Genetic Association of Lipids and Lipid Drug Targets With Abdominal Aortic Aneurysm

This meta-analysis examines data sets from 5 international genome-wide association studies to determine the association between elevated lipids and the risk for abdominal aortic aneurysm.

A bdominal aortic aneurysm (AAA) is an important cardiovascular disease (CVD) resulting in approximately 4500 deaths from AAA rupture per year in the United States. 1 Approximately 45 000 operations are carried out each year to prevent rupture, resulting in 1400 deaths. 1 Screening for AAA reduces the burden of rupture, 2 and therefore many countries now offer such screening to at-risk groups. 3,4The US Preventive Services Task Force recommends screening men aged 65 to 75 years with a history of smoking, and the American Heart Association guidelines suggest surgical repair is needed when the AAA reaches 5.5 cm in diameter.
Abdominal aortic aneurysm shares risk factors with occlusive atherosclerotic disease, but the magnitude and direction of this association is not always consistent.6][7] For example, the risk of smoking for AAA is at least 2-fold greater than that for coronary heart disease (CHD), 7 whereas type 2 diabetes appears to be protective for AAA but is a major risk factor for occlusive vascular disease. 6This example suggests that AAA may have some distinct causal pathways, and understanding these pathways is important for setting public health policies aimed at reducing the risk posed by AAA and its complications.
Genome-wide association studies (GWASs) of AAA have identified robust associations of loci that have previously been found for CHD (9p21), 8 DAB2IP (Entrez Gene 153090), 9 LDLR (Entrez Gene 3949), 10 SORT1 (Entrez Gene 6272), 11 and IL6R (Entrez Gene 3570) 12 as well as a number of variants that do not appear to be associated with other CVDs (LRP1 [Entrez Gene 4035], 13 SMYD2 [Entrez Gene 56960], ERG [Entrez Gene 2078], MMP9 [Entrez Gene 4318], and LINC00540 [Entrez Gene 100506622] 14 ).Again, these findings lend support to the hypothesis that AAA and CHD have overlapping pathophysiology, but the association with AAA and not with other CVDs suggests that discrete etiological pathways may well exist between these vascular diseases.
The role of low-density lipoprotein cholesterol (LDL-C) levels in CHD is well defined, and LDL-C lowering therapies are of clear benefit in reducing CHD risk. 159][20][21] In AAA, meta-analyses of observational studies do show a consistent inverse association of HDL-C with AAA risk, but the association with LDL-C is less clear. 22,23It is important, however, to recognize that the studies included in these metaanalyses were small case-control studies, many of which did not adjust for statin use.There is a paucity of any data reporting an association between triglycerides (TG) and AAA risk or progression.From a clinical point of view, it is important to understand the role of lipids in AAA, especially considering the excess cardiovascular risks in patients with AAA 24 and the recent publications showing low prevalence of lowering levels of LDL-C in patients with AAA. 25,26Previous genetic association studies have pointed to a potential role of lipids in AAA pathology, 10,11,27 but this current study uses a larger panel of single-nucleotide polymorphisms (SNPs), a considerably larger sample, and more advanced methods.
Mendelian randomization (MR) is an approach that uses the unique properties of genotype to investigate causal associations. 28Specifically, genotype is randomly allocated at conception (owing to Mendel's second law, a feature that is exploited to minimize confounding) and is not affected by reverse causation.Although MR has traditionally been used to explore causal associations between circulating biomarkers and disease phenotypes, it has an extension that uses genotype to validate drug targets.In this approach, variants in genes encoding potential drug targets are used as instruments to explore the utility of targeting this pathway in specific disease states. 29,30A major challenge in MR studies of complex traits such as lipid fractions is genetic pleiotropy, whereby SNPs influence circulating concentrations of multiple lipid fractions.This so-called pleiotropy may reflect an association of an SNP (or multiple SNPs in combination) with multiple discrete pathways that may have differing associations with AAA, leading to a potentially biased estimate from MR.Recent developments in the technique, such as multivariable MR, 16 weighted median MR, 31 and MR-Egger, 32 have been used to address these issues, but pleiotropy still poses a challenge.
In this study, conventional inverse-variance weighted MR, multivariable MR, weighted median MR, and MR-Egger approaches were used to investigate the role of lipids in the etiology of AAA.

Methods
From January 9, 2015, to December 21, 2016, we investigated the association of genetic risk scores (GRS) for lipid traits with AAA reported in up to 4914 cases and 48 002 controls across 5 international AAA GWASs 14 that took place in the United Kingdom and Australia, 13,14 New Zealand, 13,14 the United States, 14 the Netherlands, and Iceland. 9The GRS were composed of SNPs that are robustly associated with serum lipids in the Global Lipids Genetics Consortium meta-GWAS of circulating lipid levels. 33Data collection for this study took place between January 9, 2015, and January 4, 2016.Data analysis was conducted between January 4, 2015, and December 31, 2016.

Study Populations
We used summary SNP-AAA association statistics from the 5 published GWASs of AAA.Detailed descriptions of these GWAS analyses are provided in the eAppendix in the Supplement and previous publications. 9,13,14We supplemented the study of single variants in genes encoding lipid drug targets with data derived from the Secondary Manifestations of Arterial Diseases (SMART) study.The Table includes the number of cases and controls in each study.Descriptions of study cohorts and demographic details are presented in the eAppendix in the Supplement and previous publications. 9,13,14In all studies, the case definition of AAA was an infrarenal aortic diameter of 3 cm or more by ultrasound or computed tomographic imaging or previous AAA rupture or repair.Details of the association tests and quality control used in each study are included in the eAppendix in the Supplement and a published meta-GWAS. 14

Selection of SNPs
We identified SNPs associated with lipids in the Global Lipid Genetics Consortium 33 using the SNP selection criteria by Do et al. 16 Briefly, SNPs in association with at least 1 of the 3 lipid traits (LDL-C, HDL-C, or TG concentrations) at a genome-wide significance level (P < 5 × 10 −8 ) were selected.In Do et al 16 at loci with multiple associated SNPs, single SNPs with the strongest effect estimates were selected, and more than 1 SNP was selected only if there was evidence of minimal linkage disequilibrium (r 2 < 0.05).Data were available for the 180 of 185 SNPs (eTable 1 in the Supplement) described in Do et al. 16

Data Analysis
We first harmonized SNPs across the data sets (Global Lipids Genetics Consortium and Aneurysm Consortium) by merging SNPs on the reference SNP cluster identification or rs number.Then, we ensured that effect alleles were denoted to be the same in both data sets and double-checked the information by investigating effect-allele frequencies.We oriented all variants to ensure that the effect allele was positively associated with each lipid trait (eg, in the MR of LDL-C, all β coefficients for LDL-C were >0).This orientation resulted in a data set in which each SNP was a unique row and there were separate columns for β and SEs for each lipid trait and the log odds ratio (OR) and corresponding SE for AAA (eTable 1 in the Supplement).

Conventional MR
We conducted a conventional 2-sample MR analysis to determine the association between a 1-SD genetically elevated lipid concentration and AAA risk.For this analysis, we used the inverse-variance weighted MR method in which the SNP association estimates for the outcome (β for AAA) are regressed on the SNP association estimates for each lipid (β for LDL-C, β for HDL-C, and β for TG) individually in turn.The regression was weighted by the inverse variances of the estimated associations of the SNPs with the outcome and then was forced to pass through the origin.

Multivariable MR
To gauge some insight into potential "independent" associations of the lipids with AAA risk, we used the multivariable MR method.In this approach, a single regression model with outcome variable (β for AAA) was fitted for the predictor variables (β for LDL-C, β for HDL-C, and β for TG).The model was implemented, as described previously, 34 as a multilinear regression of SNP association estimates weighted by the inverse variances of the estimated associations of SNPs with the outcome and forced to pass through the origin.

MR-Egger
We used the MR-Egger 32 method that tests for the presence of, and provides an MR estimate that is adjusted for, unmeasured net pleiotropy.The method involves conducting an unconstrained linear regression of the SNP association estimates for the outcome on the SNP association estimates for the exposure weighted by the inverse variance of the estimated association of SNP with outcome.In MR-Egger, any net pleiotropy manifests in the intercept.Under the assumption that pleiotropic associations are independent of the associations of the SNPs with the exposure, the regression slope coefficient should represent an unbiased MR association estimate.

Weighted Median MR
As a further sensitivity analysis, we performed the weighted median MR method. 31Whereas the conventional inversevariance weighted method calculates a weighted mean of the SNP-specific causal association estimates, the weighted median method calculates a weighted version of the median of the SNP-specific causal association estimates.Because the median of a distribution is not affected by extreme values, the weighted median method is less sensitive to individual pleiotropic SNPs.The weighted median estimate is unbiased in large samples if at least 50% of the weights from SNPs are valid (eg, not pleiotropic).b This cohort was used in the mendelian randomization of drug targets analysis.
c Reflecting a single variant study only.

SNPs in Drug Target Analysis
To our knowledge, there have been no large-scale randomized trials of lipid-lowering treatments in patients with AAA, and observational studies have often been small and retrospective and yielded heterogeneous results.We examined the association of rs12916 in HMGCR (a genetic proxy for statins; Entrez Gene 3156), rs3764261 in CETP (a proxy for CETP inhibitors; Entrez Gene 1071), as well as rs2479409 and rs11206510 in PCSK9 (a proxy for PCSK9 inhibitors; Entrez Gene 255738) with AAA to identify the potential utility of pharmacological modification of these drug targets in AAA.

Statistical Calculations
The MR analyses for blood lipids were performed using the "MendelianRandomization" command in R, version 3.3.3(R Foundation for Statistical Computing), 35 and 2-tailed P values were derived from instrumental variable estimators.Given that there was only one outcome under investigation (AAA) and the lipids traits were correlated with one another, we used 2-tailed P < .05 to denote evidence against the null hypothesis (ie, P < .05provided evidence in favor of an association between the exposure and outcome).

Multivariable MR, MR-Egger, and Weighted Median MR Approaches
It is possible to remove SNPs with pleiotropic associations from the GRS, but this removal diminishes the strength of the instrumental variable 36 and can introduce bias. 37Therefore, we adopted the multivariable MR method described by Do et al 16 and modified by Burgess and Thompson 34 to gain insight into the potential independent associations of these lipid GRS with AAA risk.To account for any net unbalanced pleiotropy, we used the MR-Egger method.To reduce the influence of outlying (possibly pleiotropic) variants on the analysis, we used the weighted median MR method.None of these sensitivity MR analyses resulted in a material change to either the magnitude or significance of the estimates (Figure 1).The point estimates for concentrations of LDL-C and HDL-C remained largely unaltered, whereas for TG the point estimate diminished for the multivariable MR method; however, on the MR-Egger and weighted median MR methods, TG level remained convincingly associated with AAA.

Association of SNPs With Lipid Drug Targets
We selected rs12916 in HMGCR, rs3764261 in CETP, as well as rs2479409 and rs11206510 in PCSK9 as there are licensed drugs that target pathways associated with these genes.
The PCSK9 inhibitors are a novel class of drugs used to target LDL-C.To date, in CHD, genetic and clinical studies have had concordant results. 33,38We examined 2 independent SNPs in PCSK9 (rs2479409 and rs11206510; linkage disequilibrium r 2 = 0.07) that were used as proxies for PCSK9 inhibition in a large-scale MR analysis 39 and have strong, independent associations with both LDL-C levels and CHD.The LDL-C-lowering allele of rs2479409 was not associated with AAA risk (OR, 0.97; 95% CI, 0.92-1.02;P = .28).The LDL-C-lowering allele of rs11206510 in PCSK9 was weakly associated with AAA risk (OR, 0.94; 95% CI, 0.88-1.00;P = .04)(Figure 2).
We used rs3764261 as a proxy for CETP inhibition.Although the allele increases HDL-C levels, it is also associated with lower circulating concentrations of TG and LDL-C; thus, rs3764261 cannot be considered as an instrument for HDL-C in isolation but can be used to gauge insight into the potential associations with CETP inhibition. 30This HDLraising CETP SNP was associated with lower AAA risk (OR per HDL-C-raising allele, 0.89; 95% CI, 0.85-0.94;P = 3.7 × 10 −7 ).

Discussion
Understanding the relevance of lipid fractions in the development of AAA has important implications from both etiological and translational standpoints.In this study, we used MR to provide robust evidence that the major lipid fractions-LDL-C, HDL-C, and TG-are likely to play important roles in the etiology of AAA.A similar genetic approach has been used previously, 27 but this present study has expanded on this technique by including many more individuals and more SNPs and by using more recent developments in MR, which collec-tively increase statistical power and strengthen the validity of the association estimates reported here.
Disentangling the roles of correlated biomarkers in disease etiology continues to be an analytical challenge; to this end, we used recently developed techniques for the multivariable MR method. 16Interestingly, there appear to be independent associations between genetically instrumented levels of LDL-C, HDL-C, and TG and AAA risk.This finding is in contrast to findings in studies of CHD in which a similar approach found weaker associations between HDL-C genetic variants and CHD (after shared pathways with LDL-C and TG and pleiotropy had been taken into account 16,18,19,36 ) or aortic stenosis in which only LDL-C appeared to play a causal role. 40This finding highlights the complexity of lipid pathways across the diverse biology of CVD and suggests that results from studies focused solely on CHD (which can be defined variably) cannot always be extrapolated to other vascular diseases such as AAA.
Although it has been possible to investigate for pleiotropic associations of genetic variants used collectively in the lipid GRS employed in the MR analyses we conducted, it is not so straightforward as to disentangle the phenotypic overlap whereby many patients with AAA also harbor atherosclerotic disease in other vascular beds.Therefore, it is tempting to suggest a causal role for lipids specifically in AAA pathogenesis, but these genetic analyses do not provide definitive evidence.The data do suggest, however, that the burden of genetically influenced dyslipidemia in patients with AAA is considerable, and by extrapolation, these MR analyses lend support to the lipids playing an important role in AAA etiology and thus targeting lipids through pharmacological modification in patients with small AAAs may well be justified.This point is particularly pertinent given the recent reports of low prevalence of control of LDL-C concentrations in patients with AAA in both the United States and the United Kingdom. 25,26In addition, this group of patients should be considered in trials evaluating novel treatments of lipid-lowering medications, such as CETP or PCSK9 inhibitors.
The use of genetic data to inform drug trials and/or drug repurposing represents an important translational facet of data derived by large genome-wide consortia. 41,42In addition to the GRS for LDL-C, HDL-C, and TG, we looked at 4 loci that serve as proxies for cardiovascular drug targets that have not been subjected to clinical trials in patients with AAA.Both the LDL-C GRS and a genetic proxy for statin therapy (SNPs in HMGCR) were associated with AAA.Previous investigations on the associations of concentrations of LDL-C with AAA have used cross-sectional data sets with varying findings, and results have been hampered by concurrent LDL-C-lowering therapies. 43ndeed, there has been a suggestion that statin use may increase AAA risk. 44The collective results from this study suggest that LDL-C plays an important role in the etiology of AAA, which may explain the excess burden of CVD in patients with AAA. 24These data also support a view that patients found by screening to have AAA should be prescribed statins to reduce their CVD risk, although whether this will affect the progression of AAA cannot be answered in this study.
A recent phase 3 clinical trial showed that PCSK9 inhibitors have beneficial effects on CVD outcomes. 38Although the association we found between PCSK9 variants and AAA was weak, if PSCK9 inhibitors do prove to be a safe and costeffective means of lowering LDL-C levels, then consideration should be given to evaluating these drugs in patients with AAA.
As noted, a genetically instrumented higher HDL-C level was identified to be associated with a reduction in AAA risk.Variants in CETP have a range of results similar to pharmacological inhibition of CETP, 30 including lowering of LDL-C and raising of HDL-C levels.A trial of CETP inhibition showed modest benefit in patients following myocardial infarction, 45 and there are data to support its beneficial effects on vascular remodeling 46 that could have relevance in AAA management.Evaluation of CETP inhibition in patients with AAA may therefore be warranted.Although we cannot specifically determine whether the association between CETP polymorphisms and AAA is via HDL-C, LDL-C, or TG (or indeed all, as suggested by our GRS of lipid traits), we believe our results suggest that CETP inhibition could play a role in the management of AAA.
The findings regarding TG variants also have potential clinical implications for the development of novel treatments aimed at TG levels.They suggest that patients with AAA may benefit from lowering TG levels.As novel therapies such as APOC3 inhibitors progress from phase 2 studies to larger-scale phase 3 studies of CVD prevention, then patients with AAA could be an important CVD subphenotype in whom treatment should be evaluated.
Our study used MR, a genetic approach that has important assumptions.The SNPs used in the genetic instruments for each lipid trait were identified from recent GWASs that placed stringent thresholds on SNP discovery.As such, the genetic instruments are very unlikely to suffer from weak instrument bias; in any case, because the MR analyses used nonoverlapping data sets, such bias would tend to dilute the estimates derived from MR analyses. 47In addition, we made the assumption that the genetic instruments are not influenced by confounding and that they only associate with AAA through the exposure of interest (ie, the genetic instruments are not affected by unbalanced horizontal pleiotropy, as pictorially illustrated in Figure 1 of White et al 18 and expanded in Holmes et al 37 ).These assumptions cannot be tested with complete certainty.However, causal estimates obtained from a range of sensitivity analyses, each making different and weaker assumptions, all gave similar results.Nonetheless, residual pleiotropy could still influence our findings.

Limitations
The limitations of this study should be considered.First, we did not have data sets to evaluate AAA progression.Second, owing to limited availability of covariate data, we were unable to examine the influence of concurrent lipid-lowering therapy on the estimates derived from the GRS for blood lipid traits and AAA risk.Third, our analyses used summary-level data as described elsewhere. 16,48Use of summary-level data can hamper more refined analyses (eg, subgroup analyses by sex or age), but one of its main strengths is it facilitates 2-sample MR analyses of the type reported here.This greatly strengthens the power of the study, which enables the conduct of sensitivity analyses (such as MR-Egger and weighted median MR methods) and the investigation of certain instrumental variable assumptions such as the absence of genetic pleiotropy.Finally, although we attempted to control for pleiotropy in the analyses, we believe pleiotropy still represents a major challenge to deciphering the roles of specific lipid-based pathways.

Conclusions
Using contemporary MR approaches, we found data that lend support to the hypothesis that major lipid fractions are involved in the etiology of AAA.Consideration should be given to measures aimed at targeting lipids to reduce risk of AAA, using established and emerging therapies.

Figure
Figure 1.Association of Lipid Genetic Risk Scores With Abdominal Aortic Aneurysm (AAA) Risk

Figure 2 .Favors
Figure 2. Association of Single-Nucleotide Polymorphisms (SNPs) in Genes Encoding Drug Targets With Abdominal Aortic Aneurysm (AAA) Risk Favors LDL-C Lowering Favors LDL-C Raising

Table .
Summary of Abdominal Aortic Aneurysm Genome-Wide Association Studies Table.Up to 4914 cases and 48 002 controls were included in our analysis.The complete list of SNPs analyzed in this study, together with information on the association statistics for AAA, and for LDL-C, HDL-C, and TG levels, is included in eTable 1 in the Supplement.