Association of genetic variants related to combined lipid-lowering and antihypertensive therapies with risk of cardiovascular disease: 2 × 2 factorial Mendelian randomization analyses

Exposures and genetic instruments

The genetic instruments for proxies of statin and PCSK9 inhibitors were obtained from the publicly available summary data of a genome-wide association study (GWAS) meta-analysis of LDL-C levels in the Global Lipids Genetics Consortium [22]. Genetic variants were identified within 100 kb on either side of the target gene (HMGCR for statins and PCSK9 for PCSK9 inhibitors) and associated with LDL-C at genome-wide significance level (P < 5.0 × 10^–8) and were permitted to be in weak linkage disequilibrium (r² < 0.20) with each other to increase the proportion of variance in each respective drug target explained by the instruments, maximizing the instrument strength [23].

For the genetic instruments of antihypertensive drugs, we used genetic proxies for CCBs and BBs provided by previous studies [24]. The genes encoding pharmacological targets for the antihypertensive drugs in DrugBank (CACNA1D, CACNA1F, CACNA2D1, CACNA2D2, CACNA1S, CACNB1, CACNB2, CACNB3, CACNB4, CACNG1, and CACNA1C for CCBs, ADRB1 for BBs) were identified [24]. Genetic variants in these genes or regulatory gene regions and associated with SBP in GWAS meta-analysis of UK Biobank and International Consortium of Blood Pressure were selected as genetic instruments of the antihypertensive drugs. The GWAS meta-analysis was performed in up to 757,601 participants, with adjustment for age, age², sex, body mass index (BMI), and study-specific covariates [25]. The genetic proxies and the effect sizes of which were shown in Additional file 1: Table S1. Regarding other clinical drugs like angiotensin-converting enzyme inhibitors (ACEI), there were not enough genetic variants (only 1 single-nucleotide polymorphism [SNP]) identified in ACE gene encoding pharmacological targets [24, 26], leading to low power and the failure of naturally random allocation.

We constructed the genetic scores for statin, PCSK9 inhibitors, BBs, and CCBs by combining all variants selected for each drug to create an instrument that could overcome weak effect of single variant and that would allow us to randomly allocate the participants into approximately equal-sized groups to perform the 2 × 2 factorial analyses. The genetic scores for these four drugs were constructed using the selected instrumental variables, with the effect size of each genetic variant as the weight. And the genetic scores below the median represented genetically proxied medications of lipid-lowering or antihypertensive drugs, which were related to lower LDL-C or SBP.

In our study, we used genetic scores as genetic proxies for exposures to lipid-lowering and antihypertensive drugs. These scores have direct effects on low-density lipoprotein-cholesterol (LDL-C) and systolic blood pressure (SBP). SBP (UK Biobank fields 4080) was measured twice at baseline, following the standard protocol [27]. LDL-C (UK Biobank fields 30780), as part of the UKB Biomarker project, was determined by biochemical arrays and analyzed by enzymatic selective protection [28]. As for the individuals who were taking antihypertensive drugs (UK Biobank fields 6153 and 6177), blood pressure was adjusted by adding 15 mmHg for SBP [25]. For those on lipid-lowering medication (UK Biobank fields 6153 and 6177), we replaced their LDL-C values by LDL-C/0.7 [29].

Outcomes

The primary outcome of our study was CVD, defined as a composite of CAD, PAD and stroke. In addition, we analyzed other clinical and safety outcomes, including IS, HS, HF, DM, CKD, and dementia. These outcomes were identified using International Classification of Disease (ICD) 9 and 10 codes, non-cancer illness code (self-reported), as well as Classification of Interventions and Procedures. Detailed definitions of these outcomes were elaborated in Additional file 1: Table S2.

Statistical analyses

The application of MR must satisfy three core assumptions, including (1) the genetic instrumental variables are strongly associated with the exposure of interest (relevance assumption); (2) they are not associated with the outcome via a confounding pathway (independence assumption); and (3) the instrument variables do not affect the outcome directly, only possibly indirectly via the exposure (exclusion restriction assumption). For the relevance assumption, we evaluated the association of each drug score with LDL-C or SBP and all outcomes, using logistic or linear regression adjusted for age, sex, assessment center, SNP array, and top ten genetic principal components. And we estimated the proportion of variance for each drug explained by the instrument (R²) and F statistics. As a convention, an F statistic of at least ten indicates no weak instrument bias. To assess independence assumption, we investigated the associations of drugs scores with potential confounders (BMI, smoking, and drinking) and further adjusted the factors with significant associations (P < 0.05) in the sensitivity analyses. Regarding the exclusion restriction assumption, it is somewhat more difficult to demonstrate the validity. We assessed pleiotropy using the method of MR-Egger regression by conducting two sample MR with GWAS summary data.

We performed 2 × 2 factorial MR analyses, splitting our participants into four groups based on the scores of antihypertensive drugs and lipid-lowering drugs, as depicted in Fig. 1. Thus, individuals were categorized into: both scores ≥ median (reference), lipid-lowering drug score < median (representing statin or PCSK9 inhibitors use), antihypertensive drug score < median (representing BBs or CCBs medication), and both scores < median (proxy for combined medications). Then we investigated the associations with the outcomes between these groups in logistic regression models with adjustment for age, sex, assessment center, SNP array, and top ten genetic principal components. The mean differences of LDL-C and SBP between each two groups were directly calculated and compared to estimate the effect of antihypertensive drugs and lipid-lowering drugs separately or jointly for therapy. To avoid biased estimates attributed to arbitrary dichotomization and to maximize power, we conducted the tests of the additive interaction and multiplicative interaction using the two scores as quantitative traits. The additive interactions were evaluated via calculating relative excess risk due to interaction (RERI) and 95% CI for each combination using bootstrap. RERI or 95%CI covering 0 indicates independent and additive effect (i.e., no interaction); RERI > 0 implies positive interaction or synergism, and RERI < 0 represents negative interaction or antagonism. Their multiplicative interactions were evaluated using interaction P value for the product term.

Open in a separate window

Fig. 1

Study design of the 2 × 2 factorial Mendelian randomization analysis in the UK Biobank. BBs beta-blockers, CCBs calcium channel blockers, SBP systolic blood pressure, LDL-C low-density lipoprotein-cholesterol

All analyses were performed using R (version 3.5.1; R Project for Statistical Computing). A two-tailed P value less than 0.05 was considered statistically significant.

Instrument variables validation

For lipid-lowering drugs, we identified 5 and 11 genetic variants in the HMGCR and PCSK9 genes as proxies for statin and PCSK9 inhibitors, respectively. For antihypertensive drugs, we selected 6 and 24 genetic variants as proxies for BBs and CCBs, respectively. The detailed information of genetic variants was elaborated in Additional file 1: Table S1. As expected, the genetic scores of lipid-lowering (statin and PCSK9 inhibitors) and antihypertensive drugs (BBs and CCBs) were significantly associated with LDL-C and SBP levels (Additional file 1: Table S4), explaining 0.30%, 0.65%, 0.05%, and 0.11% of the variance, corresponding to the F statistics of 258, 253, 37, and 20, which indicated weak instrument variables were not likely to influence our results (Additional file 1: Table S5). For the independence assumption, although there was significant association of statin score with BMI, we confirmed that the allocation to the 2 × 2 groups was indeed random by comparing the characteristics between groups, indicating that potential confounders were unlikely to have effects on the results (Table 1). Moreover, the results with additional adjustment for BMI were consistent and robust (Additional file 2: Fig. S9-S11). Regarding the exclusion restriction assumption, the intercept of the MR-Egger regression was not significant, suggested no evidence of pleiotropy. Additionally, we selected reliable and widely used genetic variants for the drugs, which improved our confidence in using these variants as genetic instruments since it is also an effective strategy for avoiding pleiotropy to use the genetic variants with well-defined biological function.

Factorial analyses

Table ​Table11 shows the baseline characteristic of the four groups dichotomized by genetic risk scores. There were no significant differences in any non-lipid or non-blood pressure related baseline characteristics between the groups, which was consistent with random partitioning of participants into each group by the lipid-lowering and antihypertensive drug genetic scores, showing that the allocation was indeed random.

For CVD, we observed independent and additive effects in all combinations. In the 2 × 2 factorial MR analyses, either a lower genetic score of lipid-lowering drugs or antihypertensive drugs was associated with a lower risk of CVD, and scoring less than the median in both scores showed an approximately additive lower risk. When we performed analyses using the continuous genetic scores, there showed similar results as the dichotomized scores, and no interactions were observed. For instance, compared with the reference, 3.00 mg/dl lower LDL-C and a 4% lower risk of CVD were observed in individuals with a lower score of PCSK9 inhibitors (OR: 0.96; 95% CI: 0.94 to 0.99), and the individuals with a lower BBs score had 0.78 mmHg lower SBP and a 3% lower risk of CVD (OR: 0.97; 95% CI: 0.94 to 0.99); combined both had 2.99 mg/dl lower LDL-C, 0.85 mmHg lower SBP and was associated with a 6% lower risk of CVD (OR: 0.94; 95% CI: 0.92 to 0.97). This corresponded to a RERI index of 0 indicating an absolute lack of additive interaction (Fig. 2). In the continuous analyses, lower scores were also independently associated with a lower risk of CVD, with no evidence of multiplicative interaction between each 2 scores (P = 0.87 for interaction) (Fig. 2). Directionally consistent results for other drug combinations were similarly presented in Fig. 2. We found no significant interactions in the association of each two drugs with risk of CVD.

Open in a separate window

Fig. 2

Associations of exposure to genetically proxied lipid-lowering drugs, genetically proxied antihypertensive drugs, or both with risk of cardiovascular disease. BBs beta-blockers, CCBs calcium channel blockers, RERI relative excess risk due to interaction. Notes: The 2 × 2 analysis divided participants into four groups according to the median genetic scores of lipid-lowering drugs and antihypertensive drugs: “Both scores ≥ median” represented the group of placebos, “statin/PCSK9 score < median” was the proxy for statin or PCSK9 inhibitors use; “BBs/CCBs score < median” was the proxy for BBs or CCBs medication; and “both scores < median” was the proxy for combined medications. Continuous analysis refers to the associations where the two genetic scores were included on a continuous scale (per SD decrease) as well as the interaction between the two scores

As for other clinical and safety outcomes (including CAD, stroke, IS, HS, HF, DM, CKD, dementia, and PAD), similar additive effects were observed (Figs. ​(Figs.3,3, ​,44 and Additional file 2: Fig. S2-S8). Take CAD for example, genetically proxied medication of PCSK9 inhibitors was associated with decreased risk, and this association was not modified by BBs compared with the reference group (PCSK9 inhibitors: OR, 0.95; 95%CI, 0.93–0.98; BBs: OR, 0.97; 95%CI, 0.94–0.99; combined: OR, 0.94; 95%CI, 0.91–0.97; P = 0.51 for interaction). However, it is worth mentioning that when lipid-lowering and antihypertensive drugs were combined, there showed potential synergistic effect on DM, although it is not significant. A lower statin genetic score was associated with 4% higher risk for DM (OR: 1.04; 95% CI: 1.00 to 1.08), and a lower BBs score was associated with 5% higher risk for DM (OR: 1.05; 95% CI: 1.01 to 1.10); whereas combined both was associated with 12% higher risk for DM (OR: 1.12; 95% CI: 1.08 to 1.17) (Additional file 2: Fig. S5).

Open in a separate window

Fig. 3

Associations of exposure to genetically proxied lipid-lowering drugs, genetically proxied antihypertensive drugs, or both with risk of coronary artery disease. BBs beta-blockers, CCBs calcium channel blockers, RERI relative excess risk due to interaction. Notes: The 2 × 2 analysis divided participants into four groups according to the median genetic scores of lipid-lowering drugs and antihypertensive drugs: “Both scores ≥ median” represented the group of placebos, “statin/PCSK9 score < median” was the proxy for statin or PCSK9 inhibitors use, “BBs/CCBs score < median” was the proxy for BBs or CCBs medication, and “both scores < median” was the proxy for combined medications. Continuous analysis refers to the associations where the two genetic scores were included on a continuous scale (per SD decrease) as well as the interaction between the two scores

Open in a separate window

Fig. 4

Associations of exposure to genetically proxied lipid-lowering drugs, genetically proxied antihypertensive drugs, or both with risk of stroke. BBs beta-blockers, CCBs calcium channel blockers, RERI relative excess risk due to interaction. Notes: The 2 × 2 analysis divided participants into four groups according to the median genetic scores of lipid-lowering drugs and antihypertensive drugs: “Both scores ≥ median” represented the group of placebos, “statin/PCSK9 score < median” was the proxy for statin or PCSK9 inhibitors use, “BBs/CCBs score < median” was the proxy for BBs or CCBs medication, and “both scores < median” was the proxy for combined medications. Continuous analysis refers to the associations where the two genetic scores were included on a continuous scale (per SD decrease) as well as the interaction between the two scores

The authors would like to thank all participants and investigators who contribute to the UK Biobank. This research has been conducted using the UK Biobank Resource under Application Number 69815.