Metabolome-Wide Association Study of Multiple Plasma Metals with Serum Metabolomic Profile among Middle-to-Older-Aged Chinese Adults

Abstract

Metal exposure has been associated with risk of various cardio-metabolic disorders, and investigation on the association between exposure to multiple metals and metabolic responses may reveal novel clues to the underlying mechanisms. Based on a metabolome-wide association study of 17 plasma metals with untargeted metabolomic profiling of 189 serum metabolites among 1992 participants within the Dongfeng-Tongji cohort, we replicated two metal-associated pathways, linoleic acid metabolism and aminoacyl-tRNA biosynthesis, with novel metal associations (false discovery rate, FDR < 0.05), and we also identified two novel pathways, including biosynthesis of unsaturated fatty acids and alpha-linolenic acid metabolism, as associated with metal exposure (FDR < 0.05). Moreover, two-way orthogonal partial least-squares analysis showed that five metabolites, including aspartylphenylalanine, free fatty acid 14:1, uridine, carnitine C14:2, and LPC 18:2, contributed most to the joint covariation between the two data matrices (12.3%, 8.3%, 8.0%, 7.4%, and 7.3%, respectively). Further BKMR analysis showed significant positive joint associations of plasma Al, As, Ba, and Zn with aspartylphenylalanine and of plasma Ba, Co, Mn, and Pb with carnitine C14:2, when all the metals were at the 55th percentiles or above, compared with the median. We also found significant interactions between As and Ba in the association with aspartylphenylalanine (P for interaction = 0.048) and between Ba and Pb in the association with carnitine C14:2 (P for interaction < 0.001). Together, these findings may provide new insights into the mechanisms underlying the adverse health effects induced by metal exposure.

Keywords: Environmental health; Metabolome-wide association study; Metal−metal interaction; Multiple plasma metals; Untargeted metabolomics.

Figure 1

Scatter plot demonstrating the associations between 17 metals and 189 metabolites. P values were derived from single-metal multivariable linear regression models which adjusted for age, gender, BMI, smoking status, drinking status, and physical activity. False discovery rate (FDR) correction was performed across 3213 association tests between 17 metals and 189 metabolites.

Figure 2

Metabolic pathways associated with plasma metals. Sizes of circles were mapped to the number of significant metabolites in the pathways, and color was mapped to the FDR values (FDR < 0.05).

Figure 3

O2PLS modeling of metal–metabolite associations. The number of components included in the O2PLS model were selected by 10-fold cross-validation, and the model contains 2 joint, 10 X-orthogonal, and 10 Y-orthogonal components. (A) Loading plot for plasma metals. (B) Loading plot for serum metabolites. In this display, the relative position of each point in the two panels indicates whether regression coefficients for a given pair of metals and metabolites positively or negatively correlates to each other, while the sum of the squared loadings on the two joint components (SS_joint) indicates the magnitude of the association of the metal (metabolite) with metabolome (metallome). All 17 metals and the top 30 metabolites are labeled. O2PLS: two-way orthogonal partial least-squares.

Figure 4

Joint associations of metal mixtures with metabolites by two-stage Bayesian kernel machine regression (BKMR) analyses. The figure plots the estimated change in metabolite concentrations when all the metals at particular percentiles (x-axis) were compared to all the metals at their 50th percentile. The BKMR model adjusted for age, gender, BMI, smoking status, drinking status, and physical activity. For aspartylphenylalanine, the BKMR model included a mixture of Al (PIP = 0.09), As (PIP = 0.16), Ba (PIP = 0.37), and Zn (PIP = 0.99). For carnitine C14:2, the BKMR model included a mixture of Ba (PIP = 0.65), Co (PIP = 0.18), Mn (PIP = 0.05), and Pb (PIP = 0.88).

Figure 5

Bivariate exposure–response functions of each metal (shown by column) and metabolite concentration by two-stage Bayesian kernel machine regression (BKMR) analyses, holding concentrations of one metal (shown by row) at different quantiles (25th, 50th, and 75th) and other metals at medians. The BKMR model adjusted for age, gender, BMI, smoking status, drinking status, and physical activity. For aspartylphenylalanine, the BKMR model included a mixture of Al, As, Ba, and Zn. For carnitine C14:2, the BKMR model included a mixture of Ba, Co, Mn, and Pb. The multivariable linear regression models validated the interactions between As and Ba (P for interaction = 0.048) and Ba and Zn (P for interaction = 0.15) for aspartylphenylalanine and the interactions between Ba and Pb (P for interaction < 0.001), Ba and Co (P for interaction = 0.39), and Co and Pb (P for interaction = 0.30) for carnitine C14:2.