Integrating Metabolomics and Gut Microbiota to Identify Key Biomarkers and Regulatory Pathways Underlying Metabolic Heterogeneity in Childhood Obesity

Abstract

Background/objectives: Individuals with childhood obesity exhibit significant metabolic heterogeneity, necessitating precise biomarkers for risk stratification and assessment. This multi-omics investigation characterizes metabolic and microbial signatures underlying divergent metabolic phenotypes in the context of pediatric obesity.

Methods: We analyzed 285 Chinese children (5-7 years) stratified into five groups: wasting (WAS, n = 55), metabolically healthy/unhealthy and normal weight (MHWH, n = 54; MUWH, n = 67), and metabolically healthy/unhealthy obesity (MHO, n = 36; MUO, n = 73). Untargeted metabolomics (Orbitrap ID-X Tribrid™) and 16S rRNA sequencing were integrated with multivariate analyses (OPLS-DA with VIP > 1, FDR < 0.05; Maaslin 2 with TSS normalization and BH correction, FDR < 0.10).

Results: Analysis identified 225 differential metabolites and 12 bacterial genera. The proportion of steroids and their derivatives among differential metabolites in the MUO/MHO group was significantly lower than that in the OVOB/NOR and OVOB/WAS groups (2.12% vs. 7.9-14.1%). MUO displayed elevated C17 sphinganine and LysoPC (O-18:0) levels but reduced PI (16:0/14:1) levels. In contrast, OVOB showed upregulated glycerol phospholipids (LPCs and PSs) and downregulated PE species (e.g., PE(16:0/16:0)) as well as gut microbiota dysbiosis characterized by a higher Firmicutes/Bacteroidetes (F/B) ratio (2.07 vs. 1.24 in controls, p = 0.009) and reduced α diversity (Ace index, Chao1 index, and Shannon index values were lower in the OVOB group, Shannon index: 2.96 vs. 3.45, p = 0.03). SCFA-producing genera were negatively correlated with the OVOB group, while positively associated with PE(16:0/16:0). Internal validation showed differential metabolites had potential predictive efficacy for MUO/MHO (AUC = 0.967) and OVOB/NOR (AUC = 0.888).

Conclusions: We identified distinct lipid disruptions characterizing obesity subtypes, including steroid/terpene deficits and sphingolipid/ether lipid dysregulation in the MUO/MHO groups as well as phospholipid imbalance (↑LPC/PS↓PE) in the OVOB/NOR groups. The gut microbiota exhibited a profile characterized by low diversity, an increased F/B ratio, and a reduced abundance of SCFA-producing genera. These findings suggest potential biomarkers for childhood obesity stratification, though further validation is warranted.

Keywords: childhood obesity; gut microbiota; metabolic heterogeneity; metabolomics; multi-omics.

Figure A1

PCA scores for positive and negative (left and right, respectively) ion modes (including QC).

Figure 1

Research flow design chart.

Figure 2

OPLS-DA model and permutation testing results of samples from comparison groups. Columns 1 and 2 are the results of metabolite group difference analysis using OPLS-DA and model permutation testing under positive ion modes, respectively. (a) Description of the results of the OPLS-DA and permutation tests for the OVOB/NOR comparison group; (b) description of the results of the OPLS-DA and permutation tests for the OVOB/WAS comparison group; (c) description of the results of the OPLS-DA and permutation tests for the MUO/MHO comparison group.

Figure 3

Class distribution of differential metabolites among comparative groups. Differential metabolites were analyzed using OPLS-DA and screened according to VIP ≥ 1, FDR < 0.05, FC > 1.2, or FC < 0.83. The primary classification of human metabolites was performed using The Human Metabolome Database (HMDB Version 5.0). (a) The distribution of differential metabolites in all groups; (b) the distribution of differential metabolites in the OVOB/NOR group; (c) the distribution of differential metabolites in the OVOB/WAS group; and (d) the distribution of differential metabolites in the MUO/MHO group.

Figure 4

Heatmap analysis of different metabolites between comparative groups. Differential metabolites were screened for heatmap analysis across groups. Metabolites differentiating OVOB/NOR and OVOB/WAS were defined by an FC > 1.5 or FC < 0.67 (VIP ≥ 1, FDR < 0.05). Metabolites differentiating MUO/MHO were defined by an FC > 1.2 or FC < 0.83 (VIP ≥ 1, FDR < 0.05).

Figure 5

Pathway enrichment analysis of differential metabolites. Differential metabolites identified based on VIP ≥ 1, FDR < 0.05, and FC > 1.5/<0.67 were analyzed using MetaboAnalyst 6.0 to assess pathway enrichment. (a) Overview of metabolite enrichment sets and bubble diagram for OVOB/NOR, (b) OVOB/WAS, and (c) MUO/MHO.

Figure 6

Importance ranking of differential metabolites and efficacy of predicting OVOB. (a) Random forest model for importance ranking of metabolites. The random forest algorithm quantified variable importance based on a mean decrease in the Gini index. (b) The combined efficacy of weight, nation, CRP_H, VA, and the identified metabolites in predicting OVOB in children had an AUC of 0.888 (95% CI: 0.847, 0.928), with a sensitivity of 76.0% and a specificity of 86.2%. CRP_H, C-reactive protein; Reg, retinyl beta-glucuronide; Choe, 17:0 cholesteryl ester; CycE, cyclocalopin E; LysoPC14_1, LysoPC (14:1); VA, vitamin A; Corace, cortisone acetate; Sulsul, sulfinpyrazone sulfide; weight, birth weight.

Figure 7

Importance ranking of differential metabolites and efficacy of predicting MUO. (a) Random forest model for importance ranking of metabolites. The random forest algorithm quantified variable importance based on a mean decrease in the Gini index. (b) The combined efficacy of ALT, CRP_H, and the metabolites in predicting MUO in obese children achieved an AUC of 0.967 (95% CI: 0.936, 0.998), with a sensitivity of 88.9% and a specificity of 95.9%. ALT, alanine aminotransferase; CRP_H, C-reactive protein; PI16_14, PI (16:0/14:1); Tresul, treosulfan; Forkyn, N′-formylkynurenine; LysoPE16_0, LysoPE (16:0/0:0); Elea, elenaic acid; Hexgly, hexaethylene glycol; PS20_0, PS (P-20:0/0:0).

Figure 8

Relative abundance of gut microbiome in comparative groups. (a) Phylum level; (b) genus level.

Figure 9

Analysis of relative abundance composition of gut microbiota in different groups. The Kruskal–Wallis test was used to analyze the difference in the F/B ratio and Verrucomicrobia abundance among the three groups.

Figure 10

Box diagram of α diversity of gut microbiota in different groups. (a–d) represents the difference in the Ace, Chao1, Shannon, and Simpson indices among the three groups, respectively, and the difference was statistically significant among the three groups (p < 0.05). Kruskal–Wallis test was used to analyze the differences in the Ace, Chao1, Shannon, and Simpson indices. * indicates that the difference is statistically significant (p < 0.05), ** indicates that the difference is statistically significant (p < 0.01).

Figure 11

MaAsLin 2 analysis results of differential bacteria between comparative groups. The FDR of Colidextribacter, Dysosmobacter, and Intestinimonas was 0.02, 0.02, and 0.02, respectively; the FDR of Alistipes, Longicatena, and Butyricimonas was 0.06, 0.08, and 0.09, respectively. Associations with FDR < 0.10 were considered exploratory given the small sample size and should be interpreted with caution.

Figure 12

Spearman association between different metabolites and different bacteria in the OVOB/NOR group comparison. Square size represents the absolute value of the correlation coefficient (larger size = stronger association). * indicates that the correlation is statistically significant (p < 0.05), ** indicates that the correlation is statistically significant (p < 0.01).