Genetic association of long-chain acyl-CoA synthetase 1 variants with fasting glucose, diabetes, and subclinical atherosclerosis[S]

Long-chain acyl-CoA synthetase 1 (ACSL1) converts free fatty acids into acyl-CoAs. Mouse studies have revealed that ACSL1 channels acyl-CoAs to β-oxidation, thereby reducing glucose utilization, and is required for diabetes-accelerated atherosclerosis. The role of ACSL1 in humans is unknown. We therefore examined common variants in the human ACSL1 locus by genetic association studies for fasting glucose, diabetes status, and preclinical atherosclerosis by using the MAGIC and DIAGRAM consortia; followed by analyses in participants from the Multi-Ethnic Study of Atherosclerosis, the Penn-T2D consortium, and a meta-analysis of subclinical atherosclerosis in African Americans; and finally, expression quantitative trait locus analysis and identification of DNase I hypersensitive sites (DHS). The results show that three SNPs in ACSL1 (rs7681334, rs735949, and rs4862423) are associated with fasting glucose or diabetes status in these large (>200,000 subjects) data sets. Furthermore, rs4862423 is associated with subclinical atherosclerosis and coincides with a DHS highly accessible in human heart. SNP rs735949 is in strong linkage disequilibrium with rs745805, significantly associated with ACSL1 levels in skin, suggesting tissue-specific regulatory mechanisms. This study provides evidence in humans of ACSL1 SNPs associated with fasting glucose, diabetes, and subclinical atherosclerosis and suggests links among these traits and acyl-CoA synthesis.

replenishment, neutral lipid accumulation, and early atherosclerosis in the context of diabetes ( 6,7,11 ). Mice defi cient in ACSL1 selectively in myeloid cells or endothelial cells do not exhibit reduced blood glucose levels ( 6,12 ).
Whereas mouse studies indicate that ACSL1 might play a critical role in maintaining glucose homeostasis and atherosclerosis, the role of ACSL1 in humans is unknown. Because of the studies discussed above, we were particularly interested in the relationship between human ACSL1 SNPs with variation in fasting glucose, diabetes, and risk of atherosclerosis. We interrogated published [Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC) and Diabetes Genetics Replication and Meta-Analysis (DIAGRAM) consortium] genome-wide association study (GWAS) scans of fasting glucose ( 13,14 ) and diabetes status ( 15 ) with respect to common [minor allele frequency (MAF) >5%] SNPs in the ACSL1 locus (4q35.1). MAGIC contains >46,000 healthy nondiabetic individuals of European descent, while DIAGRAM has >56,000 individuals of primarily European descent (>12,000 with type 2 diabetes). For the common ACSL1 SNPs identifi ed through the MAGIC and DIAGRAM consortia, we conducted follow-up analyses of participants with and without diabetes from the Multi-Ethnic Study of Atherosclerosis (MESA) and in >200,000 subjects with and without type 2 diabetes included in the Penn-T2D consortium. The analyses in MESA allowed us to expand discoveries from the published GWAS effort to examine evidence of association a ) across race/ethnic groups and b ) with subclinical atherosclerosis traits, and the analysis in the Penn-T2D consortium allowed us to expand our analysis to a much larger sample size. The association between ACSL1 SNPs and coronary artery calcifi cation (CAC) was validated in African Americans ( 16 ). Furthermore, expression quantitative trait locus (eQTL) analysis was used to investigate if the identifi ed ACSL1 SNPs were associated with ACSL1 levels in skin, lymphocytes, and adipose tissue, and fi nally, DNase I hypersensitive sites (DHSs) coinciding with these SNPs were analyzed in several fetal and adult human tissues.
Together, our results demonstrate that three SNPs in the ACSL1 gene are associated with markers of fasting glucose or diabetes status and that one of these SNPs is also signifi cantly associated with subclinical atherosclerosis. This represents the fi rst indication of ACSL1 contributing to the regulation of fasting glucose and risk of diabetes and subclinical atherosclerosis in humans.
Penn-T2D consortium. The Penn-T2D consortium used publicly available DIAGRAM meta-analyses data, updated by adding additional T2D GWAS/Metabochip studies that have not previously contributed data to the DIAGRAM consortium (fi ndings from these analyses are currently submitted for publication). The Penn-T2D consortium consists of data from 56,241 type 2 diabetes cases and 187,815 controls.

Replication and validation in MESA and Penn-T2D for SNPs identifi ed in consortia studies
For each of the three ACSL1 SNPs identifi ed in MAGIC and DIAGRAM associated with fasting glucose and diabetes ( Table  1 ), analyses were focused on replicating and validating association with the corresponding trait analyzed in discovery of each SNP in MESA and the Penn-T2D consortium. Statistical significance for replication of the three SNPs was determined at a Bonferroni-corrected threshold of ␣ = 0.05/3 = 0.017, whereas ␣ = 0.05 was considered as nominal evidence of replication. For secondary analysis of subclinical atherosclerosis traits in MESA, we report fi ndings as signifi cant if they reach a Bonferroni-corrected threshold of ␣ = 0.05/12 = 0.0042, corresponding to three SNPs and four traits, or 12 tests.

Meta-analysis of CAC in African Americans
This study amassed data on 5,823 African Americans from eight US studies in CAC, examining various defi nitions of CAC: ln[CAC+1]; CAC present/absent; ln(CAC) where CAC >0; and BLOM transformed CAC, defi ned by Gomez et al. ( 28 ). Results from the eight studies were meta-analyzed in METAL ( 16 ). For this validation study, we used the results from subjects with CAC >0 on 2,520 subjects. This cohort included African Americans from MESA (29% of subjects). Statistical signifi cance for replication of the three SNPs was determined at a threshold of ␣ = 0.05/ 3 = 0.017 after Bonferroni correction.

CARDIoGRAMplusC4D consortium
In order to investigate association of the three identifi ed ACSL1 SNPs with cardiovascular events (myocardial infarction, symptomatic coronary events, or coronary stenosis), we took advantage of the CARDIoGRAMplusC4D consortium. CARDIoGRAM GWAS is a meta-analysis of 22 GWASs in participants of European descent involving 22,233 cases and 64,762 controls, and C4D GWAS is a meta-analysis of GWASs in participants of European and South Asian descent involving 15,420 coronary heart disease cases and 15,062 controls. GWAS data from both of these consortia contain nonoverlapping participants ( 29 ). We used this publicly available data and conducted a fi xed-effects meta-analysis yielding a sample size of 37,653 coronary heart disease cases and 80,182 controls.

Multiple Tissue Human Expression Resource analysis
The Multiple Tissue Human Expression Resource (MuTHER) consortium data (http://www.muther.ac.uk) were used to investigate if the three identifi ed ACSL1 SNPs were associated with ACSL1 levels, representing an eQTL analysis. MuTHER contains genome-wide expression (Illumina HT-12v3 array) profi les on lymphocytes, subcutaneous fat, and skin biopsies from ‫ف‬ 856 twins from the TwinsUK BioResource. MuTHER was interrogated for changes in probe ILMN_1684585 ( ACSL1 ). descent in 2010 ( 13 ). A subsequent effort from MAGIC performed a GWAS of the same traits with adjustment for BMI in an expanded sample of up to 51,750 nondiabetic participants of European descent ( 14 ).
DIAGRAM. The DIAGRAM consortium performed a GWAS of type 2 diabetes in 12,171 cases and 56,862 controls comprising the DIAGRAM v3 GWAS ( 15 ). The SNPs identifi ed through GWAS contributed to the design of the Metabochip, a custom genotyping array used for genetic association analysis of type 2 diabetes in an expanded sample of (primarily European descent) 34,840 cases and 114,981 controls.
Genetic association analysis of ACSL1 variants in MAGIC and DIAGRAM. We focused on ACSL1 variants within 50 kb of the ACSL1 gene, using LocusZoom ( 17 ) to identify SNPs within the region from the published GWAS and custom genotyping array results. We then conducted candidate association analyses for each of the four selected published data sets, the original MAGIC GWAS ( 13 ), the MAGIC BMI-adjusted GWAS ( 14 ), the DIA-GRAM v3 GWAS ( 15 ), and the DIAGRAM Metabochip ( 15 ). For each lookup, we applied a Bonferroni correction for the number of SNPs identifi ed within the target region. Here, we report those SNPs reaching the specifi ed Bonferroni threshold for at least one of the published data sets.

Analysis of ACSL1 SNPs in MESA
MESA. The MESA is a longitudinal cohort study of subclinical cardiovascular disease and risk factors in 6,814 men and women 45 to 84 years of age, initially free of overt disease, that predict progression to disease or progression of subclinical disease ( 18 ). The MESA Family Study (MESAFS) recruited 1,595 African American and Hispanic family members 45 to 84 years of age specifi cally for genetic analysis, and the MESA Air Pollution Study recruited an additional 257 participants ( 19 ). MESA participants had detailed medical histories and examinations for anthropometry, blood pressure, and vascular imaging [to obtain measures of subclinical atherosclerosis of CAC and common (cIMT) and internal carotid intima-media thickness (iIMT)] ( 20 ). Glucose concentrations were determined from fasting samples, and diabetes status was defi ned using the 2003 American Diabetes Association fasting criteria (fasting glucose у 126 mg/dl) ( 21 ) and/or diabetes treatment. Characterization of MESA subjects included in the present study is included in the supplementary Materials and supplementary Tables 1 and 2. MESA participants were genotyped using the Affymetrix Human SNP array 6.0 ( 22 ). Imputation was conducted with IMPUTE v2 using Phase 1 v3 of the cosmopolitan 1,000 genomes reference set ( 23 ), followed by robust quality control ( 24 ).
Genetic association analysis in MESA. MESA participants were stratifi ed by race/ethnic group using an unrelated subset of individuals from MESA and from MESAFS, by selecting, at most, one individual from each pedigree, based on inferred relationships in KING ( 25 ). Linear regression of quantitative phenotypes or logistic regression of dichotomous phenotypes was performed under an additive 1 d.f . genetic dosage model in R. For phenotypes with a substantial familial component, we performed analysis using linear mixed-effects models for quantitative traits, or generalized estimating equations for dichotomous traits, to account for familial relationships as implemented in the package R/GWAF ( 26 ). A basic model including age, gender, study site, and principal components of ancestry was used. Fixed effect meta-analysis was performed to combine race/ethnic specifi c results using METAL ( 27 ).

Identifi cation of chromatin accessibility and transcription factor binding motifs
Chromatin accessibility data were obtained from public repositories for human cell lines [ENCODE Consortium ( 30 )] as well as fetal and adult tissues and primary cell lines [Roadmap Epigenomics Consortium ( 31 ) and unpublished observations] and intersected with the three SNPs examined by this work. Transcription factor binding motif locations were determined as described by Vierstra et al. ( 32 ).

Replication of ACSL1 results identifi ed from consortia studies in MESA
In combined meta-analysis across race/ethnic groups for nondiabetic participants from MESA, we identifi ed nominal (but not Bonferroni-corrected) evidence of replication for the association of ACSL1 rs4862423 with BMIadjusted fasting glucose; further, the observed direction of effect was consistent with that seen in MAGIC ( Table 1 ;  status in the Penn-T2D Meta ( Table 2 ). The observed direction of effect was consistent with that seen in MAGIC and DIAGRAM. ACSL1 rs4862423, which was associated with fasting glucose in the BMI-adjusted MAGIC with suggestive evidence of replication in MESA ( Table 1 ), showed a suggestive association with diabetes status ( P = 0.065) in the Penn-T2D Meta for which the direction of effect was consistent with that seen with fasting glucose in MAGIC and MESA. Thus, the association of ACSL1 rs735949 and rs7681334 with fasting glucose and diabetes status is replicated in large cohorts. The association of rs4862423 with fasting glucose was signifi cant in the BMI-adjusted MAGIC in nondiabetic subjects, yet failed to reach signifi cant association with diabetes status in the Penn-T2D Meta.

ACSL1 SNP association with subclinical atherosclerosis in MESA
For subclinical atherosclerosis traits, we observed association of ACSL1 rs7681334 with CAC Agatston score (among participants with CAC >0) in combined metaanalysis across race/ethnic groups ( Table 3 ; P = 0.003). The association was stronger in participants without diabetes ( Table 3 ; P = 0.001). We further observed association of rs4862423 with CAC in participants without diabetes ( Table 3 ; P = 0.002). Race/ethnic stratifi cation provides consistent P = 0.037). The effect allele (T/C) frequency was 0.350 in MAGIC European Caucasians and 0.407 in MESA Caucasian participants. The observed effect size and strength of association for rs4862423 was most notable in MESA African Americans without diabetes compared with other race/ethnic groups ( P = 0.033; supplementary Table 3). We did not observe additional race/ethnic specifi c associations for fasting glucose-related SNPs (supplementary Tables 3 and 4). We did not observe statistically signifi cant evidence of replication for rs7681334 with fasting glucose or for rs735949 with diabetes status ( Table 1 and supplementary Table 5). Thus, these results demonstrate that the T allele in rs4862423 (T/C) associates signifi cantly with increased fasting blood glucose levels in MAGIC, with suggestive evidence of replication in MESA.

Replication and validation of ACSL1 SNP associations in Penn-T2D meta-analyses
In a second study [Penn-T2D meta-analyses (Penn-T2D Meta) data], which is larger than either of the two consortia used for discovery analysis and the MESA replication study, we were able to confi rm the association of ACSL1 rs735949 with diabetes status. Furthermore, rs7681334, which was associated with fasting blood glucose in MAGIC ( Table 1 ) was also signifi cantly associated with diabetes    Tables 6 and 7). In addition, both SNPs showed stronger effects on CAC in nondiabetics compared with those with diabetes (Cochran's Q test for heterogeneity, rs7681334, P = 0.12; rs4862423, P = 0.04).

CAC results in meta-analysis of African Americans
The association of rs4862423 with CAC observed in MESA African Americans among those with CAC scores >0 was signifi cant in the expanded sample ( Table 4 ). However, the association of rs7681334 with CAC in MESA was not observed in the African American sample. Furthermore, an association between rs735949 and CAC was observed in the African American sample, but not in MESA ( Tables 3 and 4 ). Thus, the association of rs4862423 with subclinical atherosclerosis was consistent between the two studies. The association of rs735949 and rs7681334 with CAC appeared to be infl uenced by other factors.

Analysis of association of ACSL1 SNPs in the CARDIoGRAMplusC4D consortium
In addition to testing ACSL1 SNPs for association with diabetes and subclinical atherosclerosis, we determined whether the ACSL1 SNPs modifi ed risk for clinical end points (coronary heart disease). In the CARDIoGRAM-plusC4D consortium data, there were no signifi cant associations of ACSL1 SNPs rs7681334, rs735949, or rs4862423 with coronary heart disease (supplementary Table 8).

Chromatin accessibility at the ACSL1 locus
The three trait-associated ACSL1 SNPs are located within the fi rst (rs4862423) and second (rs7681334, rs735949) introns of the ACSL1 gene. Noncoding GWAS SNPs are concentrated in regulatory DNA elements, where they may modulate transcription factor binding ( 33 ). Thus, we asked whether these ACSL1 polymorphisms coincided with DHSs, which generically indicate transcription factor binding at active regulatory elements ( 30 ). ACSL1 utilizes alternative promoters, which appear to have different levels of tissue selectivity ( Fig. 2A ). SNP rs4862423 is located in intron 1 of ACSL1 , a region that is highly accessible to DNase I in fetal heart tissue, and to a lesser extent in other fetal tissues ( Fig. 2B ) as well as adult heart, mucosal (gastric and small bowel), pancreatic, and skeletal muscle tissue (data not shown). This site does not appear to be active in the hepatocyte cell line HepG2 but is accessible in multiple cultured epithelial and skeletal muscle cell lines (data not shown). Furthermore, rs4862423 coincides with a recognition site for RREB1 (ras responsive element binding protein 1; Fig. 2C ). Although rs735949 coincides with a DHS in primary skin melanocytes (data not shown), we did not observe regulatory element activity coinciding with rs7681334 in any of the cell lines and tissues for which DHS data are available.

eQTL analysis
ACSL1 SNPs were analyzed for evidence of eQTLs using the MuTHER consortium data on three tissues: subcutaneous fat, lymphocytes, and skin. There were no significant associations for any of the three SNPs with probe ILMN_1684585 ( ACSL1 ). P values for rs7681334 were 0.45 in fat, 0.59 in lymphocytes, and 0.76 in skin. The corresponding P values for rs735949 were 0.13, 0.56, and 0.054, and for rs4862423, they were 0.35, 0.73, and 0.68 (supplementary Table 9). The most signifi cant SNP associated with probe ILMN_1684585 in skin biopsies was rs745805 ( P = 0.0014), which is in high linkage disequilibrium with rs735949 ( r 2 = 0.832 in 1000G pilot 1 CEU; supplementary Table 9). These results suggest that rs735949 might be associated with ACSL1 expression levels in some tissues.

DISCUSSION
The role of ACSL1 in humans is unknown, although mouse studies have shown that this enzyme plays an important signifi cant association (defi ned as P < 0.05) was observed for any of the three SNPs in a large-scale GWAS metaanalysis of BMI ( 34 ).
The present study also demonstrates a signifi cant association of rs4862423 with subclinical atherosclerosis, measured as CAC scores >0 in MESA and in a follow-up meta-analysis study in African Americans. The directionality of the association was different for fasting glucose and CAC. Thus, rs4862423 allele T was associated with increased fasting glucose and decreased CAC. The reason for the difference in directionality is unclear. Analysis of cardiovascular events in CARDIoGRAMplusC4D did not show association with the three ACSL1 SNPs. It is possible that the association occurs only with early atherosclerosis or with calcifi cation, rather than with more advanced lesions and clinical events. In this context, it is interesting that ACSL1 in mice is required for early lesions of atherosclerosis, with a less obvious effect on advanced lesions ( 6 ).
All three ACSL1 SNPs are located in intronic regions of ACSL1 (introns 1 and 2), and our studies demonstrate that the two SNPs most clearly associated with fasting glucose and diabetes (rs7681334 and rs735949) do not associate with identifi ed DNase I accessible regions in several cells, fetal or adult tissues most likely to be relevant for these phenotypes. However, we show that rs735949 is associated with a DNase accessible region in primary skin melanocytes. A previous study identifi ed association of rs735949 role as a metabolic rheostat mechanism in several tissues, and hence, that loss of ACSL1 results in reduced blood glucose levels due to reduced fatty acid oxidation under conditions in which excess energy is needed. We demonstrate here that specifi c SNPs in the ACSL1 gene are signifi cantly associated with fasting glucose (rs7681334 and rs4862423) or with diabetes status (rs735949) in the MAGIC and DIAGRAM consortia. We also report evidence of replication of the association of rs4862423 with fasting glucose in nondiabetic participants from MESA, and association of both rs7681334 and rs735949 with diabetes status in Penn-T2D Meta. Although the Penn-based consortium includes the DIAGRAM data, it contains twice the number of cases included in DIAGRAM. Together, these four studies make a strong case for signifi cant association of the three ACSL1 SNPs with fasting glucose and diabetes. Furthermore, whereas MAGIC and DIAGRAM included subjects primarily of European descent, MESA includes subjects of different race/ethnic groups (mostly Caucasian, African American, and Hispanic participants and a smaller fraction of Chinese participants). The smaller sample size in MESA (7,847 participants) compared with the MAGIC and DIAGRAM consortia and Penn-T2D Meta probably contributed to the lower level of signifi cance and statistical power. It is unlikely that the association of the ACSL1 SNPs with glucose levels is signifi cantly confounded by BMI or adiposity because no nominally statistically associated with increased metabolic syndrome risk ( 41 ). A recent study suggested that rs6552828 (located near rs4862423) might result in differences in ACSL1 expression ( 42 ). It is therefore possible that rs4862423 and other polymorphisms in intron 1 of the ACSL1 gene result in altered expression of ACSL1 in specifi c tissues. In summary, our study is the largest to date examining ACSL1 SNPs in humans, and the fi rst to examine the association between ACSL1 SNPs, fasting glucose, diabetes, and subclinical atherosclerosis. Our results demonstrate that three SNPs in the ACSL1 gene are associated with markers of fasting glucose or diabetes status and that one of these SNPs is also signifi cantly associated with subclinical atherosclerosis. This represents the fi rst evidence of ACSL1 association with regulation of fasting glucose and risk of diabetes and subclinical atherosclerosis in humans and suggests possible links between these traits and acyl-CoA synthesis.
We are grateful to Dr. Daniel Rader (Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA) for helpful suggestions and advice. The authors thank the participants of the MESA study, the Coordinating Center, MESA investigators, and study staff for their valuable contributions. The authors from the Candidate-Gene Association Resource (CARe) Consortium wish to acknowledge the support of the NHLBI and the contributions of the research institutions, study investigators, fi eld staff, and study participants in creating this resource for biomedical research. We thank Eric Boerwinkle with islet transcription factors ( 35 ). It is possible that rs7681334 and rs735949 coincide with regulatory elements active in other tissues, ages, or disease states relevant to fasting glucose and diabetes traits, or that these SNPs are in linkage disequilibrium with other regulatory variants at this locus. Indeed, rs735949 was in strong linkage disequilibrium with another SNP in the same ACSL1 intron 2, rs745805, which is signifi cantly associated with ACSL1 expression in skin, suggesting links between rs735949 and ACSL1 expression in a tissue-selective manner. However, rs745805 is not located within a region of appreciable DNase I hypersensitivity, nor does it coincide with a transcription factor recognition sequence (data not shown). Thus, the ACSL1 SNPs investigated in regions that do not coincide with DHSs may be in linkage disequilibrium with transcription factor binding-altering SNPs within regulatory elements nearby. Furthermore, eQTL analysis did not reveal signifi cant associations with the SNPs investigated and ACSL1 levels in fat, skin, or lymphocytes. ACSL1 expression is known to be regulated differently in different tissues ( 7 ), and the ACSL1 gene contains three alternative exon 1s controlled by different promoters ( 36 ). It can therefore not be ruled out that one or several of the SNPs are associated with ACSL1 expression levels in other tissues or in disease states. Furthermore, a long noncoding RNA, SLED1 , is located in the vicinity of rs7681334 and rs735949. The putative role of SLED1 is unknown, but it is plausible that these SNPs could affect SLED1 .
ACSL1 rs4862423, which is associated with fasting glucose and CAC, coincides with an ACSL1 DHS active in multiple fetal and adult tissues (including heart and pancreas) and is localized within a binding motif for RREB1. RREB1 has recently been implicated by GWAS as a regulatory factor associated with glucose levels ( 37 ). Mass spectrometrybased analysis of the human proteome (Proteomics DB) has demonstrated that heart, pancreas, lymphocytes, and colon express RREB1 ( 38 ). The role of RREB1 in regulation of blood glucose and atherosclerosis is unknown, but interestingly, RREB1 was recently demonstrated to modify histones by aiding in removal of histone H3 lysine 27 trimethylation (H3K27me3) marks, which often characterize silenced regions, thereby promoting Ucp1 and Cidea expression in brown adipocytes ( 39 ). This mechanism of action of RREB1 was shown to be due to recruitment of Jmjd3, an H3K27me3 demethylase. Thus, RREB1 binding to the motif associated with rs4862423 might alter chromatin modifi cation and increase ACSL1 transcription in tissues that express RREB1. This hypothesis remains speculative until tissues expressing RREB1 from subjects with rs4862423 polymorphism have been examined by electrophoretic mobility shift assays for differences in RREB1 binding and ACSL1 expression. SNP rs4862423 is located in intron 1 of human ACSL1 . Previous studies have identifi ed other SNPs located in this intron as being associated with metabolic parameters in humans ( 40,41 ): rs6552828 has been associated with gains in maximal O 2 uptake after exposure to an exercise program in sedentary adults participating in the Health, Risk Factors, Exercise Training, and Genetics (HERITAGE) Family Study ( 40 ), and rs9997745 has been