Metabolic profiling reveals heterogeneity within childhood obesity phenotypes

Background & Aims: Childhood obesity is a complex, heterogeneous disease. Children with obesity are often grouped into metabolically healthy obesity (MHO) and metabolically unhealthy obesity (MUO) phenotypes, although this classification may oversimplify the underlying metabolic diversity. We aimed to characterise the metabolic profiles associated with the conventional MHO/MUO classification and to identify data-driven metabolic profiles in 267 children and adolescents with obesity. Methods: In this cross-sectional study, we used serum 1H-NMR spectroscopy to quantify lipoproteins, glycoproteins, low-molecular-weight metabolites, and lipid fractions. Univariate comparisons and partial least squares discriminant analysis (PLS-DA) were used to characterise NMR-derived variables associated with MHO and MUO groups. We applied unsupervised k-means clustering based on principal component scores to identify data-driven metabolic profiles independent of the predefined clinical phenotypes. Results: Forty-one NMR-derived variables differed between MUO and MHO (FDR q-value < 0.05), predominantly in the triglyceride-rich lipoprotein and glycoprotein profiles, of which 31 were elevated in MUO. PLS-DA discriminated between MHO and MUO (AUC = 0.84, Q2 = 0.33; permutation p < 0.001). Unsupervised clustering identified two data-driven metabotypes that only partially overlapped with the clinical phenotypes: 27% of children with MHO clustered within the adverse metabolic profile (Cluster 2), characterised by a triglyceride-rich and inflammation-associated profile, whereas 41% of children with MUO were assigned to the comparatively favourable profile (Cluster 1). Clustering further revealed additional differences in lipoprotein subclasses and lipid composition that the MHO/MUO classification did not capture. Conclusions: The conventional MHO/MUO classification captures an important component of metabolic variation in childhood obesity but does not fully capture cardiometabolic heterogeneity. Quantitative 1H-NMR metabolomics identified putative metabotypes characterised by differences in lipoprotein composition and cholesterol distribution beyond clinical phenotypes. These findings support the potential value of metabolomic profiling to refine cardiometabolic risk stratification in children with obesity; however, the relevance of these data-driven metabotypes requires confirmation in independent cohorts and longitudinal follow-up studies.