Metabolic profiling was done on fasting serum from participants of the German KORA F4 study (n=1,768) and the British TwinsUK study (n=1,052) using ultrahigh performance liquid-phase chromatography and gas chromatography separation coupled with tandem mass spectrometry 5 (link)-7 (link). We achieved highly efficient profiling (24 minutes/sample) with low median process variability (<12%) of more than 250 metabolites, covering over 60 biochemical pathways of human metabolism (Supplemental Table 1). Based on our previous observation that ratios between metabolite concentrations can strengthen the association signal and provide new information about possible metabolic pathways 4 (link),8 (link), we included all pairs of ratios between these metabolites in the genome-wide statistical analysis. To reduce the computational and data storage burden associated with meta-analyzing over 37,000 metabolites and ratios, we applied a staged approach for selection of promising association signals (Supplemental Figure 1). In the initial screening stage we assessed associations of approximately 600,000 genotyped SNPs with over 37,000 metabolic traits (concentrations and their ratios) by fitting linear models separately in both cohorts to log-transformed metabolic traits, adjusting for age, gender and family structure (Supplemental Figure 2 & Supplemental Table 2). Next, we selected all association signals having suggestive evidence for association with a metabolic trait in both cohorts (p<10−6 in both cohorts or p<10−3 in one and p<10−9 in the other). For each of these loci, we then re-assessed the amount of association signals through fixed-effects inverse variance meta-analysis of the two cohorts for all 37,000 available traits using imputed SNPs relative to HapMap2 data (see Online Methods for details). The SNP/trait combination yielding the smallest P-value in this meta-analysis was finally selected for each locus. To account for multiple testing we applied conservative Bonferroni correction leading to an adjusted threshold for genome-wide significance of p < 2.0×10−12.