Methods Study populations We selected non-osteoporotic participants who were tested via dual-energy X-ray absorptiometry from the Center for Health Promotion and Disease Prevention of Seoul National University Hospital and the Gangnam Center for Health Promotion. The subjects underwent regular health check-ups from May 2009 to December 2013. Our sample consisted exclusively of Korean men aged between 20 and 60 years. This initial step of the procedure was performed to detect genetic defects associated with obesity and metabolic syndrome [1819]. Those with conditions that could affect body weight were excluded from the study. We also excluded individuals who had taken any medication that might influence BMD, such as bisphosphonate, within the last 3 months or had a history of surgery or foreign body insertion at the spine or femur. After applying the above criteria, 1,192 subjects were selected. After excluding 16 participants who were related to each other, our final sample consisted of 1,176 individuals. Phenotype measurement Dual-energy X-ray absorptiometry (Prodigy Advance; General Electric Company, Fairfield, CT, USA) measurements at the lumbar spine and femur were obtained from all patients, following standard protocols. We distinguished and analyzed three BMD phenotypes—lumbar spine (L1-L4) (LS-BMD), femur neck (FN-BMD), and femur total (FT-BMD)—identified here as the average value of mineral density at the neck and greater trochanter of the femur bone. Genotyping and quality control All participants underwent a microarray assay, using the Illunina HumanCore BeadChip kit (Illumina, San Diego, CA, USA). We filtered out poor quality single nucleotide polymorphisms (SNPs) and samples, which were defined as SNPs with a monographic genotype, a Hardy-Weinberg equilibrium p < 10-6, a call rate <95%, and a minor allele frequency <1%. After quality control, the remaining 264,283 SNPs were used in subsequent analysis. Genotype imputation All missing genotypes were imputed in order to examine any potential effect of the unmeasured variants. We used 1,000 Genomes ASN Phase I integrated variant set release (v3) in National Center for Biotechnology Information (NCBI) build 37 (hg19) as a reference panel, along with PLINK software (version 1.07) [20] for strand alignment, SHAPEIT2 [21] for phasing, and IMPUTE2 [22] for the imputation procedure. Only variants with Info Score ≥0.9 were included in the final count, resulting in a total of 4,414,664 SNPs, which were either genotyped or imputed with high confidence. Statistical analyses Multivariate linear regressions, adjusted for the effects of age and body mass index (BMI, weight/height2, kg/m2) [23] via an additive model, were used to further investigate the influence of SNPs on BMD. Genome-wide association patterns were analyzed by the PLINK software package version 1.07 [20], while quantile-quantile and Manhattan plots were obtained by R software 3.2.2 (2015 The R Foundation for Statistical Computing). Regional association plots of the significant SNPs were generated by LocusZoom software [24]. All results with p < 1 × 10-6 were considered statistically significant. Association with previously identified loci We attempted to determine similarities between our selected markers and five other genes, which have previously been identified as indicative of osteoporosis in Korean male populations. Tumor necrosis factor receptor superfamily member 11 (TNFSF11) on 13q14, FOXL1 on 16q24, LRP5 on 11q13.4, ZBTB40 on 1q31.3, and MEF2C on 5q14 have been connected to BMD in several different genome-wide association studies (GWASs)—an association confirmed in Korean males by a number of replication studies [1314]. We examined the most significant SNP regions of these five genes by applying our markers and considering only results with a p < 0.05. Ethics statement The study protocol was approved by the Institutional Review Board at Seoul National University Hospital (IRB No. H-0911-010-299) and all participants provided written informed consent.