Dissecting immune cell-specific genetics in migraine: a multi-omics framework for target discovery and therapeutic prioritization.

To construct robust genetic instruments for MR analysis, we began with 26,597 cis-eGenes identified across 14 immune cell types from the OneK1K sc-eQTL dataset. LD clumping was performed using a stringent threshold (r 2  < 0.001, window = 10 Mb) to ensure the independence of variants. After clumping, a total of 8,733 unique eGenes remained for downstream analysis. Gene expression coverage varied across immune cell types, with the largest sample sizes observed in B Mem, B IN, and CD8 NC ranging from 982 to 643 individuals per type. The distribution of valid eGenes was largely consistent before and after LD filtering, with CD4 NC and NK contributing the highest proportions of instrumented genes (Fig.  2 A). Gene expression data were available for 14 immune cell types, with sample sizes ranging from 643 to 982 individuals. The largest sample sizes ( n  = 982) were observed in B Mem, B IN, CD8 NC, CD8 ET, CD4 ET, and CD4 NC, ensuring robust instrument selection in these populations. Meanwhile, CD4 SOX4 and plasma exhibited relatively smaller sample sizes of 857 and 643, respectively. The selection process also categorized eGenes based on the number of independent SNPs used as genetic instruments. The majority of eGenes were associated with a single instrumental SNP, ensuring the exclusion of weak instruments that could introduce bias. In contrast, a smaller subset of eGenes had multiple independent SNPs (Fig.  2 B). Fig. 2 Summary of instrument selection for immune-cell-specific gene expressions. A  Pie charts showing the distribution of eGenes across immune cell types before (26,597 eGenes) and after (8,733 eGenes) F -statistic filtering and LD clumping. B Bar plots depicting the number of samples per immune cell type, the number of eGenes categorized by the number of independent SNPs (nSNP = 1, 2, or 3), and the distribution of eGenes with multiple independent SNPs Summary of instrument selection for immune-cell-specific gene expressions. A  Pie charts showing the distribution of eGenes across immune cell types before (26,597 eGenes) and after (8,733 eGenes) F -statistic filtering and LD clumping. B Bar plots depicting the number of samples per immune cell type, the number of eGenes categorized by the number of independent SNPs (nSNP = 1, 2, or 3), and the distribution of eGenes with multiple independent SNPs We conducted MR analyses across 14 immune cell types to investigate the causal effects of cell-type-specific gene expression on migraine. Detailed information on the instruments used in the MR analysis is summarized in Supplementary Table S1. The Manhattan plot highlights several eGenes that show significant associations with migraine risk, with multiple genes surpassing the significance threshold (Fig.  3 , Supplementary Table S2). Additionally, the chord diagram illustrates a complex network of interactions, showing how numerous genes influence migraine through various immune cell types, suggesting extensive pleiotropic effects (Fig.  4 A). Among all immune cell types, CD4 NC had the highest number of associated eGenes ( n  = 14), including key genes such as HOXB3 , CDC42 , and RPS26 (Table  2 ). CD8 ET and NK followed closely, with 13 and 11 significant eGenes respectively, underscoring the involvement of both adaptive and innate immune responses in migraine pathophysiology. Several genes were found to be associated with migraine across multiple cell types. For example, RPS26 was detected in 8 different immune cell types, indicating a potentially broad and consistent role in immune regulation related to migraine. CDC42 appeared in 6 cell types, and other genes such as BTN3A1 , HAX1 , TMEM41B , and FHL3 were also recurrently identified in several lineages. These findings emphasize the pivotal role of specific immune cell subtypes in mediating genetic risk for migraine and highlight shared causal genes as promising targets for further functional investigation and therapeutic exploration. Fig. 3 The Manhattan plot of immune-cell-specific gene expression associated with migraine. Each point represents an eGene tested for its causal association with migraine in a specific immune cell type. The red horizontal line indicates the significance threshold. Several immune cell-specific genes surpassing the threshold are highlighted The Manhattan plot of immune-cell-specific gene expression associated with migraine. Each point represents an eGene tested for its causal association with migraine in a specific immune cell type. The red horizontal line indicates the significance threshold. Several immune cell-specific genes surpassing the threshold are highlighted Fig. 4 Multi-layered analyses of prioritized migraine-associated eGenes. A  Chord diagram showing pleiotropic associations between eGenes and immune cell types. B PPI network of prioritized eGenes constructed based on STRING database. Nodes represent proteins and edges represent known or predicted interactions. C  Forest plot summarizing MR and colocalization results. Red bars indicate strong colocalization evidence (PP.H4 > 0.7), while blue bars represent moderate colocalization evidence (PP.H4 between 0.5 and 0.7) Multi-layered analyses of prioritized migraine-associated eGenes. A  Chord diagram showing pleiotropic associations between eGenes and immune cell types. B PPI network of prioritized eGenes constructed based on STRING database. Nodes represent proteins and edges represent known or predicted interactions. C  Forest plot summarizing MR and colocalization results. Red bars indicate strong colocalization evidence (PP.H4 > 0.7), while blue bars represent moderate colocalization evidence (PP.H4 between 0.5 and 0.7) Table 2 Prioritized eGenes associated with migraine across immune cell types Cell Type eGene nCount CD4 NC HOXB3 , UBE2Q2 , HAX1 , MEI1 , RPS26 , MEF2BNB , ASL , H1F0 , NELFCD , SLC2A4RG , TMEM41B , HOXB2 , MSH3 , CDC42 14 CD8 ET NUTF2 , MAD1L1 , NELFCD , HAX1 , FHL3 , BTN3A1 , TMA7 , RPS26 , CENPM , ZNF391 , ARL14EP , CDC42 , TPGS2 13 NK KLF7 , TMEM171 , BTN3A1 , TMA7 , SMPD3 , RPS26 , MEI1 , BTN2A1 , HOXB-AS1 , TPGS2 , ZSWIM7 11 B IN ZFP36L1 , TSGA10 , RPS26 , ARL14EP , TMEM41B , GINM1 , MRPL34 , CDC42 8 CD8 NC FHL3 , HAX1 , BTN3A1 , RPS26 , TMEM41B , CYP2D6 , CDC42 , TPGS2 8 B Mem SMPD3 , HIST1H3D , RPS26 , CDC42 4 CD4 ET RPS26 , ARL14EP , GCAT , CDC42 4 CD8 S100B FHL3 , RPS26 2 Mono C H1F0 1 Plasma SIRT5 1 Prioritized eGenes associated with migraine across immune cell types To investigate the potential biological interplay among the migraine-associated eGenes identified through MR, we constructed a PPI network using the STRING database with a confidence score threshold of 0.40. The resulting network visualizes both known and predicted interactions among candidate proteins, integrating evidence from curated databases, experimental studies, and computational prediction methods such as gene co-occurrence, co-expression, and text mining (Fig.  4 B). Several central nodes, including H1F0 , RPS26 , HOXB3 , and H3C12 , emerged as major hubs within the network, each interacting with multiple other proteins, suggesting their key roles in potential regulatory pathways. Notably, H3C12 showed dense connectivity, indicating possible involvement in broad transcriptional regulation mechanisms related to migraine pathophysiology. RPS26 , a ribosomal protein gene consistently implicated across immune cell types, also demonstrated a high-confidence interaction with SLC2A4RG , further supporting its relevance. Other genes such as BTN3A1 , CDC42 , NELFCD , and HAX1 were positioned within smaller clusters but maintained specific functional links, hinting at more targeted or context-specific roles. To refine candidate genes with potential causal roles in migraine, we conducted Bayesian colocalization analyses to assess whether gene expression and migraine share a common causal variant. eGenes with PP.H4 > 0.7 were considered strong evidence and PP.H4 between 0.5 and 0.7 considered moderate support (Supplementary Table S3). A total of 15 eGenes passed the strong colocalization threshold (PP.H4 > 0.7), while several others showed moderate support (Fig.  4 C). The eGene NELFCD in CD8 ET showed the highest confidence with PP.H4 = 99.88% and a risk-increasing effect on migraine (OR = 1.086, 95% CI: 1.052–1.121). Other strongly colocalized eGenes included HAX1 (CD4 NC, CD8 ET, and CD8 NC, PP.H4 > 98%), HOXB3 (CD4 NC, PP.H4 = 93.88%), and KLF7 (NK, PP.H4 = 92.15%), all indicating robust shared genetic signals. Among those with moderate evidence, CDC42 was notable for being colocalized across multiple immune cell types, including CD4 ET, CD4 NC, CD8 ET, CD8 NC, B Mem, and B IN, with PP.H4 values ranging from 57.01 to 65.94%. Although below the strong threshold, its consistent presence across diverse lineages suggests a potentially pleiotropic role in migraine-related immune regulation. To further dissect the subtype-specific genetic mechanisms underlying migraine, we performed separate MR analyses for MA and MO (Supplementary Fig. S1). For MA, several eGenes showed nominally significant associations. NELFCD in CD4 NC (OR = 1.091, P  = 7.86E − 03), BTN3A1 in CD8 ET (OR = 1.133, P  = 2.87E − 03), HIST1H3D in B Mem (OR = 0.916, P  = 2.54E − 03), and TMA7 in CD8 ET (OR = 1.116, P  = 3.74E − 02) demonstrated notable subtype-specific effects. While CDC42 in CD4 NC showed a protective trend (OR = 0.937), the result did not reach statistical significance ( P  = 7.16E − 02) (Supplementary Fig. S1A). For MO, partially overlapping but distinct associations emerged. BTN3A1 in CD8 ET (OR = 1.110, P  = 2.54E − 02), FHL3 in CD8 S100B (OR = 0.888, P  = 5.38E − 03), and HOXB3 in CD4 NC (OR = 0.900, P  = 1.39E − 02) showed significant links to migraine risk, with some displaying opposite directions of effect compared to MA (Supplementary Fig. S1B). These results highlight both shared and distinct genetic contributors to MA and MO, suggesting that certain immune-related eGenes may exert subtype-specific effects, which could inform targeted therapeutic strategies. To validate the robustness of our MR findings, we performed independent replication analyses using GWAS summary statistics from the migraine meta-analysis by Hautakangas et al. Several eGenes identified in the FinnGen dataset demonstrated consistent effect directions and nominal significance ( P  < 0.05) in the replication cohort, including GINM1 , CDC42 , NELFCD , and TMA7 . We then conducted random-effects meta-analyses to integrate evidence from both datasets. In total, 25 eGenes remained statistically significant ( P  < 0.05) and were retained for downstream analyses (Supplementary Fig. S2). Additionally, due to SNP availability and cell-type specificity constraints, sc-eQTL based MR analysis using DICE data yielded results for only three eGenes ( NELFCD , H1F0 , and HOXB3 ), all of which showed significant associations with migraine and consistent effect directions with the primary analysis (Supplementary Fig. S3). Using LDSC regression, we assessed genome-wide genetic correlations between migraine and 2,871 complex traits. In the FinnGen dataset, 357 traits showed significant correlations after Bonferroni correction, notably chronic pain (e.g., cervicalgia, rg = 0.44), psychiatric symptoms (e.g., depression, rg = 0.36), and sleep disturbances (Fig.  5 , Supplementary Table S4). In the Hautakangas et al. dataset, 245 traits were significant, including neuroticism (rg = 0.25) and irritable bowel syndrome (rg = 0.50) (Fig. S4, Supplementary Table S5). A total of 211 overlapping traits were consistently correlated with migraine in both datasets, forming a consensus set spanning pain, psychiatric, cardiovascular, gastrointestinal, and sensory domains (Fig. S5, Supplementary Table S6). These traits were used in downstream MR analyses to evaluate the pleiotropic effects of migraine-associated eGenes. Fig. 5 Genetic correlations between migraine and 2,871 phenotypes across the FinnGen migraine GWAS dataset. Each dot represents one phenotype categorized by domain. Only significantly correlated traits are colored, while non-significant ones are shown in gray. Notably, migraine was positively correlated with pain conditions (e.g., back pain), psychiatric traits (e.g., anxiety), and several cardiometabolic phenotypes Genetic correlations between migraine and 2,871 phenotypes across the FinnGen migraine GWAS dataset. Each dot represents one phenotype categorized by domain. Only significantly correlated traits are colored, while non-significant ones are shown in gray. Notably, migraine was positively correlated with pain conditions (e.g., back pain), psychiatric traits (e.g., anxiety), and several cardiometabolic phenotypes From the 211 traits that showed statistically significant and directionally consistent correlations with migraine in both datasets, we selected a refined panel of 27 migraine-related traits (Supplementary Table S7). MR analyses revealed multiple significant causal associations between migraine-associated eGenes and 27 migraine-related traits (Fig.  6 , Supplementary Table S8). Several immune cell-specific expressions of HAX1 and TMA7 were inversely associated with back pain. HOXB-AS1 showed negative effects on myalgia and chronic gastritis, while GINM1 , NELFCD , and HOXB3 were negatively associated with depression. Notably, FHL3 in CD8 + T cells was positively associated with both depression and diastolic blood pressure. SLC2A4RG and ASL influenced gastrointestinal conditions such as gastro-oesophageal reflux and irritable bowel syndrome. CDC42 and MAD1L1 were significantly associated with systolic and diastolic blood pressure, while NELFCD , HOXB3 , and TMA7 influenced sensory and respiratory traits including senile cataract, tinnitus, asthma, and COPD. These findings highlight diverse downstream phenotypic consequences and pleiotropic roles of migraine-associated eGenes. Fig. 6 Causal effects of migraine-associated eGenes on migraine-related traits. Each column represents a migraine-related phenotype and each row denotes a prioritized eGene. Bars on the right indicate the number of traits each gene is significantly associated with. * P  < 0.05, ** P  < 0.01, *** P  < 0.001 Causal effects of migraine-associated eGenes on migraine-related traits. Each column represents a migraine-related phenotype and each row denotes a prioritized eGene. Bars on the right indicate the number of traits each gene is significantly associated with. * P  < 0.05, ** P  < 0.01, *** P  < 0.001 We conducted PheWAS analyses using the AstraZeneca PheWAS portal to evaluate the potential phenotypic consequences of targeting migraine-associated eGenes (Fig.  7 ). Among the two prioritized candidates, GINM1 showed significant associations only with increased leg fat percentage ( P  = 6.55 × 10 −10 , β = 2.264), a non-pathological trait unlikely to pose clinical concerns (Supplementary Table S9). TMA7 was significantly associated with a series of adiposity-related measurements, including lower tissue fat percentage in the legs, arms, and gynoid regions ( P  < 5 × 10 −8 ), across both musculoskeletal and general health-related categories (Supplementary Table S9). Despite statistical significance, these traits reflect physiological variation in body composition rather than adverse outcomes. Notably, no strong associations were observed with severe clinical phenotypes, suggesting a favorable safety profile for therapeutic modulation of these genes. To further explore phenotypic links of prioritized eGenes, we performed enrichment analyses using the GWAS Catalog, UK Biobank, ClinVar, and PhenGenI via the Enrichr platform (Supplementary Table S10-13). Notably, MAD1L1 emerged repeatedly across multiple datasets, showing associations with traits such as generalized anxiety disorder, physical activity, mood-related phenotypes, and prostate cancer. However, most associations were with behavioral or lifestyle traits, and no strong enrichment was observed for severe adverse phenotypes. CDC42 was linked to cervical dysplasia and endometriosis, while HAX1 and ASL showed enrichment for rare metabolic or hematological disorders (e.g., severe congenital neutropenia, urea cycle disorder). Importantly, none of the top-ranked eGenes were significantly enriched for contraindicating disease risks that would raise safety concerns in therapeutic contexts. Fig. 7 The eGenes examined across significant continuous traits using the AstraZeneca PheWAS portal. A GINM1 . B TMA7 The eGenes examined across significant continuous traits using the AstraZeneca PheWAS portal. A GINM1 . B TMA7 We queried multiple drug-gene databases to identify approved or investigational compounds potentially targeting migraine-associated eGenes (Table  3 ). Several approved compounds, such as hyoscyamine ( ASL , SLC2A4RG ), bazedoxifene and flupentixol ( CDC42 ), and hydroxychloroquine ( NELFCD ), exhibited therapeutic relevance across neurological, inflammatory, and metabolic conditions. Notably, phetharbital and cerivastatin were associated with both CDC42 and NELFCD , suggesting multi-target repurposing potential. Additional agents like theophylline ( FHL3 , NUTF2 ) and penicillin G ( NELFCD ) are already widely used in clinical practice. A number of investigational or experimental compounds (e.g., 6-Azathymine, Gallic Acid, Kynurenine) may warrant further exploration. These findings highlight multiple druggable migraine-linked genes with favorable safety profiles, supporting opportunities for repositioning both existing and experimental therapeutics. Table 3 Drug repurposing for migraine-associated eGenes Drug Gene Indication Status Arginine ASL Nutritional supplementation, dietary shortage or imbalance Approved Argininosuccinate ASL Urea cycle disorders Investigational Hyoscyamine ASL , SLC2A4RG Functional gastrointestinal disorders Approved 6-Azathymine CDC42 Antiviral research Investigational Oxozinc CDC42 Topical dermatological agent Investigational Bazedoxifene CDC42 Osteoporosis Approved Flupentixol CDC42 Schizophrenia Approved Gallic Acid CDC42 Antioxidant research Investigational Guanosine CDC42 Nucleoside supplement, research Investigational Phetharbital CDC42 , NELFCD Anticonvulsant Approved Cerivastatin CDC42 , NELFCD Hypercholesterolemia Approved Theophylline FHL3 , NUTF2 Asthma, COPD Approved Gentamicin HAX1 Bacterial infections Approved Oxolinic Acid MAD1L1 Urinary tract infections Approved Mepacrine MAD1L1 Antiprotozoal Approved Pentetrazol MAD1L1 CNS stimulant (experimental) Abandoned Hydroxychloroquine Sulfate NELFCD Malaria, Lupus, Rheumatoid arthritis Approved Salbutamol Sulfate NELFCD Asthma, COPD Approved Imiquimod NELFCD Genital warts, Actinic keratosis Approved Oleanolic Acid NELFCD Liver disorders, Inflammation Approved Pulmicort Nebuamp NELFCD Asthma Approved L-arabinose NELFCD Food additive, Antidiabetic agent Investigational Penicillin G NELFCD Bacterial infections Approved Kynurenine NELFCD Tryptophan metabolism study Investigational Chromium NELFCD Nutrient supplement Approved Methacholine NELFCD Bronchial hyperreactivity test Approved Calcium Phosphate NELFCD Calcium supplement Approved Sodium Sulfate NELFCD Laxative Approved Sphingosine NELFCD Cell signaling studies Investigational Drug repurposing for migraine-associated eGenes

Overview of the study design and analysis pipeline were outlined in Fig.  1 . This study was designed following the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guidelines to ensure methodological rigor and transparency [ 24 ]. MR relies on three core assumptions: (1) the genetic variants used as instrumental variables (IVs) are robustly associated with the exposure; (2) the IVs are not associated with any confounders; and (3) the IVs affect the outcome exclusively through the exposure, not via alternative pathways. All analyses were based on publicly available summary-level data from previously published GWAS. No individual-level or identifiable participant data were used. Therefore, separate ethical approval was not required for the present research. The original GWAS studies from which the data were obtained had secured ethical approvals and informed consent from their respective participants [ 22 ]. All methods were conducted in accordance with relevant guidelines and regulations. Fig. 1 Overview of the study design and analysis pipeline. A  Data sources: sc-eQTLs from 14 immune cell types and migraine GWAS from FinnGen and Hautakangas et al. B Primary analyses: Two-sample MR to infer causal eGenes, supported by PPI networks and colocalization. C  Follow-up analyses: Genetic correlation (LDSC), pleiotropy profiling (PheWAS, GWAS Catalog, ClinVar), and drug repurposing (OpenTargets, DrugBank, DGIdb, DSigDB) Overview of the study design and analysis pipeline. A  Data sources: sc-eQTLs from 14 immune cell types and migraine GWAS from FinnGen and Hautakangas et al. B Primary analyses: Two-sample MR to infer causal eGenes, supported by PPI networks and colocalization. C  Follow-up analyses: Genetic correlation (LDSC), pleiotropy profiling (PheWAS, GWAS Catalog, ClinVar), and drug repurposing (OpenTargets, DrugBank, DGIdb, DSigDB) To define genetic instruments for immune cell-specific gene expression, we utilized cis-expression quantitative trait loci (cis-eQTLs) derived from the OneK1K sc-eQTL dataset [ 22 ]. This resource provides high-resolution gene regulatory information across 14 immune cell types. Only conditionally independent cis-eQTLs reaching genome-wide significance ( P  < 5.0 × 10 −8 ) were retained to minimize false-positive associations [ 25 ]. In addition, sc-eQTL data from the DICE database were used as an independent replication resource [ 26 ]. A multi-step filtering process was implemented to ensure instrument validity. First, linkage disequilibrium (LD) clumping was performed using the 1000 Genomes Project European reference panel, applying a stringent threshold of r 2  < 0.001 within a 10 Mb window to ensure SNP independence [ 27 ]. Second, exposure and outcome datasets were harmonized by aligning alleles and removing strand-incompatible or ambiguous SNPs. Palindromic SNPs were retained only if allele frequency information enabled unambiguous alignment. To assess instrument strength, we calculated the F -statistic for each SNP and excluded those with F  < 10 to minimize weak instrument bias [ 28 ]. After quality control and filtering, a total of 9,117 SNPs corresponding to 8,733 unique immune-cell-specific eGenes were retained as valid instruments for downstream MR analyses. Migraine-associated genetic summary statistics were obtained from the FinnGen consortium (R12), a large-scale public-private partnership that integrates genotype data from Finnish biobanks with comprehensive national health registry data ( https://www.finngen.fi/en ) [ 29 ]. The primary GWAS dataset included 55,116 migraine cases and 445,232 controls of European ancestry. Additionally, an independent replication dataset was obtained from the meta-analysis conducted by Hautakangas et al., which included 48,975 migraine cases and 540,381 controls of European ancestry [ 14 ]. This dataset integrated data from four cohorts: IHGC2016, UK Biobank, GeneRISK, and HUNT. Due to data sharing restrictions, 23andMe data were not included. In addition to the overall migraine phenotype, we also extracted GWAS summary data for two clinically relevant subtypes: migraine with aura (MA) and migraine without aura (MO). These datasets were derived from the same FinnGen release, comprising 11,757 MA cases and 9,690 MO cases, each compared against 374,605 controls (Table  1 ). These well-powered and harmonized datasets enabled robust primary and subgroup MR analyses. All outcome data were based on de-identified, aggregate-level statistics with appropriate ethical approval and participant consent obtained in the original studies. Table 1 Summary of migraine GWAS datasets used in this study Phenotype Ancestry nCase nControl Source Migraine European 55,116 445,232 FinnGen Migraine European 48,975 540,381 Hautakangas et al. Migraine with aura European 11,757 374,605 FinnGen Migraine without aura European 9,690 374,605 FinnGen Summary of migraine GWAS datasets used in this study We conducted MR analyses to estimate the causal effects of immune cell-specific gene expression on migraine using the TwoSampleMR R package (version 0.6.6) [ 30 ]. For each eGene, we selected the appropriate MR method based on the number of available instrumental SNPs. The Wald ratio method was applied when only a single SNP was available, while the inverse-variance weighted (IVW) method was used for genes with two or more independent SNPs [ 31 ]. To account for multiple hypothesis testing across thousands of genes, we applied a false discovery rate (FDR) threshold to control for type I error. Gene-trait associations that passed this threshold were deemed statistically robust and selected for downstream analyses, including colocalization, protein interaction, and drug prediction. This approach enabled a systematic evaluation of the potential causal effects of immune cell-type-specific gene expression on migraine susceptibility. Additionally, we conducted random-effects meta-analyses to combine MR estimates from the FinnGen dataset (discovery) and the meta-analysis by Hautakangas et al. (replication), using the metafor R package. To investigate the functional relationships among migraine-associated genes identified by MR analysis, we constructed a protein-protein interaction (PPI) network using the STRING database ( https://string-db.org ). Only high-confidence interactions (minimum interaction score ≥ 0.4) supported by experimental evidence or curated databases were included. The analysis was restricted to “Homo sapiens” to ensure relevance to human biology. All prioritized eGenes were uploaded into the STRING platform to visualize direct and indirect protein interactions. Network edges reflected different sources of evidence, including experimental validation, co-expression, and text mining. The resulting PPI network provided insight into the biological connectivity and potential pathway convergence of these genes, aiding in the identification of hub proteins and functional modules relevant to migraine pathophysiology. To determine whether the same genetic variants influence both gene expression and migraine risk, we performed Bayesian colocalization analysis using the coloc R package [ 32 ]. For each eGene, we assessed the posterior probabilities of five hypotheses, with a focus on PP.H4, the probability that a single shared causal variant underlies both the eQTL and the migraine association. A gene was considered to show strong evidence of colocalization if PP.H4 exceeded 70%, and moderate evidence if PP.H4 was between 50% and 70%. Colocalized eGenes passing these thresholds were prioritized for further functional interpretation and therapeutic evaluation. This approach provided additional confidence that observed associations reflect true biological mechanisms rather than linkage or confounding, helping to refine the list of candidate causal genes with immune cell-specific effects on migraine. To investigate the shared genetic architecture between migraine and other human complex traits, we performed genome-wide genetic correlation analysis using linkage disequilibrium score regression (LDSC) [ 33 ] across both the FinnGen and Hautakangas et al. migraine GWAS datasets [ 14 , 29 ]. Publicly available GWAS summary statistics for 2,871 traits were sourced from the UK Biobank (UKB) ( https://pan.ukbb.broadinstitute.org ) [ 34 ], FinnGen ( https://www.finngen.fi/en ) [ 29 ], and the Million Veteran Program (MVP) [ 35 ], encompassing a broad spectrum of phenotypic domains including pain, psychiatric, cardiovascular, and metabolic categories. We calculated genetic correlations (rg) between migraine and each trait, and employed the block jackknife method implemented in LDSC to assess statistical deviation from zero correlation. Multiple testing correction was performed using a stringent Bonferroni-adjusted threshold of P  < 1.74 × 10 −5 . To investigate the downstream phenotypic consequences of migraine-associated eGenes, we assessed their causal effects on a set of traits genetically correlated with migraine. Specifically, we leveraged results from prior LDSC-based genome-wide genetic correlation analyses conducted separately using the FinnGen and Hautakangas et al. migraine GWAS datasets. Traits that demonstrated statistically significant and directionally consistent correlations in both datasets were retained, yielding a consensus panel of migraine-related phenotypes. From this set, we selected a refined panel of migraine-related traits based on two criteria: (1) strong and concordant genetic correlation across both datasets, and (2) biological or clinical relevance to migraine pathophysiology, as supported by existing literature. The selected traits represent major domains frequently comorbid with migraine, including chronic pain conditions, psychiatric symptoms, cardiovascular traits, gastrointestinal disorders, and sensory disturbances. Subsequently, we performed MR analyses to evaluate the causal effects of the 25 prioritized eGenes on these selected traits. The full list of GWAS summary statistics used in this analysis is provided in Supplementary Table S7. Instrument selection and harmonization followed the same procedures described in earlier sections. Each gene was tested individually against each trait using appropriate MR methods based on the number of available instruments. Standardized effect sizes were calculated and visualized in a heatmap matrix to identify shared or trait-specific pleiotropic effects. To preliminarily evaluate the pleiotropic effects and potential safety profiles of migraine-associated eGenes, we conducted PheWAS using the AstraZeneca PheWAS Portal ( https://azphewas.com/ ) based on the UK Biobank 500 K WGS (v2) dataset [ 34 , 36 ]. This portal provides comprehensive evaluation on the association between protein-coding variants with ~ 13 K binary and ~ 5 K continuous phenotypes using variant-level and gene-level phenome-wide association studies in 6 ancestry groups (European, Askenazi Jewish, Admixed American, African, East Asian, South Asian). Each prioritized migraine-associated eGene was queried in this platform. Associations reaching genome-wide significance ( P  < 1.0 × 10 −8 ) were classified as strong, while P -values between 1.0 × 10 −6 and 1.0 × 10 −8 were considered suggestive. Genes showing no strong associations with unrelated phenotypes were considered to have low pleiotropic risk and were prioritized for therapeutic consideration. This analysis helped identify eGenes that may serve as safe and specific drug targets for migraine, without obvious links to adverse health outcomes. Building upon the initial screening, we conducted a comprehensive integrative annotation of prioritized migraine-associated eGenes to further refine the evaluation of phenotypic pleiotropy and clinical safety. We queried each candidate gene across the following databases: (1) GWAS Catalog (2023): To identify previously reported trait associations; (2) UKB GWAS (v1): To cross-check broader phenotype links in population-scale data; (3) ClinVar (2019): To determine whether pathogenic or likely pathogenic variants were reported; (4) PhenGenI (2021): To integrate functional evidence from dbGaP, OMIM, eQTL, and SNP databases. Queries and enrichment analyses were performed using the Enrichr platform ( https://maayanlab.cloud/modEnrichr/ ) [ 37 ], and summary tables of trait associations, variant pathogenicity, and druggability scores were compiled for each gene. This multi-source annotation helped confirm target specificity and inform drug repurposing potential. To explore potential repurposing opportunities, we integrated our eGenes with evidence from OpenTargets [ 38 ], DrugBank [ 39 ], the Drug-Gene Interaction Database (DGIdb) [ 40 ], and Drug Signatures Database (DSigDB) [ 41 ]. This multi-database strategy enabled the annotation of eGenes with known or predicted drug interactions, pharmacological mechanisms, and compound classes. Both FDA-approved medications and investigational agents were considered. Prioritization of candidate drugs was based on: (1) Evidence of existing gene-drug interactions; (2) Therapeutic relevance to neurological, inflammatory, or vascular pathways; (3) Safety and pleiotropy assessment results. eGenes with favorable target profiles and actionable compounds were nominated for potential repositioning in migraine treatment.

This study provides the first systematic integration of sc-eQTL data with MR to uncover immune cell-type-specific genetic drivers of migraine. By identifying 25 causally associated eGenes, including CDC42 , NELFCD , and HOXB3 , we highlight the significant role of both innate and adaptive immune cells in migraine pathophysiology. Subtype analyses and pleiotropy assessments further reveal shared genetic links with neuropsychiatric, gastrointestinal, and cardiovascular traits. Importantly, our drug repurposing analysis identifies several FDA-approved compounds targeting these genes, offering promising avenues for therapeutic intervention. These findings advance our understanding of migraine as an immune-influenced disorder and support the development of precision medicine strategies targeting specific immune pathways.

Migraine is a prevalent neurological disorder with limited therapeutic options, as many patients experience insufficient relief or adverse effects from current neurovascular-targeted drugs [ 42 , 43 ]. While genetic studies have advanced our understanding of migraine susceptibility, most have focused on bulk tissues, overlooking the potential contributions of immune cell-specific regulatory mechanisms. Recent evidence points to immune involvement in migraine, but the lack of cell-type resolution has impeded identification of precise targets. To overcome this, our study integrates sc-eQTL data with MR and multi-omics analyses to uncover causal immune gene effects on migraine. This approach offers higher specificity than bulk-level methods by capturing gene regulation within distinct immune lineages. We identified 25 eGenes with cell-specific causal associations to migraine, including CDC42 , NELFCD , HOXB3 , and HAX1 , many of which appeared across multiple immune cell types. These findings highlight the critical role of both innate and adaptive immunity in migraine pathogenesis and provide a foundation for developing immune-targeted therapies with improved precision and safety. Among the 25 migraine-associated immune cell-specific eGenes identified in our study, several genes exhibited strong and consistent associations across multiple immune cell types, underscoring their potential roles as central regulators in migraine pathogenesis. Notably, CDC42 stands out as a highly recurrent gene, identified across a broad spectrum of immune cell types. As a central member of the Rho GTPase family, CDC42 plays a critical role in cytoskeletal organization, vesicle trafficking, and cell polarity, all of which are essential for immune cell migration, activation, and synapse formation [ 44 , 45 ]. Its consistent presence across multiple immune lineages suggests that CDC42 may serve as a key regulator of immune cell motility and signaling in neuroinflammatory contexts. By maintaining immune cell homeostasis and preventing excessive activation or infiltration into neurovascular tissues, CDC42 may exert a protective effect against migraine initiation [ 46 ]. The fact that its effect is not confined to one cell type, but is instead observed across both innate and adaptive arms of the immune system, points to a fundamental role in modulating systemic immune tone relevant to migraine. In contrast, NELFCD appears to promote migraine risk and is predominantly expressed in CD4 NC and CD8 ET. As a core component of the negative elongation factor complex, NELFCD regulates transcriptional pausing, thus shaping the expression kinetics of genes involved in inflammatory and stress responses [ 47 ]. Its upregulation in T cells could potentiate prolonged transcriptional activation in response to triggers, leading to a more sustained pro-inflammatory milieu. In the context of migraine, this could enhance the production of cytokines or chemokines that sensitize trigeminal neurons or disrupt the blood-brain barrier [ 48 ]. Interestingly, the specificity of NELFCD to CD8 ET suggests that it may be particularly involved in recurrent or chronic forms of migraine, potentially contributing to central sensitization or cortical spreading depolarization, especially in MA. FHL3 represents another notable gene identified predominantly in CD8 + T cell subsets. This gene encodes a LIM-domain transcriptional co-regulator involved in immune cell differentiation and cytotoxic granule dynamics [ 49 ]. Its association with migraine may reflect its role in shaping the effector phenotype of CD8 + T cells, potentially modulating their interaction with vascular endothelium or neural tissues. Through regulating T cell cytotoxicity and immune surveillance, FHL3 may influence peripheral immune activation and its downstream neurological consequences. Its potential protective role suggests that balanced CD8 + T cell responses might help mitigate migraine-relevant inflammation or prevent maladaptive immune priming. HOXB3 and HAX1 also emerged as genes of interest in T cell lineages. HOXB3 , a developmental transcription factor, has recently been implicated in immune cell positioning and cytokine responses [ 50 ]. Its activity in naive T cells may influence their differentiation toward either regulatory or effector fates, which can substantially impact neuroimmune interactions [ 51 ]. Meanwhile, HAX1 is known for its roles in mitochondrial homeostasis and apoptotic regulation, particularly in neutrophils and lymphocytes [ 52 ]. In migraine, HAX1 -expressing cells might contribute to the resolution phase of inflammation or modulate energy demands during immune activation, linking cellular stress responses to headache pathophysiology. Subtype analysis of MA and MO further supports a nuanced role for immune regulation. While many genes were shared across both subtypes, their activity appeared to be more concentrated in specific immune cell types for each phenotype. For instance, TMA7 and TPGS2 were more prominently associated with MA, potentially reflecting the higher involvement of effector and memory T cells in cortical excitability, neuroinflammatory priming, or the initiation of cortical spreading depolarization, which is more commonly observed in migraine with aura. In contrast, genes such as HAX1 and FHL3 showed greater relevance to MO, suggesting that this subtype may involve distinct immune pathways, possibly related to peripheral immune signaling, altered nociceptor sensitivity, or systemic inflammatory regulation that influence migraine susceptibility in the absence of aura. Our finding that several immune cell-specific eGenes, such as NELFCD , HOXB3 , and GINM1 , influence both migraine and psychiatric traits like depression and anxiety suggests that immune-neuropsychiatric crosstalk may play a central role in the observed comorbidity. This is consistent with growing evidence that T cell subsets and myeloid cells can modulate neuroinflammation and neurotransmitter metabolism, thereby affecting cortical excitability and pain perception [ 53 , 54 ]. Transcriptional regulators such as NELFCD and FHL3 may serve as upstream hubs in these immune-brain signaling networks, helping to explain their pleiotropic effects across both neurological and psychiatric domains. Interestingly, HAX1 was inversely associated with both back pain and migraine, indicating it may represent a shared protective factor and a candidate for dual-therapeutic targeting. Similarly, TMA7 showed positive associations with irritable bowel syndrome and mouth ulcers, which align with its risk-enhancing effects in migraine and may reflect a common inflammatory or mucosal immune component. However, not all pleiotropic relationships point toward unified mechanisms. TMA7 also exhibited negative associations with back pain, neuroticism, mood swings, and acute inflammation of the orbit, traits that trend oppositely to its migraine effect, raising concerns about potential trade-offs in therapeutic applications. Likewise, the positive associations of MAD1L1 , CDC42 , and FHL3 with elevated blood pressure contrast with their migraine-linked directions, suggesting potential risks if these genes are modulated systemically. These observations emphasize the need for directional pleiotropy analysis not only to identify shared mechanisms but also to highlight safety-critical contradictions. Encouragingly, some genes such as MEF2BNB demonstrated strong trait specificity, with minimal associations beyond benign traits like body fat distribution. Such “clean” targets are particularly attractive for therapeutic development, offering biological relevance without broad off-target effects. Finally, subtle associations linking genes like CDC42 and MAD1L1 to gynecological or neoplastic traits should be interpreted with caution, as they may arise from tissue-shared regulatory networks rather than direct pathogenic roles [ 55 , 56 ]. Nonetheless, they underscore the importance of cell-type-specific safety profiling and long-term monitoring when advancing immune-based interventions for migraine. The integration of genetically supported eGenes with existing pharmacological databases highlights several promising avenues for drug repurposing in migraine. Notably, multiple identified targets, such as CDC42 , NELFCD , and ASL , are already linked to FDA-approved or clinically studied compounds, offering a potential shortcut to therapeutic development by leveraging known safety profiles and pharmacokinetics. Among the most compelling candidates is hydroxychloroquine, a well-established immunomodulator currently used for lupus and rheumatoid arthritis, which targets NELFCD [ 57 ]. Given NELFCD ’s role in T cell activation and transcriptional regulation, hydroxychloroquine may attenuate immune-driven neuroinflammatory cascades involved in migraine onset. Similarly, bazedoxifene, approved for osteoporosis and also targeting CDC42 , may offer therapeutic benefit by modulating cellular signaling pathways implicated in both immune and vascular function [ 58 ]. Importantly, several drugs such as phetharbital and cerivastatin interact with multiple migraine-linked genes ( CDC42 and NELFCD ), suggesting a potential for polypharmacological synergy [ 59 ]. Targeting overlapping immune pathways through a single agent may enhance efficacy or allow for lower dosing thresholds, thereby reducing side effects. This multi-target engagement could be particularly valuable in treating complex, immune-involved migraine subtypes. While a subset of candidate compounds (e.g., 6-Azathymine, sphingosine) remains investigational or abandoned, they still hold mechanistic interest and could inform future compound screening or structure-based drug design. The identification of FHL3 and NUTF2 as targets of theophylline, a bronchodilator used in asthma, further raises the possibility that some agents with established CNS or immune effects may have previously unrecognized applications in migraine management [ 60 ]. Altogether, these findings reinforce the utility of genomics-informed drug repurposing. By aligning genetic evidence with existing pharmacological assets, we highlight feasible and biologically relevant candidates for translation into migraine therapeutics, warranting further preclinical validation and safety-focused clinical investigation. Unlike previous migraine genetic studies that largely relied on bulk tissue eQTL analyses from brain or whole blood, this study is the first to systematically integrate sc-eQTL data with MR to explore cell-type-specific causal mechanisms in migraine. By resolving gene expression regulation at the level of distinct immune cell types, our approach overcomes the limitations of tissue-averaged analyses and enables more precise identification of functionally relevant and druggable targets. Previous studies have been unable to pinpoint which immune lineages contribute to migraine susceptibility, thus limiting progress in immune-targeted therapies [ 17 , 18 , 61 ]. Our findings support the emerging concept of an “immune-regulated subtype of migraine,” where dysregulated gene expression in specific immune cell populations, such as T cells or B cells, may drive neuroinflammation, vascular dysregulation, or sensory hypersensitivity. This insight offers a potential framework for redefining migraine subtypes based on molecular immunological signatures rather than symptoms alone, and may inform more targeted therapeutic interventions. From a public health perspective, this study provides a foundation for developing biologically informed, personalized approaches to migraine management, especially for individuals who respond poorly to existing neurovascular-based treatments [ 62 ]. It highlights the value of integrating immune genomics into migraine research to guide safer, more effective, and more precise treatment strategies. This study has several limitations that should be acknowledged. First, our analyses were based on GWAS datasets predominantly derived from individuals of European ancestry, which may limit the generalizability of our findings across diverse populations. Future studies in more ancestrally diverse cohorts are essential to validate the identified targets globally. Second, we used summary-level genetic data, which precludes fine-grained analyses such as sex- or age-specific effects, gene-gene interactions, and temporal dynamics of gene expression. Third, while we incorporated multiple layers of publicly available functional data, including sc-eQTL, PheWAS, and drug-gene interaction datasets, the lack of experimental validation limits causal inference and translational conclusions. Fourth, while PheWAS is a valuable tool for assessing pleiotropy and preliminary safety, we explicitly acknowledge that PheWAS alone is insufficient for comprehensive safety evaluation. Electronic health record-based datasets often underrepresent rare diseases, subclinical phenotypes, drug response traits, and nuanced neurological or immune conditions. Furthermore, most PheWAS resources prioritize European ancestry and may lack clinical granularity. Expanded or complementary phenotypic resources may reveal additional off-target associations not captured in our initial analysis. Therefore, the absence of strong associations in our PheWAS results should be interpreted cautiously and not as definitive evidence of target safety. Future work may incorporate computational pleiotropy prediction tools such as PleioPred to further quantify gene-level trait sharing and refine target specificity. Fifth, while our drug repurposing analysis integrated multiple curated resources including Open Targets, DrugBank, DGIdb, and DSigDB, we acknowledge that quantitative prioritization using Open Targets genetic evidence scores was not performed in the current version. Future efforts may formally incorporate these scores to enhance the objectivity of target selection and ensure stronger alignment with genetic evidence for migraine. Finally, the sc-eQTL data used in this study were derived from peripheral immune cells, which may not fully capture gene regulation in the central nervous system or neurovascular interface, regions critically relevant to migraine pathophysiology. Despite these limitations, our findings provide a biologically informed framework for prioritizing immune-mediated targets in migraine and lay the groundwork for future functional and translational research.

Migraine is a prevalent and disabling neurological disorder characterized by recurrent episodes of moderate to severe headaches, often accompanied by nausea, vomiting, and sensitivity to light or sound [ 1 ]. It affects over one billion people worldwide and poses a major public health burden. Despite its high prevalence, the underlying biological mechanisms remain incompletely understood, limiting the development of targeted and effective treatment strategies [ 2 , 3 ]. Current anti-migraine therapies primarily target neurovascular and neuronal signaling pathways, such as serotonin (5-HT) receptors and calcitonin gene-related peptide (CGRP) receptors [ 4 , 5 ]. Several drugs, including triptans and CGRP antagonists, have demonstrated clinical efficacy [ 6 ]. However, a significant proportion of patients do not respond adequately or experience adverse effects, underscoring the need for continued research into novel molecular mechanisms and drug targets to broaden the therapeutic landscape [ 7 ]. Recent studies have suggested a potential immunological component in migraine pathogenesis. Evidence points to the involvement of immune cells such as T cells, B cells, and natural killer cells (NK), along with cytokines and chemokines that may contribute to neuroinflammation during migraine attacks [ 8 – 10 ]. Immunomodulators and inflammatory markers have been associated with migraine frequency and severity, indicating a link between immune system dysregulation and migraine susceptibility [ 8 , 11 , 12 ]. Despite these findings, the molecular basis of immune involvement remains poorly characterized, and the precise regulatory genes and pathways remain largely unexplored. Understanding how immune cell-specific gene expression influences migraine could open new avenues for targeted treatment. Therefore, there is an urgent need to systematically uncover immune-mediated mechanisms and identify potential immunological drug targets for migraine, which could complement or surpass current neurovascular-based strategies. Human genetics offers a powerful tool for validating drug targets and improving the success rates of clinical drug development. Genetic evidence supporting a drug target has been shown to significantly increase the likelihood of approval in clinical trials. In the context of migraine, large-scale genome-wide association studies (GWAS) have identified over 40 common DNA sequence variants that affect risk for migraine types, providing critical insights into its polygenic architecture [ 13 – 15 ]. However, translating these findings into actionable therapeutic targets remains challenging. Mendelian randomization (MR), a method that uses genetic variants as instrumental variables, has emerged as a promising approach to infer causal relationships between gene expression and complex traits like migraine [ 16 ]. Previous MR studies leveraging bulk tissue expression quantitative trait loci (eQTL) datasets have identified several candidate genes potentially involved in migraine, including in tissues such as brain and vascular endothelium [ 17 , 18 ]. While informative, these bulk-level analyses cannot resolve cell-type-specific regulatory effects, particularly within the immune system. This limitation is especially important given the increasing recognition of immune involvement in migraine. Recent advances in single-cell eQTL (sc-eQTL) technologies provide a refined resolution by linking genetic variants to gene expression in specific immune cell types, which enable a more precise dissection of regulatory mechanisms and improve our ability to identify functionally relevant and cell-type-specific therapeutic targets [ 19 , 20 ]. This approach has proven invaluable for pinpointing causal genes and elucidating their roles in disease pathogenesis, particularly in immune-related disorders and cardiometabolic diseases [ 21 – 23 ]. However, despite its potential, there remains a gap in research integrating sc-eQTL data with MR analysis to systematically investigate the causal relationships between immune-related genes and migraine. In this study, we performed a sc-eQTL-based MR analysis to systematically explore the causal effects of immune cell-specific gene expression on migraine. We further integrated colocalization analysis, protein interaction networks, phenome-wide association studies (PheWAS), and drug enrichment prediction to prioritize therapeutic targets. This comprehensive approach aims to reveal immune-mediated mechanisms underlying migraine and identify safe, genetically supported, and targetable candidates for drug development.

Additional file 1: Table S1. Summary of genetic instruments used in MR analysis for eGenes. Table S2. eGenes significantly associated with migraine in MR analysis. Table S3. Colocalization results of migraine-associated eGenes. Table S4. Genetic correlation results (LDSC) using FinnGen migraine GWAS. Table S5. Genetic correlation results (LDSC) using Hautakangas migraine meta-GWAS. Table S6. Traits consistently correlated with migraine in both GWAS datasets. Table S7. GWAS summary statistics for migraine-related traits used in downstream analysis. Table S8. MR results of migraine eGenes on migraine-related traits. Table S9. Phenome-wide associations of prioritized eGenes from AstraZeneca PheWAS. Table S10. GWAS Catalog-based enrichment analysis of eGenes using Enrichr. Table S11. UK Biobank trait enrichment analysis of eGenes using Enrichr. Table S12. ClinVar pathogenic variant enrichment of eGenes using Enrichr. Table S13. PhenGenI-based enrichment analysis of prioritized eGenes using Enrichr. Additional file 1: Table S1. Summary of genetic instruments used in MR analysis for eGenes. Table S2. eGenes significantly associated with migraine in MR analysis. Table S3. Colocalization results of migraine-associated eGenes. Table S4. Genetic correlation results (LDSC) using FinnGen migraine GWAS. Table S5. Genetic correlation results (LDSC) using Hautakangas migraine meta-GWAS. Table S6. Traits consistently correlated with migraine in both GWAS datasets. Table S7. GWAS summary statistics for migraine-related traits used in downstream analysis. Table S8. MR results of migraine eGenes on migraine-related traits. Table S9. Phenome-wide associations of prioritized eGenes from AstraZeneca PheWAS. Table S10. GWAS Catalog-based enrichment analysis of eGenes using Enrichr. Table S11. UK Biobank trait enrichment analysis of eGenes using Enrichr. Table S12. ClinVar pathogenic variant enrichment of eGenes using Enrichr. Table S13. PhenGenI-based enrichment analysis of prioritized eGenes using Enrichr. Additional file 2: Fig. S1. Subtype-specific analysis of prioritized eGenes in migraine. (A) Forest plot of MR effect estimates for migraine with MA. (B) Forest plot of MR effect estimates for MO. Fig. S2. Estimates of meta-analysis from discovery dataset and replication dataset for 25 prioritized eGenes on migraine. Discovery dataset: GWAS summary data from FinnGen. Replication dataset: Meta-GWASs from Hautakangas et al. Fig. S3. Forest plot of MR results for three eGenes using sc-eQTL data from the DICE cohort. Fig. S4 Genetic correlation volcano plot between migraine and 2,871 human phenotypes using LDSC across migraine meta-GWASs from Hautakangas et al. Only traits passing Bonferroni correction are highlighted, revealing substantial pleiotropy, especially with pain and psychiatric traits. Additional file 2: Fig. S1. Subtype-specific analysis of prioritized eGenes in migraine. (A) Forest plot of MR effect estimates for migraine with MA. (B) Forest plot of MR effect estimates for MO. Fig. S2. Estimates of meta-analysis from discovery dataset and replication dataset for 25 prioritized eGenes on migraine. Discovery dataset: GWAS summary data from FinnGen. Replication dataset: Meta-GWASs from Hautakangas et al. Fig. S3. Forest plot of MR results for three eGenes using sc-eQTL data from the DICE cohort. Fig. S4 Genetic correlation volcano plot between migraine and 2,871 human phenotypes using LDSC across migraine meta-GWASs from Hautakangas et al. Only traits passing Bonferroni correction are highlighted, revealing substantial pleiotropy, especially with pain and psychiatric traits.