Mapping the brain cell–specific regulatory architecture of migraine: a single-cell causal framework nominating inhibitory-neuronal BTBD16 and astrocytic RIMS1 as therapeutic targets

Abstract

Background

Migraine is a common, disabling neurological disorder. Genome-wide association studies have mapped numerous migraine risk loci, but the causal genes and their cell-type context remain unclear. Prior work linked migraine GWAS to bulk brain eQTLs; however, tissue-average signals obscure cell-specific regulation.

Methods

We extended these findings to single-cell resolution. Cis-eQTL instruments from 183 human donors across eight brain cell types were filtered by genome-wide significance, LD pruning, and instrument strength, yielding 1,746 independent eGenes. Two-sample Mendelian randomization (MR) tested effects on migraine risk in FinnGen R12 (discovery) with replication in UK Biobank (GCST90473326). To control for multiple testing, we applied within–cell-type Bonferroni correction and global false discovery rate (FDR) adjustment. Bayesian colocalization was performed in both discovery and replication cohorts to evaluate shared causal variants. We also performed a phenome-wide association screen (PheWAS) and profiled regional brain RNA.

Results

Eleven eGenes were significant in FinnGen. Protective associations were observed for BTBD16 in excitatory and inhibitory neurons, RRP15 in excitatory neurons, CCDC146 and GSTM3 in oligodendrocytes, and PDE4B in microglia. Risk-increasing associations were found for GSTM2 (excitatory neurons), RIMS1 and DPH1 (astrocytes), AADAC (microglia), and RBM20 (endothelium). Replication supported signals for inhibitory-neuronal BTBD16 and astrocytic RIMS1. Colocalization analyses indicated shared causal variants at both loci in the discovery cohort (PP.H4 > 0.80). PheWAS showed no genome-wide liabilities for either gene. Regional expression suggested white-matter enrichment for BTBD16 and a cerebellar peak for RIMS1.

Conclusions

Cell-type–specific MR sharpens migraine mechanisms beyond bulk tissue and prioritizes inhibitory-neuronal BTBD16 (protective) and astrocytic RIMS1 (risk-increasing) for mechanistic validation and therapeutic exploration.

Supplementary Information

The online version contains supplementary material available at 10.1186/s10194-025-02241-7.

Introduction

Migraine is a common primary neurological disorder affecting 14% of the global population [1] and characterized by recurrent, throbbing headaches with photophobia, phonophobia, nausea and vomiting [2]. It ranks among the leading causes of years lived with disability worldwide, imposing substantial social and economic burden [3]. Large genome-wide association studies (GWAS) have mapped over a hundred susceptibility loci [4], advancing genetic discovery but leaving the causal genes, effector cell types, and mechanisms incompletely resolved [5, 6].

Recent work in The Journal of Headache and Pain integrated migraine GWAS with plasma pQTLs and bulk brain eQTLs, prioritizing therapeutic candidates (e.g., PNKP) via Mendelian randomization and colocalization [7]. While informative, bulk eQTLs average signals across heterogeneous tissue and cannot assign regulatory effects to specific brain cell types—an important limitation for mechanistic inference in the CNS.

Here we extend and complement that study by moving from bulk tissue to single-cell resolution. We test the hypothesis that cell-type–specific gene expression in the brain causally influences migraine risk. Using single-cell eQTL instruments across eight major brain cell classes, we perform two-sample Mendelian randomization in a discovery cohort with replication in an independent cohort, conduct Bayesian colocalization in the discovery data to evaluate shared causal variants, and undertake a phenome-wide association screen (PheWAS) to assess potential on/off-target liabilities. By enabling cell-type–specific causal inference—rather than tissue-average effects—our analysis refines the mechanistic context of migraine biology and provides targeted candidates for follow-up. An overview of our analytic workflow is provided in Figure S1.

Methods

Exposure definition and tool selection

To identify genetic instruments for MR, we utilized cis-expression quantitative trait loci (cis-eQTLs) from a single-cell eQTL dataset comprising 183 neuropathologically normal individuals [8].

This dataset was generated from post-mortem adult human brain samples and included transcriptomic profiles of eight major brain cell types: excitatory neurons, inhibitory neurons, astrocytes, oligodendrocytes, oligodendrocyte precursor cells (OPCs), microglia, endothelial cells, and pericytes. Cell types were defined based on canonical markers as reported by Bryois et al. [8], such as SLC17A7 and SATB2 for excitatory neurons, GAD1 and GAD2 for inhibitory neurons, AQP4 and FGFR3 for astrocytes, RGS5 for pericytes, CLDN5 for endothelial cells and C1QB and CSF1R for microglia. Notably, the dataset does not distinguish between resting and activated glial states. A multi-step selection process was employed to ensure robust instrument quality. First, we retained conditionally independent cis-eQTLs exceeding a genome-wide significance threshold (P < 5 × 10⁻⁸). Linkage disequilibrium (LD) pruning was then performed using the 1000 Genomes European reference panel (r² < 0.001, 10 Mb window) to ensure SNP independence. The exposure and outcome datasets were harmonized, with strand-incompatible or ambiguous SNPs excluded, and palindromic SNPs retained only if allele frequencies permitted clear alignment. SNPs with an F-statistic < 10 were excluded to mitigate weak instrument bias. Following these steps, 1,746 SNPs from 1,745 cell-specific eGenes were selected for downstream analysis (Table S1).

GWAS data

Results were generated from two datasets: the discovery cohort from FinnGen Release 12, comprising 55,116 migraine cases and 445,232 controls [9], and the replication cohort from the UK Biobank GCST90473326, with 25,393 non-Finnish European ancestry cases and 433,047 controls [10]. This two-cohort approach validated our findings across diverse populations. In the discovery stage, we utilized the FinnGen “MIGRAINE_TRIPTAN” phenotype, which is defined by prescription records for triptan medications. While this phenotype reflects treatment usage, in the Finnish healthcare system triptans are prescription-only and typically indicate a physician-confirmed diagnosis of migraine. As such, this phenotype has been widely accepted as a reliable proxy for migraine in large-scale genetic studies and has been used in several prior publications [6, 7, 11].

Mendelian randomization based on single-cell eQTLs

Two-sample MR was conducted using the TwoSampleMR R package to evaluate the causal effects of brain cell-type-specific gene expression on migraine risk. For eGenes identified by a single SNP, causal inference was based on the Wald ratio. For genes with multiple independent SNPs, inverse variance weighting (IVW) was applied. To account for multiple comparisons, Bonferroni correction was initially applied within each cell type. Additionally, a global false discovery rate (FDR) correction was conducted across all gene–cell type combinations using the Benjamini–Hochberg procedure to control for multiplicity at the study-wide level. To validate the assumed direction of causality (gene expression → migraine), we performed Steiger directionality tests across all instrument–outcome pairs. These tests assess whether the genetic instruments explain more variance in the exposure than in the outcome, providing additional support for the causal inference.In the replication stage, we evaluated only those eGenes that were identified as significant in the discovery cohort. Given the targeted and hypothesis-driven nature of this analysis, further correction for multiple testing was not applied in the replication phase.

Colocalization analysis

Bayesian colocalization analysis was performed using the coloc R package [10] to evaluate whether migraine-associated loci share a causal variant with eQTL signals for cell-type-specific gene expression. Analyses were conducted using three symmetric genomic windows (± 100 kb, ± 250 kb, and ± 500 kb) centered on lead SNPs to assess the sensitivity of results to window size. Default prior probabilities were applied throughout: p₁ = 1 × 10⁻⁴ for the association with gene expression, p₂ = 1 × 10⁻⁴ for the association with migraine, and p₁₂ = 1 × 10⁻⁵ for the probability of a shared causal variant. For each gene–trait pair, posterior probabilities were computed for five competing hypotheses, with a particular focus on PP.H4—the probability that both traits share a single causal variant. Gene–trait pairs with PP.H4 > 0.80 were considered to show strong evidence of colocalization and prioritized for further interpretation, while PP.H4 values between 0.50 and 0.80 were interpreted as indicating moderate support.

Trait pleiotropy assessment via GeneATLAS

To evaluate potential horizontal pleiotropy of lead SNPs identified in MR analyses, we queried each variant in the GeneATLAS database (http://geneatlas.roslin.ed.ac.uk/) [12], which includes GWAS summary statistics for complex traits from the UK Biobank.

Phenotype-wide association study (PheWAS)

PheWAS was performed using the AstraZeneca PheWAS portal (https://azphewas.com/) to assess pleiotropic effects and potential adverse outcomes of therapeutic targets [13]. This analysis utilized data from 15,500 binary and 1,500 continuous phenotypes, with exome-sequencing data from the UK Biobank. The results provide valuable insights into the genetic underpinnings of complex traits and the safety and efficacy of drug targets.

Expression in different brain regions

Gene expression data for candidate targets across various human tissues were obtained from the Human Protein Atlas (https://www.proteinatlas.org) [14]. Investigating gene expression in different brain regions may offer insights into the potential mechanisms by which these targets could act as therapeutic options for migraine.

Results

Instrument selection for cell-type–specific MR

We identified 1,746 independent eGenes from a single-cell eQTL dataset covering eight brain cell types (excitatory and inhibitory neurons, astrocytes, oligodendrocytes, OPCs, microglia, endothelial cells, and pericytes; Fig. 1A, Table S1). After applying stringent criteria (P < 5 × 10⁻⁸, LD pruning: r² < 0.001, F-statistic > 10), these eGenes were selected as instruments for downstream MR analysis. The distribution of eGenes was cell-type specific, with excitatory neurons contributing the highest number (n = 625), followed by oligodendrocytes (423), inhibitory neurons (234), astrocytes (185), OPCs (129), microglia (82), endothelial cells (49), and pericytes (17). Most eGenes were instrumented by a single SNP, with a subset containing multiple independent SNPs for IVW estimation.

Fig. 1.

Selection of cell-type–specific eGenes for migraine MR analysis. (A) Distribution of cell-type–specific eGenes before and after clumping. Left: 133,492 eGenes identified across eight major brain cell types. Right: After applying clumping criteria (genome-wide significance, LD pruning, and F-statistic > 10), 1,746 independent eGenes were retained for further analysis. (B) Number of eGenes per cell type after clumping. Excitatory neurons contributed the most eGenes (625), followed by oligodendrocytes (423), inhibitory neurons (234), astrocytes (185), OPCs (129), microglia (82), endothelial cells (49), and pericytes (17). Color coding represents single (light) and multiple (blue) independent SNPs used to instrument each eGene

Cell-type–specific MR in the discovery cohort

Using FinnGen R12 migraine GWAS (MIGRAINE_TRIPTAN), two-sample MR identified 11 eGenes after within–cell-type Bonferroni correction (Fig. 2, Tables S2–S3): six were protective and five were risk-increasing. In excitatory neurons, BTBD16 (OR = 0.953, 95% CI 0.934–0.973; P = 3.8 × 10⁻⁶) and RRP15 (0.925, 0.892–0.958; 1.9 × 10⁻⁵) were protective, whereas GSTM2 increased risk (1.092, 1.047–1.140; 5.0 × 10⁻⁵); the same protective direction for BTBD16 was seen in inhibitory neurons (0.949, 0.925–0.973; 3.8 × 10⁻⁵). In oligodendrocytes, CCDC146 (0.931, 0.903–0.960; 4.4 × 10⁻⁶) and GSTM3 (0.911, 0.869–0.955; 1.0 × 10⁻⁴) were protective. Astrocytic signals were risk-increasing for RIMS1 (1.048, 1.025–1.072; 3.9 × 10⁻⁵) and DPH1 (1.063, 1.031–1.097; 1.2 × 10⁻⁴). In microglia, PDE4B was protective (0.951, 0.928–0.975; 8.1 × 10⁻⁵), whereas AADAC increased risk (1.067, 1.029–1.107; 4.7 × 10⁻⁴); endothelial RBM20 also conferred higher risk (1.035, 1.015–1.055; 5.8 × 10⁻⁴). Overall, BTBD16 showed concordant protection across neuronal compartments, whereas risk-increasing associations were observed for GSTM2, RIMS1, DPH1, AADAC, and RBM20. Steiger directionality tests confirmed that all of instruments explained greater variance in gene expression than in migraine risk, supporting the assumed causal direction (Table S4).

Fig. 2.

Two-sample MR results for migraine risk across brain cell types. (A) Volcano plots showing the results of two-sample Mendelian randomization (MR) for eight brain cell types. Each point represents an eGene, with the x-axis showing log₂(OR) (effect size per SD increase in gene expression) and the y-axis showing − log₁₀(P) (statistical significance). (B) Summary of significant eGenes identified in the FinnGen R12 discovery cohort (MIGRAINE_TRIPTAN). For each cell type, eGenes with significant associations are listed along with the statistical method (Wald ratio), P-value, and odds ratio (OR) with 95% confidence intervals (CI). Significant results were observed for genes in astrocytes (RIMS1, DPH1), endothelial cells (RBM20), excitatory neurons (BTBD16, RRP15, GSTM2), inhibitory neurons (BTBD16), microglia (PDE4B, AADAC), and oligodendrocytes (CCDC146, GSTM3).

Replication and colocalization

We validated these results in the UK Biobank cohort (GCST90473326), with BTBD16 in inhibitory neurons and RIMS1 in astrocytes showing consistent associations (Fig. 3). To visualize replication overlap, we generated Venn diagrams for the discovery and replication cohorts (Figure S2). To further evaluate the robustness of these associations, we queried the lead SNPs—rs10736311 for BTBD16 in inhibitory neurons and rs982565 for RIMS1 in astrocytes—using the GeneATLAS platform. Neither variant demonstrated strong associations with unrelated traits, suggesting limited evidence of horizontal pleiotropy at these loci (Figure S3).

Fig. 3.

Two-sample MR results for migraine risk in the UK Biobank replication cohort.Forest plot summarizing the results of two-sample Mendelian randomization (MR) for migraine risk using the UK Biobank replication dataset. Each point represents the odds ratio (OR) for a one-SD increase in gene expression, with 95% confidence intervals (CI) indicated by the horizontal lines. Significant eGenes are marked in red. Key significant findings include RIMS1 in astrocytes (P = 2.89 × 10⁻²), and BTBD16 in excitatory neurons (P = 1.01 × 10⁻²), which are consistent with the discovery cohort

Bayesian colocalization analysis, conducted in the discovery cohort (FinnGen R12), was performed using a ± 100 kb genomic window centered on each lead SNP. This analysis revealed strong colocalization at the BTBD16 and RIMS1 loci (BTBD16, PPH4 = 0.89; RIMS1, PPH4 = 0.85), suggesting that both gene expression and migraine risk are likely influenced by the same genetic variants (Figure S4A-B, Supplementary Tables 5–6). The lead SNPs were rs982565 for BTBD16 and rs10736311 for RIMS1. To assess the robustness of these findings, we performed sensitivity analyses using extended genomic windows (± 250 kb and ± 500 kb). The colocalization signals remained stable, with PPH4 values ranging from 0.84 to 0.88 across both genes. Specifically, under a 250 kb window, PPH4 = 0.88 for BTBD16 and 0.84 for RIMS1(Figure S4C-D, Table S7-8); under a 500 kb window, PPH4 = 0.88 for BTBD16 and 0.84 for RIMS1 (Figure S4E-F, Table S9-10).

To replicate our colocalization findings, we conducted Bayesian colocalization analysis in the UK Biobank (GCST90473326) using the same prior settings and genomic windows (± 100 kb, ± 250 kb, ± 500 kb) as in the discovery stage. However, PPH4 values were generally lower in UKB compared to FinnGen. Specifically, PPH4 values dropped below 0.5 for BTBD16 and RIMS1 (Figure S5; Table S11–16). These weaker colocalization signals in UKB may reflect multiple factors: (1) reduced statistical power due to a smaller number of migraine cases (~ 25,000 in UKB vs. ~55,000 in FinnGen); (2) increased diagnostic heterogeneity from self-reported or ICD-based migraine definitions; (3) differences in allele frequency and LD structure across populations; and (4) the dense SNP architecture in UKB (> 80 million SNPs), which increases data sparsity and reduces SNP–eQTL overlap. Together, these limitations may attenuate colocalization signals despite consistent MR directions across datasets.

To further investigate the specificity of our top findings, we assessed whether the lead SNPs used in MR analyses for BTBD16 and RIMS1 might also act as eQTLs for other genes within the same cell type. In astrocytes, 193 genes had genome-wide significant SNPs, but none overlapped with rs982565, the lead SNP for RIMS1(Table S17). Although rs982565 is annotated as an eQTL for LINC00472 in bulk brain tissue (GTEx), its association in astrocytes was weak (P = 0.29; F = 1.13), suggesting insufficient instrument strength to support multivariable MR (Table S18). Similarly, in inhibitory neurons, rs10736311 was not shared with any of the 233 eGenes identified, and GTEx data confirmed its exclusive association with BTBD16(Table S19). These observations indicate a low likelihood of horizontal pleiotropy via shared cis-regulation.

PheWAS

Using the AstraZeneca PheWAS Portal, we screened BTBD16 and RIMS1 instruments across the phenome to probe potential on-/off-target liabilities. No phenotype reached genome-wide significance (P < 5 × 10⁻⁸) for either gene (Figures S6A–B). At nominal thresholds, we observed no coherent cross-system pattern. These findings support the specificity of the MR signals and argue against broad horizontal pleiotropy or obvious safety concerns, while modest effects cannot be excluded.

Regional expression of BTBD16 and RIMS1

Using the Human Protein Atlas brain RNA dataset (nTPM), both genes were detected across multiple regions with clear heterogeneity (Figure. S7A–B). RIMS1 showed a pronounced cerebellar peak, with intermediate expression in cortex, hippocampal formation, thalamus, and brainstem (midbrain/pons), and minimal signal in choroid plexus. BTBD16 displayed low–moderate, widespread expression, relatively enriched in white matter and brainstem/thalamic regions, with lower levels in limbic cortices and choroid plexus. These spatial profiles support CNS relevance of the MR findings and nominate cerebellar–brainstem circuits (RIMS1) and white-matter/brainstem pathways (BTBD16) as plausible contexts for target engagement.

Discussion

Migraine is a complex neurobiological disorder shaped by genetic, neurovascular and immune influences. Translating GWAS signals into biology requires resolving which brain cells mediate risk. Here, we integrated single-cell transcriptome-wide Mendelian randomization (scTWMR) with Bayesian colocalization to map cell-type–specific regulation. Across eight brain cell classes, we identified 11 MR-significant eGenes and, notably, prioritized inhibitory-neuronal BTBD16 (protective) and astrocytic RIMS1 (risk-increasing); both were supported by replication and colocalization. These results indicate that migraine susceptibility is not a tissue-average phenomenon but is sculpted by cell-contextual regulatory networks, refining and extending prior bulk-eQTL observations.

To better understand the biological plausibility of these prioritized genes, we examined their known cellular functions in the context of migraine-relevant neurobiology. RIMS1 encodes a RIM protein, an essential component of presynaptic active zones that orchestrates synaptic vesicle docking and calcium channel coupling. In the hippocampal CA1 region, postsynaptic RIM is crucial for NMDAR-mediated synaptic responses, playing a role in synaptic plasticity and memory processing [15, 16]. Functionally, RIM-mediated Ca²⁺-dependent synaptic vesicle fusion facilitates excitatory neurotransmission—processes closely associated with cortical spreading depression (CSD) and central sensitization, both hallmark features in migraine pathophysiology [17, 18].

Moreover, NMDA receptors (NMDARs) contribute to sustained depolarization in a subset of trigeminal ganglion (TG) neurons, enhancing trigeminal nociceptive signaling and cortical excitability. Sensitization of CGRP-dependent, NMDAR-expressing neurons further links glutamatergic signaling with migraine biology [19]. NMDARs are considered central to the initiation of CSD, particularly in migraine aura. Although RIMS1 has not yet been directly implicated in migraine in prior literature, our results suggest that experimental studies targeting RIMS1 modulation may illuminate its role in regulating NMDAR activity and calcium dynamics in migraine-related circuits.

Interestingly, RIMS1 also exhibits high expression in the cerebellum—a region increasingly implicated in migraine pathophysiology through its role in pain modulation and sensorimotor integration. The cerebellum interacts with the prefrontal cortex via thalamic circuits and contributes to the inhibition of nociceptive signaling [20, 21]. Given the established involvement of CGRP and NMDA receptor–mediated pathways in both cerebellar excitability and migraine development, RIMS1 may also exert functional effects in this region [19]. These observations expand the potential relevance of RIMS1 beyond hippocampal circuits, suggesting that its role in calcium-regulated synaptic transmission may influence multiple brain networks implicated in migraine.

BTBD16, encoding a BTB/POZ-domain protein implicated in protein-protein interactions, has shown associations with migraine and hypothyroidism, with shared genetic components enriched in immune regulatory pathways [22]. Genetic variants in BTBD16 have also been linked to bipolar disorder [23]—a condition frequently comorbid with migraine—as well as to type 2 diabetes [24]. Intriguingly, epidemiological studies in Chinese populations suggest a reduced risk of migraine among individuals with type 2 diabetes, with some patients experiencing alleviation of headache symptoms [25]. These observations raise the possibility that BTBD16 may exert a protective effect against migraine, potentially through immune or metabolic pathways.

Methodologically, moving from bulk to single-cell eQTLs marks a step-change for migraine genetics. To our knowledge, this is the first study to apply sc-eQTLs systematically to a single-cell brain atlas to interrogate migraine mechanisms. Whereas prior work (e.g., Lou et al.) [7] used MR and colocalization on bulk brain eQTLs and identified targets such as PNKP, bulk approaches aggregate heterogeneous cell populations and obscure cell-specific regulation. Although we did not replicate those bulk-derived genes, we uncovered a distinct set of cell-type–specific migraine-associated eGenes using sc-eQTL instruments—an expected divergence that likely reflects the finer resolution of single-cell analyses. Together, these observations highlight the complementarity of bulk and single-cell strategies for resolving the cellular genetic architecture of the migraine brain. By coupling sc-eQTLs with MR and colocalization, we provide a higher-resolution, biologically specific framework that strengthens causal inference, mitigates confounding, and directly links genetic variation to migraine risk through cell-type–specific gene expression. This approach offers a practical blueprint for future single-cell studies and for developing more precise therapeutic hypotheses.

Several limitations warrant consideration. First, our analyses primarily included individuals of European ancestry, which may limit generalizability; multi-ancestry cohorts will be essential to test transferability across populations with different genetic and environmental backgrounds. Second, single-cell transcript levels do not necessarily mirror protein abundance or activity: post-transcriptional regulation, degradation, and subcellular localization can decouple mRNA from protein function [26]. Accordingly, orthogonal validation—such as proteomics, CRISPR-based perturbation, and experimental model systems—will be important to confirm the biological significance of our findings. Third, although Bayesian colocalization supported the causal involvement of BTBD16 and RIMS1, the lead variants identified—rs982565 and rs10736311—have not been previously described in genome-wide association or functional studies. While this highlights the potential novelty of our findings, it also underscores the need for further experimental work to clarify the molecular roles of these variants and their regulatory impact in relevant brain cell types. Finally, while the FinnGen MIGRAINE_TRIPTAN phenotype is a widely used and pragmatic proxy for migraine diagnosis, it is based on prescription records. As such, it may partially reflect treatment access or healthcare-seeking behavior. Although triptans in Finland are strictly prescription-based and commonly reflect physician-confirmed diagnoses, some degree of misclassification or treatment bias cannot be ruled out.

Conclusion

This brief report extends bulk brain eQTL–based work to single-cell resolution, refining the cellular context of migraine genetics. Using sc-eQTL instruments across eight brain cell types, we identified 11 cell-type–specific eGenes in FinnGen and replicated key signals in UK Biobank. In particular, inhibitory-neuronal BTBD16 showed a protective association, whereas astrocytic RIMS1 increased risk; discovery-stage colocalization supported shared causal variants at both loci, and PheWAS revealed no genome-wide liabilities, strengthening target specificity. Regional expression offers plausible sites for target engagement. Together, these data provide cell-type–resolved, testable hypotheses for mechanism and therapeutics, complementing prior bulk-tissue analyses and prioritizing BTBD16 and RIMS1 for functional validation and translational follow-up.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Co–first authors: Hong Ye, Yajing Huang, and Cheng Wang contributed equally to this work Corresponding authors: Junjie Fang and Qiuhan Xu.Conceptualization: Junjie Fang, Qiuhan Xu; Methodology: Hong Ye, Yajing Huang, Cheng Wang; with input from Jiancheng Jin and Chaoya Jiang; Data curation: Hong Ye, Yajing Huang; Formal analysis (MR, colocalization, replication): Hong Ye, Yajing Huang, Cheng Wang; Visualization: Hong Ye, Cheng Wang; Validation/Interpretation: Jiancheng Jin, Chaoya Jiang; Writing – original draft: Hong Ye, Yajing Huang, Cheng Wang; Writing – review & editing: Jiancheng Jin, Chaoya Jiang, Junjie Fang, Qiuhan Xu; Supervision & project administration: Junjie Fang, Qiuhan Xu; Funding acquisition: Junjie Fang, Qiuhan Xu. All authors read and approved the final manuscript.

Funding

This work was supported by the Zhejiang Provincial Traditional Chinese Medicine Science and Technology Program (Grant No. 2024ZL986).

Data availability

Summary statistics for the discovery GWAS (FinnGen, Release 12) are accessible via the FinnGen results portal. Replication summary statistics (UK Biobank; accession GCST90473326) are available from the GWAS Catalog. Processed single-cell eQTL instruments, harmonization code, and per-cell-type MR/colocalization outputs will be deposited as Supplementary Materials and in a public repository upon acceptance. Where relevant, access details for commonly used resources are provided here for convenience: FinnGen results portal, GWAS Catalog, and other QTL resources as exemplified in prior JHP work. Note: We will also provide the exact SNP lists (effect alleles, weights, F-statistics) for all instruments as Supplementary Table S1, consistent with JHP’s data-sharing guidance.

Declarations

Ethics approval and consent to participate

This study used de-identified, summary-level genetic data from publicly available GWAS/eQTL resources. All necessary consents and ethics approvals were obtained by the original studies; no new approval was required for the present analyses.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.