Untargeted Metabolomics Analysis of the Serum Metabolic Signature of Childhood Obesity

Abstract

Obesity rates among children are growing rapidly worldwide, placing massive pressure on healthcare systems. Untargeted metabolomics can expand our understanding of the pathogenesis of obesity and elucidate mechanisms related to its symptoms. However, the metabolic signatures of obesity in children have not been thoroughly investigated. Herein, we explored metabolites associated with obesity development in childhood. Untargeted metabolomic profiling was performed on fasting serum samples from 27 obese Caucasian children and adolescents and 15 sex- and age-matched normal-weight children. Three metabolomic assays were combined and yielded 726 unique identified metabolites: gas chromatography-mass spectrometry (GC-MS), hydrophilic interaction liquid chromatography coupled to mass spectrometry (HILIC LC-MS/MS), and lipidomics. Univariate and multivariate analyses showed clear discrimination between the untargeted metabolomes of obese and normal-weight children, with 162 significantly differentially expressed metabolites between groups. Children with obesity had higher concentrations of branch-chained amino acids and various lipid metabolites, including phosphatidylcholines, cholesteryl esters, triglycerides. Thus, an early manifestation of obesity pathogenesis and its metabolic consequences in the serum metabolome are correlated with altered lipid metabolism. Obesity metabolite patterns in the adult population were very similar to the metabolic signature of childhood obesity. Identified metabolites could be potential biomarkers and used to study obesity pathomechanisms.

Keywords: childhood obesity; lipidomics; obesity biomarkers; obesity pathogenesis; obesity pathomechanisms; untargeted metabolomics.

Figure 1

Volcano plot of metabolites across groups, with log-transformed adjusted p-values and fold changes. Red circles represent metabolites with increased expression in the obesity group. Blue circles represent metabolites with decreased expression in subjects with obesity. Grey circles represent non-significant metabolites (adjusted p-values ≥ 0.05). The top 15 top significant metabolites are labeled. SM, sphingomyelin; PC, phosphatidylcholine; CE, cholesteryl ester; TG, triacylglycerol; 9,10-EpOME, 9,10-epoxyoctadecenoic acid; PI, phosphatidylinositol.

Figure 2

Metabolite classes with significantly different concentrations between patient groups.

Figure 3

Box-plots of selected metabolites with significantly different concentrations between children with (green) and without (red) obesity. Fold changes and p-values are provided in Table S1. (A) Sphingomyelin (SM) d36:1 (d18:1/18:0); (B) Phosphate; (C) Phosphatidylcholine (PC) 40:6 (18:1/22:5); (D) Cholesteryl eicosapentaenoic acid (CE (20:5); (E) Triacylglycerol (TG) 58:8; (F) 9,10-epoxyoctadecenoic acid (9,10-EpOME).

Figure 4

Box-plots of BCAAs with significantly different concentrations between children with (green) and without (red) obesity. Fold changes and p-values are provided in Table S1. (A) Leucine; (B) Isoleucine; (C) Valine.

Figure 5

Univariate ROC curve results for phosphate and sphingomyelin SM (d18:1/18:0). (A,B) ROC curves for (A) phosphate (A) and (B) SM (d18:1/18:0). Sensitivity and specificity are shown on the y- and x-axes, respectively. The area-under-the-curve (AUC) is in blue, and 95% CIs are shown. (C,D) Box-plots of (C) phosphate and (D) SM (d18:1/18:0) between children with (green) and without (red) obesity. The red horizontal line represents the optimal cut-off.

Figure 6

PCA analysis between patient groups. (A) Two-dimensional (2D) score plots between PC1 and PC2. Patients with obesity are shown in green, and those without obesity are shown in red. (B) Scree plot showing the variance explained by PCs 1−5.

Figure 7

OPLS-DA analysis between patient groups. (A) Score plot of all metabolite features. (B) Permutation analysis with observed and cross-validated R2Y and Q2 coefficients. (C) Important metabolites identified by OPLS-DA. Colored boxes on the right indicate metabolite concentrations in each patient group. SM, sphingomyelin; PC, phosphatidylcholine; CE, cholesteryl ester; TG, triacylglycerol; 9,10-EpOME, 9,10-epoxyoctadecenoic acid.

Figure 8

Model performance of six SVM classifiers with an increasing number of metabolites. (A) ROC curves for each SVM classifier, based on average cross-validation performance. AUCs and 95% CIs are presented in the figure legend. (B) Predictive accuracy for each SVM. The model with the highest accuracy is highlighted in red.

Figure 9

Variable importance from the SVM model with 25 metabolites. Metabolites are ranked from most to least important. The colored boxes on the right indicate metabolite concentrations in each patient group. SM, sphingomyelin; PC, phosphatidylcholine; PI, phosphatidylinositol; CE, cholesteryl ester; TG, triacylglycerol; 9,10-EpOME, 9,10-epoxyoctadecenoic acid; LPC, lysophosphatidylcholine; FA, fatty acid; Cer, ceramide; G-Glu-Glu, gamma-glutamyl-glutamic acid.

Figure 10

Heatmap of the Pearson correlation coefficient matrix: (A) Correlations between selected metabolites and (B) between selected metabolites and clinical variables. Red and blue indicate negative and positive correlations, respectively. Color intensity indicates the absolute correlation value. Empty cells indicate non-significant correlations (adjusted p-values > 0.05). SM, sphingomyelin; PC, phosphatidylcholine; CE, cholesteryl ester; TG, triacylglycerol; 9,10-EpOME, 9,10-epoxyoctadecenoic acid; BMI, body mass index; LDL, low-density lipoprotein cholesterol; SBP, systolic blood pressure.