Identification of comorbid genes between type 2 diabetes and migraine through peripheral blood single-cell and Mendelian randomization analysis

Abstract

Background

Type 2 diabetes mellitus (T2DM) is a metabolic disorder characterized by hyperglycemia and insulin resistance, Migraine is a common chronic neurological disease caused by increased excitability of the central nervous system, both exerting substantial health burdens. However, the shared genetic basis and underlying molecular mechanisms remain largely unexplored. This study integrates single-cell data and Mendelian randomization (MR) analysis to identify comorbidity-associated genes and elucidate potential mechanistic links between these two conditions.

Methods

Single-cell datasets from T2DM and migraine were analyzed to identify differentially expressed genes (DEGs). MR analysis was employed to prioritize key causal genes, followed by network-based functional characterization, disease-drug association analysis, cell annotation, and pseudo-time trajectory modeling.

Results

Analysis of single-cell data identified 2,128 migraine-associated and 3,833 T2DM-associated genes, with 714 genes shared between the two diseases. MR analysis highlighted AP4E1 and HSD17B12 as key regulators implicated in both conditions. Network analysis further linked these genes to lipid metabolism and vesicle transport pathways. Computational predictions revealed common comorbidities, including metabolic dysregulation and chemical-induced liver injury, as well as potential therapeutic agents such as valproic acid and bisphenol A. Single-cell annotation identified six major immune cell types in T2DM (T cells, NK cells, B cells, CD14 monocytes, CD16 monocytes, and dendritic cells), with T cells emerging as central players. In migraine, five immune cell types were identified (CD4 T cells, CD8 T cells, B cells, NK cells, and monocytes), with monocytes being the predominant cell type. Pseudo-time analysis delineated seven subpopulations of T cells and four subpopulations of monocytes, suggesting distinct functional trajectories in disease pathogenesis. However, due to the use of peripheral blood-derived single-cell data, genes primarily expressed in the central nervous system, such as CALCA and RAMP1, could not be detected, limiting the identification of certain migraine-specific pathways.

Conclusions

This single-cell data and MR analysis investigation identifies AP4E1 and HSD17B12 as pivotal genetic determinants in T2DM-migraine comorbidity, shedding light on their molecular interplay and potential therapeutic relevance.

Supplementary Information

The online version contains supplementary material available at 10.1186/s10194-025-02090-4.

Background

Diabetes mellitus (DM) is a complex metabolic disorder influenced by both genetic susceptibility and environmental factors. It is primarily classified into type 1 and type 2 diabetes mellitus (T1DM and T2DM), with T2DM accounting for over 90% of cases [1–3]. Characterized by chronic hyperglycemia, insulin secretion deficits, and impaired insulin signaling, T2DM represents a significant global health challenge. Current estimates indicate that approximately one in ten adults worldwide is affected, contributing to increased morbidity, mortality, and economic burden. With accelerating urbanization and lifestyle transitions, the prevalence of T2DM is projected to exceed 700 million cases by 2045 [4]. Notably, the metabolic dysregulation preceding T2DM onset progresses over years, highlighting the window of opportunity for early intervention through the modification of risk factors [5]. However, identifying molecular targets for early-stage intervention remains a critical challenge in T2DM prevention.

Migraine, a prevalent and disabling neurological disorder, arises from central nervous system hyperexcitability and ranks among the leading causes of disability globally [6]. Epidemiological studies suggest a higher prevalence in women, affecting approximately 18% of females and 6% of males, with chronic migraine impacting around 2% of the population and imposing substantial healthcare and socioeconomic burdens [7, 8]. While available therapies include pharmacological and non-pharmacological interventions, adherence to prophylactic treatment remains suboptimal, with fewer than 13% of eligible patients receiving preventive medication [7, 9]. Emerging evidence implicates distinct molecular signatures, including inflammatory mediators and protein biomarkers, in migraine pathophysiology, offering new avenues for targeted interventions [10].

Migraine exhibits extensive comorbidities, spanning neurological, cardiovascular, psychiatric, and metabolic disorders [11, 12]. Intriguingly, epidemiological studies suggest an inverse association between T2DM and migraine, raising the possibility of shared genetic determinants and overlapping biological pathways [13–15]. However, the molecular mechanisms underpinning this relationship remain largely uncharacterized. Integrating Mendelian randomization (MR) analysis with single-cell data analysis enables systematic identification of comorbid genes and elucidation of their roles within relevant pathophysiological networks.

In this study, we leverage MR and single-cell data analysis to identify and prioritize shared molecular signatures between migraine and T2DM. We further investigate the biological pathways, regulatory networks, and pharmacological implications of these comorbid genes, aiming to provide novel insights into disease pathogenesis and potential therapeutic targets. Nonetheless, as this study relies on peripheral blood single-cell data, genes with central nervous system-specific expression may be underrepresented, which could limit the detection of core migraine-associated pathways.

Materials and methods

Data source

Single-cell datasets for T2DM and migraine were obtained from the Gene Expression Omnibus (GEO) under accession numbers GSE280401 and GSE269117, respectively. The GSE280401 dataset comprises single-cell data from two T2DM patients and three healthy controls, while the GSE269117 dataset includes data from five migraine patients and five healthy controls. To assess the impact of candidate genes on T2DM and migraine, MR analysis was conducted using candidate genes as exposure factors, with T2DM and migraine as separate outcome events. The genome-wide association studies (GWAS) data for T2DM-related genes were obtained from the database ebi-a-GCST90018926 (Population: European), while the GWAS data for migraine-related genes were sourced from ebi-a-GCST90038646 (Population: N/A).

Single-cell data processing

Single-cell data were processed using the Seurat package [16]. A Seurat object was created for each dataset, followed by quality control (QC) filtering. QC thresholds were defined as follows: for GSE280401, cells were retained if the number of detected genes (nFeature_RNA) ranged between 800 and 5000, and mitochondrial gene expression (percent.mt) was below 10%; for GSE269117, cells with nFeature_RNA between 600 and 6000 and percent.mt below 20% were retained. Data normalization was performed using the NormalizeData function, followed by identification of highly variable genes using the FindVariableFeatures function with the variance-stabilizing transformation (vst) method, selecting the top 2000 variable genes [17]. The ScaleData function was applied to standardize expression values before dimensionality reduction. Principal component analysis (PCA) was conducted, and the significance of principal components was evaluated using the JackStraw and ScoreJackStraw functions. The optimal number of principal components was determined based on scree plots and elbow plots. Cell clustering was performed at a resolution of 0.3, followed by Uniform Manifold Approximation and Projection (UMAP) to visualize cellular distributions in a reduced-dimensional space. Cell-type annotation was achieved using established marker genes [18] and the CellMarker 2.0 database, enabling classification of cell clusters into distinct immune populations. The expression patterns of marker genes across cell populations were visualized using heatmaps, and differential cell-type abundances were compared between disease and control groups.

Identification of differentially expression genes (DEGs) and enrichment analysis

Differentially expressed genes (DEGs) were identified across distinct cell types using the FindMarkers function in Seurat. A gene was considered differentially expressed if it met the criteria |log fold change (logFC)| > 0.5 and p < 0.05. To identify shared molecular signatures between T2DM and migraine, overlapping DEGs were extracted. Functional enrichment analysis was conducted using the clusterProfiler package [19], including Gene Ontology (GO) annotation, including biological processes (BP), cellular components (CC), and molecular functions (MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis.

MR analysis identifies key genes

Two-sample MR analysis was conducted using R packages, leveraging publicly available genome-wide association study (GWAS) summary statistics from the IEU OpenGWAS database. To infer causal relationships between exposure and outcome traits, MR analysis was performed using five complementary methods: MR Egger [20], weighted median [21], inverse variance weighted (IVW) [22], simple mode [23], and weighted mode [24]. The IVW method served as the primary statistical approach, where a p < 0.05 indicated a significant causal association between the exposure and the outcome. The odds ratio (OR) was calculated, with OR > 1 suggesting a risk factor and OR < 1 indicating a protective factor. To identify genes with significant causal relationships with T2DM and migraine, MR analysis was applied to genetic variants associated with both diseases. Heterogeneity analysis was conducted using the mr_heterogeneity function in the TwoSampleMR package to assess variability across instrumental variables. Horizontal pleiotropy was evaluated using the mr_pleiotropy_test function to detect potential biases due to pleiotropic effects. Leave-one-out sensitivity analysis was performed using the mr_leaveoneout function to assess the influence of individual genetic variants on the overall MR results. MR-Steiger filtering analysis was applied to infer the directionality of causation between the exposure and outcome variables.

Analysis of regulatory networks of key genes and drug prediction

To investigate the functional interactions of key genes, a gene-gene interaction (GGI) network was constructed using the GeneMANIA platform. The top 20 genes exhibiting the highest correlation with the key genes were identified, alongside the seven most significantly enriched pathways associated with these genes. To explore the regulatory mechanisms at the post-transcriptional level, microRNA (miRNA) predictions were performed using the miRWALK (http://mirwalk.umm.uni-heidelberg.de) and miRDB (http://www.mirdb.org) databases. Common miRNAs targeting the key genes were identified, providing insights into potential regulatory interactions influencing disease pathogenesis. Gene-disease associations were further analyzed using the Comparative Toxicogenomics Database (CTD), which systematically predicts gene-disease interactions. The top five diseases associated with the key genes were selected for visualization. Additionally, to identify potential therapeutic targets, the top 10 diseases linked to the key genes based on CTD predictions were analyzed to explore their relevance in drug discovery.

Acquisition of putative relevant cells types and pseudo-time analysis

To assess the differential expression of key genes across distinct cellular populations, the Wilcoxon test was applied to all samples in the single-cell datasets GSE280401 and GSE269117. Expression levels of key genes were compared between disease and control groups within each identified cell type. Cell types exhibiting significant differential expression of key genes (p < 0.05) were defined as putative relevant cell types. To investigate the cellular dynamics along the disease trajectory, pseudo-time analysis was performed on putative relevant cell clusters. Uniform Manifold Approximation and Projection was utilized to refine the clustering of putative relevant cell types into distinct subpopulations. The FindAllMarkers function in Seurat was applied with parameters min.pct = 0.2 and only.pos = TRUE to identify marker genes for each subpopulation. To model the temporal evolution of cellular states, Branched Expression Analysis Modeling (BEAM) from the Monocle package [25] was employed to infer cell-state transitions along the pseudotime axis. Dynamic gene expression patterns were visualized using the plot_pseudotime_heatmap function, providing insights into the temporal regulation of key prognostic genes during disease progression.

Statistical analysis

Statistical analyses were conducted using R software (version 4.2.2) and images were modified using Adobe Illustrator (version 2023). The p < 0.05 was considered statistically significant.

Results

Analysis of single-cell data for T2DM

Following QC filtering, the GSE280401 dataset retained a total of 6,702 cells (4,116 from controls and 2,586 from T2DM patients), encompassing 25,509 genes (Figure S1 A). Using the FindVariableFeatures function, the top 2,000 highly variable genes were selected by analyzing the relationship between mean expression and variance. Among them, the top 10 most variable genes were highlighted (Fig. 1A). PCA dimensionality reduction analysis showed that the principal components (PCs) of the control and T2DM samples were well integrated (Figure S1 B). Dimensionality reduction and clustering analysis revealed that the first 50 PCs contributed substantially to overall variance, with a stabilization trend observed beyond PC 30. This suggests that the primary biological signals originate predominantly from the first 30 PCs, which were selected for downstream analyses (Figure S1 C). After dimensionality reduction analysis, 11 cell clusters were identified (Fig. 1B). Subsequently, the cells were annotated into six distinct cell types, including T cells, NK cells, B cells, CD14 monocytes, CD16 monocytes, and Dendritic cells (Fig. 1C, D). The abundance of cell types showed that NK cells accounted for a higher proportion in the control group, while T cells accounted for a higher proportion in both groups (Fig. 1E). Between T2DM and control groups, there were 501 upregulated and 532 downregulated genes in B cells, 78 upregulated and 218 downregulated genes in Basal cells, 485 upregulated and 384 downregulated genes in CD14 monocytes, 274 upregulated and 735 downregulated genes in CD16 monocytes, 232 upregulated and 780 downregulated genes in NK cells, and 199 upregulated and 553 downregulated genes in T cells. The union of DEGs across all cell types was designated as T2DM_gene, comprising a total of 3,833 genes for subsequent analysis (Fig. 1F).

Fig. 1.

Analysis of single-cell data for T2DM. A Identification of highly variable genes. B UMAP dimensionality reduction clustering analysis and grouped dimensionality reduction clustering analysis. C Expression levels of highly expressed genes specific to each cell cluster. D Cell annotation analysis and grouped cell annotation analysis. E Proportion of each cell type in different samples. F Manhattan plot of DEGs among different cell types

Analysis of single-cell data for migraine

In the GSE269117 dataset, we obtained 25,012 cells (12,430 controls and 12,582 with migraine) and 18,530 genes were retained after QC (Figure S2 A). Using the FindVariableFeatures function, the top 2,000 highly variable genes were selected by analyzing the relationship between mean expression and variance, and the top 10 genes were labeled (Fig. 2A). Normalization results showed that the PCs of the two samples merged well (Figure S2 B). The cellular dimensionality reduction and clustering analysis revealed the contribution of the first 50 PCs to the variation, with a gradual stabilization after the 30th PC. This suggested that the true signals mainly originate from the first approximately 30 PCs, which can be selected for subsequent analysis (Figure S2 C). After dimensionality reduction analysis, 14 cell clusters were identified (Fig. 2B). The cells were annotated into five distinct cell types, including CD4 T cells, CD8 T cells, B cells, NK cells, and monocytes (Fig. 2C, D). The abundance of cell types showed that B cells accounted for a lower proportion in CRC group samples (Fig. 2E). Between migraine and controls, there were 172 upregulated and 93 downregulated genes in B cells, 211 upregulated and 100 downregulated genes in CD4 T cells, 162 upregulated and 87 downregulated genes in CD8 T cells, 516 upregulated and 1,119 downregulated genes in monocytes, and 203 upregulated and 33 downregulated genes in NK cells. The union of differential genes among cell types was designated as migraine gene, with a total of 2,128 genes used for subsequent analysis (Fig. 2F).

Fig. 2.

Analysis of single-cell data for migraine. A Identification of highly variable genes. B UMAP dimensionality reduction clustering analysis and grouped dimensionality reduction clustering analysis. C Expression levels of highly expressed genes specific to each cell cluster. D Cell annotation analysis and grouped cell annotation analysis. E Proportion of each cell type in different samples. F Manhattan plot of DEGs among different cell types

Identification of candidate genes

A total of 714 candidate genes associated with both T2DM and migraine were identified (Fig. 3A). GO enrichment analysis revealed 428 enriched terms, with 355 BP terms, 16 MF terms, and 57 CC terms (Fig. 3B-D). The most significantly enriched terms included positive regulation of cytokine production, MHC protein complex binding, and cytosolic ribosome, among others. KEGG pathway enrichment analysis identified 48 significantly enriched pathways, primarily involving Leishmaniasis, Coronavirus disease-COVID-19, and Yersinia infection, and other pathways (Fig. 3E).

Fig. 3.

Identification of candidate genes. A Identification of candidate genes. B-D GO enrichment analysis. E KEGG enrichment analysis

MR analysis identifies key genes

Using the candidate genes as exposure factors and T2DM as the outcome event for MR analysis, a total of 34 genes with significant causal relationships with T2DM were identified. Among them, BIRC3, ST6GAL1, RPLP0, ADGRE2, CD68, ELP2, WDR45B, CCT6A, CAMLG, RPL13, TRIM56, SLC9A9, EXT1, LIMK2, TMEM50A, ZFP36L1, KCNQ5, RNPS1, and PPM1N are risk factors (OR > 1), while the remaining are protective factors (OR < 1) (Fig. 4A). To assess the correlation between the exposure factors and the outcome, a scatter plot was generated, combining the SNP-exposure effect and the SNP-outcome effect (Fig. 4B). A forest plot was also constructed to visualize the diagnostic efficacy of the estimated exposure factors from each SNP on the outcome (Fig. 4C). A randomness test was conducted, and the results showed a symmetrical distribution of single-nucleotide polymorphisms (SNPs) for the identified genes (Fig. 4D). Heterogeneity analysis indicated no heterogeneity among the instrumental variables (IVs), with p > 0.05 (Table S1). Horizontal pleiotropy test further confirmed the absence of confounding factors, as all p > 0.05, supporting the validity of the analysis (Table S2). Leave-one-out sensitivity analysis revealed that excluding individual SNPs did not significantly alter the effect of the remaining SNPs on the outcome, reinforcing the reliability and stability of the MR results (Fig. 4E). Finally, Steiger’s directionality test was performed to infer the direction of causality between the exposure factors and the outcome. The results confirmed that all 34 exposure factors showed a consistent causal direction (TRUE), i.e., all were retained (Table S3). Based on the above results, the 34 genes selected by MR were designated as genes with a significant causal relationship with diabetes and are denoted as T2DM_MR_genes.

Fig. 4.

MR analysis of T2DM. A Exposure outcome MR analysis. B Scatter plot. C Forest plot. D Funnel plot. E Leave-one-out analysis

MR analysis using candidate genes as exposure factors and migraine as the outcome identified 32 genes with significant causal associations with migraine. Among these, ABCB4, SLC2A3, ME1, GPI, MICU1, GNLY, BCL11A, RUNX2, RPL13A, PDGFC, PPP4C, VOPP1, SCIMP, IFI16, and GHITM were identified as risk factors (OR > 1), while the remaining genes were classified as protective factors (OR < 1) (Fig. 5A). To evaluate the correlation between the exposure factors and the outcome, a scatter plot was created, combining the SNP-exposure effect and the SNP-outcome effect (Fig. 5B). A forest plot was also generated to assess the diagnostic efficacy of the estimated exposure factors from each SNP on the outcome (Fig. 5C). A randomness test was performed to assess whether the analysis could be attributed to random variation. The results showed a symmetrical distribution of SNPs for the identified genes, confirming the robustness of the findings (Fig. 5D). Heterogeneity analysis showed no evidence of heterogeneity among the IVs, with all Q_pval values exceeding 0.05 (Table S4). Horizontal pleiotropy test was carried out, and the results confirmed that no pleiotropic effects were present (p > 0.05), further ensuring the reliability of the analysis results (Table S5). Leave-one-out analysis was conducted, the results revealed that excluding individual SNPs had minimal impact on the effect of the remaining SNPs on the outcome, further supporting the reliability and stability of the findings (Fig. 5E). To explore the causal direction between the exposure factors and the outcome, Steiger’s directionality test was performed, and the results showed that all 32 exposure factors retained a consistent causal direction (TRUE) (Table S6). Based on the above results, the 32 genes selected by MR were designated as migraine_MR_genes with significant causal relationships with Migraine.

Fig. 5.

MR analysis of migraine. A Exposure outcome MR analysis. B Scatter plot. C Forest plot. D Funnel plot. E Leave-one-out analysis

Analysis of regulatory networks of key genes and drug prediction

The intersection of T2DM_MR_genes and migraine_MR_genes revealed two key genes: AP4E1 and HSD17B12 (Fig. 6A). A GGI network analysis was performed to explore their functional associations, identifying several genes, including AP4B1, AP4S1, and TEPSIN, that are closely linked to these key genes. Additionally, enriched pathways related to these genes included the AP-type membrane coat adaptor complex, membrane coat, and coated membrane (Fig. 6B). Regulatory network analysis indicated that AP4E1 was predicted to interact with 82 miRNAs across two databases, whereas HSD17B12 was predicted to interact with 22 miRNAs in the same databases (Fig. 6C). Disease prediction analysis associated both genes with conditions such as weight loss, chemical and drug-induced liver injury, necrosis, and hyperplasia (Fig. 6D). Furthermore, these key genes were linked to potential drug interactions with Aflatoxin B1, Bisphenol A, and Valproic Acid (Fig. 6E).

Fig. 6.

Analysis of regulatory networks of key genes and drug prediction. A Screening of key genes. B Construction of GGI network for key genes. C Regulatory network analysis. D Disease prediction based on key genes. E Drug prediction based on key genes

Putative relevant cell types acquisition and pseudo-time analysis

To identify differential cell types associated with T2DM, we analyzed all samples from the GSE280401 dataset. Our findings revealed that T cells were notably significant across both genes, leading to their designation as the putative relevant cell population (Fig. 7A). Cell clustering analysis further classified T cells into seven distinct subpopulations (Fig. 7B). Among these, subpopulation 2 was identified as the starting point in the developmental trajectory and was labeled as the initial cell. Over time, this subpopulation differentiates into other T cell types (Fig. 7D). To explore the expression dynamics of key genes during the differentiation process, we examined the expression of prognostic genes at various stages within the T cell developmental continuum. The analysis revealed that these two genes were divided into two clusters based on their expression patterns across cellular stages (Fig. 7C).

Fig. 7.

Obtain differential cells related to T2DM and pseudo-time analysis. A Expression levels of the AP4E1 and HSD17B12 in different samples. B Clustering analysis of T cells. C Pseudo-time trajectory dynamic heatmap. D Pseudo-time trajectory analysis of T cells. Note: ‘ns’ represents that it is not significant; ‘*’ represents that p < 0.05; ‘**’ represents that p < 0.01; ‘***’ represents that p < 0.001

To investigate the differential cell populations associated with migraine, we analyzed all samples from the GSE269117 dataset. The results highlighted monocytes as the putative relevant cell population, as they exhibited significant relevance across both genes (Fig. 8A). Cell clustering analysis further subdivided monocytes into four distinct subpopulations (Fig. 8B), with subpopulation 1 identified as the initial cell, serving as the starting point for differentiation. Over time, this subpopulation undergoes differentiation into other monocyte subsets (Fig. 8D). To examine the expression dynamics of key genes during the differentiation process, we assessed the expression of prognostic genes at various stages of monocyte differentiation within the dataset. The analysis revealed that the two genes were grouped into two clusters (Fig. 8C).

Fig. 8.

Obtain differential cells related to migraine and pseudo-time analysis. A Expression levels of the AP4E1 and HSD17B12 in different samples. B Clustering analysis of monocytes. C Pseudo-time trajectory dynamic heatmap. D Pseudo-time trajectory analysis of monocytes. Note: ‘ns’ represents that it is not significant; ‘*’ represents that p < 0.05; ‘**’ represents that p < 0.01; ‘***’ represents that p < 0.001

Discussion

Our study identified AP4E1 and HSD17B12 as the comorbid genes associated with both T2DM and migraine, offering new insights into the pathogenesis of these two conditions. The AP4E1 gene encodes a protein that is a member of the adaptor protein complex large subunit family. These proteins are integral components of heterotetrameric adaptor protein complexes, which play crucial roles in the secretory and endocytic pathways. They mediate vesicle formation and sorting of integral membrane proteins, particularly from the trans-Golgi network to the endosomal-lysosomal system. Disruption of AP4E1 has been linked to a range of disorders, including Spastic Paraplegia 51, Cerebral Palsy, Familial Persistent Stuttering, AIg12-Congenital Disorder of Glycosylation, and AP4 Deficiency Syndrome [26–28], emphasizing its critical role in neuronal function and axonal transport. Previous genetic studies have demonstrated several pathological processes initiated by AP4E1 mutations [29–32]. First, abnormal lysosomal enzyme sorting leads to substrate accumulation and triggers a disorder of the autophagy-lysosomal system. Second, impaired clearance of misfolded proteins such as β-amyloid promotes the accumulation of toxic substances in neurons. Third, transport of organelles (mitochondria and synaptic vesicles) in neuronal axons is blocked, resulting in decreased synaptic transmission efficiency. Fourth, abnormal lysosomal function releases proinflammatory factors, exacerbating neuroinflammation and oxidative stress.

Protein expression analyses from various existing databases, including the Human Protein Atlas, GTEX, and UniProt, indicate that AP4E1 is predominantly expressed in neurons, particularly in the cerebral cortex, hippocampus, and cerebellum. This expression pattern is consistent with its role in regulating synaptic vesicle transport, neurotransmitter release, and axonal transport. Additionally, AP4E1 alleles are specifically enriched in active enhancers of monocytes, macrophages, and microglia. While there is currently no direct literature examining the relationship between AP4E1 and both T2DM and migraine, its functional roles in vesicle trafficking and lysosomal regulation suggest a potential indirect association with the pathophysiology of these conditions. Gene-gene interaction studies have demonstrated that SNPs in one gene can influence the expression of other genes [33]. Some SNPs or mutations in genes involved in solute carrier regulation, ion channels, glutamatergic neurotransmission, and cortical excitability have been linked to both polygenic and monogenic forms of migraine [34]. GWAS have identified that the AP4E1 rs3751591 SNP is associated with estradiol levels, which in turn can influence the susceptibility to cortical spreading depression [35]. As calcitonin gene-related peptide (CGRP) receptors, the endoplasmic reticulum is expressed in the same brain regions, including trigeminal, indicating that estrogen can participate in the regulation of CGRP signaling pathways in migraine pathophysiology [36]. This hormonal theory is supported that gene variants of estrogen metabolism and receptors have been related with menstruation-related migraine [35]. Given AP4E1’s involvement in intracellular vesicle trafficking and protein sorting, its expression patterns are likely closely tied to these functions. Insulin secretion by pancreatic β-cells relies heavily on vesicular transport mechanisms, and the AP4 complex may be critical for the trafficking of insulin-containing vesicles within these cells [4]. Functional impairments in AP4E1 could disrupt insulin secretion, ultimately leading to hyperglycemia. Although direct evidence connecting AP4E1 to diabetes remains sparse, its known biological functions suggest it may play a pivotal role in insulin secretion, glucose transport regulation, and inflammatory responses. Future studies should integrate approaches from genetics, molecular biology, and clinical research to further explore its mechanisms.

The HSD17B12 gene encodes an important enzyme, 17beta-hydroxysteroid dehydrogenase (17β-HSD), which is involved in the production of very long-chain fatty acids. These fatty acids serve as precursors for membrane lipids and lipid mediators, playing critical roles in various biological processes. HSD17B12 also catalyzes the conversion of estrone to estradiol, thus contributing to estrogen formation. Diseases associated with HSD17B12 include Congenital Central Hypoventilation Syndrome, Neuronal Ceroid Lipofuscinosis 1, Adrenoleukodystrophy, and Neuroblastoma [37–39]. It participates in key pathways such as the synthesis of very long-chain fatty acyl-CoAs, lipid metabolism, and steroid hormone metabolism. Recent studies have shown that a deficiency in HSD17B12 dosage leads to premature ovarian failure in both humans and mice [40]. Recombinant HSD17B12 has been shown to catalyze the reduction of estrone to estradiol in HEK293 cells [41]. HSD17B12 was previously reported that it regulate female ovarian function and fertility via the metabolic processes of arachidonic acid involved in the prostaglandin synthesis pathway [42]. These findings suggest that HSD17B12 is a predominant source of estrogens. Estrogen is well-known for its modulatory effects on pain perception, especially within the trigeminovascular system, which is considered the anatomical and physiological basis for the pathogenesis of migraine. However, it remains uncertain whether HSD17B12 variants contribute to the modest increase in migraine risk. The limited evidence on the genetic basis of migraine is largely due to the use of small cohort sizes and variant case definitions. Therefore, more extensive genetic studies are necessary to determine the role of HSD17B12 in migraine pathogenesis.

Furthermore, this study highlights HSD17B12 as a potential causal gene for T2DM. The gene has been identified as playing an important role in controlling body fat size, fat cell numbers, and diabetes risk, positioning it as a significant genetic risk locus for T2DM [43]. RNA-seq analysis of non-diabetic islets revealed that HSD17B12 is highly expressed in human pancreatic islets, with upregulated expression observed in donors with T2DM. Differential expression analysis indicated that HSD17B12 expression was significantly decreased in diabetic islets compared to non-diabetic islets [44]. These findings suggest that HSD17B12 may serve as a novel candidate gene for assessing pancreatic β cell function. Supporting this, published studies have shown elevated expression levels of HSD17B12 in tissues involved in lipid metabolism, such as the liver, kidney, skeletal muscle, and adipose tissue in both humans and mice [37, 44, 45]. HSD17B12 could thus potentially serve as an early-stage biomarker in blood samples from prediabetic individuals. Further functional research is needed to clarify the role of HSD17B12 in the pathogenesis of T2DM and to facilitate the development of targeted therapeutic interventions.

We also observed that diseases such as weight loss, chemical and drug-induced liver injury, necrosis, and hyperplasia, along with drugs like aflatoxin B1, bisphenol A, and valproic acid, were commonly associated with both T2DM and migraine. Weight loss has been shown to have dose-dependent ameliorative effects on T2DM [46]. A random-effects meta-analysis revealed that weight loss significantly reduced migraine frequency. The underlying mechanisms connecting weight loss and migraine may involve chronic inflammation and obesity-related comorbidities [47]. A previous study reported that female pubertal T2DM mice were more prone to liver injury induced by Di-(2-ethylhexyl) phthalate [48]. Additionally, a case report described a migraine patient who developed drug-induced liver injury associated with rizatriptan [49]. Recent data suggested that the rebalancing of Nrf2/NOX2 pathways prevented neuronal loss in the brain and reduced glial hyperplasia by inhibiting the activation of NF-kB and NLRP3 inflammasome, thereby mitigating nervous system injury during migraine [50].

Bisphenol A, one of the most prevalent endocrine-disrupting chemicals, is strongly implicated in the development of T2DM. Exposure to bisphenol A has been linked to damage in organs targeted by T2DM and may exacerbate the progression of several chronic complications associated with the disease [51]. Bisphenol A is known to exhibit estrogenic activity and may influence the intensity and duration of migraine attacks via estrogen receptor signaling pathways [52]. Valproic acid has been shown to significantly reduce blood glucose levels and decrease the expression of gluconeogenic genes in hyperglycemic Otsuka Long-Evans Tokushima Fatty rats [53]. In addition, valproic acid was associated with a 50% or greater reduction in the number of migraine attack days per month in adult patients, with the treatment being generally well-tolerated [54]. These findings suggest potential therapeutic avenues for combined treatments targeting both T2DM and migraine, which could guide the development of more effective and integrated treatment strategies.

Through single-cell analysis, our study identified T cells as putative relevant cell types in T2DM pathology, its specific role still requires further functional studies for validation. Increasing evidence highlights the abnormal immune response of CD4 + T cells in T2DM. For instance, the proportion of Th1 cells, which are involved in inflammation, is elevated in adipose tissue and peripheral circulation. Similarly, the proportion of the Th17 subset is increased, accompanied by higher levels of IL-17 in the peripheral blood. IL-17 can promote the production of TNF-α, leading to the development of obesity and insulin resistance in the body [55–58]. In contrast, the defective Th 2 immune response with anti-inflammatory effects and the decreased proportion of regulatory T lymphocytes (Treg cells) further lead to the reduction of Treg/Th 17 and Treg/Th 1 [57, 59–61].

The pathogenesis of T2DM is primarily driven by insulin resistance and insufficient insulin secretion. IFN-γ, TNF-α, and IL-2 cytokines secreted by CD4 + T cells inhibit lipoprotein lipase activity, promote adipocyte fat breakdown, and increase free fatty acid levels. These free fatty acids impair insulin signaling by inhibiting the activity of phosphoinositide 3-kinase, suppressing the expression of insulin receptor substrate-1, and inhibiting tyrosine phosphorylation. This results in reduced insulin sensitivity in tissues and organs, contributing to the development of insulin resistance [55, 57, 58]. Additionally, IFN-γ and TNF-α can indirectly exacerbate the inflammatory response by synergizing with other cytokines like IL-1β. Co-culturing mouse pancreatic β cells with IFN-γ, TNF-α, and IL-1β promotes the expression of inducible nitric oxide synthase, which increases nitric oxide production. This impairs glucose-induced insulin secretion by the pancreatic β cells [62]. Furthermore, the combination of TNF-α and IL-1β can induce Fas-mediated apoptosis, leading to pancreatic β cell death. IL-29 has been shown to upregulate TNF-α mRNA expression in Th1 cells through NFATc1 activation, amplifying the Th1-mediated inflammatory response in T2DM [55].

Our single-cell analysis also identified monocytes as the putative relevant cells types in the pathogenesis of migraine, its specific role still requires further functional studies for validation. Monocytes are a subset of mononuclear leukocytes (white blood cells) that play critical roles in both innate and adaptive immunity. These cells can differentiate into macrophages or dendritic cells and are recruited into tissues, where they contribute to immune responses and inflammation [63]. Mounting evidence links migraine to neurogenic inflammation, in which inflammatory mediators activate and sensitize peripheral nociceptors [64–66]. Elevated levels of inflammatory agents can provoke the activation of trigeminal nerves and the release of various vasoactive neuropeptides such as CGRP, substance P (SP) and neurokinin A, that induce vasodilatation, secondary extravasation, edema and mast cell degranulation, consequently contributing to neurogenic inflflammation [66, 67]. Gallai et al. observed a significant increase in the phagocytic and chemotactic responses of monocytes in both migraine with aura and migraine without aura patients, compared to those with tension-type headaches [68]. Further studies have shown that monocyte count and the monocyte-to-lymphocyte ratio were notably higher during migraine attacks than in healthy controls [69]. Han et al. reported excessive spontaneous release of IL-1β, IL-6, and TNF-α by monocytes in migraine patients [70]. Previous studies also demonstrated that TNF-α exerts proinflammatory effects by inducing procoagulant activity, activating plasminogen, and promoting the adhesion of leukocytes to endothelial cells during migraine attacks [71]. SP is considered as one principal mediator of migraine, which stimulates monocyte chemotaxis and the release of proinflammatory compounds such as prostaglandin E and thromboxane A2 derived from guinea-pig macrophages and cultured astrocytes [72, 73]. The release of SP from sensory nerve fibers upon activation can further stimulate monocytes to release cytokines such as IL-1β, IL-6 and TNF-α, which may represent one of the pathways through which sensory nerve fibers lead to neurogenic inflammation. The synergistic interaction between SP and these mediators in enhancing monocytes responses may be an important mechanism underlying the initiation and maintenance of neurogenic inflammation during migraine attacks. In addition, monocytes infiltrate into the meninges or perivascular regions during migraine attacks. Their released proinflammatory factors can activate trigeminal nerve endings, inducing neurons to release CGRP, which modulates the activity of monocytes through its cognate receptor, receptorassociated modifying protein and calcitonin receptor-like receptor, thereby forming an “inflammation-neuropeptide” positive feedback loop that exacerbates vascular dilation and pain [70, 71]. CGRP is now recognized as a central player in neurogenic inflammation and the sensitization of the trigeminovascular system, a key mechanism underlying migraine. CGRP has gained prominence as a potential biomarker for migraine. Recently, the FDA approved CGRP receptor antagonists and monoclonal antibodies targeting the CGRP signaling pathway, which have demonstrated significant efficacy in improving both acute and preventive migraine treatments [74, 75].

This study provides novel insights into the shared genetic architecture of T2DM and migraine through single-cell data and MR analysis, identifying AP4E1 and HSD17B12 as key genes with T2DM and migraine, suggesting shared genetic susceptibility contribute to the comorbidity between T2DM and migraine. The finding confirms previous study that showed a higher co-occurrence of both diseases [13]. Despite these findings, limitations remain, including the need for larger and more diverse cohorts to account for gender and ethnic heterogeneity, as well as the challenge of fully capturing complex exposure-response dynamics beyond the linear assumptions of MR analysis. Another limitation is the data source of single cells. The analysis in this study was based on peripheral blood samples, which might not be able to fully capture the expression characteristics of genes that are predominantly expressed in the central nervous system or the trigeminal nerve. For example, key genes in the CGRP signaling pathway (such as CALCA, RAMP1) [76]. These genes have been confirmed in the pathophysiology of migraine, but were not identified in this study, possibly because their expression levels in peripheral blood cells were low or they were not expressed at all. The single-cell dataset used in this study did not provide donor-specific information such as BMI and age, which limited our ability to perform conventional covariate adjustment analyses. Future studies are warranted to validate our findings using datasets with more comprehensive clinical annotations. Nevertheless, the identification of AP4E1 and HSD17B12 as potential therapeutic targets provides a compelling basis for drug development, with the prospect of repurposing or designing interventions that simultaneously address both T2DM and migraine. Understanding the genetic and molecular convergence of these conditions not only refines disease classification but also opens avenues for precision medicine strategies aimed at mitigating their shared pathophysiological mechanisms.

Conclusion

By integrating single-cell data and MR analysis, this study uncovers the shared genetic architecture of T2DM and migraine, identifying AP4E1 and HSD17B12 as key comorbidity-associated genes. Single-cell data analysis reveals that T cells are the predominant immune cell type implicated in T2DM, with 3,833 associated genes (TM2D_gene), while monocytes emerge as the critical cell type in migraine, with 2,128 associated genes (MI_gene). Cross-referencing these datasets identifies 714 overlapping genes, and MR analysis further pinpoints AP4E1 and HSD17B12 as the most significant shared genetic factors contributing to the co-occurrence of these diseases. These findings provide new insights into the molecular interplay underlying T2DM and migraine comorbidity, offering potential therapeutic targets for precision medicine. Future studies should further investigate the regulatory mechanisms of these genes and explore targeted interventions to improve the integrated management of both conditions.

Abbreviations

T2DM

Type 2 diabetes mellitus

MR

Mendelian randomization

DEGs

Differentially expressed genes

GEO

Gene Expression Omnibus

QC

Quality control

PCA

Principal component analysis

UMAP

Uniform Manifold Approximation and Projection

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

GWAS

Genome-wide association study

OR

Odds ratio

GGI

Gene-gene interaction

CTD

Comparative Toxicogenomics Database

BEAM

Branched Expression Analysis Modeling

SNPs

Single-nucleotide polymorphisms

CGRP

Calcitonin gene-related peptide

SP

Substance P

Authors' contributions

BBY and GGL contributed to the study conception and design. BBY and RRM downloaded the database. JRL, QS and XYW performed the statistical analysis. BBY drafted the manuscript. All authors commented on previous versions of the manuscript. All authors contributed to the article and approved the submitted version.

Funding

This study was funded by Key Research and Development Program Project of Shaanxi province (2020SF-335), Young Talent Development Program of Xi’an Ninth Hospital (2020-5).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All data analyzed during this study have been previously published.

Competing interests

The authors declare no competing interests.