Replication of newly identified genetic associations of abdominal aortic aneurysm with SMYD2, LINC00540, PCIF1/MMP9/ZNF335, and ERG

Abstract

Objective:

A recently published genome-wide association study of abdominal aortic aneurysm (AAA), based on pooled case-control data of European ancestry, identified four new loci for AAA: SMYD2 (top SNP rs1795061), LINC00540 (rs9316871), PCIF1/MMP9/ZNF335 (rs3827066), and ERG (rs2836411). Of the four, rs1795061 and rs2836411 showed significant heterogeneity across studies and the p value for rs9316871 did not reach the genome-wide significance threshold until discovery and replication data were pooled together in that study. The objective of this study was to replicate these newly identified genetic associations for AAA in a US-based prospective cohort study, the Atherosclerosis Risk in Communities (ARIC) Study, and a Greece-based case-control study.

Methods:

ARIC identified 408 clinically-diagnosed AAAs among 8,962 individuals of European ancestry during a median of 22 years of follow-up. The Greek case-control study included 341 AAAs of European ancestry recruited in a tertiary referral center and 292 geographically- and ethnically-matched controls recruited from the same institution. We used a Cox proportional hazards model to analyze the ARIC data and logistic regression to analyze the Greek data.

Results:

In ARIC, rs9316871 and rs3827066 were significantly associated with AAA risk (hazard ratio (p) was 0.77 (0.004) and 1.22 (0.03), respectively), rs2836411 was associated at borderline significance (1.13 (0.08)), while rs1795061 was not associated (p=0.55). In the Greek case-control study, rs1795061 and rs2836411 were significantly associated with AAA (odds ratio (p) was 1.66 (<0.001) and 1.29 (0.04), respectively) while rs9316871 was not (p=0.81). Genotyping of rs3827066 did not succeed. In the meta-analysis of the two studies, the association for rs9316871and rs2836411 was statistically significant and consistent between the two studies: p=0.02 and 0.007, respectively.

Conclusions:

We replicated the associations of rs9316871and rs2836411 with AAA risk in the meta-analysis of the two independent cohorts, providing further support for the importance of these loci in the etiology of AAA.

INTRODUCTION

Significant genetic influence has been demonstrated for AAA, with heritability of up to 77%.^{1, 2} Recently, Jones et al reported a large genome-wide association study (GWAS) of AAA that pooled data from several case-control studies of populations of European ancestry (EA).³ This study identified four new genes/loci: SMYD2 (top SNP rs1795061), LINC00540 (rs9316871), PCIF1/MMP9/ZNF335 (rs3827066), and ERG (rs2836411). Of the four, rs1795061 and rs2836411 showed significant heterogeneity across studies and the p value for rs9316871 did not reach the genome-wide significance threshold until discovery and replication data were pooled together. It is important to replicate these new findings in independent populations and in prospective data. We sought to replicate the newly identified associations of these SNPs with AAA in two independent studies that were not part of the study by Jones et al: a prospective cohort study of European Americans with 24 years of follow-up in the US, the Atherosclerosis Risk in Communities (ARIC) Study, and a case-control study using a white Caucasian population from Greece.

MATERIALS AND METHODS

Study populations and genotyping

The ARIC Study recruited a population-based cohort of 15,792 persons (73% EAs) aged 45 to 64 years in 1987–1989 and followed them through 2011 for clinically-diagnosed AAAs, identified by searching hospitalization and death records as well as Medicare data.⁴ Clinically-diagnosed AAAs were defined as ICD-9-CM codes of 441.3 (ruptured AAA) or 441.4 (AAA without mention of rupture), or procedure codes of 38.44 (AAA resection and replacement) or 39.71 (AAA endovascular repair), or the following cause of death codes: ICD-9 441.3 or 441.4 or ICD-10 code I71.3 (ruptured AAA) or I71.4 (AAA without mention of rupture).⁴ Details of AAA ascertainment and risk factor measurement in ARIC have been described elsewhere.⁴ We excluded participants who reported AAA surgery prior to baseline (n = 11 without information on ICD codes) and uncertain AAA status during follow-up (n = 30).

ARIC extracted genomic DNA from blood and conducted genotyping with the Affymetrix Genome-Wide Human SNP array 6.0. To expand the number of genetic markers beyond the genotype array for association analyses,⁵ ARIC conducted race-specific imputation of variant dosages to the 1000 Genomes Project Phase I version 3 reference panel. Before imputation, individuals were removed for being first-degree relatives, genetic outliers, or not matching existing genotype data. Detailed information on GWAS genotyping, quality control and imputation procedures has been provided elsewhere.⁶ Principal components (PCs) based on the GWAS data were generated by EIGENSTRAT⁷ to reflect the population structure or genetic ancestry of the ARIC participants. Of 11,447 European Americans who were at risk of AAA at baseline in 1987–1989, 8,962 EAs had the GWAS array genotyping and imputation data. Among them, 408 individuals were diagnosed with AAA from 1987–1989 through 2011, with a median of 22 years of follow-up for the cohort.

At the end of follow-up for clinically-diagnosed AAAs, ARIC conducted an ultrasound exam in 2011–2013 among its surviving participants (n=5,911) and identified an additional 75 asymptomatic AAAs. We did not include the ultrasound-detected AAAs in this study because doing so would require a separate analysis of these AAAs with adjustment for the potential selection bias caused by differential attrition of the cohort members who had the ultrasound, which is underpowered.

The Greek case-control study prospectively recruited 418 consecutive AAA patients between July 2009 and December 2012 in a tertiary referral vascular surgery center in Northern Greece and a group of 447 geographically matched controls recruited within the same time interval.⁸ An AAA was defined as maximal anterior-posterior diameter of the infra-renal aorta exceeding 3.0 cm. The AAA patients, who were referred to a vascular surgeon (on an outpatient basis) for consideration of surgical treatment, either had an incidental diagnosis of AAA whilst undergoing investigations for other conditions or were referred via a local informal screening program. While rupture was not an exclusion criterion, none of the AAA patients presented with a rupture (given logistical difficulties with consent). All controls were recruited from the same surgical unit and had cross-sectional imaging of their abdomen within 4 weeks of recruitment to exclude an AAA. Cases and controls were all Caucasians born in northern Greece. Patients with inflammatory aneurysms or connective tissue disorders related with aneurysmosis were excluded. A full medical and surgical history was recorded at baseline, including anthropometric assessments and a list of medications. Of the 418 AAAs and 447 controls, 341 AAAs and 292 controls had DNA samples available for genotyping for the current study.

In the Greek case-control study, DNA samples were extracted from peripheral venous blood and stored at 4°C until analysis. The candidate SNPs were identified using LightSNip (TIB MOLBIOL GmbH, Berlin, Germany) on the Roche LightCycler 480 system (Roche Diagnostics Corp, Indianapolis, Ind) based on real-time polymerase chain reaction and melting-curve analysis. The LightCycler 480 instrument was programmed according to the instructions for use for each LightSNip. DNA samples of both AAAs and controls were included in each plate. Clinical data were not available during genotyping. Genotyping for rs3827066 failed due to insufficient quality and quantity of DNA samples. Sixty-four duplicate samples were included in the genotyping, yielding an overall concordance rate of 98.6% between the duplicates. We excluded 3 AAAs from all analyses due to questionable allele calls or missing data for more than one of the SNPs. After further exclusion of missingness for each individual SNP, the final sample size was 333 AAAs and 292 controls for rs1795061, 336 and 292 for rs9316871, and 323 and 291 for rs2836411.

In both studies, all participants provided informed consent and the Institutional Review Board or Ethics Committee at each institute approved the study. None of the data on the associations between AAA and the four genetic markers in ARIC or the Greek study have been published anywhere before.

Statistical analysis

In ARIC, we used a Cox proportional hazards regression to test the association of SNP allele dosages with AAA, with time-to-event calculated from baseline to the time of first AAA event, date of death, loss to follow-up, or through December 31, 2011, whichever came first. Hazard ratios (HRs) and 95% confidence intervals (CIs) were computed with adjustment for age, sex, field center (i.e., study site), and the first five PCs for population stratification to reduce potential confounding by these characteristics.

In the Greek case-control study, logistic regression was performed to calculate odds ratios (ORs) and 95% CIs to assess the association between the allele dosage of each SNP and AAA status, with the adjustment for age and sex.

In both studies, the at-risk allele was chosen to be consistent with the Jones et al report.³

Meta-analysis:

To reduce the influence of variation of individual studies and increase statistical precision, we used a sample size-weighted meta-analysis to pool p-values and direction of genetic effects from the two studies for rs1795061, rs9316871, and rs2836411. The p-value-based approach was chosen because we used different study designs which yielded different effect estimates (HR vs OR) in the two studies. The number of AAAs in each study was used as the weight. An I² was estimated in a heterogeneity test to assess the difference of test statistics and the direction of association between the two studies. Furthermore, we pooled the data of our two studies with those from Jones et al to evaluate if adding the ARIC and Greek data strengthened the overall GWAS associations of Jones et al. The meta-analyses were conducted using the METAL package.⁹

Power analysis:

we estimated the power to replicate the associations for the three SNPs reported in Jones et al in the combined ARIC and Greek samples. At an alpha of 0.05, we had acceptable power to replicate the reported associations in the pooled ARIC and Greek samples: rs1795061 (MAF=0.34, OR=1.13, power=61%), rs9316871 (0.20, 0.86, 62%), and rs2836411 (0.37, 1.13, 60%).

RESULTS

Table 1 presents baseline characteristics of the two study populations.

Table 1.

Baseline characteristics in ARIC cohort study (1987–2011) and Greek case-cohort study: mean ± SD or percentage.

	ARIC Cohort Study			Greek Case-Control Study
Baseline characteristics	AAA (N=408)	Non-AAA (N=8,554)	P value	AAA (N=338†)	Controls (N=292‡)	P value
Age, years	56.6 ± 5.2	54.7 ± 5.7	<0.001	69±8	73±5	0.001
Male, %	72.2	45.3	<0.001	91.8	75.0	0.006
BMI, kg/m2	26.7 ± 4.0	27.0 ± 4.9	0.10	27.8 ± 3.3	27.4 ± 3.2	0.73
Smoking status			<0.001			0.43
Current smoker, %	50.5	23.2		19.6	21.6
Former smoker, %	37.0	35.5		56.6	51.1
Never smoker, %	12.5	41.3		23.8	27.3
Pack-years smoking*	35.0 ± 24.3	16.2 ± 21.4	<0.001	NA	NA	NA
Hypertension, %	36.5	26.3	<0.001	76.8	77.4	0.87
PAD, %	4.8	1.9	<0.001	21.1	17.5	0.27
Diabetes, %	5.9	8.7	0.05	22.3	23.3	0.78
Hypercholesterolemia, %	36.1	26.0	<0.001	58.2	53.1	0.20

^†N was 333 AAAs for rs1795061, 336 AAAs for rs9316871, and 323 AAAs for rs2836411;

^‡N was 292 controls for both rs1795061 and rs9316871, and 291 controls for rs2836411;

^*Including non-smokers.

PAD=peripheral arterial disease; NA = not assessed or available.

In ARIC, rs2836411 and rs9316871 were genotyped while rs1795061 and rs3827066 were imputed. Imputation quality, reflected by the ratio of observed to expected variance of the dosage statistic, was excellent for these two SNPs (0.98 and 0.99, respectively). In the whole study population, a test of genotype frequencies against Hardy-Weinberg equilibrium was marginally significant for rs2836411(p=0.05) and not significant for the other three SNPs. Minor allele frequencies (MAF) for the 4 SNPs agreed with those in Jones et al.³ After adjustment for age, sex, field center, and the first five PCs, rs9316871 and rs3827066 were significantly associated with AAA (HR (95% CI) =0.77 (0.65, 0.92) and 1.22 (1.02, 1.46), respectively, per one copy increase in the effect allele dose), and rs2836411 was close to the threshold for significance (HR (95% CI) =1.13 (0.99 1.31), p=0.08) (Model 1 in Table 2). Notably, the association for rs9316871 suggests a protective effect associated with the minor allele G. The direction of association for the 3 SNPs was consistent with that of Jones et al.³ Rs1795061 was not associated with AAA in ARIC (p=0.55) (Model 1 in Table 2).

Table 2.

Association of novel AAA SNPs with AAA in ARIC cohort study (1987–2011) and Greek case-control study.

				ARIC Cohort Study			Greek Case-control Study			Meta-analysis_1			Meta-analysis_2
SNP	Gene	A1	A2	MAF	HR (95% CI)*	P	MAF	OR (95% CI)§	P	Dir	P	I2	Dir	P	I2
Model 1
rs1795061	SMYD2	T†	C	0.31	0.96 (0.82, 1.12)	0.55	0.21	1.66 (1.27, 2.17)	1.96×10−4	−+	0.04	90‡	−++	1.1×10−11	80‡
rs9316871	LINC00540	G†	A	0.23	0.77 (0.65, 0.92)	0.004	0.25	0.97 (0.75, 1.25)	0.81	−−	0.02	68	−−−	3.8×10−11	42
rs3827066	PCIF1-ZNF335-MMP9	T†	C	0.16	1.22 (1.02, 1.46)	0.03	NA	NA	NA	NA	NA	NA	++	2.2×10−18	0
rs2836411	ERG	T†	C	0.34	1.13 (0.99, 1.31)	0.08	0.33	1.29 (1.01, 1.65)	0.04	++	0.007	0	+++	2.6×10−10	0
Model 2
rs1795061	SMYD2	T†	C	0.31	0.95 (0.81, 1.11)	0.52	0.21	1.71 (1.30, 2.23)	9.6×10−5	−+	0.03	91	−++	8.8×10−12	82‡
rs9316871	LINC00540	G†	A	0.23	0.75 (0.63, 0.90)	0.002	0.25	0.97 (0.75, 1.26)	0.83	−−	0.01	74	−−−	3.0×10−11	54
rs3827066	PCIF1-ZNF335-MMP9	T†	C	0.16	1.20 (1.01, 1.43)	0.04	NA	NA	NA	NA	NA	NA	++	2.8×10−18	0
rs2836411	ERG	T†	C	0.34	1.12 (0.97, 1.30)	0.11	0.33	1.27 (0.99, 1.62)	0.06	++	0.01	0.1	+++	3.9×10−10	0

A1=effective allele, A2=alternative allele; Meta-analysis_1: ARIC and the Greek case-control study (not for rs3827066); Meta-analysis_2: ARIC, the Greek case-control study (not for rs3827066), and Jones et al;³ MAF=minor allele frequency; Dir=direction; I²=heterogeneity test statistics; NA=not available or applicable;

^*Hazards ratios (HR) were modeled per 1 unit increment in the effect allele dose in Cox regression, with covariate adjustment in Model 1 (adjusted for age, sex, field center, and the first five PCs for population stratification) or Model 2 (Model 1 further adjusted for pack-years of smoking and total cholesterol);

^§Odds ratios (OR) were modeled per 1 unit increment in the effect allele dose in logistic regression, with covariate adjustment in Model 1 (adjusted for age and sex) or Model 2 (Model 1 further adjusted for smoking status and binary hypercholesterolemia);

^†Minor allele;

^‡p<0.05, other I² without this label had p>0.05.

In the Greek case-control study, genotype frequencies for rs1795061, rs9316871, and rs2836411 were in accordance with Hardy-Weinberg equilibrium (p>0.05) in the controls. MAF for rs1795061 was lower than that in ARIC and Jones et al, while MAF for the other two SNPs was similar to that of ARIC and Jones et al. rs1795061 and rs2836411 were significantly associated with the risk of AAA (OR (95% CI) = 1.66 (1.27 2.17) and 1.29 (1.01 1.65), respectively) (Model 1 in Table 2), rs9316871 was not associated with AAA (p=0.81).

Associations of rs1795061, rs9316871, and rs2836411 with AAA were statistically significant in the meta-analysis of the two studies (Model 1 in Table 2). Heterogeneity (I²>0) was observed for rs1795061 and rs9316871, with that for rs1795061 being statistically significant (p<0.05). The direction of association for rs1795061 was not consistent between the two studies. Pooling the data of our two studies with those from Jones et al in a second meta-analysis strengthened the statistical significance of the GWAS associations for the four SNPs (Model 1 in Table 2).

In a secondary analysis, we further adjusted for two important risk factors for AAA, smoking (pack-years of smoking in ARIC and smoking status in the Greek study) and cholesterol (total cholesterol in ARIC and binary hypercholesterolemia in the Greek study). The additional adjustments did not result in appreciable changes in either individual study or the meta-analyses (Model 2 in Table 2). As a note, 12 AAAs and 99 non-AAAs in ARIC were excluded from this analysis due to missing information on the smoking variable.

DISCUSSION

In this study we replicated the previously reported new associations of AAA risk with rs9316871 and rs3827066 in the US-based prospective study, and with rs1795061 and rs2836411 in the Greek case-control study. The direction of association with rs9316871 and rs2836411 was consistent between the two studies and the associations were significant in the meta-analysis combining the two studies. The ARIC data showed the association of rs9316871, rs3827066, and rs2836411 with longitudinal AAA risk based on 24 years of followed-up. To the best of our knowledge, there are few data in the literature that reported longitudinal AAA risk associated with genetic variants in a community-based study. While the effect size for the replicated variants was modest, their public health implication is not trivial because their at-risk alleles are common in the general population.

By extensive look-up in expression quantitative trait locus and RNA sequencing data, Jones et al identified functional relevance for three of the variants: rs3827066 with PLTP expression, rs2836411 with ERG expression, and rs9316871 with FGF9 expression.³ They also searched for these putative AAA risk variants in published GWAS data of other cardiometabolic phenotypes but did not find associations of the new variants with coronary artery disease (CAD), hypertension, lipid traits, or diabetes. However, a newly published GWAS for CAD identified rs3827066 in PCIF1-ZNF335-MMP9 as a new locus for CAD (OR=1.04).¹⁰ This new GWAS finding further highlights the overlap between genetic risk factors for AAA and other forms of cardiovascular disease.

In Jones et al, the p for rs9316871 in LINC00540 reached the genome-wide significance threshold only when the discovery and replication data were pooled together. Therefore, the replication of this locus in ARIC provides further corroboration that it is related to the risk of AAA. The association at this locus showed modest heterogeneity between ARIC and the Greek study, with consistent direction. Unlike the analysis in ARIC, the Greek study did not control for possible population stratification using ancestry-informative markers or principal components. However, the threat of confounding by population stratification is likely small, although cannot be completely ruled out, because both cases and controls were Caucasians who were recruited from and born in a small area of northern Greece.

The association between rs1795061 and AAA risk was replicated in the Greek study but not in ARIC (p=0.55). In fact, the direction of association in ARIC was opposite to that in the Greek study and the GWAS by Jones et al. Imputation error was an unlikely explanation because rs1795061 was imputed with high confidence in ARIC (imputation quality score=0.98). The MAF was similar in ARIC and Jones et al (0.31 vs 0.34) but lower in the Greek study (0.21 among the controls). The association of this SNP with AAA also showed high heterogeneity across studies in Jones et al. Since rs1795061 is located at an intergenic region near SMYD2 and GWAS is an indirect mapping strategy that relies on variants that are in linkage disequilibrium (LD) with the true causal variants to capture the genetic associations,^{11, 12} it is possible that the rs1795061-AAA association reflects the effect of an underlying, causal variant that was not directly measured by the GWAS arrays. If so, the extent of LD between rs1795061 and the underlying causal variant may differ across populations. Another interpretation is that different genetic backgrounds and environmental exposures may have modified the effect of this locus in different populations.

In summary, based on the data from two independent populations of European ancestry, our study consistently replicated the newly reported associations for AAA at LINC00540 and ERG, further supporting the importance of these loci in the etiology of AAA. Improved understanding of AAA etiology might benefit the management of AAA and suggest new primary prevention or therapeutic strategies. Future studies in diverse populations are needed to further validate these genetic loci as risk factors for AAA as well as to identify the underlying causal variants and mechanisms for validated associations. Furthermore, with the increasing number of genetic loci and variants that might be identified for AAA in future studies, it may be worthy to evaluate the efficiency of genetic testing in relatives of AAA patients for the identification of individuals at increased risk of developing AAA. Notably, the potential value of using genomic information in cardiovascular risk prediction has recently been demonstrated in the field, for example, on CAD.¹³ Future studies with comprehensive assessment of genomic variants in large populations will be needed for the evaluation of clinical application of AAA genomic findings.

What does this study/review add to the existing literature and how will it influence future clinical practice.

A recently published genome-wide association study of abdominal aortic aneurysm (AAA) identified four new loci for AAA: SMYD2 (top SNP rs1795061), LINC00540 (rs9316871), PCIF1/MMP9/ZNF335 (rs3827066), and ERG (rs2836411). Based on data from a US-based prospective cohort study and a Greece-based case-control study, we replicated the associations of rs9316871and rs2836411 with AAA risk in the meta-analysis of the two independent cohorts, providing further support for the importance of these loci in the etiology of AAA. Findings from the study contributed to the improved understanding of the pathophysiology of AAA and may aid in the management of patients with this condition.