ABSTRACT:
Aims We aimed to estimate the effect of blood pressure and blood pressure lowering medications (via genetic proxies) on peripheral artery disease.
Methods and Results GWAS summary statistics were obtained for BP (International Consortium for Blood Pressure + UK Biobank GWAS; N = up to 757,601 individuals), peripheral artery disease (PAD; VA Million Veteran Program; N = 24,009 cases, 150,983 controls), and coronary artery disease (CAD; CARDIoGRAMplusC4D 1000 Genomes; N = 60,801 cases, 123,504 controls). Genetic correlations between systolic BP (SBP), diastolic BP (DBP), pulse pressure (PP) and CAD and PAD were estimated using LD score regression. The strongest correlation was between SBP and CAD (rg = 0.36; p = 3.9 × 10−18). Causal effects were estimated by two-sample MR using a range of pleiotropy-robust methods. Increased SBP, DBP, and PP increased risk of both PAD (SBP OR 1.25 [1.19–1.31] per 10mmHg increase, p = 3 × 10−18; DBP OR 1.27 [1.17–1.39], p = 4 × 10−8; PP OR 1.51 [1.38–1.64], p = 1 × 10−20) and CAD (SBP OR 1.37 [1.29–1.45], p = 2 × 10−24; DBP OR 1.6 [1.45–1.76], p = 7 × 10−22; PP OR 1.56 [1.4–1.75], p = 1 × 10−15). The effects of SBP and DBP were greater for CAD than PAD (pdiff = 0.024 for SBP, pdiff = 4.9 × 10−4 for DBP). Increased liability to PAD increased PP (beta = 1.04 [0.62–1.45] mmHg per 1 unit increase in log-odds in liability to PAD, p = 1 × 10−6). MR was also used to estimate the effect of BP lowering through different classes of antihypertensive medications using genetic instruments containing BP-trait associated variants located within genes encoding protein targets of each medication. SBP lowering via calcium channel blocker-associated variants was protective of CAD (OR 0.38 per 10mmHg decrease in SBP; 95% CI 0.19–0.77; p = 0.007).
Conclusions Higher BP is likely to cause both PAD and CAD but may have a larger effect on CAD risk. BP-lowering through calcium-channel blockers (as proxied by genetic variants) decreased risk of CAD.
INTRODUCTION
Peripheral artery disease (PAD) is a common manifestation of atherosclerotic cardiovascular disease (ASCVD), estimated to affect more than 12 million individuals in the United States, and more than 120 million individuals worldwide (1,2). PAD shares a number of risk factors with other forms ASCVD like coronary artery disease (CAD) and ischemic stroke (3). These risk factors include smoking, diabetes, hypertension, hyperlipidemia, and obesity (2–4). Observational studies have identified hypertension as one of the strongest risk factors for incident and prevalent PAD (5–11), although these studies may be limited by residual environmental confounding or reverse-causality. While randomized controlled trials of antihypertensive medications have demonstrated broad protection from coronary artery disease and death from cardiovascular causes, whether lower blood pressure reduces risk of PAD specifically has not been reliably established. Similarly, the relative effect of blood pressure on PAD and CAD has not been fully investigated.
Recent genome-wide association studies (GWAS) of PAD, CAD, and blood pressure including more than 700,000 individuals have identified hundreds of genetic variants associated with these traits (12,13). The Mendelian randomization (MR) framework (under certain assumptions) can leverage this genetic variation (which is randomly assorted during meiosis, mimicking a randomized trial), to provide unconfounded causal estimates of the relationship between traits (14). MR assumes that genetic variants are likely to be independent of many confounders of the exposure-outcome relationship. This assumption is plausible because genetic variants are randomly inherited by offspring from parents during meiosis and conception, analogous to treatment allocation in a randomized trial. Because large randomized trials evaluating the relationship between treatment of hypertension and PAD outcomes may be unfeasible, other study designs are needed to fill this evidence gap. Here, we leverage population-scale genetic variation within the Mendelian randomization framework to 1) establish the relationship between blood pressure and risk of PAD, 2) quantify differences in the effect of blood pressure on CAD and PAD, and 3) estimate the effect of blood pressure lowering (using genetic proxies of antihypertensive medications) on PAD risk.
METHODS
Study Exposures
The 2018 Evangelou et al. International Consortium for Blood Pressure + UK Biobank GWAS, which included measurements of systolic blood pressure, diastolic blood pressure, and pulse pressure in up to 757,601 individuals was used to identify genetic variants associated with the primary blood pressure phenotypes (15). This study included up to 299,024 European participants from 77 independent studies genotyped with various arrays and imputed to either the 1000 Genomes Reference Panel or the HRC panels, and 458,577 participants from the UK Biobank imputed to UK10K + 1000 Genomes reference panel. Measurement of blood pressure varied among cohorts, and study-specific details are presented in the supplemental material of Evangelou et al., 2018. Summary statistics for the blood pressure genome wide association study are publicly available, and may be downloaded from the NHLBI GRASP catalog (https://grasp.nhlbi.nih.gov/FullResults.aspx).
Study Outcomes
The 2019 Million Veteran Program genome wide association study of Peripheral Artery Disease by Klarin et al. identified 24,009 cases and 150,983 controls of European Ancestry (12). This study defined cases and controls based on electronic health record phenotyping within the Veterans Affairs (VA) Healthcare System and was validated against ankle brachial index measurement and manual chart review. The current analysis focused on participants of European ancestry. MVP PAD genome wide association study summary statistics are available on dbGAP (Accession phs001672.v2.p1).
Genome-wide association study summary statistics for coronary artery disease were obtained from the Nikpay et al. 2016 CARDIoGRAMplusC4D 1000 Genomes-based GWAS. This study was a meta-analysis including 60,801 CAD cases and 123,504 controls, with genotypes imputed using the 1000 Genomes phase 1 version 3 reference. Summary statistics were downloaded from http://www.cardiogramplusc4d.org/data-downloads/.
Genetic Correlation
LD-score regression was used to estimate the genetic correlation between blood pressure traits, and between blood pressure traits and ASCVD traits (https://github.com/bulik/ldsc) (16,17). GWAS summary statistics were obtained for systolic blood pressure, diastolic blood pressure, pulse pressure, peripheral artery disease, and coronary artery disease (12,13,15). Summary statistics were filtered to HapMap3 SNPs, and SNP heritability and genetic correlations were estimated using pre-computed 1000 Genomes European LD-scores.
Mendelian Randomization
Two-sample Mendelian randomization analyses were performed in R using the TwoSampleMR package (https://github.com/MRCIEU/TwoSampleMR) (18). Genetic instruments for SBP, DBP, and PP were constructed using linkage-disequilibrium independent (r2 < 0.001, distance = 10,000kb; 1000 Genomes European reference panel), genome-wide significant (p < 5×10−8) variants identified using GWAS summary statistics for each trait (Supplemental Table 3). For bi-directional MR analysis, additional instruments were constructed for CAD and PAD using the same procedure (Supplemental Table 6). For each variant included in the genetic instruments, proportion of variance explained was calculated using the formula (where MAF represents the effect allele-frequency, beta represents the effect estimate of the genetic variant in the exposure GWAS, se represents the standard error of effect size for the genetic variant, and N represents the sample size)(19). F-statistics were calculated for each variant using the formula
(where R2 represents the variance in exposure explained by the genetic variant, and N represents the number of individuals in the exposure GWAS) to assess the strength of the selected instruments (Supplemental Table 4)(20). For the primary analysis, power calculation was performed to identify the minimum-detectable effect size maintaining 80% power with two-sided alpha of 0.05. The primary MR analyses used inverse-variance weighting with random effects. The MR-Egger intercept test was used to evaluate for evidence of horizontal pleiotropy. Leave-one-out, single-SNP, and funnel-plot diagnostic MR analyses were performed. Sensitivity analyses were performed using MR methods that make different assumptions about the presence of pleiotropy (weighted median, penalized weighted median, and weighted mode) (21).
Multivariable MR (MVMR) was used in additional sensitivity analyses to jointly estimate the direct effects of multiple blood pressure traits, using genetic instruments including genomically independent (r2 < 0.001, distance = 10,000kb) variants that were associated at genome-wide significance (p < 5×10−8) with any exposure, weighted by the effect of each SNP on each exposure (Supplemental Table 8)(22). Effect estimates were scaled to correspond to a 10mmHg change in blood pressure.
Antihypertensive Drug MR
Two MR analyses were performed to estimate the effect of 10mmHg lower of blood pressure by antihypertensive drugs. In the first analysis, genetic instruments consisting of variants within genes encoding protein targets of ACE-inhibitors, beta-blockers, and calcium-channel blockers were obtained from Gill et. al., with each variant weighted by its effect on systolic blood pressure (Supplemental Table 5)(23). In a sensitivity analysis, genetic instruments were constructed that mimic the action of 12 antihypertensive medication classes. In contrast to the first method which selected genetic variants based on their proximity to genes encoding protein targets, this method prioritized variants representing expression quantitative trait loci (eQTL) for genes encoding protein targets of antihypertensive medications that were demonstrated to effect systolic blood pressure (24). For both methods, inverse-variance weighted and weighted-median MR was performed, with MR-Egger intercept test used to assess for horizontal pleiotropy. For instruments with only 1 variant, Wald-ratio MR was performed.
Statistical Analysis
All statistical analyses were performed using R version 3.6.2 (R Foundation for Statistical Computing).
RESULTS
Genetic Correlation
Cross-trait LD-score regression was used to estimate the genetic correlation between blood pressure traits. All blood pressure traits (SBP, DBP, PP) were strongly, positively correlated (Supplemental Figure 1; Supplemental Table 1). The strongest correlation among blood pressure traits was between systolic blood pressure and pulse pressure (rg = 0.85; p < 1.0 × 10−300).
Genetic correlation between blood pressure traits and ASCVD traits (CAD and PAD) was then assessed. All BP traits were strongly, positively correlated with ASCVD traits (Supplemental Figure 2; Supplemental Table 2). The strongest correlation among BP and ASCVD traits was between systolic blood pressure and CAD (rg = 0.36; p = 3.9 × 10−18).
Effects of Blood Pressure on ASCVD: Mendelian Randomization
To determine whether the genetic correlations between BP and ASCVD traits would be consistent with causal effects, we performed two-sample mendelian randomization using genome-wide association study summary statistics. Genetic instruments for blood pressure contained between 342 and 410 independent genetic variants, explaining between 3.5% and 4.3% of the variability in blood pressure, with F-statistics ranging from 29.6 to 670.6 (consistent with low risk of weak-instrument bias) (Supplemental Tables 3–4). The primary analysis maintained power to detect a 7% – 11% increase in risk of ASCVD per 10mmHg increase in blood pressure (Supplemental Table 4). In inverse-variance weighted analyses, each 10mmHg increase in SBP increased risk of both PAD (OR 1.25; 95% CI 1.19–1.31; p = 3 × 10−18) and CAD (OR 1.37; 95% CI 1.29–1.45; p = 2 × 10−24), though the effect was stronger for CAD than PAD (pdifference = 0.024) (Figure 1, Supplemental Table 5). Each 10mmHg increase in DBP increased risk of both PAD (OR 1.27; 95% CI 1.17–1.39; p = 4 × 10−8) and CAD (OR 1.6; 95% CI 1.45–1.76; p = 7 × 10−22), with a stronger effect for CAD than PAD (pdifference = 4.9 × 10−4) (Figure 1, Supplemental Table 5). Each 10mmHg increase in PP increased risk of both PAD (OR 1.51; 95% CI 1.38–1.64; p = 1 × 10−20) and CAD (OR 1.56; 95% CI 1.4–1.75; p = 1 × 10−15), with similar effects on CAD and PAD (pdifference = 0.60) (Figure 1, Supplemental Table 5). MR-Egger bias intercept term was p > 0.05 for all trait-outcome pairs except PP-PAD (p = 0.012) (Supplemental Table 5). Results remained robust in sensitivity analyses using MR methods that make different assumptions about the presence of pleiotropy (Supplemental Table 5).
In inverse variance weighted Mendelian randomization analyses, elevations in each blood pressure trait increased risk of both coronary and peripheral artery disease. Results scaled to reflect odds of outcome per 10mmHg increase in blood pressure. n SNP = number of single nucleotide polymorphisms in the exposure instrument; OR = odds ratio; 95% CI = 95% confidence interval; P = pvalue SBP = Systolic blood pressure; DBP = Diastolic blood pressure; PP = Pulse pressure; PAD = Peripheral artery disease; CAD = coronary artery disease.
Effects of Liability to ASCVD on Blood Pressure: Mendelian Randomization
Because stiffening of peripheral vessels in the setting of peripheral artery disease may affect blood pressure, raising the possibility of reverse-causation in assessment of the relationship between blood pressure and PAD, bi-directional MR analysis was performed. Genetic instruments for PAD and CAD were selected and used to estimate the effect of liability to ASCVD on blood pressure traits (Figure 2; Supplemental Tables 6–7). In inverse-variance weighted analysis, liability to PAD increased PP (beta = 0.97 mmHg per 1 log-odds increase in risk of PAD; 95% CI 0.4–1.5; p = 8 × 10−4), and CAD increased PP (beta = 0.542 per 1 log-odds increase in risk of CAD; 95% CI 0.006–1.1; p = 0.05. Neither PAD nor CAD affected SBP or DBP. The MR-Egger bias intercept term had p > 0.05 for all analyses, indicating no positive evidence for bias. In sensitivity analysis applying MR methods making different assumptions about the presence of pleiotropy, there was weak evidence that both PAD and CAD increase SBP and PP while decreasing DBP (Supplemental Table 7).
In inverse variance weighted Mendelian randomization analyses, both peripheral artery disease and coronary artery disease increased pulse pressure. Results reflect increase in blood pressure (mmHg per 1 log-odds unit increase in risk of the exposure). n SNP = number of single nucleotide polymorphisms in the exposure instrument; OR = odds ratio; 95% CI = 95% confidence interval; P = p-value SBP = Systolic blood pressure; DBP = Diastolic blood pressure; PP = Pulse pressure; PAD = Peripheral artery disease; CAD = coronary artery disease.
Multivariable Mendelian Randomization
Because blood pressure traits are highly correlated and unlikely to affect cardiovascular outcomes in isolation, we performed multivariable MR to jointly estimate the direct effects of each blood pressure trait on ASCVD outcomes (Supplemental Table 8). Each 10mmHg increase in SBP increased risk of both PAD (direct OR 1.67; 95% CI 1.19–2.34; p = 0.003) and CAD (direct OR 1.44; 95% CI 1.02–2.03; p = 0.04). After accounting for the effect of SBP, we did not observe residual evidence of a significant direct effect of DBP on PAD or CAD.
Antihypertensive Drug Mendelian Randomization
Two Mendelian randomization analyses were performed to estimate the effect of blood pressure lowering on ASCVD outcomes. The first analysis focused on the effect of SBP lowering using genetic variants located within genes encoding protein targets of common classes of blood pressure lowering medications (ACE-inhibitors, beta-blockers, and calcium channel blockers) (Figure 3; Supplemental Tables 10–11). This analysis identified protective effects of beta-blocker-associated variants on CAD (OR 0.59 per 10mmHg decrease in SBP; 95% CI 0.44- 0.78; p = 3 × 10−4), and protective effects of calcium channel blocker associated variants on CAD (OR 0.69 per 10mmHg decrease in SBP; 95% CI 0.59–0.80; p = 2 × 10−6). No effect of blood pressure lowering via genetic proxies of these medication classes on risk of PAD was detected. Results were consistent in weighted-median MR sensitivity analyses (Supplemental Table 11).
Results scaled to reflect odds of outcome per 10mmHg decrease in blood pressure. n SNP = number of single nucleotide polymorphisms in the exposure instrument; OR = odds ratio; 95% CI = 95% confidence interval; P = p-value SBP = Systolic blood pressure; DBP = Diastolic blood pressure; PP = Pulse pressure; PAD = Peripheral artery disease; CAD = coronary artery disease.
In the second analysis, the effects of both SBP and DBP lowering conferred by genetic variants located within targets of a broader set of antihypertensive medications were considered (Supplemental Tables 12–13). Results for SBP were similar to the first analysis, identifying protective effects of calcium channel blocker associated variants on CAD (OR 0.64 per 10mmHg decrease in SBP; 95% CI 0.50–0.81; p = 2 × 10−4), vasodilator associated variants on CAD (OR 0.62 per 10mmHg decrease in SBP; 95% CI 0.40–0.95; p = 0.03), and angiotensin II receptor antagonist associated variants on CAD (OR 0.30 per 10mmHg decrease in SBP; 95% CI 0.097–0.95; p = 0.04). Though not associated with CAD, SBP lowering via beta-blocker associated variants decreased risk of PAD (OR 0.51 10mmHg decrease in SBP; 95% CI 0.27–0.97; p = 0.04). Only results for calcium channel blocker associated variants were robust in weighted-median sensitivity analyses.
For diastolic blood pressure, lowering via calcium channel blocker associated variants decreased risk of CAD (OR 0.52 per 10mmHg decrease in DBP; 95% CI 0.34–0.78; p = 0.002), lowering via angiotensin II receptor antagonist associated variants decreased risk of CAD (OR 0.11 per 10mmHg decrease in DBP; 95% CI 0.014–0.91; p = 0.04), lowering via beta-blocker associated variants decreased risk of CAD (OR 0.43 per 10mmHg decrease in DBP; 95% CI 0.19–1; p = 0.05), and lowering via thiazide diuretic associated variants increased risk of PAD (OR 2.97 per 10mmHg decrease in DBP; 95% CI 1.1–8; p = 0.03). Only results for calcium channel blocker associated variants were robust in weighted-median sensitivity analyses (Supplemental Table 13).
DISCUSSION
This Mendelian randomization study leveraged natural genetic variation in blood pressure in up to 757,601 individuals to examine the relationship between blood pressure and both PAD and CAD. The principal findings were: 1) Lifetime exposure to elevated SBP, DBP, and PP all increased risk of PAD and CAD; 2) Elevated SBP and DBP more strongly increased risk of CAD compared to PAD; 3) PAD led to a small but significant increase in PP; 4) Based on genetic proxies, the optimal antihypertensive regimens for prevention/treatment of PAD remains unclear. There are several implications from the results of this study.
First, this study supports observational findings that elevated blood pressure is associated with increased risk of PAD. Multiple observational studies have identified elevated SBP and clinical diagnosis of hypertension as strong risk factors for PAD, while the relationship between DBP and PAD has remained less clear (5–11,25–27). Unlike other observational studies, our MR study leveraged genetic variants as instrumental variables for SBP, DBP, and PP. Because genetic variants are randomly inherited by offspring from their parents, mimicking a trial randomizing individuals to a lifetime of increased blood pressure, the Mendelian randomization framework is less susceptible to residual environmental confounding than traditional observational studies (14). The finding of our MR analysis that elevated SBP increases risk of both PAD and CAD is consistent with prior studies. We also find a strong effect of DBP on both PAD and CAD, clarifying discrepant findings in prior observational studies. Overall, the MR findings of our study are consistent with a causal relationship between blood pressure traits and both PAD and CAD.
Next, we found that elevated SBP and DBP each increased risk of CAD more than PAD. These findings are in contrast to a prior observational analysis that found that SBP or DBP had similar effects on CAD and PAD (7). While broad recommendations for lifestyle modification and treatment of ASCVD risk factors are clearly important at both the population level and individual level, understanding the impact of interventions on specific ASCVD outcomes may further inform treatment and prevention guidelines and discussions with patients. Particularly in light of our recent finding that smoking more strongly increases risk of PAD in comparison to CAD or ischemic stroke (28), this study adds further nuance to the relationship between traditional ASCVD risk factors and specific ASCVD outcomes.
Our finding that increased pulse pressure increases PAD risk is consistent with findings from multiple prior observational studies (29–32). Because increased pulse pressure is a marker of increased arterial stiffness and may be caused by PAD, the observational studies investigating the relationship between these traits may have been limited by the possibility of reverse causality. Using bi-directional MR we were able to overcome this limitation, finding elevated PP to be a risk factor for PAD, and PAD to be a risk factor for increased PP. Our multivariable MR analysis of SBP and DBP suggests the effect of elevated DBP may be attenuated after accounting for the genetic effect of SBP. In other words, holding diastolic blood pressure constant, the effect of blood pressure on ASCVD is due to the effects of increased SBP (and indirectly, increased PP). These findings lend further support for the role of increased PP on PAD.
Finally, we used antihypertensive drug MR to estimate the effect of 10mmHg lowering of blood pressure by different classes of medication. In this analysis, we identified a protective effect of calcium-channel blockers on risk of CAD, consistent across all sensitivity analyses. Confidence intervals for other drug classes were wide, and did not exclude meaningful effects, which may reflect the small number of genetic variants included in the genetic instruments for each antihypertensive drug class. More robust genetic instruments may ultimately reveal additional antihypertensive drug classes that robustly lower ASCVD risk. Similarly, because increased SBP and DBP more strongly affected risk of CAD than PAD, the lack of effect of medication-specific SBP or DBP-lowering instruments and PAD is not surprising. While small beneficial genetic effects may compound over a lifetime leading to protection from ASCVD, the effects of antihypertensive medications occur on a much shorter timescale. Our findings do not exclude beneficial effects of potent antihypertensive medications on risk of PAD and CAD, particularly given the strong causal effects of each BP trait and each ASCVD outcome.
The overall findings of our study have implications for PAD prevention and treatment guidelines. The current 2016 American Heart Association/American College of Cardiology (AHA/ACC) and 2017 European Society of Cardiology (ESC) PAD guidelines make strong recommendations for the treatment of hypertension to prevent cardiovascular events.(3,4) The trials cited to support these recommendations focused on cardiovascular events broadly, or differences in safety and efficacy between different antihypertensive classes, rather than PAD-specific outcomes (33–41). A Cochrane Review found poor evidence for the use of antihypertensive medications specifically for PAD, though recognized the large benefit of these medications for prevention of cardiovascular events and mortality more broadly (42). Our MR study provides strong evidence consistent with a casual effect of increased blood pressure on PAD. In the absence of large randomized trials of antihypertensive medications focused on PAD-specific outcomes, these results add support for current guideline recommendations. Further, these results may help calibrate the expected benefit that programs to treat hypertension may have on the global burden of PAD.
This study has several potential limitations. The genetic studies of blood pressure, CAD, and PAD used in our analysis were primarily composed of individuals of European ancestry. Further study of BP and ASCVD genetics in diverse ancestral populations is necessary to improve the generalizability of our findings. Mendelian randomization relies on a number of assumptions in order for causal estimates to be valid (14). While we have employed multiple MR methods and sensitivity analyses to assess for and address potential violations of these assumptions, we cannot completely exclude the possibility of confounding. Future study on the role of hypertension treatment in the prevention and treatment of PAD focused on PAD-specific outcomes is warranted.
Overall, we find strong evidence consistent with a causal effect of blood pressure traits on ASCVD outcomes, with a stronger effect of SBP and DBP on CAD in comparison to PAD. Further study of the differential effects of traditional ASCVD risk factors on specific ASCVD outcomes may help guide prevention and treatment strategies for these common diseases.
Data Availability
Summary statistics for the blood pressure genome wide association study are publicly available, and may be downloaded from the NHLBI GRASP catalog (https://grasp.nhlbi.nih.gov/FullResults.aspx). MVP PAD genome wide association study summary statistics are available on dbGAP (Accession phs001672.v2.p1). Data on coronary artery disease have been contributed by CARDIoGRAMplusC4D investigators and have been downloaded from www.cardiogramplusc4d.org/data-downloads/.
FUNDING
This work was supported by US Department of Veterans Affairs grants IK2-CX001780 (Damrauer), and I01-BX003362 (Tsao/Chang). This research is based on data from the MVP, Office of Research and Development, Veterans Health Administration and was supported by award no. MVP000. This publication does not represent the views of the Department of Veterans Affairs or the United States government. This work was also supported by the National Institute of Diabetes and Digestive and Kidney Diseases R01-DK101478 (Voight), and a Linda Pechenik Montague Investigator Award (Voight). The Medical Research Council (MRC) and the University of Bristol support the MRC Integrative Epidemiology Unit [MC_UU_12013/1, MC_UU_12013/9, MC_UU_00011/1]. NMD is supported by a Norwegian Research Council Grant number 295989.
DATA AVAILABILITY
Summary statistics for the blood pressure genome wide association study are publicly available, and may be downloaded from the NHLBI GRASP catalog
(https://grasp.nhlbi.nih.gov/FullResults.aspx). MVP PAD genome wide association study summary statistics are available on dbGAP (Accession phs001672.v2.p1). Data on coronary artery disease have been contributed by CARDIoGRAMplusC4D investigators and have been downloaded from http://www.cardiogramplusc4d.org/data-downloads/.