CDH13 promoter SNPs with pleiotropic effect on cardiometabolic parameters represent methylation QTLs

CDH13 encodes T-cadherin, a receptor for high molecular weight (HMW) adiponectin and low-density lipoprotein, promoting proliferation and migration of endothelial cells. Genome-wide association studies have mapped multiple variants in CDH13 associated with cardiometabolic traits (CMT) with variable effects across studies. We hypothesized that this heterogeneity might reflect interplay with DNA methylation within the region. Resequencing and EpiTYPER™ assay were applied for the HYPertension in ESTonia/Coronary Artery Disease in Czech (HYPEST/CADCZ; n = 358) samples to identify CDH13 promoter SNPs acting as methylation Quantitative Trait Loci (meQTLs) and to investigate their associations with CMT. In silico data were extracted from genome-wide DNA methylation and genotype datasets of the population-based sample Estonian Genome Center of the University of Tartu (EGCUT; n = 165). HYPEST–CADCZ meta-analysis identified a rare variant rs113460564 as highly significant meQTL for a 134-bp distant CpG site (P = 5.90 × 10−6; β = 3.19 %). Four common SNPs (rs12443878, rs12444338, rs62040565, rs8060301) exhibited effect on methylation level of up to 3 neighboring CpG sites in both datasets. The strongest association was detected in EGCUT between rs8060301 and cg09415485 (false discovery rate corrected P value = 1.89 × 10−30). Simultaneously, rs8060301 showed association with diastolic blood pressure, serum high-density lipoprotein and HMW adiponectin (P < 0.005). Novel strong associations were identified between rare CDH13 promoter meQTLs (minor allele frequency <5 %) and HMW adiponectin: rs2239857 (P = 5.50 × 10−5, β = −1,841.9 ng/mL) and rs77068073 (P = 2.67 × 10−4, β = −2,484.4 ng/mL). Our study shows conclusively that CDH13 promoter harbors meQTLs associated with CMTs. It paves the way to deeper understanding of the interplay between DNA variation and methylation in susceptibility to common diseases. Electronic supplementary material The online version of this article (doi:10.1007/s00439-014-1521-6) contains supplementary material, which is available to authorized users.


Supplemental
Genetic association between four overlapping common SNPs identified in HYPEST -CADCZ study groups and DNA methylation in the ECGUT sampleset Supplemental Table S8 Results of HYPEST-CADCZ meta-analysis for association tests between the DNA methylation level at targeted CDH13 promoter CpG sites/units and cardio-

Study inclusion criteria, measurement of blood pressure and serum lipids in HYPEST/CADCZ participants
HYPEST subjects (full sample: n=1,966; aged 18-85 yrs) have been recruited across Estonia during 2004-2007 by North Estonia Medical Center (Cardiology andBlood Centres), and by Tartu University Hospital (Cardiology Clinic, Blood Centre) with the main aim to analyze genetic-epidemiological risk factors for cardiovascular disease (CVD) in Estonian population (Org et al. 2011). The CADCZ subjects (full sample: n=869; 18-67 yrs) have been recruited across Czech Republic during 1998-2000 by the Cardiology Department of the 2 nd Clinic of Internal Medicine, Faculty Hospital Královské Vinohrady in Prague with the main aim to study genetic factors relating to homocysteine metabolism in coronary artery disease (CAD) (Janosíková et al. 2003). For the current study, the HYPEST and CADCZ patients were selected based on the following criteria: (i) clinically diagnosed CVD (MI, CAD, hypertension) or (ii) determined metabolic risk-factors CVD (serum LDL >3 mmol/L, total cholesterol >5 mmol/L; blood pressure >140/90 mmHg).
For all HYPEST and CADCZ subjects resting blood pressure (BP) had been measured by trained clinicians during recruitment. In both studies three BP measurements per subject were obtained after a rest in a sitting position using a standard mercury column sphygmomanometer and size-adjusted cuffs and the median value was recorded. Although longitudinal clinical data for past BP measurements is available for HYPEST subjects (Org et al. 2011), the current study used BP values documented at the recruitment prior the blood draw for DNA extractions. This allowed correct incorporation of subject's age as a potential confounder for methylation level of some CpG sites (Bell et al. 2012;Gomes et al. 2012).
Biochemical parameters (total cholesterol, LDL, HDL and triglycerides) from fasting serum of each participant were measured at the recruitment. Clinical chemistry assays for HYPEST samples were performed using standardized assays (Cobas Integra 800®  Among 192 HYPEST and 166 CADCZ subjects 15.6% and 32.5% received lipidlowering medications, and 86.4% and 25.5% antihypertensive treatment, respectively. For the subjects under treatment, corrected values for blood lipids and for BP were used in subsequent genetic data analysis. Serum lipid levels were corrected according to Wu et al. (2007) (total cholesterol +1.31 mmol/L; HDL -0.059 mmol/L; LDL +1.243 mmol/L; triglycerides +0.222 mmol/L) and BP values according to Tobin et al. (2005) (SBP +15 mmHg; DBP +10 mmHg).

Measurement of serum HMW adiponectin in HYPEST
Serum samples of HYPEST subjects were stored at -86 o C immediately after blood draw. The concentration of high molecular weight (HMW) adiponectin in serum was measured for the study subjects with available stored serum samples (n=184) using Human HMW Adiponectin/Acrp30 ELISA assay (R&D Systems) protocol as recommended by the manufacturer. HMW Adiponectin concentration in serum samples was measured in duplicate on one ELISA reaction plate and the average of two measurements was calculated and used in the following association analysis. Intra-and interassay coefficients of variation were < 5% and < 15%, respectively.

Data collection, blood pressure and lipid measurements in EGCUT participants
The population-based biobank of the Estonian Genome Center of the University of Tartu (EGCUT, http://www.geenivaramu.ee/en/) includes epidemiological-clinical datasets and DNA samples extracted from blood for Estonian adults across all age groups (n=51,515; Leitsalu et al. 2014). Each participant filled out a Computer Assisted Personal interview (http://www.geenivaramu.ee/sites/default/files/geenivaramu/ksimustiku_tutvustus.pdf) during 1-2 hours at a recruitment office, including personal data, genealogical data, lifestyle data and medical history (diseases recorded in ICD-10 systems and the use of medicaments). Height, weight and BP phenotypes (SBP and DBP) were measured at the recruitment office according to standard medical procedures. BP was measured three times per subject after a rest in a sitting position and the median values were used in the analysis. Serum lipid parameters (total cholesterol, LDL, HDL and triglycerides) were assessed from fresh blood at the United Laboratories, Tartu University Clinics using identical assays and platforms as for the HYPEST samples. The whole project is conducted according to the Estonian Genes Research Act (http://www.geenivaramu.ee/for-scientists/human-genes-research-act.html; Dec 13 th , 2000) and all participants have signed the informed consent.
In the current study, in silico data for the CDH13 region was extracted and analyzed for the EGCUT samples (n=165; aged 18-84; 41.6±22.7 years) with available datasets for both, genome-wide DNA methylation (Infinium HumanMethylation450 BeadChip, Illumina) and genome-wide imputed genotypes (HumanOmniExpress BeadChips, Illumina). Among the included 165 EGCUT subjects 5.5% received lipid-lowering and 24.2% antihypertensive treatment. In the statistical analyses, serum lipid and BP values of these individuals were corrected using the identical approach as described for the HYPEST and CADCZ subjects.

CDH13
Whole blood consists of many different cell types, which may be characterized by differential CpG-methylation at several loci (Adalsteinsson et al. 2012). In order to assess the suitability of DNA extracted from whole blood for the reliable CpG methylation profiling at the CDH13 locus, a publicly available genome-wide DNA methylation dataset for purified human blood cells was used (Reinius et al. 2012; http://publications.scilifelab.se/kere_j/methylation). This dataset comprises of DNA methylation profiles of six healthy male subjects determined for whole blood, mononuclear cells, granulocytes, and cells from seven selected purified lineages. DNA methylation analysis was performed with the Infinium Human Methylation 450K bead chip technology (Illumina, USA). For the current study DNA methylation values for 69 CpG sites present at the 450K chip for the CDH13 genic region (GRCh37/hg19, Chr16:82,638,838,786) were comparatively assessed in different blood cell types and whole blood ( Figure S2). All analyzed blood cell types demonstrated similar CpG methylation profile across the entire CDH13 promoter and genic region, supporting the suitability of using whole blood cells for the inter-individual DNA methylation studies of CDH13.

Experimental and analytical details, and quality control of CpG methylation measurements in HYPEST/ CADCZ
Four EpiTYPER TM assays were designed to cover 110 CpG sites (13-40 CpG-sites/assay) within a ~1.5 kb target region (1162 bp; GRCh37/hg19, Chr16: 82,660,652 -82,661,813; Figure S3) using the EpiDesigner software as instructed by the manufacturer. The sequence context (e.g. presence of GT-microsatellite repeats) did not allow the inclusion of six CpGsites. Genomic DNA (400 ng) extracted from whole blood was bisulfite treated using the EZ-96 DNA Methylation TM kit (Zymo Research). Converted DNA was amplified (length of PCR products 289-470 bp) with primers (Table S1) containing T7-promoter tag and treated with shrimp alkaline phosphatase (Sequenom Inc.). Purified products were in vitro transcribed and fragmented using T-specific cleavage. Fragments (1-57 bp) were analyzed with MassARRAY Analyzer 4 (Sequenom Inc.), and methylation values were calculated by EpiTYPER Analyzer v.1.0 (Sequenom Inc.). Fragments containing >1 CpG sites were named as CpG-units (60 CpG-sites formed 21 CpG-units with 2-6 CpG sites/unit) and the methylation value for a CpG-unit was calculated as average methylation across the CpG-sites forming a unit. The remaining 50 CpG-sites within the targeted region contained 1 CpG-site per fragment and were assessed individually.
EpiTYPER Analyzer assesses the quality for every measured mass spectrum representing one EpiTYPER TM assay, and confidence values (CV; 0 = low and 5 = high confidence, respectively) for the spectrums are calculated. Confidence is calculated by analyzing the recalibration of the spectrum, the number of missing signals, and the number of additional signals. After empirical estimation, only assays with CV≥3 (88.4% across all assays and analyzed individuals) were included for subsequent analysis. The challenge of the EpiTYPER TM assay method is reliable measurement of CpG-methylation for the genomic regions with a high density of CpG sites, such as CDH13 promoter CpG-island ( Figure S3).
During Quality Control (QC), EpiTYPER TM Analyzer program automatically excluded from methylation calculations the fragments with too low (<1500 Da) or too high (>7000 Da) mass for technically reliable in mass-spectrometry measuring range (20 CpG-sites). Fragments with indistinguishable mass or low CpG-methylation call rate were excluded from statistical tests (24 CpG-sites).
Protocol for DNA methylation detection by the EpiTYPER TM assay is detailed in (http://bioscience.sequenom.com/sites/bioscience.sequenom.com/files/EpiTYPER%20Applic ation%20Note.pdf). Its principles are described by Ehrich et al (2005). instruments under following conditions: initial denaturation at 95°C for 3 min, followed by 10 'touchdown' cycles of 95°C for 15s, and annealing from 66°C to 57°C for 30s, and 68°C for 2.5 min, then an additional 25 cycles of 95°C for 15s, 56°C for 30s and 68°C for 2.5 min.

Long-range PCR for CDH13 promoted region amplification
Starting from the 11th cycle the extension time was prolonged by 1s for each remaining cycle.
The reaction was ended by final incubation of 68°C for 10 min.
Detection P-values were obtained for every genomic position in every sample. Failed positions were defined as signal levels lower than background from both the methylated and unmethylated channels. Samples with non-significant detection p-values (> 0.01) in more than 10% of the probes were discarded. Using MixupMapper we also corrected for mixups in the methylation data (Westra et al. 2011). After SWAN normalization Beta-values were transformed to M-values: M = log 2 (beta/1-beta). Further, we applied a principal component analysis (PCA) on the methylation correlation matrix to correct for physiological or environmental variation (e.g. phenotype difference) as well as systematic experimental variation (e.g. batch and technical effect). In order to target the difference in the genetic variation of methylation, we removed the global variation in methylation by using the residual methylation for each probe after removing the optimal number of PCs. This optimum was determined based on the maximum number of cis-meQTLs.

Imputation of genome-wide genotyping data in EGCUT
Imputing of unobserved genotypes was implemented in IMPUTE v2 (Howie et al. 2009)  In subsequent data analysis the study subjects were divided into two groups based on the determined length of their microsatellite alleles ( Figure S4). Individuals carrying one or two extremely long (≥299 bp; HYPEST, n=18; CADCZ, n=29) alleles formed the 'long allele' group, subjects with two <299 bp alleles formed the 'normal allele' group. The effect of the microsatellite length on methylation level on neighboring CpG sites/unites was tested using logistic regression implemented in R software. Age and gender were used as covariates in all analyses. Length of the microsatellite showed no evidence of association with DNA methylation (data not presented). Numbering and the precise localization of CpG sites/units within CDH13 promoter region is provided in Figure S2. CpG sites 1-15 are located upstream from CpG island, sites 16-116 are within the CpG island and sites 18-26 are in the first exon. b Genomic position on chromosome 16 (GRCh37/hg19). c Effects and P-values are calculated using linear regression, including gender and experiment series as covariates to the model. Effect of age is determined as increase or decrease in CpG site methylation level in percentile scale (1-100) per annum (0= no methylation; 100= full methylation; positive beta-values= gain of methylation; negative beta-values= loss of methylation). d Results were combined using the inverse-variance method under fixed-effects model. Triglycerides -0.008632 0.008828 -0.025933 0.00867 0.32816 a Numbering and the precise localization of CpG sites/units within CDH13 promoter region is provided in Figure S2. CpG sites 1-15 are located upstream from CpG island, sites 16-116 are within the CpG island and sites 18-26 are in the first exon. b Genomic position on chromosome 16 is according to GRCh37/hg19. c Effects (methylation on scale from 0-100% effect on cardio-metabolic traits) and P -values are calculated using linear regression, including age, gender and experiment series as covariates to the model. d Results were combined using the inverse-variance method under fixed-effects model. Estimated Bonferroni threshold α=0.05/ [5 (independent parameters) x 46 (CpG sites/units) x 2 (HYPEST and CADCZ studies)]=1.09x10 -4 . SE, standard error; CI, confidence interval.