|Home | About | Journals | Submit | Contact Us | Français|
The Vitamin D endocrine system is essential for calcium homeostasis, and low levels of vitamin D metabolites have been associated with cardiovascular disease risk. We hypothesized that DNA sequence variation in genes regulating vitamin D metabolism and signaling pathways might influence variation in coronary artery calcification (CAC).
We genotyped single nucleotide polymorphisms (SNPs) in GC, CYP27B1, CYP24A1, and VDR and tested their association with CAC quantity, as measured by electron beam computed tomography. Initial association studies were carried out in a discovery sample comprised of 697 Amish subjects and SNPs nominally associated with CAC quantity (4 SNPs in CYP24A1, P = 0.008-0.00003) were then tested for association with CAC quantity in two independent cohorts of subjects of European Caucasian ancestry (Genetic Epidemiology Network of Arteriopathy (GENOA) Study (n = 916) and The Penn Coronary Artery Calcification (PennCAC) sample (n = 2,061)). One of the four SNPs, rs2762939, was associated with CAC quantity in both GENOA (P = 0.007) and PennCAC (P = 0.01). In all three populations the rs2762939 C allele was associated with lower CAC quantity. Meta-analysis for the association of this SNP with CAC quantity across all three studies yielded a P value of 2.9 × 10-6.
A common SNP in the CYP24A1 gene was associated with CAC quantity in three independent populations. This result suggests a role for vitamin D metabolism in the development of CAC quantity.
Coronary artery calcification (CAC) is a measure of coronary subclinical atherosclerosis that predicts risk for cardiovascular disease (CVD) in the general population1. The potential role of circulating levels of vitamin D in influencing the initiation and/or progression of CAC is of great clinical interest given the large proportion of the population deficient in vitamin D2, the fact that the vitamin D endocrine system is essential for calcium homeostasis, and that low vitamin D levels have been associated with risk for CVD2. However, circulating levels of vitamin D metabolites have not been consistently associated with CAC, although causal relationships may not be apparent from simple cross-sectional studies.
The effects of vitamin D are myriad, including enhancement of cellular differentiation and immunomodulation, in addition to the well known effects on calcium homeostasis3. Vitamin D effects are mediated by activation of the nuclear vitamin D receptor (VDR), which regulates transcription of a large number of genes in many tissues. The metabolism of vitamin D, providing active 1,25(OH)2D to the VDR, requires the involvement of several key proteins in a tightly regulated process. Vitamin D (made in the skin keratinocytes in response to UVB light or from exogenous supplements) is sequentially hydroxylated by the mitochondrial cytochrome P450 enzymes 25-hydroyxlase (the product of CYP2R1) in the liver and 1-α hydroxylase (the product of CYP27B1) in the kidney (among other tissues) into the active metabolite 1,25(OH)2D. After 1,25(OH)2D binds to the ubiquitous VDR, the retinoid X receptor4 is recruited, forming a heterodimer complex that regulates gene transcription by interacting with vitamin D response elements (VDRE) in the promoter region of genes. Vitamin D metabolites are transported through the circulation by vitamin D binding protein (GC).
Concentrations of 1,25(OH)2D are tightly regulated by both 1-hydroxylase and the catabolic enzyme 24-hydroxylase (gene CYP24A1). CYP24A1 is induced by both 1,25OH)2D and 25(OH)D and is one of the most highly inducible genes in humans, capable of increasing its transcription by 20,000 fold5. The high inducibility of CYP24A1 is likely to be a critical factor in the large therapeutic window of vitamin D.
Thus, the key genes involved in vitamin D metabolism pathway include CYP2R1, CYP27B1, CYP24A1, VDR, and GC. We hypothesized that DNA sequence variation in these genes and levels might influence CAC quantity variation. To test this hypothesis we obtained genotypes of single nucleotide polymorphisms (SNPs) in four available genes of the five major genes involved in the regulation of vitamin D levels (GC, CYP27B1, CYP24A1, and VDR) and tested their association with CAC quantity. Initial association studies were carried out in a discovery sample comprised of 697 Amish subjects and SNPs nominally associated with CAC quantity in this cohort were then tested for association with CAC quantity in two independent cohorts of subjects also of European Caucasian ancestry (Genetic Epidemiology Network of Arteriopathy (GENOA) Study and The Penn Coronary Artery Calcification (PennCAC) sample).
The discovery sample was comprised of participants of the Amish Family Calcification Study (AFCS), a community-based study initiated in 2001 to identify the joint determinants of bone mineral density and vascular calcification. Details of recruitment procedures have been previously published6. Many related individuals were recruited (indeed, all Amish are related) and recruitment efforts were made without regard to CVD health status in this generally healthy population. Because vascular calcification occurs infrequently in young adulthood, only men age 30 years and older or women age 40 years or older were recruited into the AFCS. The protocol was approved by the Institutional Review Board of the University of Maryland and other participating institutions. Informed consent, including permission to contact relatives, was obtained before participation.
All AFCS participants underwent a detailed medical history interview and assessment of potential risk factors for CVD at the Amish Research Clinic in Strasburg, PA. Physical examinations were conducted in the early morning following an overnight fast. Blood samples were obtained for biochemical measurements and DNA extraction.
Images of the coronary arteries were obtained by electron beam computed tomography (EBCT) using an Imatron C-150 scanner. The protocol included 30 to 40 three mm contiguous transverse slices between the aortic root and the apex of the heart, gated to 80% of the R-to-R interval (the cycle between two consecutive R-waves) and obtained during a single breath hold. The extent of calcification in the thoracic aorta was assessed by scanning between the superior aspect of the aortic arch and the superior pole of the kidney at six mm intervals. The presence of calcification was defined as a density >130 Hounsfield Units (HU) with area > 1 mm2. CAC was quantified using the Agatston method which incorporates both density and area7. The sum of the scores in the left main, left anterior descending, circumflex, and right coronary arteries was considered the CAC score. All scans were scored by a single experienced cardiologist.
A total of 786 AFCS study participants were genotyped using the HumanCVD BeadChip V2 (Illumina, Inc., San Diego, CA), which includes ~50,000 SNPs within ~2,000 cardiovascular disease (CVD) candidate genes8. GC, CYP27B1, and VDR were considered in the chip design as “group 1” genes, with denser SNP coverage (inclusive of the intronic, exonic, untranslated regions (UTRs) and 5 kb of the proximal promoter regions), while CYP24A1 was considered as a “group 2” gene, with sparser SNP coverage (inclusive of intronic, exonic and flanking UTRs). In the chip design, tag SNPs were selected from group 1 genes to capture known variation across all of the four representative HapMap populations and SeattleSNPs (where available) for SNPs having minor allele frequency (MAF) > 0.02 and an r2 of at least 0.8. Tag SNPs included in group 2 genes were designed to capture known variation for SNPs having MAF > 0.05 and an r2 of at least 0.58. Carriers for the known monogenic mutation R3500Q in APOB (n=81)9, responsible for Familial Defective apob-100, were excluded from the analysis. R3500Q has been associated with CAC quantity in the Amish.
The HumanCVD BeadChip included 11, 13, 21, and 116 SNPs in GC, CYP27B1, CYP24A1, and VDR genes, respectively. The initial genotype calls were generated with the cluster file provided by Illumina BeadStudio software. Individuals with call rates less than 95% were excluded (n=8). SNPs with low call rate (<95%) were excluded. Of successfully genotyped SNPs, 3, 10, 3,and 47 were monomorphic in the Amish across GC, CYP27B1, CYP24A1, and VDR, respectively, and 5 SNPs (3 from CYP24A1 and 2 from VDR) were excluded due to low call rate. This left a total of 93 SNPs (8 in GC, 3 in CYP27B1, 15 in CYP24A1, and 67 in VDR included in this analysis (SNP names are provided in Supplemental Table 1). The observed distribution of genotypes was tested for deviation from Hardy-Weinberg Equilibrium (HWE) using Pearson's χ2 test.
Genetic Epidemiology Network of Arteriopathy (GENOA) Study. The Family Blood Pressure Program (FBPP), established by the National Heart Lung and Blood Institute in 1996, joined existing research networks that were investigating hypertension and CVD10. One of the four FBPP networks is GENOA, which recruited sibships with hypertensive adults to investigate genetic contributions to hypertension and hypertension-related target organ damage. Sibships containing at least two individuals with clinically diagnosed essential hypertension before age 60 were recruited from Rochester, Minnesota. Participants were diagnosed with hypertension if they had either 1) a previous clinical diagnosis of hypertension by a physician with current anti-hypertensive treatment, or 2) an average systolic blood pressure ≥ 140 mm Hg or diastolic blood pressure ≥ 90 mm Hg based on the second and third readings at the time of their clinic visit as stipulated by the Joint National Committee-7 guidelines11. Exclusion criteria were secondary hypertension, alcoholism or drug abuse, pregnancy, insulin-dependent diabetes mellitus, active malignancy, hypercalcemia, renal insufficiency (creatinine>1.3 mg/dl) and known metabolic bone disease other than osteoporosis. After identifying the hypertensive sibships, all members of the sibship were invited to participate regardless of their hypertension status.
Between December 2000 and February 2004, 1,241 of the original GENOA study participants in the Rochester field center returned to undergo risk factor and target organ damage measurement. Individuals with a history of coronary revascularization (n=83) were not eligible for measurement of CAC. After further excluding participants with a history of myocardial infarction (n=19), stroke (n=27), or a positive angiogram (n=31), and those who self-reported not being of European ancestry (n=2) and those missing risk factor data (n=2) there were 1,077 GENOA participants with CAC and risk factor measures.
Participants were imaged with an Imatron C-150 electron beam computed tomography scanner (Imatron Inc., South San Francisco, California) as previously described12. CAC was defined as a hyperattenuating focus in a coronary artery area > 1.0 mm2 and having CT number > 130 Hus. Quantity of CAC was defined as the total CAC score summed from the four major epicardial arteries using the method of Agatston7. The average CAC score from two sequential scans was used.
SNPs to be tested for replication (rs4809960, rs2296241, and rs2245153) had been previously genotyped in the GENOA Study on the Affymetrix® Genome-Wide Human SNP Array 6.0 platform. All passed the overall quality control filters (e.g., SNP call rate >95%). SNP imputation was performed using MACH version 1.0.16 (http://www.sph.umich.edu/csg/abecasis/MaCH) based on the publicly available phased haplotypes from HapMap release 22, build 36, CEU population. Association analysis of GENOA SNP (rs2762939) was based on the imputed genotype dosage that ranges from 0 to 2, which is the expected number of effective alleles. The imputed genotype of rs2762939 had an imputation quality score of 0.80 and r2 of 0.62 with a genotyped SNP.
Of the 1,077 GENOA participants with CAC and risk factor data, 916 had genotype data available for analyses. These 916 participants were in 422 sibships including 115 singletons, 204 sibships of size 2, 64 sibships of size 3, 16 sibships of size 4, 13 sibships of size 5, 5 sibships of size 6, 1 sibship of size 7, 2 sibships of size 8, 1 sibship of size 9, and 1 sibship of size 10. The study was approved by the institutional review boards of all the participating institutions and participants gave written informed consent.
The Penn Coronary Artery Calcification (PennCAC) sample included European Caucasian ancestry subjects from recruited from three separate parallel studies: the Study of Inherited Risk of Coronary Atherosclerosis (SIRCA; n =799), the Penn Diabetes Heart Study (PDHS; n =782), and the Philadelphia Area Metabolic Syndrome Network (PAMSyN; n =480)13. PennCAC sample were genotyped using the HumanCVD BeadChip V2 (Illumina, Inc., San Diego, CA).
SIRCA is a cross-sectional study of factors associated with CAC in a community-based sample of asymptomatic subjects and their families14. Subjects were healthy adults aged 30-75 yrs who had a family history of premature coronary artery disease (CAD). Subjects were excluded if they reported evidence of CAD on screening questionnaire, reported a history of diabetes mellitus, or had a serum creatinine >3.0 mg/dl.
PDHS is an ongoing, cross-sectional community-based study of type-2 diabetic subjects without clinical evidence of coronary heart disease (CHD) or overt chronic kidney disease15. Subjects were aged 35 to 75 years; had a clinical diagnosis of type-2 diabetes (defined as fasting blood glucose ≥ 126 mg/dl, 2-h postprandial glucose ≥ 200 mg/dl, or use of oral hypoglycemic agents/insulin in a subject greater than age 40 yr); and had negative pregnancy test if female. Subjects were excluded if they had evidence of clinical CHD, a clinical diagnosis of type 1 diabetes (insulin use prior to age 35), a serum creatinine >2.5 mg/dl, or weight >300 lbs.
PAMSyN is a cross-sectional study of patients with one or more metabolic syndrome risk factors. PAMSyN participants were recruited between 2004-2009 via the University of Pennsylvania Health System primary care providers, word of mouth in the community, and Penn health fairs for CVD risk factors. Subjects were aged 18-75 yrs. Subjects with known type 1 diabetes or clinical atherosclerotic CVD were excluded.
SIRCA, PDHS, and PAMSyN are all are single center studies that used the same general clinical research center (GCRC), nursing staff, CT scanner, and research laboratories. All study subjects were evaluated at the GCRC at Penn as previously described14. CAC scores were determined by electron beam computed tomography (Imatron, San Francisco, CA) according to the method of Agatston7 from forty continuous 3-mm thick computed tomograms. All Penn study protocols were approved by the Penn IRB and all subjects provided written informed consent.
Because the distribution of CAC scores was positively skewed, the scores were natural log-transformed after adding 1. To evaluate the distribution of log-transformed CAC scores after regressing out the effects of age, sex, and age×sex, we obtained the residualized values of transformed CAC score in the AFCS sample. The residualized values were approximately normally distributed.
We followed a two-staged approach in which we first identified a small number of promising SNPs in the Amish Discovery Set for follow-up, which we would then take forward for more rigorous statistical testing in the two independent replication cohorts. Our criteria for identifying SNPs of potential interest in the Amish discovery sample was that they be associated with CAC quantity at a liberal threshold of P < 0.01. To establish statistical significance, we then required that these SNPs be associated in both of the replication cohorts at p-values taking into account the number of SNPs tested in each cohort (i.e., Bonferroni-corrected significance thresholds of 0.05/4 = 0.0125.) Finally, we combined evidence for association across all three studies in a meta-analysis, using a Bonferroni correction for 93 SNPs.
In the Amish Discovery sample, association analyses were carried out using a variance component approach suitable for analysis of large pedigrees. The approach used a mixed effect model to assess the effect of genotype (included as a fixed effect) on CAC score while adjusting for covariates (age, sex, age×sex) and including a polygenic component (as a random effect) to that accounted for the correlations in phenotype existing among related individuals. The polygenic component was modeled using the relationship matrix derived from the complete Amish pedigree structure. These analyses were carried out using Mixed Model Analysis for Pedigree (MMAP) software developed by one of the coauthors (JR O'Connell). Pairwise linkage disequilibrium (LD) correlation statistics (r2) were computed in the Amish using the Haploview software (www.broad.mit.edu)16.
The SNPs that nominally associated with CAC score in the Amish sample were then tested for association in the GENOA and PennCAC samples. In GENOA, association analysis was carried out in R using linear mixed effect models and included a random effect for family structure (ie, sibship). The effect of the CYP24A1 SNPs on log-transformed (CAC Score + 1) was assessed with an indicator variable for sibship included as a random effect to account for the family structure (siblings) with adjustment for age, sex, and an age-by-sex interaction. The covariance structure was defined as compound symmetry. In PennCAC, the SNP association with log-transformed (CAC Score + 1) was tested using a linear regression model with CYP24A1 genotype included as an independent variable and age, sex, and sex × age included as covariates. The analysis was carried out using PLINK v 1.06 software program.
For SNPs tested for association in all three cohorts, we combined association results across cohorts in a meta-analysis using a weighted fixed effects meta-analysis approach as implemented in the METAL software program (http://www.sph.umich.edu/csg/abecasis/Metal/index.html).
The characteristics of participants in the Amish Discovery sample (n = 697), and GENOA (n = 916) and PennCAC (n = 2,061) replication samples were summarized in Table 1. Mean ages of participants in the Amish and PennCAC samples were comparable (53-54 yrs) and slightly younger than mean age in the GENOA sample (58 yrs). The proportion of men in the samples ranged from 35-45% and the mean body mass index was 28.1 kg/m2 in the Amish compared to 30.2 and 30.7 kg/m2 in PennCAC and GENOA. By design, there was a substantially higher proportion of subjects with diabetes in the PennCAC sample (40.2%) compared to the Amish (2.4%) or GENOA (13.4%) samples. By design, there was also a substantially higher proportion of subjects with hypertension in GENOA sample (70.2%) compared to the Amish (17.5%) or PennCAC (43.7%) samples. In the GENOA sample, blood pressure measurements were recorded on blood pressure lowering treatment.
Table 2 presents the sizes of the three genes, the numbers of SNPs genotyped, and summary results of the association testing. A total of 8 SNPs in GC (4 exons, 6,822 bp), 3 SNPs in CYP27B1 (9 exons, 4,860 bp) 15 SNPs in CYP24A1 (12 exons, 20,529 bp), and 67 SNPs in VDR (11 exons, 63,493 bp) were successfully genotyped and tested for association. We estimated the proportion of allelic variation in the genes captured by these SNPs in the HapMap CEU population. The 8 SNPs in GC, 3 SNPs in CYP27B1 and 67 SNPs in VDR (both IBC Group 1 genes) captured 86%, 100% and 89% of all alleles having frequency of 0.02 or higher at an r2 > 0.80. The 15 SNPs in CYP24A1 (a Group 2 gene) captured 67% of all alleles having a frequency of 0.02 or higher at an r2 > 0.80. There was no statistical evidence for deviation from HWE for any of the 93 successfully genotyped SNPs in the Amish based on a threshold of P < 0.0005 (0.05/93). In the discovery sample, no SNPs in GC, CYP27B1 or VDR gene were associated with CAC score at even a nominal P value of 0.01 (see Supplemental Table 1 for individual SNP association results). In contrast, four of the 15 SNPs in CYP24A1 were associated with CAC score, with nominal P values ranging from 0.008 to 0.00003. The 4 SNPs in CYP24A1 nominally associated with CAC quantity were 146 bp - 5155 bp apart from each other. SNP rs2762939 was located in one haplotype block and SNPs rs4809960, rs2296241, and rs2245153 in another block (all pairwise LD among rs4809960, rs2296241, and rs2245153: r2 =0.06-0.47, D’=0.99). Rs2762939 was not in LD with the other three SNPs (pairwise LD between rs2762939 and other three SNPs: r2 < 0.07, D′ < 0.55). Figure 1 shows the LD (r2) structure of the 15 SNPs in CYP24A1 in the Amish sample and p-values for association of these SNPs with CAC score.
The four SNPs in CYP24A1 with nominal P value < 0.01 in the Amish were then tested for association with CAC score in the GENOA and PennCAC populations. One of the four SNPs, rs2762939, was associated with CAC score in both GENOA (P = 0.007) and PennCAC (P = 0.01) (Table 3). In all three populations the C allele (minor allele frequency ranging from 0.25 to 0.31 across the three populations) was associated with lower CAC score. A formal test of heterogeneity indicated no significant differences in effect size among across the three populations (Het p-value = 0.45). Combining the three P values in a meta-analysis for the rs2762939- CAC score association across all three studies yielded a meta-analysis P value of 2.9 × 10-6. Combining results from the GENOA and PennCAC samples only (that is, excluding results from the Amish Discovery sample) yielded a P value of 2.8 × 10-4. There was no evidence for association of CAC score with any of the other three SNPs in the GENOA and PennCAC samples.
The relation between rs2762939 and traditional risk factors was tested in AFCS. This SNP was not significantly associated with any of the blood pressure or serum lipid phenotypes, nor was it associated with variation in serum 25(OH)D levels (data not shown). Consistent with these results, the association of rs2762939 with CAC quantity was essentially unchanged with additional adjustment for LDL cholesterol and diabetes.
We have identified a common SNP rs27629393 in CYP24A1 that is associated with quantity of coronary artery calcification. The gene product of CYP24A1 is the major enzyme responsible for the catabolism of 1,25 (OH)2D and 25(OH)D. Our data thus suggest that genetic variability in vitamin D homeostasis may contribute to the pathogenesis of coronary artery calcification.
CYP24A1 plays a pivotal role in maintaining vitamin D homeostasis. Deletion of Cyp24a1 in mice causes 1,25-(OH)2D excess and hypercalcemia with severe bone mineralization defects and ectopic vascular calcification (renal calcium deposition) after chronic treatment of 1,25-(OH)2D17. On the other hand, transgenic rats that constitutively express Cyp24a1 develop atherosclerotic lesions in the aorta, which greatly progress with high-fat and high-cholesterol feeding18.
Although there is much circumstantial evidence, the role of vitamin D in cardiovascular health remains controversial19. There is some epidemiological evidence in humans supporting a role of vitamin D metabolites in the development of CAC. Watson et al. measured serum 1,25-(OH)2D in 153 asymptomatic individuals with high risk of CHD and 13 patients with familial hypercholesterolemia and reported in both groups, an inverse correlation between serum 1,25-(OH)2D and CAC20. Doherty TM et al. reported that serum 1,25-(OH)2D independently and inversely predicted CAC quantity in a sample of 283 asymptomatic subjects with risk factors for CHD21. Low 25(OH)D levels were also associated with subsequent development of CAC in the Multi-Ethnic Study of Atherosclerosis (MESA)22. By contrast, there were no association between serum 25(OH)D levels and coronary calcification in a smaller study of 50 patients undergoing angiography23. In this Amish population there is no correlation between season-adjusted 25(OH)D levels and prevalent CAC24. The discordant results observed could be due to limited power to detect associations of small magnitude, differences in patient characteristics between studies, or differences in assays used in these studies.
Although 25(OH)D is the best marker of vitamin D stores, 1,25 (OH)2D is the active metabolite, and 25(OH)D levels might not reflect circulating 1,25 (OH)2D levels. In addition, the unknown local 1,25 (OH)2D production in the vasculature or in macrophages and other cell types involved in plaque formation might be more important than circulating levels for pathogenesis of vascular calcification.
This is the first demonstration of association between CYP24A1 gene variants and vascular calcification in human population studies. CYP24A1 SNPs or haplotypes have also been associated with other diseases in humans, including cancer 25 and asthma26. In their asthma study, Wjst et al reported that the CYP24A1 haplotype that was moderately associated with asthma was also associated with 1,25 (OH)2D, 25(OH)D, and IgE26. A recent genome-wide association analysis based on ~34,000 individuals identified an association between rs6013897, located ~27.5 kb centromeric of CYP24A1, and 25(OH)D levels27. The SNP is 38.8 kb away from rs2762939, the SNP associated with CAC quantity in the present study, and there is little LD between rs6013897 and rs2762939 (HapMap CEU r2 = 0.04, D’ = 0.27).
Despite the primal role of CYP24A1 to vitamin D metabolism, the mechanism linking rs2762939 to coronary calcification remains unclear. In neither the Amish study nor in an independent sample of 546 premenopausal Caucasian women from Indiana, USA (data not shown) was there any evidence for an association of rs2762939 with 25(OH)D levels. However, while an association of rs2762939 with 25(OH)D levels would have provided a clear potential mechanism linking this SNP to variation in CAC levels, there are several ways through which this SNP could affect CAC without influencing mean 25(OH)D levels. For example, this SNP (or a SNP in high LD with this SNP) could influence the conversion of active 1,25(OH)2D into its inactive metabolite without affecting 25(OH)D levels. One way to test this hypothesis would be to assess the association of rs2762939 with serum 1,25(OH)2D, but unfortunately we do not have these measurements in our study. Another potential mechanism is that this genetic variant may affect 24-hydroxylase expression and vitamin D metabolism locally in the endothelium, and these effects may not be correlated with circulating levels of vitamin D metabolites.
In summary, a common variant in the CYP24A1 gene was associated with CAC quantity in three independent populations. This result suggested a role for vitamin D metabolism in coronary atherosclerosis. Future studies should elucidate the underlying mechanisms and signaling pathways that entwine vitamin D metabolism and vascular health. These studies may lead to novel screening and therapeutic options for the identification and treatment of individuals at increased risk for CVD events.
The Amish component of this study was supported by NIH research grants U01 HL72515 and R01 HL088119, F32AR059469, University of Maryland General Clinical Research Center grant (M01 RR 16500), an AHA Scientist Development Grant (0830146N), and an AHA Grant-In-Aid (0855400E). Partial funding was also provided by the Mid-Atlantic Nutrition and Obesity Research Center (P30 DK072488) and the Baltimore Veterans Administration Medical Center Geriatric Research and Education Clinical Center.
GENOA was funded by NIH research grants R01 HL087660, U10-HL54457 (Genetic Determinants of High BP in Three Racial Groups), R01 Hl091988, and an NIH General Clinic Research Center Grant (MO1-RR00585) awarded to Mayo Clinic Rochester.
The PennCAC resource was supported by a Clinical and Translational Science Award (UL1RR024134) from the National Center for Research Resources (NCRR) and a Diabetes and Endocrine Research Center (P20-DK 019525) award, both to the University of Pennsylvania. M.P.R. is supported in by R01-DK071224 from the National Institutes of Health.
The study of Indiana Women was supported by NIH research grants P01 AG-18397 and UL1 RR-025761.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.