Association of Genetic Instrumental Variables for Lung Function on Coronary Artery Disease Risk
A 2-Sample Mendelian Randomization Study
Background: Lung function, assessed by forced expiratory volume in 1 second (FEV1) and forced vital capacity (FVC), is inversely associated with coronary artery disease (CAD), but these associations could be because of confounding or reversed causality. We conducted a 2-sample Mendelian randomization study, using publicly available data from relevant genome-wide association studies, to examine the role of FEV1 or FVC on CAD.
Methods: We used the most recent genome-wide association studies on lung function to extract genetic instruments related to FEV1 and FVC (n=92 749). Data on the association between genetic instruments and CAD were obtained from Coronary Artery Disease Genome wide Replication and Meta-analysis plus The Coronary Artery Disease Genetics 1000 Genomes-based genome-wide association studies (60 801 CAD cases and 123 504 controls). We used inverse-variance weighting with a multiplicative random effect to estimate the genetic instrumented association of FEV1 and FVC on CAD. Sensitivity analyses included weighted median and MR-Egger methods.
Results: Each SD greater FEV1 was associated with a lower risk of CAD (odds ratio, 0.78 per SD; 95% confidence interval, 0.62–0.98) with a similar magnitude for FVC on CAD risk (odds ratio, 0.82 per SD; 95% confidence interval, 0.64–1.06). Estimates for FEV1 were similar when using MR-Egger method (odds ratio, 0.80 per SD; 95% confidence interval, 0.33–1.94) although the magnitude was smaller for weighted median method (odds ratio, 0.93 per SD; 95% confidence interval, 0.75–1.17). Estimates for FVC in the sensitivity analyses were attenuated (median) or changed direction (MR-Egger).
Conclusions: Our study suggested an inverse relation between FEV1 and CAD, but for FVC, evidence is less clear.
- coronary artery disease
- forced expiratory volume
- Mendelian randomization analysis
- odds ratio
- vital capacity
See Editorial by Nowak
Lung function, including forced expiratory volume in 1 second (FEV1) and forced vital capacity (FVC), predicts coronary artery disease (CAD) in observational studies although it is uncertain whether these associations are confounded or because of reverse causality. Mendelian randomization, which uses genetic markers of FEV1 and FVC, may clarify the role of FEV1 and FVC in CAD because this design is less susceptible to confounding. In this study, we conducted a Mendelian randomization study to assess the role of FEV1 and FVC in CAD using data from large genetic studies on lung function (n=92 749) and on CAD (60 801 CAD cases and 123 504 controls). We found a potential inverse association between FEV1 and CAD risk, with directionally consistent estimates across various sensitivity analyses, whereas the association between FVC and CAD risk was less clear. As such, this study provides more credible evidence that improving FEV1 may reduce the risk of CAD. Further research to determine the underlying mechanisms, which FEV1 may impact CAD risk, may identify novel targets for CAD prevention.
Reduced lung function is consistently associated with higher risk of cardiovascular diseases, including coronary artery disease (CAD), in observational studies.1–5 Possible mechanisms linking reduced lung function and cardiovascular diseases include increased inflammation because people with higher lung function tend to have lower levels of inflammatory markers, such as C-reactive protein and fibrinogen,6 inefficient cardiotoxic substance removal, or a mismatch of ventilation/perfusion ratio.6 However, these inverse associations could be a result of residual confounding or reverse causality. Smoking and height, which are both strongly related to lung function and cardiovascular disease in directions that would result in the direction of association between these 2, are particularly important sources of confounding.2,7 Furthermore, early CAD may result in reduced perfusion and reduced lung function, that is, reverse causation.8 If improving lung function can reduce CAD risk, this is a potentially modifiable risk factor that could have important public health benefit beyond that of improving lung function.
Mendelian randomization studies may provide more credible evidence concerning the causal role of lung function in CAD than that possible with conventional observational approaches. For example, previous studies have shown that Mendelian randomization studies generate comparable results with randomized controlled trials.9 A Mendelian randomization study to test the effect of lung function on CAD would use genetic variants that are robustly associated with lung function as instrumental variables for its causal effect. Because these variants are randomly allocated from parents to their offspring at conception, they are less susceptible to reverse causation than observational studies.10 Furthermore, the random allocation also implies that the genetic variants were less likely to be affected by the many factors that commonly confound observational studies, which is supported by empirical evidence.10,11 To our knowledge, no previous Mendelian randomization studies have explored the effect of lung function on CAD. Therefore, we conducted a 2-sample Mendelian randomization study to clarify the relation of lung function (forced expiratory volume in 1 second [FEV1] and forced vital capacity [FVC]) and CAD risk using summary statistics from large genome-wide association studies (GWAS).12,13 We hypothesized higher lung function (higher FEV1 and FVC) would be associated with lower risk of CAD. Further, because FVC is particularly related to height (more so than FEV1), we hypothesized that inverse association between FVC and CAD would be weaker than that between FEV1 and CAD in our Mendelian randomization study, if height is a key confounder of FVC with CAD.
The data and methods are publicly available.
Summary of Study Design
We used 2-sample Mendelian randomization with summary data from publicly available GWAS databases.12,14 First, we identified independent (r2<0.05) genetic instruments (single-nucleotide polymorphisms [SNPs]) strongly associated with FEV1 or FVC (P<5×10−8) from GWAS of lung function and obtained the estimates between each of these SNPs and lung function parameters (ie, difference in mean FEV1 or in FVC and their corresponding standard errors)—sample 1.15–17 Second, we obtained the corresponding genetic associations for each of the lung function SNPs with CAD risk from Coronary Artery Disease Genome wide Replication and Meta-analysis plus The Coronary Artery Disease Genetics (CARDIoGRAMplusC4D) 1000 Genome-based GWAS—sample 2.18 We then used these summary results to calculate the Mendelian randomization estimates of FEV1 and FVC on CAD risk, using various statistical methods as described in the statistical analysis section.19,20
GWAS of Lung Function (FEV1 and FVC—Sample 1)
For this study, we used data from the most recent GWAS (which included both FEV1 and FVC as outcomes),15 which was built on previous GWAS16,17 and had a combined discovery and replication sample size of 92 749 (discovery, 38 199; replication, 54 550). It was undertaken in European-origin participants and was imputed based on the 1000 Genomes Project phase 1 reference panel.15 Variants with imputation information <0.3, assessed using r2.hat (MACH and minimac) or.info (IMPUTE2), were excluded. The GWAS used linear regression adjusted for age, age 2, sex, height, and principal component for population structure, separately for never and ever smokers. The residuals from FEV1 and FVC were converted to ranks and then transformed to normally distributed z scores, which were then used in an additive genetic model, separately for never and ever smokers, with the results from these 2 (never and ever smokers) strata and then meta-analyzed. In the GWAS, only SNPs previously not identified were genotyped and analyzed in both discovery and replication stages, whereas previously known SNPs were only genotyped and analyzed in the discovery stage. As such, we extracted information for 17 SNPs related to FEV1 (10 from both stages and 7 from stage 1) and 11 SNPs related to FVC (5 from both stages and 6 from stage 1).15
GWAS of CAD: CARDIoGRAMplusC4D 1000 Genomes (Outcome—Sample 2)
We obtained summary data on the estimates of each lung function GWAS SNP with CAD from CARDIoGRAMPLUSC4D (downloading these data from www.CARDIOGRAMPLUSC4D.ORG).18 CARDIoGRAMplusC4D 1000 Genomes-based GWAS is a meta-analysis of GWAS of CAD case–control studies of people of mainly European (≈91%), South Asian, and East Asian descent imputed using the 1000 Genomes phase 1 v3 training set with 38 million variants. The study interrogated 9.4 million variants and included 60 801 CAD cases and 123 504 controls.18 Case status was defined as having CAD if they had received a diagnosis of myocardial infarction, acute coronary syndrome, chronic stable angina, or coronary stenosis >50%. Diagnoses were ascertained in various ways in different studies; extracted from medical records, clinical diagnosis at the time of study, procedures (coronary angiography results or bypass surgery), medications or symptoms that indicate angina, or self-report of a doctor diagnosis, as described elsewhere.18 Covariate adjustment in the GWAS included study-specific covariates (eg, age and sex) and genomic control. The GWAS reported the corresponding log odds and standard error on CAD for each SNP effect allele using an additive model.18
Harmonization of the 2 Samples
To ensure correct allele harmonization, SNPs were orientated so that the effect allele was the same for both the exposure and the outcome datasets. Palindromic SNPs were checked and corrected using the effect allele frequency reported in the exposure and outcome GWAS to identify corresponding strands between the 2 GWAS. If effect allele frequency is >0.42 and ≤0.50, a proxy nonpalindromic SNP (R2=1) would be used instead to avoid ambiguity in strand direction because alleles of palindromic SNPs are the same on both strands.21 Linkage disequilibrium of the SNPs were checked via 1000 Genomes project to avoid inclusion of correlated SNPs. If 2 SNPs had an R2 of >0.05, the SNP with the smaller P value was retained and the other SNP was discarded.
Main Mendelian Randomization Analysis
Inverse-variance weighting (IVW) with multiplicative random effects was used to give our main Mendelian randomization effect estimate, which is a weighted regression of SNP-outcome association (log odds of CAD per effect allele) on SNP-exposure association (mean difference in FEV1 or FVC per effect allele), with the intercept constrained to zero.22 One key assumption for this method to produce a valid estimate is that there is no other way SNPs could affect the outcome (CAD here) than through the risk factor that the SNPs are instrumenting for (FEV1 or FVC). Specifically with MR, it assumes no horizontal pleiotropy.12 We also calculated Wald ratio for each SNP (ie, SNP-outcome association divided by SNP-exposure association), with the standard error calculated by dividing the standard error of the SNP-outcome association by the effect size of the SNP-exposure association. We then meta-analyzed these ratios using a fixed-effect method and compared the results with the multiplicative random effects IVW. Both methods are supposed to estimate the same overall effect, but standard errors will be larger when using multiplicative random effects IVW in the presence of heterogeneity of effect across SNPs. We also estimated the I2 statistics to explore whether there was heterogeneity in effects between SNPs, which could be suggestive that ≥1 SNPs are invalid instruments.
We undertook several sensitivity analyses to test the robustness of our main analyses. Because the IVW multiplicative random effects models can give biased estimates if ≥1 instruments are invalid because of horizontal pleiotropy,20 we particularly focus on sensitivity analyses that allow the assumption of no horizontal pleiotropy to be relaxed. These analyses have different assumptions to the IVW (main analyses) method and also to each other. In comparison with the IVW method, they have less statistical efficiency; this is particularly the case with MR-Egger. Therefore, the aim of these sensitivity analyses was mostly to examine whether the magnitude and direction of effect estimates were consistent across methods.
The IVW estimates were recalculated removing 1 SNP at a time aiming to explore whether the overall estimate was primarily driven by any (outlying) SNP.
Similar to the IVW, the MR-Egger method is a weighted regression of the SNP-outcome on the SNP-exposure association, but unlike the IVW method, the intercept is not constrained to be zero.19 This means that if there is unbalanced pleiotropy, the Mendelian randomization (slope of the regression line) effect estimate will differ from the IVW estimate, and the intercept of the regression line will be nonzero. The MR-Egger method slope will give unbiased estimate even if all instruments are invalid (ie, because of the presence of horizontal pleiotropy).19 However, it has an additional assumption known as the Instrument Strength Independent of Direct Effect, which requires no correlation between the strength of the direct effect of instrument on outcome and the instrument strength (strength of association between genetic variants for lung function and lung function). One possible scenario through which the instrument strength independent of direct effect assumption could be violated is if the pleiotropic effects of genetic instruments are mediated via the same exposure-outcome confounder.20 The MR-Egger method is statistically inefficient and expected to have considerably wider confidence intervals (CIs) than our main analyses or other sensitivity analyses. As such, we focus on whether the odds ratio (OR) is consistent in terms of magnitude and direction with our main analysis results and also whether the intercept value suggests evidence of pleiotropy.
Weighted Median Method
The weighted median method is a further method that allows some relaxation of the horizontal pleiotropy assumption. It weights each of the Wald ratio effects of each SNP by the corresponding precision and then takes the median weighted ratio as the causal effect.20 It will provide an unbiased estimate so long as >50% of the weights were derived from valid instruments.20 This means that the estimate will be biased if 1 single SNP is pleiotropic and contributes >50% of the weight for the overall effect or if a group of SNPs are pleiotropic and together they contribute >50% of the weight.
Difference in the Imputation Reference Panel for Lung Function GWAS
To examine whether the results were robust to the difference in the imputation reference panel, we also extracted genome-wide significant SNPs for lung function from the most recent GWAS, which used HapMap reference populations (CEU), and repeated the analyses.16,17 For FEV1, the SNPs were extracted from a GWAS with a sample size of 94 612 (discovery, 48 201; replication, 46 411) of mainly European origin.17 For FVC, the SNPs were extracted from a GWAS with a sample size of 85 170 (discovery, 52 253; replication, 32 917) of mainly European origins.16 Similar adjustments, stratification by smoking and analytical methods, and imputation quality control threshold were used in these GWAS as in the one used in our main analyses with imputation using the 1000 Genome Project phase 1 reference panel although the FVC GWAS used absolute FVC (mL) as the outcome.16 To improve comparability with the other analyses, we rescaled the Mendelian randomization estimates of FVC (HapMap) into per SD. The SD was derived by pooling the SDs reported in the European cohorts in the FVC GWAS.
Exclusion of Potential Pleiotropic SNPs
Considering that height and smoking are key potential confounders, we checked whether any of our selected FEV1 or FVC genetic instruments were related to height or smoking (ever smoking status and cigarettes smoked per day) using summary statistics from the relevant GWAS consortia (Genetic Investigation of Anthropometric Traits and Tobacco and Genetics, respectively).23,24 Similarly, we explored whether any of our FEV1 or FVC genetic instruments were related to established cardiovascular risk factors by looking for their associations using data from relevant GWAS: triglycerides and low density lipoprotein cholesterol (from Global Lipids Genetics Consortium),25,26 body mass index (Genetic Investigation of Anthropometric Traits),27 and systolic blood pressure (UK Biobank).28,29 We considered a SNP as potentially pleiotropic if there was evidence of it being associated with height, smoking, or cardiovascular risk factors after applying a Bonferroni correction to account for multiple testing (P<0.0004), calculated by 0.05 (nominal P value)/126 (number of comparisons). We repeated the main analysis (IVW estimate) after excluding the SNPs classified as pleiotropic.
Multivariable Mendelian Randomization Analysis
Because some of the SNPs predicted both FEV1 and FVC (ie, with P value <5×10−8 for both parameters) based on the most recent GWAS,15 we also conducted a multivariable Mendelian randomization analysis to adjust effects of FEV1 by FVC, and vice versa, to disentangle the effects of each exposure on CAD risk.30
Exploring Instrument Strength
Instrument strength was assessed by presenting the first-stage (regression of FEV1 and FVC on genetic instruments) F statistics and R2 values. For both of these statistics, higher values related to stronger instruments.
We calculated the first-stage F statistics for the main analyses using Equation 1 in the Data Supplement. The GWAS only reported R2 of the instruments (known and novel) for the discovery stage.15 We directly extracted the R2 for known loci as reported in the GWAS discovery stage. However, the R2 for novel loci in the discovery stage could be susceptible to winner’s curse.31 As such, we calculated the R2 for novel loci using Equation 2 in the Data Supplement using parameters from each novel loci based on the combined analyses (discovery+replication). Then, we summed up the R2 and computed the F statistics using the sample size as stated in the GWAS.
All analyses were performed using R, version 3.3.2 (R Development Core Team, Vienna, Austria), with the R package (TwosampleMR). This study only used publicly available data, and hence, no ethical approval was required to conduct this study.
Among 17 SNPs for FEV1 (1000 Genomes imputation), rs1985524 was palindromic and was replaced by rs1989154 (R2=1) to avoid ambiguity in the strand direction, rs11383346 was removed because this was an indel and was not available in CARDIoGRAMplusC4D, hence 16 SNPs were used. Among 11 SNPs for FEV1 (HapMap imputation), rs17331332 and rs3995090 were in linkage disequilibrium with other SNPs and were removed, and hence, 9 SNPs remained in these analyses. Among 11 SNPs for FVC (1000 Genomes imputation), rs1430193 was palindromic and was replaced by rs1430194 to avoid ambiguity in the strand direction; rs11383346 was removed because this was an indel and was not available in CARDIoGRAMplusC4D, thus 10 SNPs remained in these analyses. Among 6 SNPs for FVC (HapMap imputation), rs1430193 was palindromic and was replaced by rs1430194.
Tables I and II in the Data Supplement summarize the genetic associations as reported in the relevant GWAS after allele harmonization. The first-stage F statistics and variance were explained by the genetic instruments and were 49.5 and 0.8% for FEV1 and 38.4 and 0.4% for FVC.
Table III in the Data Supplement shows the associations of all SNPs that were used as instrumental variables with FEV1 and FVC with smoking, height, body mass index, low density lipoprotein cholesterol, triglycerides, and systolic blood pressure. After correcting for multiple testing, 5 of the SNPs we were using as genetic instruments were associated with height (rs1032296, rs134041, rs1430194, rs6923462, and rs7155279), 1 SNP (rs2036527) associated with the number of cigarettes smoked per day, 1 with body mass index (rs4237643), and 4 with systolic blood pressure (rs1989154, rs2571445, rs2863171, and rs7068966); none of our genetic instruments were associated with ever smoking, low density lipoprotein cholesterol, or triglycerides. Based on the lung function GWAS, rs6681426, rs6441207, rs2274116, and rs10850377 were associated with both FEV1 and FVC.15
FEV1 was inversely associated with CAD risk in the IVW with multiplicative random effects model (OR, 0.78 per SD increase; 95% CI, 0.62–0.98; I2=62%; Table). Findings for the association of FEV1 with CAD risk were consistent across the range of sensitivity analyses. The leave-one-out plot suggested the overall estimate was not driven by any single outlying SNP (Figure 1). The estimate from MR-Egger, which is more robust to invalid variants, was consistent with the main estimate (OR, 0.80; 95% CI, 0.33–1.94), although had wider confidence as expected because of the lower statistical power of this approach. In addition, there was no evidence for directional pleiotropy based on the MR-Egger intercept value (1000 Genomes imputation: −0.0009, P=0.96; HapMap imputation: −5.4×10−5, P=0.998; Table). The weighted median estimate was somewhat attenuated in comparison with the main IVW analyses, but consistent with those and the MR-Egger results (OR, 0.93 per SD; 95% CI, 0.75–1.17). Similar findings were observed when we used estimates from the GWAS imputed using HapMap reference panel (Table).
There was also an inverse association of genetic instruments for FVC with CAD risk in the IVW with multiplicative random effects model using 1000 genome imputation (OR, 0.82 per SD increase; 95% CI, 0.64–1.06; I2=35%; Table). This point estimate is similar to that for the association of FEV1 with CAD but with wider CIs. The estimate was weaker (closer to the null) using the weighted median method (OR, 0.92; 95% CI, 0.69–1.23), and the MR-Egger method yielded a directionally different estimate with wide CIs (OR, 1.14; 95% CI, 0.36–3.61). There was no evidence for directional pleiotropy based on the MR-Egger intercept value (1000 Genomes imputation: −0.011, P=0.59; HapMap imputation: −0.022, P=0.48; Table). The leave-one-out plot suggested the overall estimate was not driven by any single SNP, although it was somewhat attenuated when rs4237643 or rs6681426 were removed (Figure 2). When we used estimates from the GWAS with the HapMap imputation, the main IVW estimates were similar to the 1000 genome imputation (OR, 0.85; 95% CI, 0.49–1.48), as were the results for weighted median analyses (attenuating toward the null) and MR-Egger (showing a positive, rather than inverse, association; Table).
Repeating the analyses with the exclusion of SNPs related to height, smoking phenotypes, body mass index, and systolic blood pressure did not notably change the results (Tables IV and V in the Data Supplement). The fixed-effects Wald ratio meta-analysis gave similar results (Table VI in the Data Supplement). The multivariable Mendelian randomization analysis produced directionally similar results compared with the MR-Egger regression for FEV1 (OR, 0.82 per SD increase; 95% CI, 0.53–1.28) and FVC (OR, 1.84 per SD increase; 95% CI, 0.61–5.50).
To our knowledge, this is the first Mendelian randomization study examining the role of lung function on CAD. Consistent with previous multivariable regression analyses, we found an inverse relation between FEV1 and CAD.1,2 In our main analyses, the association of FVC on CAD was similar to that of FEV1. However, although the association of FEV1 with CAD was robust to numerous sensitivity analyses, those for FVC were less so, with MR-Egger and the multivariable regression analyses (in which we adjusted for effect of SNPs on FEV1) suggesting a possible positive association of FVC on CAD, albeit with wide CIs including the null. Thus, our study does not corroborate previous multivariable regression studies of FVC, which suggested an inverse association of FVC with CAD risk.6 Overall, our findings suggest that higher FEV1 may reduce CAD risk but that previous inverse associations of FVC with CAD from multivariable regression analyses are possibly explained by residual confounding, in particular because of height.
The inverse association of FEV1 with cardiovascular disease risk has been consistently seen in observational studies.1–4 However, smoking is a strong determinant of FEV1 and CAD, and it is difficult to fully adjust for it without detailed information on life course frequency, amount, and intensity of smoking. Therefore, residual confounding by smoking might have explained those previous results.2,7 Our Mendelian randomization study, which is less susceptible to such confounding,10 suggests that greater FEV1 may be related to reduced CAD risk. This conclusion is particularly supported by the consistency of results across several different Mendelian randomization approaches, including a meta-analysis of Wald ratio estimates for each SNP, MR-Egger, and weighted median methods, as well as when results from FEV1 GWAS with a different imputation reference panel used to identify genetic instruments, and when we removed SNPs with known associations with smoking, height, and body mass index (for which we found some possible evidence of pleiotropy), and multivariable Mendelian randomization analysis. These different approaches each have differing underlying assumptions, and hence key sources of bias, and thus the triangulation of findings across them increases confidence that FEV1 is causally related to CAD risk.32 We did not find robust and consistent (across our sensitivity analyses) evidence of an association of FVC on CAD. It is possible that the null findings are because of lack of statistical power since some of these methods are less statistically efficient and with FEV1 CIs were wide and included the null is some sensitivity analyses. The difference between the sensitivity analyses of FEV1 and CAD and of FVC and CAD is that the former were all directionally consistent and had broadly similar magnitudes, whereas with FVC, 2 of the sensitivity analyses suggested positive rather than inverse associations, which suggests less robust Mendelian randomization evidence for an inverse association.
That our study suggests possible differences in the association of FEV1 and FVC on CAD is perhaps not surprising given they reflect different aspects of lung function, as reflected by the differing genetic architecture observed in GWAS.16,17 FEV1 represents severity of airway obstruction,33 whereas FVC is a measure of overall lung capacity. These differences result in somewhat different confounding structures between the 2 measurements, with smoking being more strongly associated with FEV1 and height with FVC. Thus, our results suggest that inability to fully account for confounding by height in previous multivariable regression analyses might have resulted in spurious inverse associations of it with CAD. Our results also suggest that airway obstruction, rather than lung size or capacity, may increase CAD risk.
Strengths and Limitations
Key strengths of this study are the large sample size, with >60 000 CAD cases, and the use of Mendelian randomization to minimize the impact of confounding and reverse causation. The limitations of our study in inferring causal relation relate to the extent to which the key Mendelian randomization assumptions were satisfied—that the genetic instruments are (1) robustly associated with risk factor of interest (here FEV1 and FVC), (2) not associated with confounders of the risk factor–outcome association, and (3) not related to the outcome via any path other than through the risk factor (ie, that there is no horizontal pleiotropy). In 2-sample Mendelian randomization as used here, weak instrument bias is expected to bias findings toward the null (rather than toward the confounded result as occurs in 1-sample Mendelian randomization). However, this will depend on the extent of overlap between the 2 samples, with substantial overlap resulting in similar bias to that seen for 1-sample Mendelian randomization.12 Only 1 cohort (n=837; 0.5%) was used in both the lung function and CAD GWAS samples, and the high F statistic and R2 for the combined genetic instrumental variables, suggest that weak instrument bias is unlikely to have impacted our results. Horizontal pleiotropy is unlikely given the similarity of results across the different methods that we have used, and the intercept values from the MR-Egger being close to zero. One limitation of using summary GWAS data is that we are unable to explore the association of the genetic instruments with a broad range of observed potential confounders, but we did find similar results when we excluded SNPs that are known to be related to smoking and height, the 2 confounders we were mostly concerned about, 1 SNP associated with body mass index, and 4 SNPs associated with systolic blood pressure. We did not find evidence for any of our genetic instrumental variables being robustly statistically associated with low density lipoprotein cholesterol and triglycerides, which are important risk factors for CAD. In these analyses, the fact that the GWAS of both FEV1 and FVC adjusted for height and smoking could introduce spurious associations through a phenomenon known as collider bias. However, results were consistent after removing SNPs that had a residual association with height and smoking (Table IV in the Data Supplement), suggesting that collider bias is unlikely to fully account for the inverse association between FEV1 and CAD. Nevertheless, research using UK Biobank may help verify the findings because some of these biases can only be explored with individual data.34 Last, we were unable to explore potential reverse causation using a bidirectional Mendelian randomization design because summary data across the whole genome for the lung function GWAS are currently not available. Although there may be a bidirectional association (ie, a propensity to CAD causing reduced FEV1, as well as vice versa), this would not negate the association of FEV1 in CAD that we have identified here.
In this first Mendelian randomization study of the relation of FEV1 and FVC with CAD risk that we are aware of, we find evidence for an inverse association of FEV1 with CAD risk but less robust evidence for an association of FVC with CAD risk. Although it has been suggested that increased systemic inflammation in those with poor lung function, inefficient cardiotoxic substance removal, or a mismatch of ventilation/perfusion ratio are potential ways in which lower FEV1 is related to increased CAD risk, the exact mechanisms are unclear.6 Because our findings now provide more concrete evidence that there may be a causal effect that is not explained by confounding because of smoking or height, further research to explore potential mechanisms could identify novel targets for the prevention of CAD.
Data on coronary artery disease have been contributed by CARDIoGRAMplusC4D investigators and have been downloaded from www.CARDIOGRAMPLUSC4D.ORG. We acknowledged the Tobacco and Genetics Consortium, Genetic Investigation of Anthropometric Traits Consortium, Global Lipids Genetics Consortium, and the UK Biobank data on blood pressure, analyzed by Neale Laboratory and deposited at MR Base (Work conducted under the UK Biobank applications 18597 and 11898). Dr Jack Bowden (MRC Integrative Epidemiology Unit at the University of Bristol) provided valuable insights at the planning stage of this study.
The views expressed in this article are those of the authors and not necessarily any of the funders. The work presented here is the responsibility of the authors and not anyone acknowledged.
Sources of Funding
Drs Borges and Lawlor work in a unit that receives funding from the UK Medical Research Council (MC_UU_12013/5), and Dr Lawlor is a UK National Institute of Health Research Senior Investigator (NF-SI-0611-10196). Dr Borges is supported by an Medical Research Council Skills Development Fellowship MR/P014054/1. Dr Au Yeung was supported by the Health and Medical Research Fund Research Fellowship Scheme (01150037), Food and Health Bureau, Hong Kong Special Administrative Region, People’s Republic of China. The funders had no role in the design, analyses, interpretation of results, or writing of the article.
Dr Lawlor receives support from Roche Diagnostics and Medtronic for biomarker research that is unrelated to the content of this article. The other authors report no conflicts.
The Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGEN.117.001952/-/DC1.
The views expressed in this article are those of the authors and not necessarily any of the funders. The work presented here is the responsibility of the authors and not anyone acknowledged.
- Received August 2, 2017.
- Accepted February 21, 2018.
- © 2018 American Heart Association, Inc.
- Johnston AK,
- Mannino DM,
- Hagan GW,
- Davis KJ,
- Kiri VA
- Wannamethee SG,
- Shaper AG,
- Papacosta O,
- Lennon L,
- Welsh P,
- Whincup PH
- Li J,
- Agarwal SK,
- Alonso A,
- Blecker S,
- Chamberlain AM,
- London SJ,
- et al
- Borges MC,
- Lawlor DA,
- de Oliveira C,
- White J,
- Horta BL,
- Barros AJ
- Hartwig FP,
- Davies NM,
- Hemani G,
- Davey Smith G
- Bowden J,
- Del Greco M F,
- Minelli C,
- Davey Smith G,
- Sheehan N,
- Thompson J
- Wood AR,
- Esko T,
- Yang J,
- Vedantam S,
- Pers TH,
- Gustafsson S,
- et al
- Liu DJ,
- Peloso GM,
- Yu H,
- Butterworth AS,
- Wang X,
- Mahajan A,
- et al
- Locke AE,
- Kahali B,
- Berndt SI,
- Justice AE,
- Pers TH,
- Day FR,
- et al
- Hemani G,
- Zheng J,
- Wade KH,
- Laurin C,
- Elsworth B,
- Burgess S,
- et al
- Churchhouse C,
- Neale B
- Rosner B
- Lawlor DA,
- Tilling K,
- Davey Smith G
- Young RP,
- Hopkins R,
- Eaton TE