Introduction Testosterone, the most important testicular androgen in men, is largely bound to two plasma proteins. Most of the circulating testosterone (∼50–60%) is bound with high affinity to sex hormone-binding globulin (SHBG), while a smaller fraction (40–50%) is bound loosely to albumin, and 1–3% is unbound and termed free testosterone [1]. In prospective cohort studies, low serum testosterone concentrations are associated with cardiovascular morbidity, metabolic syndrome [2], [3], dyslipidemia [4], hypertension [5], type 2 diabetes mellitus [6], stroke [7], atherosclerosis [8]–[10], osteoporosis, sarcopenia, and increased mortality risk [11]–[13]. Thus, there is growing evidence that serum testosterone is a valuable biomarker of men's overall health status. Since age, body mass index (BMI), and smoking are known to affect serum testosterone concentrations [14], we used these parameters as common set of covariates in all association models. Studies in male twins indicate that there is a strong heritability of serum testosterone, with genetic factors accounting for 65% of the variation in serum testosterone [15]. However, the genetic determinants of serum testosterone and the genetic risk factors for low concentrations are poorly understood. Given the current gap in knowledge of the genetic factors that contribute to the inter-individual variability in serum testosterone concentration in men we conducted a meta-analysis of genome-wide association studies (GWAS). This two-stage meta-analysis included data from 14,429 Caucasian men from 10 independent cohorts within the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium. In stage one, the discovery stage, genome-wide association data from seven cohorts were meta-analyzed (n = 8,938) and all genome-wide significant findings that fulfilled the criteria described in the methods section were followed up in the three replication cohorts: one in silico replication cohort (n = 871) and two replication cohorts with de novo genotyping (n = 4,620). All association analyses of the discovery stage were conducted both with and without additional adjustment for serum SHBG concentrations. Our primary aim was to identify genetic variants reproducibly associated with serum testosterone concentrations in men, evaluated as a continuous trait. We also assessed whether the lead single-nucleotide polymorphisms (SNPs) from the continuous trait analyses had a significant influence on the risk of having low serum testosterone, defined as 0.4 at the individual cohort level during meta-analysis. 10.1371/journal.pgen.1002313.t001 Table 1 Meta-analyses of discovery and replication cohorts. SNPs rs12150660 and rs6258 (on chromosome 17 in SHBG) identified in GWAS for total testosterone Discovery Replication Combined A1/A2 FREQ* beta se p n beta se p n beta se p n Testosterone (ng/dl) rs12150660 T/G 0.23 26.4 3.1 1.9E-17 8938 38.8 3.6 2.3E-27 5429 31.8 2.3 1.2E-41 14367 rs6258 T/C 0.02 −74.7 9.9 4.1E-14 8938 −102.9 16.3 2.9E-10 5483 −82.3 8.5 2.3E-22 14421 SHBG (nmol/l) rs12150660 T/G 0.23 3.6 0.3 3.0E-42 8366 4.4 0.4 8.5E-36 5682 3.9 0.2 2.1E-75 14048 rs6258 T/C 0.02 −6.6 0.8 1.2E-15 8366 −9.5 1.3 6.7E-14 5733 −7.4 0.7 3.5E-27 14099 Testosterone (SHBG-adjusted) rs12150660 T/G 0.23 11.1 3.0 2.5E-04 8366 11.6 3.0 9.9E-05 5414 11.3 2.1 9.0E-08 13780 rs6258 T/C 0.02 −41.8 9.4 8.2E-06 8366 −33.2 13.8 1.6E-02 5467 −39.1 7.7 4.5E-07 13833 Calculated Free Testosterone (ng/dl) rs12150660 T/G 0.23 −0.1 0.1 9.6E-02 8366 0.1 0.1 1.6E-02 5414 0.0 0.0 3.9E-01 13780 rs6258 T/C 0.02 −0.2 0.2 3.2E-01 8366 −0.5 0.3 9.0E-02 5467 −0.3 0.2 6.5E-02 13833 SNP rs5934505 (on chromosome X near FAM9B) identified in GWAS for SHBG-adjusted total testosterone Discovery Replication Combined A1/A2 FREQ* beta se p n beta se p n beta se p n Testosterone (ng/dl) C/T 0.26 14.1 3.2 1.1E-05 5067 27.2 6.0 5.4E-06 3816 17.0 2.8 1.7E-09 8883 SHBG (nmol/l) C/T 0.26 −0.2 0.3 5.9E-01 4607 0.5 0.7 4.7E-01 4072 −0.1 0.3 8.5E-01 8679 Testosterone (SHBG-adjusted) C/T 0.26 18.1 3.1 8.5E-09 4599 27.7 4.7 4.4E-09 3801 21.0 2.6 5.6E-16 8400 Calculated Free Testosterone (ng/dl) C/T 0.26 0.4 0.1 4.0E-07 4607 0.6 0.1 8.7E-10 3801 0.5 0.1 6.7E-15 8408 Effects size is given per minor allele. All seven discovery cohorts (n = 8,938) were included in the GWAS of chromosomes 1–22 while only the two largest cohorts (FHS and SHIP. n = 5,067) had GWAS data available for the X chromosome. A1 = allele 1. A2 = allele 2. FREQ* = Frequency of allele 1. In the KORA cohort, testosterone was measured using plasma but the analyses after excluding KORA yielded similar results. Calculated free testosterone was calculated for all subjects with both testosterone and SHBG available by using a modified law of mass action equation. The concentrations of testosterone and SHBG and a fixed value for SHBG's dissociation constant were used in these calculations. Replication of autosomal hits The associations of rs12150660 and rs6258 were confirmed in the three replication cohorts (in silico replication in YFS and de novo replication in MrOS Sweden and EMAS), demonstrating a combined p-value in the discovery and the replication cohorts of 1.2×10−41 and 2.3×10−22, respectively (Table 1 [SNPs rs12150660 and rs6258]). Both SNPs showed considerable heterogeneity of results across the studies as measured by the I 2 statistic [18]. The I 2 values for the discovery meta-analysis using the untransformed total testosterone values were 76.7% and 81.6% for rs12150660 and rs6258, respectively. The heterogeneity was reduced to 39.3% and 75.5% for rs12150660 and rs6258, respectively, by meta-analysing the z-score based untransformed total testosterone values and to 30.9% and 78.0%, respectively, by meta-analysing the inverse-normal transformed testosterone values. For rs12150660, a substantial amount of heterogeneity could be explained by phenotypic variation among the cohorts, whereas for rs6258 one cohort (InCHIANTI) showed consistent opposite effect directions in all models used. To take into account this heterogeneity, we additionally calculated a random effects model for untransformed total testosterone values. The association for rs12150660 remained genome-wide significant in the combined discovery and replication stage meta-analysis, the association for rs6258 reached genome-wide significance after excluding the InCHIANTI cohort (Table S3). The genetic influence on low serum testosterone concentrations In Table 2, the serum testosterone concentrations according to genotype are given for the three replication cohorts. As expected, mean serum testosterone concentrations were found to be lower in men with GG than in those with TT genotype for rs12150660. Similarly, men with the CT genotype for rs6258 had lower serum testosterone concentrations than those with CC genotype. The TT genotype of rs6258 was extremely rare and only found in two subjects in the replication cohorts. The two autosomal SNPs identified by GWAS had a significant influence on the risk of having low serum testosterone (serum testosterone 0.4 at the individual cohort level during meta-analysis. All seven discovery cohorts (n = 8,938) were included in the GWAS of chromosomes 1–22 while only the two largest cohorts (FHS and SHIP, n = 5,067) had GWAS data available for the X chromosome. (PDF) Click here for additional data file. Figure S2 Quantile-quantile plot of the genome-wide association results of the inverse-variance weighted meta-analysis of untransformed serum testosterone including all SNPs (black) and after removal of the SNPs of the SHBG locus (blue). (PDF) Click here for additional data file. Figure S3 Associations for (A) rs12150660 and (B) rs6258 with testosterone and for (C) rs5934505 with SHBG-adjusted testosterone. Effects sizes are given per minor allele. Beta estimates and their 95% confidence intervals are given. The size of the data markers is proportional to the weight (inverse of the variance) of each study. (PDF) Click here for additional data file. Figure S4 Risk of low serum testosterone concentrations (serum testosterone <300 ng/dl), according to the number of combined risk alleles for rs12150660 (G = risk allele) and rs6258 (T = risk allele) in the three replication cohorts (MrOS Sweden, EMAS, and YFS). Bars indicate 95% confidence intervals. Only two individuals in the three replication cohorts had four risk alleles and therefore individuals with three and four risk alleles were pooled into one group with ≥3 risk alleles. Two risk allele counts were used as reference, since this is the most prevalent amount among the cohorts. (PDF) Click here for additional data file. Figure S5 Subjects heterozygous for the SHBG allele containing an rs6258 SNP have lower serum SHBG steroid-binding capacity (Y-axis) when compared to the concentrations of SHBG measured by immunoassay (X-axis). Serum SHBG concentrations from 10 individuals homozygous for the wild type SHBG allele (CC, dashed regression line r2 = 0.872) or heterozygous for the rs6258 variant SHBG allele (CT, solid regression line r2 = 0.866) were measured by a time-resolved immunofluorometric assay[33], and a steroid-binding capacity assay using [3H]DHT as the labelled ligand.[34] (PDF) Click here for additional data file. Table S1 Characteristics of 14,429 men from 10 cohorts included in the genome-wide association study meta-analysis. (PDF) Click here for additional data file. Table S2 Additional genotyping information for the 10 cohorts included in the genome-wide association study meta-analysis. (PDF) Click here for additional data file. Table S3 Meta Analysis of untransformed total testosterone using Random Effect Model. (PDF) Click here for additional data file. Text S1 Supplemental methods. (DOC) Click here for additional data file.
