
| PAM | NA | NA | NA | NA | FUSION, JTI-PrediXcan | β0.481055165 | 0.603875519769939 | 1.14298651236614 |
| PPP2R5A | NA | NA | NA | NA | FUSION, JTI-PrediXcan | β0.085714562 | 0.165926735029786 | 0.217291415404394 |
| RHOF | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.148481422282098 | β0.14003072 | 0.507150954767214 |
| SREBF2 | NA | NA | NA | NA | FUSION, JTI-PrediXcan | β0.160664588 | 0.426017810166225 | 0.435117715897074 |
| TEF | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.217670506634366 | 0.234955018252937 | 0.236393961733179 |
| TMEM170A | NA | NA | NA | NA | FUSION, JTI-PrediXcan | 0.00148289361765374 | 0.136070609755002 | 0.120312782524287 |
| TNFRSF13C | NA | NA | NA | NA | FUSION, JTI-PrediXcan | β0.454208941 | β0.383619352 | β0.128668956 |
| ZC3H7B | NA | NA | NA | NA | FUSION, MAGMA, JTI-PrediXcan | 0.137018599759484 | β0.050569323 | 0.00247931245476753 |
| annotation | (MA)pcauchy | (MA)pmedian | (MO)pcauchy | (MO)pmedian |
|---|---|---|---|---|
| Cartilage | 0.001989966 | 0.038852604 | 0.000381139 | 0.050527718 |
| Jaw and tooth | 0.003904829 | 0.141825932 | 0.000583311 | 0.016264954 |
| Cartilage primordium | 0.012272543 | 0.122737728 | 0.000664256 | 0.021000919 |
| Connective tissue | 0.024432631 | 0.096181151 | 0.002125124 | 0.038464403 |
| Mucosal epithelium | 0.027639135 | 0.172755272 | 0.00232748 | 0.031748097 |
| Lung | 0.031452614 | 0.332030211 | 0.003264737 | 0.026035984 |
| Meninges | 0.032544189 | 0.090544602 | 0.005124826 | 0.079225257 |
| Kidney | 0.033560443 | 0.177016842 | 0.006535275 | 0.031672725 |
| GI tract | 0.03598681 | 0.134574184 | 0.010118752 | 0.082811706 |
| Inner ear | 0.049208145 | 0.195626826 | 0.010658896 | 0.032342653 |
| Submandibular gland | 0.060975461 | 0.159003375 | 0.011595545 | 0.106249565 |
| Adipose tissue | 0.06386762 | 0.139983882 | 0.013616844 | 0.073440821 |
| Muscle | 0.088866265 | 0.204478963 | 0.019531163 | 0.087727565 |
| Adrenal gland | 0.093269614 | 0.234082467 | 0.035138463 | 0.119015463 |
| Choroid plexus | 0.095712793 | 0.287723657 | 0.044570075 | 0.100558032 |
| Smooth muscle | 0.107115547 | 0.240173689 | 0.045962044 | 0.138612758 |
| Brain | 0.115143314 | 0.221362095 | 0.068530434 | 0.222520925 |
| Epidermis | 0.132507572 | 0.323154373 | 0.087729798 | 0.272284051 |
| Sympathetic nerve | 0.14419019 | 0.31973193 | 0.128177824 | 0.188094815 |
| Dorsal root ganglion | 0.149228734 | 0.329010516 | 0.226310633 | 0.272710625 |
| Spinal cord | 0.160259216 | 0.312185247 | 0.235648717 | 0.276595869 |
| Liver | 0.215698278 | 0.285288797 | 0.296550361 | 0.415234415 |
| Heart | 0.224998398 | 0.348309827 | 0.418357735 | 0.46242447 |
| Bone | 0.313597694 | 0.358625962 | 0.996361999 | 0.586744823 |
| Cavity | 0.741609388 | 0.472535142 | 0.997954333 | 0.292827 |
Tissue enrichment specificity of MA and MO in sc-ST
Fig. 5 displays genome-wide spatial association signals of MA- and MO-associated genes identified by the gsMap algorithm. The gsMap-derived gene-specific scores (GSS) are cataloged in Supplementary Tables S14-S15, with spatially resolved LDSC enrichment of tissue-specific MA and MO across murine organ regions detailed in Supplementary Tables S16-S17. Spatial organ mapping results (Supplementary Tables S18-S19) and their integrative visualization (Fig. 6) further delineate MetS-associated cellular patterns in E16.5 mouse embryos at single-cell resolution. Statistical validation using the Cauchy combination test (Table 2) identified 25 organs/tissues showing significant enrichment associations. Single-cell spatial transcriptomic analysis revealed distinct enrichment patterns between MA and MO across E16.5 murine embryonic organs, demonstrating complementary associations with genetic enrichment profiles from LDSC-SEG and MAGMA analyses. MO exhibited a pronounced peripheral tissue tropism in spatial mapping, with significant associations in cartilage (MO: pCauchy = 0.00038 vs. MA: pCauchy = 0.00199), jaw/tooth (MO: pCauchy = 0.00058 vs. MA: pCauchy = 0.0039), and meninges (MO: pCauchy = 0.0051). These findings align with LDSC-SEG and MAGMA results, which consistently prioritized vascular tissues (e.g., tibial artery) and gastrointestinal systems (transverse colon, gastroesophageal junction), suggesting MO pathogenesis may center on peripheral vascular dysfunction and mucosal barrier disruption. In contrast, while MA partially overlapped with MO in cartilage primordium (pCauchy = 0.012) and mucosal epithelium (pCauchy = 0.0276), its weaker meningeal association (pCauchy = 0.0325) and adrenal divergence (MA: pCauchy = 0.093 vs. MO: pCauchy = 0.035) contrasted with MAGMA-implicated neuroimmune mechanisms (basal ganglia, hypothalamus) and the absence of LDSC-SEG signals. This discrepancy implies MAβs neuroregulatory abnormalities may involve non-heritable pathways, such as epigenetic modifications or cell-cell interactions. Notably, neither subtype reached significance in brain tissue (MA: p = 0.115; MO: p = 0.0685), though MOβs lower p-value synergized with MAGMAβs vascular tissue signals, reinforcing its peripherally-driven hypothesis. Integrating spatial transcriptomic and genetic enrichment analyses, we propose MO mechanisms converge on heritable peripheral tissue dysregulation, whereas MA likely involves dynamic modulation of central neuroimmune microenvironments.
Genome-wide spatial association signals of MA and MO-associated genes identified by gsMap. From top to bottom, the Manhattan plots of MA and MO, respectively. Chromosomal positions (x-axis) and -log10p values (y-axis) are shown. MA: migraine with aura; MO: migraine without aura; gsMap: genetically informed spatial mapping of cells for complex traits
Spatial mapping of MA and MO-associated cellular patterns in E16.5 mouse embryonic single-cell spatial transcriptomics (ST) data, generated by gsMap algorithm across 25 organs. MA: migraine with aura; MO: migraine without aura; gsMap: genetically informed spatial mapping of cells for complex traits
Discussion
This study systematically dissects the genetic heterogeneity and molecular mechanisms underlying migraine and its subtypes (MA and MO) through an integrative framework combining large-scale GWAS, multi-omics functional annotations, cross-method validation, and sc-ST. Leveraging subtype-stratified data from FinnGen R11 and global GWAS repositories, we confirmed robust genetic correlations between MA and MO while revealing subtype-specific divergence in functional genomic annotations via partitioned heritability (S-LDSC). Crucially, sc-ST mapping of E16.5 murine embryos (25 organs) resolved spatially enriched patterns that complemented and extended genetic findings: MO exhibited pronounced peripheral tropism in vascular tissues and mucosal epithelia, aligning with LDSC-SEG/MAGMA-prioritized pathways (e.g., ITPKB, CFDP1), whereas MA showed preferential enrichment in hypothalamic nuclei and microglia-adjacent regions, corroborating its neuroimmune etiology (CACNA1A, STAT6). By integrating sc-ST with multi-omics profiling (TWAS, SMR, PoPS), we uncovered microenvironment-specific co-localization of cross-subtype core genes (e.g., PHACTR1-STAT6 clusters in meningeal vascular niches) and developmental origins of subtype divergence, such as prenatal enrichment of migraine-associated loci in neural crest-derived tissues (e.g., jaw primordium). The methodological innovation of combining FUSION, MAGMA, JTI-PrediXcan, and sc-ST not only enhanced gene prioritization robustness but also bridged spatial-genetic hierarchies, while the expanded data ecosystemβspanning 53 tissue transcriptomes, methylation QTLs, and embryo-wide spatial atlasesβprovided unprecedented resolution into neurovascular coupling, neuroinflammation, and epigenetic regulation. For clinicians, these insights translate into three actionable priorities: [1] nuanced patient counseling emphasizing MAβs neuroimmune aura mechanisms versus MOβs vascular-peripheral pain pathways; [2] subtype-guided therapeutic selection (e.g., neuronal excitability modulators for MA, vascular/CGRP-targeted agents for MO); and [3] early adoption of emerging biologics targeting prioritized genes like STAT6 (MA) or TRPM8 (MO) as clinical evidence evolves.
The distinct tissue enrichment patterns of MA and MO, revealed through integrated analyses of LDSC-SEG, MAGMA, and sc-ST, underscore their divergent pathophysiological mechanisms. This complements the comprehensive framework presented by Raggi et al. [36] in their hallmark review of migraine pathophysiology, which systematically catalogues how such molecular pathways translate to clinical phenotypes. While MA exhibited potential associations with central nervous and immune microenvironments in sc-ST mapping, these signals lacked statistical significance in LDSC-SEG, contrasting with MAGMAβs tentative links to basal ganglia and hypothalamic regulation [37], implicating neuronal hyperexcitability and cortical spreading depression [38, 39]. MAGMA and sc-ST also revealed weak associations with cultured fibroblasts in MA, possibly indicating indirect roles of glial or peripheral immune cells [40β42]. However, LDSC-SEG did not support these signals, suggesting MAβs neuroimmune mechanisms may involve epigenetic or post-transcriptional regulation rather than genetic enrichment alone. Conversely, MO showed consistent enrichment in peripheral vascular tissues and gastrointestinal regions. These findings suggest MO pathology involves vascular smooth muscle dysfunction (e.g., TRPM8-mediated cold sensitivity [43β45]) and gut-brain axis interactions (e.g., CGRP signaling via CALCA [46, 47]). Petschner et al. [48] provide additional clinical context, demonstrating that MO patients exhibit downregulated retinoic acid signaling (via CYP26B1) and immune-metabolic pathways in blood transcriptomics, which may explain the comorbidity between MO and gastrointestinal disorders. Their identification of LRP1 polymorphisms as regulators of retinoic acid availability offers a mechanistic link to our observed vascular dysregulation in MO, suggesting dietary or supplemental retinoic acid modulation as a potential adjunct therapy. Genetically, MA overlaps with epilepsy-related loci (e.g., PRRT2 [49, 50]), while MO is enriched in pain-modulatory genes (e.g., LRP1) and CGRP pathways (CALCA) [47, 51].
Our integrative multi-omics analysis systematically reveals significant genetic heterogeneity between migraine subtypes. The low-density lipoprotein receptor gene LRP1 plays a central role in maintaining blood-brain barrier integrity while also regulating neuronal calcium signaling and synaptic plasticity in MA [52]. Recent studies demonstrate that LRP1 polymorphisms significantly influence the propagation and duration of cortical spreading depression (CSD), potentially explaining the visual aura in MA patients [53]. In MO, LRP1 primarily modulates peripheral vascular tone and pain signaling through calcium channel regulation in vascular smooth muscle cells. The dual role of LRP1 in both subtypes, as highlighted by Sun et al. [54] and Petschner et al. [48], underscores its potential as a pleiotropic target for migraine therapy, particularly given its involvement in oxidative stress (MO) and retinoic acid metabolism (MA-MO overlap). The vascular regulator PHACTR1 shows strong associations with cerebrovascular abnormalities (e.g., arterial dissection) in MA [55]. GWAS data indicate that the PHACTR1 rs9349379 polymorphism not only increases MA risk but also correlates with post-stroke migraine [56]. In MO, PHACTR1 appears to influence headache initiation through peripheral vascular contraction-dilation balance. The STAT6 gene provides novel insights into migraineβs neuroimmune mechanisms. In MA, STAT6 promotes neurogenic inflammation and blood-brain barrier disruption via mast cell activation [57], while STAT6-deficient mice show attenuated CSD responses [47]. In MO, STAT6 mediates peripheral nociceptor sensitization through IL-4/IL-13 signaling in macrophages. Although methodological support varies across subtypes, REEP3 plays important roles in neuronal plasticity. Transcriptomic analysis reveals upregulated REEP3 expression in MA prefrontal cortex, potentially contributing to aura through glutamatergic signaling [37]. In MO, REEP3 associates with peripheral nociceptor sensitization [58]. Among MO-specific genes, ITPKB enhances trigeminovascular sensitivity by regulating calcium release in nociceptive neurons [58]. Single-cell sequencing shows ITPKBβs specific expression in small-diameter trigeminal ganglion neurons, potentially explaining MO patientsβ mechanical and thermal hypersensitivity. The transcriptional regulator FHL5 interacts with voltage-gated calcium channels (e.g., CACNA1A) to modulate neuronal excitability in MA [59], while influencing pain-related gene expression through histone deacetylation in MO [60]. Additional genes contribute through diverse mechanisms: MEF2D regulates synaptic plasticity and pain pathways [61] while NAB2 modulates NGF signaling in nociceptive neurons [52]. Our novel findings also implicate UFL1 and ZBTB39 in migraine chronification - UFL1 stabilizes inflammatory proteins [37], and ZBTB39 epigenetically modifies pain-related genes [60]. The distinct molecular pathways underlying MA and MO pathogenesis highlight the need for subtype-specific treatment approaches. Most critically, our integrated framework bridges the βgenotype-phenotype gapβ that has long hindered migraine management. By systematically identifying and validating subtype-specific molecular networks, we directly address the unmet need for refractory migraine biomarkers highlighted in Raggi et al.βs [36] comprehensive review.
Our study has several limitations requiring consideration. First, GWAS was conducted exclusively in European ancestry populations, potentially limiting the generalizability of findings to other geographic cohorts. Second, TWAS relied on post-mortem expression profiles from GTEx v8, introducing potential biases due to tissue-specific degradation or confounding factors unrelated to in vivo metabolic states. Third, the PoPS framework excluded genes with incomplete functional annotations, introducing selection bias in candidate prioritization. Finally, while our multi-omics framework revealed divergent molecular architectures between migraine subtypes (MA: neuroimmune-epigenetic dysregulation; MO: vascular-metabolic perturbations), the mechanistic roles of the identified genes require rigorous functional validation to establish them as drivers of clinically distinct disease. This necessitates integrated approachesβincluding single-cell sequencing, animal models, and multi-ethnic replicationβto inform clinical translation. Critically, although prior clinical studies suggest differential treatment responses (e.g., triptan efficacy), our study did not directly correlate the identified genetic and spatial signatures with therapeutic outcomes. Therefore, future work integrating subtype-stratified clinical trial data or experimental models (e.g., MA/MO patient-derived cellular systems) is essential to bridge this gap and confirm whether the proposed pathomechanisms fully account for the observed phenotypic heterogeneity. These limitations underscore the need for broader population sampling and functional genomics to refine precision strategies for migraine management.
Conclusion
This study delineates the genetic and spatial architecture of migraine subtype (MA/MO) divergence through integrative multi-omics, combining cross-validated genetic analyses (LDSC-SEG, MAGMA) with sc-ST. Despite shared genetic correlations, sc-ST identified distinct spatial pathophysiological niches: MA showed prenatal enrichment in neural crest-derived tissues and hypothalamic microglial niches, implicating neuroimmune/epigenetic regulation. Conversely, MO exhibited peripheral tropism within vascular smooth muscle and gut-brain axis interfaces, validated by pathway analyses. Multi-modal frameworks prioritized core genes and subtype-exclusive loci, highlighting divergent neurovascular, inflammatory, and epigenetic mechanisms. Resolving spatial-genetic hierarchies via sc-ST uncovered developmental origins of divergence and microenvironmental drivers invisible to bulk analyses. These findings provide a nuanced pathophysiological understanding to inform patient counseling, guide subtype-specific therapeutic strategies, and support adoption of emerging targeted therapies. Critically, our results reconcile key clinical insights: [1] Oxidative stress and neurovascular coupling bridge MA/MO mechanisms; [2] The gut-brain axis critically modulates peripheral nociception in MO; [3] Stratified management is imperative for refractory cases, especially MA with central neuroimmune dysregulation. By integrating spatial-genetic hierarchies with clinical heterogeneity, this work establishes a roadmap for next-generation migraine therapeutics.
Supplementary Information
Supplementary Material 1.
Supplementary Material 2.
Supplementary Material 3.
Supplementary Material 4.
Supplementary Material 5.
Supplementary Material 6.
Supplementary Material 7.
Supplementary Material 8.
Supplementary Material 9.