Methods

Study materials
We used Affymetrix Genomewide SNP 5.0 genotyping data provided by the Korea Association Resource (KARE) consortium, Korean Genome Epidemiology Study (KoGES). As a control, we purchased a HapMap cell line, GM10851, from Coriell Institute for Medical Research (Camden, NJ, USA) and extracted genomic DNA using the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany).

Pre-processing SNP array data
Before CNV calling procedures, all required pre-processing procedures, including allele correction, summarization, and background correction, were performed as described previously by using the software provided by Affymetrix [5]. In brief, background-corrected data were normalized using quantile normalization and summarized by median polish.

CNV calling
For this study, Affymetrix 5.0 SNP array (Affymetrix, Santa Clara, CA, USA) data of 10 subjects were randomly selected from the KARE dataset for CNV calling. Affymetrix 5.0 data of NA10851 was used as a control. For defining CNVs, we choose 3 different segmentation algorithms, PennCNV [25], QuantiSNP [26] and BirdSuite [27]. PennCNV implements a hidden Markov model (HMM) and considers SNP allelic ratio distribution in addition to signal intensity. We only used CNVs from PennCNV with samples that had a standard deviation of log R ratio (LRR) smaller than 0.2, drift values of b-allele frequency (BAF_drift) smaller than 0.01, and waviness factor between -0.05 and 0.05. For QuantiSNP, we did not apply any criteria for sample filtering, since QuantiSNP uses an Objective Bayes Hidden-Markov Model (OB-HMM) for calibrating the model for fixing the false positive error rate and maximum marginal likelihood to set other hyperparameters. It was originally developed for detecting CNVs from BeadArray SNP genotyping data of Illumina, but Affymetrix data also could be processed as long as they have probe signal intensities and B-allele frequencies. Birdsuite uses exclusive copy number analysis routine (Canary) for CNP locus and HMM for finding rare CNVs. It generates a logarithm of the odds ratio (LOD) value for each CNV that describes the likelihood of a CNV relative to no CNV over a given interval, including flanking sequences. Of the BirdSuite calls, only the CNVs that have an LOD value smaller than 2 were used for further analysis. All 3 methods were used with default values, and no other option was added or changed.

Validation
To estimate the false discovery rate of our CNV-calling algorithm, CNVs were validated by genomic real-time qPCR. For this purpose, we randomly selected 25 CNV loci and performed genomic qPCR using DNAs from study subjects who showed corresponding CNVs on that loci. The PCR primers used in this study were designed using the PrimerQuest program (http://www.idtdna.com/Scitools/Applications/Primerquest). To verify the specificity of the PCR reactions under the unified denaturation temperature (60℃), we performed PCR and agarose gel electrophoresis for each primer set. We also screened the University of California Santa Cruz (UCSC) database (http://genome.ucsc.edu/) to confirm the unique sequence without any repeat sequences in the primers. Sequence information of the primers for genomic qPCR validation is listed in Table 1.
Ten microliters of the reaction mixture contained 10 ng of genomic DNA, SYBR Premix Ex Taq TM II (TaKaRaBio, Shiga, Japan), 1×ROX (Toyobo, Osaka, Japan), and 10 pmol of primers. Thermal cycling conditions consisted of 1 cycle of 10 s at 95℃, followed by 40 cycles of 5 s at 95℃, 10 s at 61℃, and 20 s at 72℃. All PCR experiments were repeated twice, and amplification efficiencies for both target and reference genes were evaluated using a standard curve over 1:5 serial dilutions. The copy number of each target was defined as 2-ΔΔCT, where ΔCt is the difference in threshold cycles for the sample in question normalized against the reference gene (RNaseP) and expressed relative to the value obtained by calibrator DNA (NA10851 and Promega DNA), as described elsewhere [28].