Anis Engberg posted an update 4 months, 1 week ago
Fig. S1). The molecular identities of the most strongly replicating features for each locus were determined according to a workflow described in Materials and Methods (validation of metabolite ITF2357 cell line identifications are presented in Supp. Fig. S3). Molecular identification was also attempted for the five loci that did not replicate between the sample set 1 and sample set 2 results. Quantile–quantile (Q–Q) plots of all SNP associations at P < 10−4 in each of sample sets 1 and 2 are shown in Supp. Figure S4. Validation of the discovery of association of PYROXD2 with caprolactam is shown in Figure 1. All SNP-feature associations truncated a P < 10−7 in both samples 1 and 2 are provided as Supporting Information Excel spreadsheets. Although we were unable to replicate a number of previously described genes in the context of multiple testing in the present study, the ACADS and ACADM loci that were strongly highlighted in Illig et al. (2010) were regarded as the most important to pursue. We therefore scrutinized the specific markers previously highlighted, rs2014355 in ACADS and rs211718 in ACADM, to characterize the features they were associated with in our discovery sample set (sample set 1). For rs2014355 in ACADS, the strongest association was with a molecular feature with m/z 232.155, corresponding to the [M+H] of butyrylcarnitine and with a P value of 4.34 × 10−6 (ANOVA F-test with disease status and age as covariates and genotype as the main effect). For rs211718 in ACADM, the strongest association was with a molecular feature at m/z 260.182, corresponding to the [M+H] of hexanoylcarnitine with a P value of 3.04 × 10−6 (ANOVA F-test as above). For rs2014355 and rs211718, there were only 2 molecular features each with P < 1 × 10−4. Gieger et al. (2008) introduced a strategy involving genetic association using metabolite ratios, under the supposition that ratios would serve as proxies for enzymatic reactions. We attempted a similar strategy here for three genes, ACADS, ACADM, and FADS1 (rs2014355 in ACADS, rs211718 in ACADM, and rs174549 in FADS1). We sought an improvement in association using ANOVA (F-test) with genotype as a main effect, the strongest associated phenotype as above as the dependent variable (butyrylcarnitine for ACADS, hexanoylcarnitine for ACADM, and glycerolphosphocholine [18:0/20:4] for FADS1) and with each of the 6,137 remaining traits as covariates. This analysis was conducted in sample set 1 only. For FADS1, we observed the strongest improvement with adjustment for a molecular feature at m/z 799.605 (possibly corresponding to C22:2-OH sphingomyelin) with a genotype main effect P = 6.81 × 10−34. For ACADS (rs2014355), the best association was found when a molecular feature at m/z 114.068 (possibly corresponding to creatinine) was added as a covariate with P = 3.27 × 10−11. For ACADM (rs211718), the best association was observed for a feature at m/z 370.