The SMR and TSMR methodologies were applied to assess the influence of antidiabetic medications on stroke and related subtypes. This analysis followed three key assumptions: (1) genetic variants should exhibit a strong correlation with the exposure; (2) these variants must remain independent of confounding factors that might distort the exposure‒outcome relationship; (3) any effects on the outcome must be mediated solely through exposure, without direct influence. Colocalization analyses were subsequently conducted to strengthen evidence for significant associations. Fig. 1 presents the flowchart of the study design.

Fig. 1.

Flowchart of the study design. Abbreviations: eQTLs, expression quantitative trait loci; GWAS, genome-wide association study; FI, fasting insulin; HbA1c, glycated hemoglobin; 2hGlu, 2-hour glucose postchallenge; T2D, type 2 diabetes; FG, fasting glucose; AIS, any ischemic stroke; CES, cardioembolic stroke; LAS, large-artery atherosclerotic stroke; SVS, small-vessel stroke; SMR, summary-based Mendelian randomization; TSMR, two-sample Mendelian randomization; HEIDI, heterogeneity in dependent instruments; IVW, inverse-variance weighted; WM, weighted median.

This study considered principal antidiabetic pharmacotherapies, including metformin, GLP-1 receptor agonists, SGLT2 inhibitors, DPP-4 inhibitors, insulin and its analogs, thiazolidinediones, and sulfonylureas. Genes encoding the target proteins for these drugs were identified via the DrugBank and ChEMBL databases (Table 1). Nine genes were associated with HbA1c (Supplementary Table 3), T2D (Supplementary Table 4), fasting blood glucose (Supplementary Table 5), or 2hGlu (Supplementary Table 6), whereas no genes were associated with FI (Supplementary Table 7).

According to the SMR analysis, genetic variations in the target genes for sulfonylureas (KCNJ11) and SGLT2 inhibitors (SLC5A2) were significantly associated with AIS risk (KCNJ11: odds ratio (OR) = 1.11, 95% confidence interval (CI) = 1.01–1.21; p = 0.033; SLC5A2: OR = 1.05, 95% CI = 1.01–1.10; p = 0.017). Furthermore, the upregulation of insulin receptor (INSR) expression in the tibial artery was associated with LAS mitigation (OR = 0.66, 95% CI = 0.46–0.95; p = 0.026) (Fig. 2). The results for other target genes and their associations with stroke subtypes are detailed in Supplementary Tables 8–11. Additionally, the SMR analysis with HEIDI tests revealed no evidence of pleiotropy for the three target genes of the respective antidiabetic drugs in relation to stroke.

Fig. 2.

Forest plots of the effects of the three antidiabetic drug targets on the risk of the four stroke subtypes (results based on the SMR analysis). * indicates p 0.05. SMR, summary-based Mendelian randomization; AIS, any ischemic stroke; CES, cardioembolic stroke; LAS, large-artery atherosclerotic stroke; SVS, small-vessel stroke; SLC5A2, solute carrier family 5 member 2; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; INSR, insulin receptor.

The TSMR analysis findings corroborated those of the SMR analysis, showing that elevated KCNJ11 expression in the blood and SLC5A2 expression in the tibial artery increased AIS risk (KCNJ11: OR_ivw = 1.105, 95% CI = 1.009–1.210; p = 0.030; SLC5A2: OR_ivw = 1.043, 95% CI = 1.005–1.082; p = 0.026). Similarly, upregulated INSR expression in the tibial artery reduced LAS risk (OR_ivw = 0.687, 95% CI = 0.511–0.922; p = 0.012; Fig. 3). Additional TSMR results are provided in Supplementary Tables 12–15. No SNP agent was inadequate for heterogeneity tests, and the leave-one-out analysis results are presented in Supplementary Tables 16–19.

Fig. 3.

Forest plots of the effects of the three antidiabetic drug targets on the risk of the four stroke subtypes (results based on the TSMR analysis). * indicates p 0.05. TSMR, two-sample Mendelian randomization; AIS, any ischemic stroke; CES, cardioembolic stroke; LAS, large-artery atherosclerotic stroke; SVS, small-vessel stroke; SLC5A2, solute carrier family 5 member 2; KCNJ11, potassium inwardly rectifying channel subfamily J member 11; INSR, insulin receptor.

Next, colocalization analysis was conducted to assess whether SNPs linked to stroke and eQTLs share causal genetic variants. These results suggested that in the region, most identified genes likely possessed a common causal variant. Notably, significant colocalization was observed for INSR and SLC5A2 across datasets (PPH4 = 0.77 and 0.60, respectively), whereas KCNJ11 did not reach statistical significance (Supplementary Table 20).

The KCNJ11 gene encodes the Kir6.2 subunit, a key component in the ATP-sensitive potassium (K-ATP) channel. Sulfonylureas inhibit the K-ATP channel by binding to its SUR1 subunit, thereby promoting insulin secretion from pancreatic beta cells. Despite the longstanding use of sulfonylureas in the management of T2D, which is due to the efficacious glucose-lowering properties of these compounds, the relationship between these compounds and stroke remains controversial. A previous study demonstrated that sulfonylureas did not affect the severity of stroke in patients with diabetes [25]. Some studies suggest that sulfonylureas increase the risk of stroke, as tolbutamide has been shown to exacerbate neuronal damage in ischemia models, including a mouse middle cerebral artery occlusion (MCAO) model and an in vitro neuronal oxygen–glucose deprivation (OGD) model [26].

However, our findings align with those of a previous study that employed a two-sample MR approach, colocalization analyses, and independent blood glucose GWAS summary data to explore the associations between antidiabetic medications and stroke risk. This study demonstrated that sulfonylureas were linked to a decreased risk of both stroke and ischemic stroke, which is consistent with our findings, whereby elevated KCNJ11 (the target gene of sulfonylureas) expression is associated with an increased risk of AIS [27]. Although the KCNJ11 colocalization analysis in our study (PPH4 0.6) did not reach statistical significance, the persistent MR signal supports a causal link, warranting a mechanistic investigation.

In addition to reducing the incidence of stroke, several studies have indicated that sulfonylureas may offer potential benefits for patients with diabetes experiencing acute ischemic stroke. These benefits include associations with reduced stroke severity and decreased mortality rates [28, 29]. Clinical trial results have demonstrated that sulfonylurea compounds, particularly glibenclamide, have a significant effect on reducing cerebral edema in patients with stroke. A randomized, double-anonymized, placebo-controlled GAMES-RP phase 2 trial revealed that intravenous glibenclamide significantly reduced biomarkers of cerebral edema and showed a non-significant mortality trend. However, intravenous glibenclamide failed to meet the primary endpoint established by the trial in patients with large hemispheric infarction. Nonetheless, compared with the placebo, glibenclamide significantly improved survival and tended to improve functional outcomes at 90 days and 12 months among patients aged $\le$70 years with large hemispheric infarction [30, 31]. A retrospective cohort study revealed that pre-admission sulfonylurea use in diabetic patients with acute spontaneous intracerebral hemorrhage (ICH) was independently associated with smaller admission hematoma volumes and improved functional outcomes at discharge [32]. Another study demonstrated that intravenous glyburide is related to attenuation of vasogenic edema biomarkers (reduced T2 fluid attenuated inversion recovery (T2 FLAIR) signal intensity, diminished apparent diffusion coefficient (ADC) pseudonormalization, and lower plasma matrix metalloproteinase (MMP-9) levels) in patients with large hemispheric stroke [33]. Additionally, two studies demonstrated that the sulfonylurea compound glibenclamide could mitigate damage induced by ischemia‒reperfusion (IR) injury in the hippocampus [34] and reduce cerebral edema and the infarct volume in rats subjected to MCAO [35]. Consequently, the impact of sulfonylureas on stroke incidence and poststroke recovery, including functional outcomes and mortality, warrants further investigation.

SGLT2, which is encoded by the SLC5A2 gene, is highly expressed in the proximal renal tubules, facilitating the active transport of D-glucose against its concentration gradient across the electrochemical gradient of sodium ions. SGLT2 inhibitors, which inhibit glucose reabsorption, are oral antidiabetic drugs that successfully reduce serum glucose levels. Our study revealed that a greater risk of ischemic stroke was linked to higher SLC5A2 gene expression levels. This result is consistent with those of earlier studies showing that SGLT2 inhibitors could lower the incidence of stroke [36, 37, 38, 39, 40].

The EMPA-REG OUTCOME trial reported that empagliflozin, a SGLT2 inhibitor, significantly reduced the risk of cardiovascular mortality, all-cause mortality, and heart failure hospitalization in high cardiovascular risk patients with T2D compared to placebo [36]. A secondary analysis of the EMPA-REG OUTCOME trial demonstrated that empagliflozin consistently reduced cardiovascular risk in patients with T2D and atherosclerotic cardiovascular disease [37]. Lv et al. [38] investigated the causal relationship between SGLT1/2 inhibition and clinical signs of cerebral small vascular disease (CSVD) via a two-sample, two-step MR approach. By altering 4-acetamidobutanoate levels and cholesterol metabolism, SGLT2 inhibitors reduced the risk of SVS and increased the integrity of the white matter architecture [38]. To evaluate the association between SGLT2 inhibition and cardiovascular disease (CVD), as well as the mediating effects of blood lipids, Li et al. [39] performed a two-sample, two-step MR investigation. SGLT2 inhibition via non-high-density lipoprotein cholesterol (non-HDL-C) was linked to a lower risk of both ischemic stroke and any stroke [39]. A separate meta-analysis revealed that, compared with patients with T2D receiving alternative therapeutic drugs, patients treated with SGLT2 inhibitors had lower risks of stroke, cardiovascular events, and mortality [40]. Moreover, numerous studies have suggested that SGLT2 inhibitors may confer a neuroprotective effect against critical pathological alterations associated with CSVD [41]. Numerous studies examining the underlying mechanisms have documented the effects of SGLT2 inhibitors on blood lipid profiles [42, 43], showing reductions in total cholesterol (TCH), low-density lipoprotein cholesterol (LDL-C), and triglyceride levels [44], which are associated with a greater risk of CVD [45]. In diabetic murine models, SGLT2 inhibitors attenuate atherogenesis, a key contributor to stroke [46]. Moreover, the concurrent use of SGLT1 and SGLT2 inhibitors may significantly reduce the risk of stroke [47].

In addition to decreasing the incidence of stroke, several studies suggest that SGLT2 inhibitors may confer potential benefits in the context of ischemic stroke. In both nondiabetic and diabetic murine models, administering the SGLT2 inhibitor empagliflozin or luseogliflozin was associated with improved neurological outcomes and reduced infarct volumes. These effects may be attributed to the antioxidant, anti-inflammatory, and antiapoptotic properties, as well as a reduction in pericyte loss [48, 49, 50]. Specifically, empagliflozin was shown to mitigate the ultrastructural remodeling of the neurovascular unit and neuroglia induced by T2D, including the attenuation or loss of endothelial tight and adherens junctions within the blood‒brain barrier [51]. Additionally, canagliflozin, another SGLT2 inhibitor, was demonstrated to effectively reduce astrocyte and cerebral swelling in cerebral ischemia [52]. The energy-dependent transport of D-glucose via SGLT2 may exacerbate the energy deficit observed during cerebral ischemia, suggesting that SGLT2 inhibition could provide therapeutic benefits in this context [53].

However, discrepancies also exist among various study findings. Another meta-analysis indicated that SGLT2 inhibitors did not significantly reduce the incidence of ischemic stroke compared with placebo or other oral hypoglycemic agents [54]. A separate meta-analysis suggested a neutral effect of SGLT2 inhibitors on the risk of stroke and associated stroke subtypes while highlighting a potential protective effect against hemorrhagic stroke [55]. These results are consistent with those of other studies [56, 57, 58]. This apparent contradiction may arise from fundamental differences, as our MR analysis serves as a proxy for lifelong SGLT2 inhibition through SLC5A2 expression, whereas clinical trials evaluate the effects of shorter-term interventions. Therefore, further studies are required to elucidate these findings and clarify the therapeutic potential of SGLT2 inhibitors in stroke prevention and treatment.

The upregulation of INSR expression in the tibial artery was associated with a reduced risk of LAS. Insulin resistance, which is prevalent among patients with ischemic stroke, is linked to increased mortality, increased stroke recurrence rates, and adverse outcomes [59, 60]. Studies have shown that insulin resistance independently predicts atherosclerotic plaque progression. Key underlying mechanisms involve the disruption of insulin signaling across multiple cell types. In endothelial cells, impaired signaling reduces the production of nitric oxide (NO)—a vital vasodilator—and increases the expression of adhesion molecules (e.g., vascular cell adhesion molecule 1 (VCAM-1)), fostering immune cell recruitment and inflammation. Insulin resistance in vascular smooth muscle cells promotes proliferation and migration, contributing to arterial wall thickening. Meanwhile, resistance in macrophages impairs survival and anti-inflammatory functions, hindering cholesterol clearance and increasing plaque inflammation. Furthermore, insulin resistance adversely affects lipid metabolism, elevating triglyceride levels and lowering HDL cholesterol levels, thereby further compounding the risk of atherosclerosis [61]. The insulin-sensitizing agent pioglitazone exerts anti-atherogenic effects by ameliorating insulin resistance and influencing multiple components of insulin resistance syndrome (INRS) [62]. Our findings support the potential benefits of targeting the insulin receptor or using insulin sensitizers in this patient population.

The strengths and innovations of our study are as follows. First, we employed an SMR analysis, which helps avoid confounding bias and reverse causality, to explore potential drug targets and identify drugs to be repurposed for stroke treatment. Second, repurposing antidiabetic drugs for stroke treatment offers notable advantages, as these drugs have well-established mechanisms of action and clinical data to facilitate faster development of effective therapies. Third, to minimize spurious associations caused by population stratification, we restricted our analyses to individuals of European ancestry, which ensured that differences among populations did not confound the genetic associations observed. However, our study has several limitations. First, this study inferred drug mechanisms solely from gene expression associations, without conducting functional validation, thereby lacking direct mechanistic evidence. Second, this study was limited to evaluating the genetic influences on the incidence of stroke. The effects of antidiabetic medications on poststroke recovery, including functional outcomes and mortality, necessitate further exploration through longitudinal clinical studies. Third, the causal conclusions drawn from MR are contingent upon the following key assumptions: the genetic variants used as instruments must be associated with the drug target, exert an influence on stroke risk exclusively through that target, and not be linked to any confounders in the relationship between the drug and stroke. Additionally, genetic variations reflect the long-term effects of antidiabetic drug targets, which may differ from the effects observed during short-term use of these drugs. Furthermore, the study sample consisted predominantly of individuals of European ancestry; this demographic limitation not only risks introducing bias into the MR results but also restricts the generalizability of our findings to other ethnic groups. Finally, while FI served as a positive control, the absence of FI-associated loci (Supplementary Table 7) likely reflects insufficient statistical power in the source GWAS—notably smaller cohort sizes relative to polygenic complexity—rather than biological irrelevance. Future investigations would benefit from large-scale FI-linked GWAS consortia to resolve such modest effects. These limitations should be carefully considered when interpreting our findings.