Safety of beta-blocker and calcium channel blocker antihypertensive drugs in pregnancy: a Mendelian randomization study | BMC Medicine


Source data

A flowchart depicting the statistical analysis plan is displayed in Fig. 3. Publicly available genetic association summary data were used for all analyses. Appropriate ethical approval and participant consent were obtained in the original studies that generated the data. A summary of the data sources used in this study is presented in Table 3. Genetic association estimates were extracted from summary data of GWAS studies on 757,601 participants for systolic blood pressure (SBP; corrected for blood pressure-lowering medication use and body mass index) [14]4743 pre-eclampsia or eclampsia cases (identified using registry data on International Classification of Diseases (ICD) 8,9, and 10 codes) and 136,325 controls (women without a hypertensive disorder in pregnancy) [15]7676 gestational diabetes cases (identified using registry data on ICD codes 9 and 10) and 130,424 controls (women without any pregnancy-related morbidity) [15]and 155,202 women who have given birth at least once for birthweight of the first child [16]. The birthweight of the first child variable was split into three bins, which are coded in 3 categories as follows: 0: birthweight of the first child below 7 pounds, 1: birthweight of the first child is 7 pounds, 2: birthweight of the first child is above 7 pounds. All studies were performed in European ancestry populations. All GWASs were corrected for age, sex (where applicable) and principal components.

Fig. 3
picture 3
Table 3 Information on the studies and consortia from which genetic association data were obtained

instrumental variables

For each drug target, single-nucleotide polymorphisms (SNPs) in the drug target gene region (± 10kB) that were associated with SBP (p < 5 × 10−8) were selected as proxies for drug target perturbation after clumping to a pairwise linkage disequilibrium threshold of r2< 0.1 using the 1000 Genomes European reference panel. For SNPs reflecting SBP reduction by any mechanism, we took all genome-wide significant hits from the GWAS. These variants were also clumped a to a pairwise linkage disequilibrium threshold of r2< 0.1 using the 1000 Genomes European reference panel, to mimic the approach taken for the drug targets as closely as possible. A flowchart showing the number of SNPs at each stage is shown in Additional file 2: Fig. S1. The full list of SNPs for the different targets are presented in Additional file 1: Tables S1, and S12-S17. The gene locations for BB and CCB drug targets are reported in Table 1.

Mendelian randomization

MR estimates for SBP reduction through the two classes of antihypertensive medications on pregnancy outcomes were generated using the two-sample inverse-variance weighted (IVW) model [17] with standard error correction for under dispersion. This was implemented in the TwoSampleMR package version 0.5.6 [18]in R version 4.1.0 [19]. If there was only a single variant available, we used the Wald ratio to estimate the effect. The results are reported as odds ratios (OR) or beta coefficients scaled per 10 mmHg SBP reduction with 95% confidence intervals (95%CI). MR is a method that, under three core assumptions, intends to estimate causal effects. These three core assumptions are: each variant must be associated with the exposure (relevance assumption), each variant is not associated with any potential confounding variable (independence assumption), and each variant must be only associated with the outcome via the risk factor (exclusion restriction) [20]. An F-statistic value above 10 is used as a rule of thumb in the instrumental variables literature to support that the relevance assumption seems to be met. For the two other assumptions (independence assumption and exclusion restriction), it is somewhat more difficult to demonstrate their validity. Since we study known drug targets for lowering SBP, this improves our confidence in using variants associated with SBP in the genes coding for these drug targets. Additionally, we implement methods more robust to violations of the conventional Mendelian randomization assumptions: MR-Egger analysis, the weighted median method and MR-PRESSO. The methods are described in the Additional file 2: Supplementary Note.

Since 3 different outcomes (pre-eclampsia or eclampsia, gestational diabetes, and birthweight category of the first child) were tested, a Bonferroni correction was applied when interpreting the p-value for statistical significance 0.05/3 = 0.017.


For the BB gene region (ADRB1), a Bayesian test for genetic colocalization was also performed to investigate the likelihood that the exposure-outcome pair had a shared versus distinct causal variants [21]. For these analyses, the R package roommate (version 5.1.0) was used [21]. Approximate Bayes Factor Colocalization analyzes (using the coloc.abf function) were carried out, which assumes that there is at most 1 causal SNP in the gene region for each trait. The method allows for the following hypotheses:

  • H0: no causal variants in the gene region.

  • H1: only trait 1 has a causal variant in the gene region.

  • H2: only trait 2 has a causal variant in the gene region.

  • H3: both traits have a different causal variant in the gene region.

  • H4: both traits have a shared causal variant in the gene region.

Default priors were used as prior probabilities of the different hypotheses: ({p}_{1}={p}_{2}={10}^{-4}),({p}_{12}={10}^{-5}). Using the data and priors, the method calculates the posterior probabilities for the different hypotheses [21]. In Additional file 2: Supplementary Note, we discuss the use of priors in more depth and rerun our analyzes using different values ​​for the joint prior ({p}_{12}).

Evidence for a shared causal variant was concluded when the posterior probability for H4 surpasses a value of 0.5 (this means that a posteriori a shared causal variant is more likely than distinct causal variants). If a shared causal variant was found for the pair, this contributes further evidence that the antihypertensive drug target and outcome share a causal mechanism, and that any identified MR association is unlikely to be attributable to genetic confounding through a variant in linkage disequilibrium. This analysis was not performed for the CCB target as a whole, because this is made up of several proteins that are coded for by different genes. However, we do inspect each CCB region separately for completeness.

Given that colocalization is being performed to explore the robustness of an MR association to possible genetic confounding, a H4 > 0.5 was considered to represent evidence of colocalization, as this would suggest that there is more evidence of a shared causal variant than distinct causal variants underlying the MR association between the exposure and the outcome.

Sensitivity analyzes

To further explore the robustness of our findings to consideration of blood pressure genetic association estimates specific to females, or related to diastolic blood pressure (DBP) rather than SBP, we further repeated the analyzes using instruments selected from GWAS data obtained in females, and on DBP rather than SBP. Full details are provided in the Additional file 2: Supplementary Note.

Leave a Comment