Genetic loci associated with circulating levels of very long-chain saturated fatty acids[S]

Very long-chain saturated fatty acids (VLSFAs) are saturated fatty acids with 20 or more carbons. In contrast to the more abundant saturated fatty acids, such as palmitic acid, there is growing evidence that circulating VLSFAs may have beneficial biological properties. Whether genetic factors influence circulating levels of VLSFAs is not known. We investigated the association of common genetic variation with plasma phospholipid/erythrocyte levels of three VLSFAs by performing genome-wide association studies in seven population-based cohorts comprising 10,129 subjects of European ancestry. We observed associations of circulating VLSFA concentrations with common variants in two genes, serine palmitoyl-transferase long-chain base subunit 3 (SPTLC3), a gene involved in the rate-limiting step of de novo sphingolipid synthesis, and ceramide synthase 4 (CERS4). The SPTLC3 variant at rs680379 was associated with higher arachidic acid (20:0 , P = 5.81 × 10−13). The CERS4 variant at rs2100944 was associated with higher levels of 20:0 (P = 2.65 × 10−40) and in analyses that adjusted for 20:0, with lower levels of behenic acid (P = 4.22 × 10−26) and lignoceric acid (P = 3.20 × 10−21). These novel associations suggest an inter-relationship of circulating VLSFAs and sphingolipid synthesis.

as markers of peroxisomal disorders ( 1 ), VLSFAs also play important roles in normal physiology ( 2 ). VLSFAs are components of sphingolipids ( 3 ), such as sphingomyelin and ceramides, and impart specifi c biological activities to the sphingolipids ( 4 ). For example, sphingolipids with VLSFAs promote the formation of lipid microdomains in plasma membranes ( 5 ). In addition, ceramides with VLS-FAs show biological activities opposite to ceramides with palmitic acid (16:0), a long-chain saturated fatty acid ( 6 ).
Recent studies in human populations suggest benefi cial effects of circulating levels of VLSFAs (7)(8)(9). We previously reported that higher levels of VLSFAs in plasma phospholipids were associated with a lower risk of incident atrial fi brillation in the Cardiovascular Health Study (CHS) ( 7 ). More recently, we showed that higher levels of erythrocyte VLSFAs were associated with lower risk of incident sudden cardiac arrest in a population-based case-control study, where VLSFAs were measured in blood samples collected at the time of cardiac arrest ( 8 ). In addition, a recent report from the European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct case-cohort study showed an association of higher levels of plasma phospholipid VLSFAs with a lower risk of incident type 2 diabetes ( 9 ). In contrast, higher levels of circulating 16:0 are associated with higher risks of atrial fi brillation ( 7 ), sudden cardiac arrest ( 8 ), and diabetes ( 9 ). These studies suggest a need to differentiate between saturated fatty acids of different lengths and a need to study the determinants of circulating VLSFAs.
We have previously found evidence of high heritability of erythrocyte levels of 24:0 in the Kibbutzim Family Study ( 18 ). To identify common genetic variants that might infl uence levels of circulating 20:0, 22:0, and 24:0, we conducted a meta-analysis of genome-wide association studies (GWASs) of plasma phospholipid/erythrocyte VLSFAs in 10,129 participants of European ancestry in seven cohort studies, as part of the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium.
The SPTLC3 variant at rs680379 was associated with higher arachidic acid (20:0 , P = 5.81 × 10 ؊ 13 ). The CERS4 variant at rs2100944 was associated with higher levels of 20:0 ( P = 2.65 × 10   SNPs with minor allele frequency (MAF) р 1% or effective degree of freedom (2 × MAF × sample size × observed divided by expected variance for imputed allele dosage ) р 50 were excluded from the meta-analyses. Additional details on genotyping and imputation per cohort are provided in the supplementary text. Association analysis between genotype and each fatty acid was performed separately within each study cohort according to a prespecifi ed analysis plan. All studies conducted linear regression analysis measuring the additive effect of the number of effect alleles, or equivalently the imputed number of effect alleles for imputed genotypes. In absence of a known model, we chose the additive model a priori because it has good power for all "monotone" models, including recessive and dominant ( 23 ). The analyses used robust standard errors and were adjusted for age, sex, site of recruitment where appropriate, and where needed, principal components to account for possible population genetic substructure.

Ethics statement
Each cohort that participated in this study secured approval from their respective institutional review boards, and all participants provided written informed consent in accordance with the Declaration of Helsinki.

Study samples
We conducted a meta-analysis of data from seven cohorts comprising 10,129 individuals of European ancestry (

Meta-analysis
Because circulating levels of the fatty acids of interest differed across the cohorts ( Table 1 ) and because combining effect sizes requires that the trait is measured on exactly the same scale in each study, we performed a z score-based meta-analysis of each fatty acid, as implemented in METAL (Developed by Goncalo Abecasis at the University of Michigan). Genomic control correction was applied to each study prior to the meta-analysis and correction factors ranged from 0.96 to 1.03 (20:0), 1.00 to 1.03 (22:0), and 0.99 to 1.03 (24:0). P values less than 5 × 10 Ϫ 8 were considered statistically signifi cant.
Our approach, to include all the studies in a meta-analysis, has more power than splitting the studies into a discovery sample and a replication sample ( 24 ).
In sensitivity analyses, we performed fi xed-effects meta-analyses in which studies were stratifi ed according to whether fatty acids were measured in plasma phospholipids or erythrocytes. The fi xed effect meta-analyses were conducted for the associations of three SNPs, discovered with the z score-based meta-analysis, and using standard deviation units for the fatty acids to minimize study to study differences in the measurement of the VLSFAs. We tested for a difference between plasma phospholipids and erythrocytes using fi xed effect meta-analyses regression ( 25 ). Table 1 presents demographic characteristics and mean levels of circulating 20:0, 22:0, and 24:0 for the study sample of up to 10,129 subjects of European ancestry in the ARIC, CARDIA, CHS, HPFS, MESA, NHS, and WGHS cohorts. The cohorts differed by age (mean age ranged from 45.6 to 75.0 years) and gender (percent of men ranged from 0 to 100%). Levels of VLSFAs varied with the fatty acid compartment measured (erythrocyte vs. plasma phospholipid) and by cohort. The higher levels of VLSFAs in the CHS cohort, compared with other cohorts using plasma phospholipids, may be due to a higher recovery of these hydrophobic fatty acids with the method used in the CHS for the extraction of fatty acid methyl esters (details in the supplementary text).

RESULTS
The meta-analysis of GWAS results revealed two genetic loci associated with circulating levels of 20:0 at genomewide signifi cance ( Table 2 Table 2 ). The directions of associations of the variant alleles were consistent in all cohorts where results for these variants were available (supplementary Table 1 ).
In sensitivity analyses, we repeated the meta-analysis of the associations of rs2100944, rs11666913, and rs680379 with VLSFAs to compare results in plasma phospholipids and erythrocytes. We saw evidence for smaller effect sizes in erythrocytes than in plasma phospholipids in the associations of rs2100944 and rs11666913 with each VLSFA, but not in the association of rs680379 with 20:0 (supplementary Tables 5, 6).

DISCUSSION
In this large meta-analysis, we report for the fi rst time that variation in two genes involved in sphingolipid biosynthesis is signifi cantly associated with circulating levels of VLSFAs. Specifi cally, levels of the fatty acid 20:0 were associated with common variation in SPTLC3 and in CERS4 . After adjustment for levels of 20:0, levels of 22:0 and 24:0 were also associated with variation in CERS4 , but in a direction opposite to that of 20:0 ( Fig. 2 ).
SPTLC3 codes for the enzyme of the fi rst step of sphingolipid de novo biosynthesis ( 27 ) ( Fig. 3 ). The enzyme condenses 16:0 or myristic acid with the amino acid serine resulting in the "sphingoid" backbone of all sphingolipids. This step is a rate-limiting step that infl uences the fl ow through the de novo synthesis pathway. VLSFAs are used at a downstream step to form ceramides ( Fig. 3 ), and ceramides can be metabolized into sphingomyelin, a phospholipid. Interestingly, the SNP in SPTLC3 that was most strongly associated with circulating levels of 20:0, rs680379, was also associated with circulating levels of ceramides containing 22:0 and ceramides containing 24:0 in the European Special Population Research Network (EUROSPAN) ( 28 ). These observations suggest that plasma phospholipid/ erythrocyte levels of 20:0 may refl ect sphingomyelin produced by de novo synthesis.
CERS4 encodes for a ceramide synthase, an enzyme that introduces a fatty acid, such as 20:0, into the sphingoid backbone (generated by sptlc1, 2, or 3), thereby producing dihydroceramide, a precursor of ceramides and sphingomyelins ( Fig. 3 ). Ceramide synthases also produce ceramide from sphingosine that is recycled from the breakdown of sphingomyelin and other sphingolipids, although the relative contribution of the de novo synthesis and recycling pathways to ceramide production is not well-understood ( 29 ). Humans have six genes encoding ceramide synthases with different specifi city for saturated fatty acids of different lengths ( 30 ). CERS4 uses preferentially 20:0 or 18:0 to produce ceramide containing 20:0 or ceramide with 18:0. Several of genetic loci most closely associated with circulating 20:0 have also been found associated with circulating ceramides containing VLSFAs in the GWAS of plasma sphingolipids in the EUROSPAN ( 28 ). In particular, rs11666866 (a CERS4 SNP also in our top hits, with r 2 of 0.97 with rs11666913) was associated with higher levels of plasma ceramide containing 20:0, mirroring the association of rs11666866 and rs11666913 with higher levels of circulating 20:0 that we observed. Likewise, rs1466448 (another CERS4 SNP in our top hits, with an r 2 of 0.69 with rs2100944) was associated with lower levels of ceramide with 20:0 in EUROSPAN, again similar to the association of rs1466448 and rs2100944 with lower levels of circulating 20:0 in our study. The associations of CERS4 variants, in the same direction, with both circulating levels of ceramides containing 20:0 and overall plasma phospholipid/ erythrocyte levels of 20:0, which include sphingomyelin in addition to phosphoglycerolipids, suggest that incorporation into sphingolipids is an important metabolic fate of 20:0. This observation is generally consistent with the known fatty acid composition of sphingolipids ( 31,32 ).
We observed opposite association of CERS4 variants with 20:0 versus 22:0 and 24:0 ( Fig. 2 ), with a set of variants associated with higher 20:0 and lower 22:0 and 24:0, and another set of variants associated with lower 20:0 and higher 22:0 and 24:0. An intriguing possibility is that genetic variation in CERS4 might alter the specifi city of the enzyme, with some variants favoring 20:0 over 22:0 or 24:0 and other variants favoring 22:0 or 24:0 over 20:0. Our results highlight the need to identify the functional allele(s) in these loci that may lead to alterations in VLSFA levels.
We recently reported genome-wide associations with the saturated fatty acids 16:0 and 18:0 ( 33 ). None of the reported fi ndings for 16:0 and 18:0 overlapped with the genome-wide association fi ndings with VLSFAs in this study. In particular, 18:0 was associated with variation in LPGAT1 , an enzyme involved in phosphatidylglycerol synthesis, whereas VLSFAs were associated with genes involved in sphingolipid synthesis in the current study. These observations further highlight likely differences in the metabolism of saturated fatty acids of different lengths, and the need for more targeted research of individual saturated fatty acids to understand their roles in health and disease.
Little is known of the biological activities of VLSFAs, yet emerging evidence suggests possible benefi cial properties. We previously reported an inverse association of plasma phospholipid VLSFAs with risk of incident atrial fi brillation ( 7 ), and more recently we found an inverse association of erythrocyte VLSFAs with risk of sudden cardiac arrest ( 8 ). In addition, Forouhi et al. ( 9 ) reported an inverse association of plasma phospholipid VLSFAs with risk of incident type 2 diabetes. Interestingly, the VLSFAs that are incorporated into ceramides also impart protective activities ( 6 ). Ceramides are best known for promoting apoptosis ( 34,35 ); however, the saturated fatty acid that is N-acylated to the sphingoid backbone appears to modify ceramide biological activities ( 6 ). For example, in the worm Caenorhabditis elegans , ceramide with 16:0 promotes apoptosis ( 36 ), but  The fi gure shows the pathway location of the two genes associated with levels of VLSFAs. SPTLC3 produces 3-ketosphinganine (sphingoid backbone) in the fi rst rate-limiting step of the pathway. CERS4 introduces a VLSFA into the backbone to produce dihydroceramide, the precursor of ceramide and sphingomyelin. ceramides with 20:0 or 22:0 are protective against hypoxiainduced apoptosis ( 37 ). Similarly, in a mouse model of lipotoxic cardiomyopathy, treatment with myriocin lowers total ceramides and enhances cardiomyocyte survival ( 34 ); however, in genetically engineered mice with lower levels of ceramides with 18:0, 20:0, and 24:0, the heart shows more (not less) apoptosis ( 38 ). Involvement of VLSFAs in sphingolipid metabolism, highlighted in the present study fi ndings, suggests sphingolipids may mediate the reported VLSFA associations. Drug therapies exist that infl uence global levels of sphingolipids [e.g., Fenretinide ( 39 )]. If sphingolipids with specifi c fatty acids are shown to be protective of arrhythmias and diabetes, it may spur the development of more selective drug therapies, and promote drug and dietary trials to alter sphingolipid composition and disease risk.
Our study has several strengths that include the use of samples from population-based cohort studies to enhance generalizability of fi ndings; the measurement of VLSFAs in plasma or erythrocyte phospholipids, objective biomarkers of tissue membrane phospholipids; and the large sample size that allowed detection of genome-wide significant associations. Additionally, we took advantage of correlations between the VLSFAs to increase the power of detecting associations with 22:0 and 24:0.
Potential limitations should also be considered. The top SNPs may tag less common SNPs that are responsible for the observed associations, and resequencing of the loci may be needed to identify potentially causal variants. The VLS-FAs were measured in plasma phospholipids in some cohorts and erythrocytes in the others, and the levels varied with the methodology that was used. Although we addressed this limitation by using z scores in our meta-analysis, residual heterogeneity remained possible. It is reassuring to see that the directions of the associations for the top SNPs were consistent in all the cohorts. The study only included participants of European ancestry and further studies are needed to confi rm the associations in other ethnic groups.
In conclusion, we identifi ed novel associations of common variants in two sphingolipid genes with circulating levels of VLSFAs. These novel associations suggest an inter-relationship of circulating VLSFAs and sphingolipid endogenous synthesis.
The authors thank the staff, participants, and other investigators of the contributing studies. A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org. A full list of participating MESA investigators and institutions can be found online at http://www.mesa-nhlbi.org . Summary data from the CHARGE Fatty Acids Working Group genomewide meta analyses can be found online at http://www. chargeconsortium.com/main/results