Our research employed a 2-sample MR design based on the 3 key MR assumptions (18). Figure 1 illustrates the strategy used in this study. The MR studies were reported in accordance with the suggestions made by Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization reporting guidelines (19). No further ethical approval was required because the genome wide association study (GWAS) summary data utilized in this study were already publicly available and the first scientific investigation passed ethical assessment. Supplementary Table S1 lists the download links for the GWAS summary data used in this study.

Pooled GWAS data for HDP (n=296,824) and PE (130,207) were obtained from a recent GWAS meta-analysis involving 4 cohorts (20). In this study, the term ‘PE’ encompasses not only preeclampsia and eclampsia but also chronic hypertension with superimposed preeclampsia. For clarity, all three conditions are referred to as ‘PE’ in the subsequent sections of this paper. The GWAS summary data for GH (n= 202,768) were extracted from the FinnGen Consortium (R9) (21, 22). The MEGASTROKE consortium, which performed a meta-analysis of 29 European-descent GWAS studies, provided a pooled set of GWAS data for stroke, including 40,585 stroke cases and 406,111 controls (8). All stroke cases were confirmed using clinical and imaging criteria. including any stroke. A stroke encompasses all types of strokes, whereas an IS specifically refers to those diagnosed as IS. According to the TOAST classification criteria, IS is categorized into three subtypes: LAA, SVS, and CES. Summary data on lacunar stroke (n= 254,959) were obtained from the most recent GWAS study (23). Lacunar stroke was defined according to the TOAST criteria or brain magnetic resonance imaging. Pooled GWAS data for intracranial hemorrhage (ICH) (n=473,513) were obtained from a study on genetic associations in human disease phenotypes (24). Supplementary Table S1 provides details of the included GWAS summary data, encompassing disease diagnostic criteria and demographic data. The exposure and outcome datasets had virtually no sample overlap.

To proxy HDP, PE, and GH, single-nucleotide polymorphisms (SNPs) were used as genetic instruments at a genome-wide significance level (P<5×10⁻⁸). We conducted linkage disequilibrium (LD) clumping between SNPs (r² = 0.001, window size=10,000 kb) according to European data from the 1000 Genomes Project. F-statistics were calculated by the formula F=((N−K−1)×R²)/(K×(1−R²)) (25). R² was the proportion of variation in exposure explained by the genetic instruments, N was the sample size of the exposure GWAS, and K was the number of genetic instruments used. An F-statistic greater than 10 indicated the absence of a weak instrumental variable. SNPs with incompatible alleles, palindromic SNPs with intermediate allele frequencies, and SNPs directly associated with outcome (P<5×10⁻⁸) were excluded. Additionally, the MR Steiger filtering test was used to determine directionality of the exposure effect on outcomes (26). The “TRUE” result predicted the expected effect direction of exposure on the outcome. “FALSE” results represented SNPs that explained a greater proportion of variance in the outcome than in the exposure, and such SNPs were excluded to reduce bias due to reverse-causality instrument variables.

Two-sample univariate MR analyses were performed separately to estimate the genetically predicted effects of exposure on outcomes. The random effects inverse variance-weighted multiplication (IVW) method, which assumes that all SNPs are valid instruments and allows for pleiotropy while providing the most precise estimates, was used as the primary MR analysis method (27). Additional sensitivity analyses included MR-Egger, weighted median, maximum likelihood (ML), constrained ML and model averaging (cML-MA), MR pleiotropy residual sum and outlier (MR-PRESSO), and MR Steiger test of directionality (28–31). Pleiotropy was detected via the MR-Egger intercept test, where P > 0.05 indicated the absence of horizontal pleiotropy. The MR-Egger approach yields estimates after accounting for horizontal pleiotropy; however the statistical power is weak (28). When valid instrument variants account for more than 50% of the weight, the weighted median technique can produce consistent estimate (29). The ML method assumes that there is no pleiotropy or heterogeneity; if this assumption holds, the estimated results are unbiased (30). The cML-MA approach can be used to control for related and unrelated pleiotropy (26). The MR-PRESSO method can identify outliers and potential pleiotropy, and provide the same results as an IVW after removing outliers (31). The MR Steiger test of directionality was used to test whether the hypothesis that exposure caused the outcome was true (26). Heterogeneity among the different instrumental variables was evaluated using Cochran’s Q test. Burgess’s online calculator was used to calculate the power of the MR estimates (32).

Previous studies have found that a genetic predisposition to systolic blood pressure (SBP), diastolic blood pressure (DBP) or type 2 diabetes mellitus (T2DM) increases the risks for HDP, PE, and GH (33, 34). Therefore, we performed a multivariate MR (MVMR) analysis to estimate the direct effects of HDP, PE, and GH on stroke, especially for any significant exposure-outcome correlation identified in the univariate MR analysis (P_IVW <0.05) (35). Our primary analysis used the IVW technique, and the sensitivity analyses for MVMR used MR-Egger, MR-LASSO, and MVMR-robust. MR-lasso and MVMR-robust methods can provide valid estimates with minimal bias and reduce the type I error risk in the presence of pleiotropic SNPs (36, 37). Cochran’s Q test and the MR-Egger test were used to detect heterogeneity and horizontal pleiotropy, as described previously. The strength of the instrumental variables was evaluated using the Sanderson-Windmeijer conditional F-statistic. Weak instrumental variables were identified in the MVMR analysis if the F-statistic was less than 10 (12).

The Bonferroni method was used to adjust the significance threshold. For the univariate MR analysis, the significance threshold was set to 2.38×10^-3 (calculated as 0.05/21, with 3 exposures and 7 outcomes, totaling 21 analyses). P-values between the adjusted threshold for significance and 0.05 were considered suggestive of an association between exposure and outcome. For the MR analysis, the estimates from different MR methods were in the same direction and considered significant, at least when the IVW method estimates were significant. For the MVMR analysis, the significance threshold was set at 0.05. The association results are presented as odds ratios (OR) with 95% confidence intervals (CI).

Genetic correlations between HDP and stroke phenotypes were estimated by LD score regression (LDSC) (12). This analysis leveraged genetic covariance and the LD score to determine the genetic correlation between 2 phenotypes based on GWAS summary-level data. Here, we used the ‘ldscr’ R package to assess genome-wide pairwise associations between 2 different traits (38). A strong correlation was considered when the genetic correlation coefficient (rg) was greater than 0.7 The significance threshold for the LDSC analysis was set to 2.38×10^-3 (0.05/21).

R software (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria) was used for all statistical analyses. The R package for MR analysis consists of the TwoSampleMR (version 0.5.6), MVMR (version 0.3), MR-PRESSO (version 1.0), and ldscr (version 0.1.0) (31, 38, 39).

Thirteen, 9, and 9 SNPs were selected as instrumental variables for HDP, GH, and PE, respectively ( Supplementary Table S2 ). The F-statistic for each instrument was greater than 10, indicating sufficient instrumental strength. The details of the SNPs associated with exposure to the outcomes are provided in Supplementary Table S3. Univariate MR results for the effects of HDP, PE, and GH on stroke are shown in Figure 2 ; Supplementary Table S4 .

Genetically predicted susceptibility to HDP was associated with an increased risk of LAA (OR=1.50, 95% CI: 1.17–1.91, P=1.13×10^-3) and SVS (OR=1.29, 95% CI: 1.20–1.50, P=1.52×10^-3); and possibly IS (OR=1.13, 95% CI: 1.04–1.23, P=4.99×10^-3) and stroke (OR=1.11, 95% CI: 1.03–1.20, P=8.85×10^-3) based on the IVW method. The remaining 4 sensitivity analyses yielded consistent estimates in the same direction. No evidence of heterogeneity, horizontal pleiotropy, or outliers was found in the remaining analyses, except for the LAA analysis had heterogeneity. No associations were found between HDP and CES, lacunar infarction and ICH.

Genetically predicted susceptibility to GH may be associated with an increased risk of LAA (OR=1.22, 95% CI: 1.01–1.47, P=4.07×10^-2) and SVS (OR=1.20, 95% CI: 1.01–1.43, P=3.74×10^-2). No evidence of heterogeneity or horizontal pleiotropy was observed. Although the MRPRESSO test suggested potential horizontal pleiotropy (MRPRESSO Global Test: P=1.00×10^-2), no outliers were detected in the LAA analysis. In addition, the results of the ML-based and cML-MA methods were consistent with those of the IVW method. No association was found between the genetically predicted GH susceptibility and stroke or other stroke subtypes.

Genetically predicted susceptibility to PE may be associated with an increased risk of stroke (OR=1.10, 95% CI: 1.02–1.19, P=1.16×10^-2) and IS (OR=1.10, 95% CI: 1.02–1.20, P= 1.84×10^-2). The remaining 4 sensitivity analyses yielded consistent estimates in the same direction. No evidence of heterogeneity, horizontal pleiotropy, or outliers were observed. In the association analysis between PE and the LAA, SVS, or CES, there were suggestive associations based on the IVW method. Although the results of the ML and cML-MA methods were consistent with those of the IVW method, the variations in the conclusions obtained using the MR-Egger method and the heterogeneity and potential horizontal pleiotropy indicated the estimates were not robust.

In all univariate MR analyses, the Steiger test results suggested that the direction of causality was consistent with the hypothesized direction ( Supplementary Table S5 ). The statistical power was ≥80% in most of the MR analyses regarding associations between HDP and stroke.

The MVMR results of the effects of HDP, PE, and GH on stroke are presented in Figure 3 ; Supplementary Tables S6 - S8 . After adjustment for SBP, the results of MVMR suggested that genetically predicted susceptibility to HDP (OR=1.17, 95% CI: 1.01–1.36, P=3.56×10^-2) and PE (OR=1.19, 95% CI: 1.02–1.37, P=2.26×10^-2) was associated with an increased LAA risk.

After adjustment for DBP, the MVMR results suggested that genetically predicted susceptibility to HDP (OR=1.07, 95% CI: 1.00–1.14, P=3.08×10^-2) and PE (OR=1.08, 95% CI: 1.01–1.15, P=1.51×10^-2) were associated with an increased risk of IS; and HDP (OR=1.34, 95% CI: 1.17–1.55, P=3.40×10^-5), PE (OR=1.33, 95% CI: 1.16–1.52, P=3.47×10^-5) and GH (OR=1.19, 95% CI: 1.05–1.36, P=8.30×10^-3) were associated with an increased LAA risk.

After adjusting for T2DM, the MVMR results suggested that genetic susceptibility to HDP or PE was associated with an increased risk of stroke, IS, LAA, or SVS, and that GH was associated with an increased SVS risk.

The instrumental variable strength was significantly reduced in the MVMR compared to that in the univariate MR analysis. There was evidence of heterogeneity in almost all relationships. After adjusting for SBP and DBP, the evidence of pleiotropy was tested in the association analysis between PE and LAA. Therefore, we used the MR-lasso and robust MVMR methods to correct for the bias caused by weak instrumental variables or pleiotropic SNPS. The sensitivity analyses had similar estimation results for most MVMR analyses.

We found genetic associations between HDP, PE, and GH and stroke, IS, LAA, and SVS ( Figure 4 ; Supplementary Table S9 ). No genetic associations were observed between HDP, PE, or GH and CES, lacunar infarction, or ICH.