Association of FADS1/2 Locus Variants and Polyunsaturated Fatty Acids With Aortic Stenosis

This genome-wide association study identifies genetic loci and pathways associated with aortic stenosis.

© 2020 Chen HY et al. JAMA Cardiol. The effect size for FADS1 gene expression is terms of the normalized effect size while the effect 3 size for aortic valve calcium is expressed as the natural logarithm of the odds ratio. 4 The cohort is ethnically diverse and its racial composition has been previously described 1 .

Stenosis and Their Associations With Aortic Stenosis in the GERA
Genotyping of DNA extracted from saliva was performed on customized, ethnicity-specific Affymetrix Axiom arrays 2 .
In the present study, we restricted our analyses to unrelated individuals of self-reported infarction or coronary artery disease (ICD-9 410-414) or a procedure code for coronary artery bypass surgery or percutaneous coronary intervention in the electronic health records, or selfreported revascularization, angina, or myocardial infarction. Hypertension, smoking, and diabetes were self-reported in questionnaire data and the body mass index was calculated from self-reported height and weight. Ages greater than 90 years had been rounded down to 90, to enhance the privacy of these participants (n=389).
Our genome-wide association study was performed among 44,703 unrelated participants

Replication Cohorts
All participants provided written, informed consent and each study was approved by the relevant institutional review boards.

Malmö Diet and Cancer Study
The Malmö Diet and Cancer Study is a prospective, population-based cohort of 30,447 individuals living in Malmö, Sweden. This cohort has been described previously 9 , but briefly, blood samples from a nested random sub-cohort and sets of cases for a range of incident diseases collectively comprising more than half the cohort were genotyped on the Illumina Human Omni

Vanderbilt DNA Biobank (BioVU)
The Vanderbilt DNA Biobank, or BioVU, is a biorepository featuring de-identified linkage of electronic health records and genetic data for patients undergoing treatment at the mild-moderate AS were not included as cases and were also excluded from the control group.
Controls were defined as subjects older than 55 years with at least one TTE, but who had an aortic valve peak velocity < 1.5 m/s and no degree of AS. Among 8,314 unrelated participants of European ancestry (n=759 AS cases), we modelled the association of variants with AS in logistic regression models adjusted for age, age 2 , and sex using SNPTEST version 2.5.4-beta3 7 .

UK Biobank
The UK Biobank is a longitudinal cohort of more than 500,000 predominantly white recruited throughout the University of Pennsylvania Health System who have consented to genotyping/sequencing and access to electronic health record phenotype data. This cohort has been described previously 15 . Cases of CAVD and disease-free controls were ascertained through electronic health record text mining, genotyping was performed using the Illumina Quad Omni Genotyping Chip, and imputation was performed using the Michigan Imputation Server 4 using the Haplotype Reference Consortium version r1.1 5 as the reference panel. AS cases were defined as the presence of a diagnosis code for AS (ICD-9 424.1 or ICD-10 I35.0) or a procedure code for AVR in the EHR, or a validated procedure of text mining echocardiography reports 16 . The remaining individuals were designated AS controls. Participants with congenital heart disease (ICD-9 746-747 or ICD-10 Q20-Q22) were excluded. We restricted our analysis to 6,143 European-ancestry participants (n=1,593 AS cases) and modeled the association of variants with AS in PLINK 2.0 alpha 10 using logistic regression models adjusted for age, age 2 , and sex.

European Prospective Investigation of Cancer and Nutrition -Norfolk
The European Prospective Investigation of Cancer and Nutrition -Norfolk is a prospective, population-based cohort of 25,639 individuals living in Norfolk, United Kingdom and the construction of this cohort has been described in detail elsewhere 17 . AS cases were defined as individuals coded with AS (ICD-10 I35) or who died with AS as an underlying cause according to their death certificate. All other participants were defined as controls. Genotyping of blood samples was performed using the Affymetrix UK Biobank Axiom Array platform and following standard quality control, imputation was performed using IMPUTE software 18

Statistical Analysis
Among 28,041 participants with complete diet data from the MDCS, we examined whether dietary linoleic acid or α-linolenic acid intake was associated with incident AS in a Cox proportional-hazards model adjusted for age, sex, energy intake, season, and diet method.
Linoleic acid intake was measured by a modified diet history methodology combining a 168item dietary questionnaire, a 7-day menu book, and a 1-h diet history interview specifically designed for the MDCS, as previously described 30

Risk Score Construction
The arachidonic acid GRS was constructed from independent variants (r 2 ≤0.001) associated with plasma arachidonic acid (percentage of total fatty acids) at a genome-wide level of significance (p≤5×10 -8 ) in a previous publication 31 . If a variant was not available, a proxy in high linkage disequilibrium (r 2 ≥0.95) was used if available.
The estimates used for aortic stenosis were obtained through meta-analysis of estimates from seven cohorts: GERA (imputed using the Michigan Imputation Server 4 with the Haplotype Reference Consortium version r1.1 5 as the reference panel), QUEBEC-CAVS, MDCS, BioVU, UKB, PMBB, and Umeå. To account for differences in minor allele frequencies, the GRS was limited to 32 variants with a minor allele frequency≥0.01 in both the AA publication and the GERA cohort. Meta-analysis was performed using PLINK version 1.9 10 .
The aortic valve calcium estimates for each variant in the GRS were obtained from a prior genome-wide meta-analysis 3 . The Mendelian randomization analysis for aortic valve calcium was performed using 24 variants with available association results and a minor allele frequency≥0.01 in both the arachidonic and AVC studies.

Risk Score Construction and Analysis
We performed Mendelian randomization to assess whether elevated FADS1 or FADS2 expression was causally associated with AS and AVC. Estimates for significant FADS1 expression quantitative trait loci (eQTL) in the liver were extracted from the Gene-Tissue Expression (GTEx) Project, release V8 (dbGaP accession phs000424.v8.p2) 32 . No significant FADS2 liver eQTL were available. Variants were included in the genetic risk score if they were non-ambiguous, bi-allelic, not in very high linkage disequilibrium with another variant in the risk score (r 2 <0.9) in the GERA cohort, and an effect estimate was also available for the outcome (5 variants for FADS1 with AS [eTable 11], 6 variants for FADS1 with AVC [eTable 12]). The association of these variants with AS were estimated in the GERA cohort using logistic regression models adjusted for age, age 2 , and sex, while their associations with AVC were extracted from a prior meta-analysis 3 . All variants were well imputed in the GERA cohort (info score≥0.3). Imputation quality metrics were not available for the AVC meta-analysis.
To estimate the causal association of FADS1 expression with the odds of AS and AVC, we applied the inverse variance-weighted meta-analysis method of Mendelian randomization and accounted for correlation between variants in the risk score using r values computed in the GERA cohort using PLINK 2.0 alpha 10 . This approach to account for a correlated genetic risk score has been described previously 33 . As sensitivity analyses, we performed inverse-variance weighted, penalized weighted median, and Egger extension methods if the primary analysis was significant (eTable 13).