Global metabolomic profiling targeting childhood obesity in the Hispanic population^,

Abstract

Background: Metabolomics may unravel important biological pathways involved in the pathophysiology of childhood obesity.

Objectives: We aimed to 1) identify metabolites that differ significantly between nonobese and obese Hispanic children; 2) collapse metabolites into principal components (PCs) associated with obesity and metabolic risk, specifically hyperinsulinemia, hypertriglyceridemia, hyperleptinemia, and hyperuricemia; and 3) identify metabolites associated with energy expenditure and fat oxidation.

Design: This trial was a cross-sectional observational study of metabolomics by using gas chromatography–mass spectrometry and ultrahigh-performance liquid chromatography–tandem mass spectrometry analyses performed on fasting plasma samples from 353 nonobese and 450 obese Hispanic children.

Results: Branched-chained amino acids (BCAAs) (Leu, Ile, and Val) and their catabolites, propionylcarnitine and butyrylcarnitine, were significantly elevated in obese children. Strikingly lower lysolipids and dicarboxylated fatty acids were seen in obese children. Steroid derivatives were markedly higher in obese children as were markers of inflammation and oxidative stress. PC6 (BCAAs and aromatic AAs) and PC10 (asparagine, glycine, and serine) made the largest contributions to body mass index, and PC10 and PC12 (acylcarnitines) made the largest contributions to adiposity. Metabolic risk factors and total energy expenditure were associated with PC6, PC9 (AA and tricarboxylic acid cycle metabolites), and PC10. Fat oxidation was inversely related to PC8 (lysolipids) and positively related to PC16 (acylcarnitines).

Conclusions: Global metabolomic profiling in nonobese and obese children replicates the increased BCAA and acylcarnitine catabolism and changes in nucleotides, lysolipids, and inflammation markers seen in obese adults; however, a strong signature of reduced fatty acid catabolism and increased steroid derivatives may be unique to obese children. Metabolic flexibility in fuel use observed in obese children may occur through the activation of alternative intermediary pathways. Insulin resistance, hyperleptinemia, hypertriglyceridemia, hyperuricemia, and oxidative stress and inflammation evident in obese children are associated with distinct metabolomic profiles.

INTRODUCTION

Childhood obesity is associated with increased risk of glucose intolerance, hypertension, dyslipidemia, insulin resistance, chronic inflammation, hyperuricemia and nonalcoholic fatty liver disease (1). Although genetics, diet, physical activity, and other behavioral and environmental factors contribute to these ailments, the metabolic pathways that underlie the development and progression of these metabolic diseases remain uncertain. Metabolomics, which is the measurement of low–molecular weight metabolites such as nutrient intermediates, hormones, and other signaling molecules, may provide insight into their cause.

Metabolomic studies in obese adults have consistently indicated a metabolomic profile associated with insulin resistance, diabetes, and cardiovascular disease (2–7). A pattern of increased plasma concentrations of branched-chain amino acids (BCAAs)⁸ (Val, Leu, and Ile), aromatic amino acids (AAs) (Phe and Tyr), C3 and C5 acylcarnitines, and Glx and Ala was shown to be strongly associated with insulin resistance. The accumulation of incompletely oxidized lipid species in the mitochondria also has been implicated in the development of insulin resistance (3, 4, 6).

Metabolomic profiling in obese and nonobese children has been performed with conflicting findings (8–11). In a comparison study of normal-weight and obese adolescents, there was little evidence of defective fatty acid or AA metabolism associated with obesity (8). In another study, increased plasma AA concentrations (BCAAs, Phe, Met, His, Arg, Ser, Gly, and acylcarnitines) were positively associated with β cell function relative to insulin sensitivity in contrast to findings in adults (11). Acylcarnitines (12:1 and 16:1) were higher, and 3 AAs (Glu, Met, and Pro) and acyl-alkyl phosphatidylcholines (34–38) and lysophosphatidylcholines (18:1, 18:2, and 20:4) were lower, in obese children, which were indicative of oxidative stress and changes in β oxidation and sphingomyelin metabolism (9). Higher concentrations of BCAAs and related intermediates C3 and C5 acylcarnitines, androgens, and large neutral AAs were seen in obese children (10). Thus, existing data are equivocal as to whether metabolomic alterations or adaptive metabolic plasticity occurs in children with obesity.

Because of the potential of metabolomics to reveal important biological pathways involved in the pathophysiology of childhood obesity and to reconcile the existing conflicting evidence in children, we applied nontargeted metabolomic profiling in 803 Hispanic children enrolled in the Viva la Familia Study (12). Obese children in this cohort were shown to be at increased risk of glucose intolerance, hypertension, dyslipidemia, insulin resistance (13), chronic inflammation (14), hyperuricemia (15) and nonalcoholic fatty liver disease (16). Our research aims were to 1) identify metabolites that differ significantly between nonobese and obese Hispanic children; 2) collapse metabolites into principal components (PCs) associated with obesity and metabolic risk, specifically, hyperinsulinemia, hypertriglyceridemia, hyperleptinemia, and hyperuricemia; and 3) identify metabolites associated with energy expenditure and fat oxidation.

METHODS

Subjects and design

Global metabolomic profiling was performed on fasting plasma samples from 803 Hispanic children enrolled in the Viva la Familia Study, which was a cross-sectional observational study designed to identify genetic variants influencing childhood obesity and its comorbidities. Eligibility and enrollment were described in detail elsewhere (12). Family recruitment and phenotyping were conducted in 2000–2005 in Houston, Texas. All enrolled children provided written assent, and parents gave written informed consent. The protocol was approved by the Institutional Review Boards for Human Subject Research for Baylor College of Medicine and Affiliated Hospitals and for the Texas Biomedical Research Institute.

Methods

In-depth phenotyping included anthropometric measures and body composition, 24-h calorimetry, and fasting blood sampling for clinical biochemistries. Anthropometric measures were performed by using standardized techniques according to Lohman et al. (17). Body composition, including fat-free mass (FFM) and fat mass (FM), was determined by using dual-energy X-ray absorptiometry. Methods used to measure fasting blood biochemistries were described elsewhere (12–14, 18). Room respiration calorimetry was used to make 24-h measurements of energy expenditure, the respiratory quotient (RQ), and substrate oxidation (19).

Metabolomic profiling

Fasting EDTA plasma samples were taken from 803 Hispanic children and stored at −80°C. Analyses were performed on the plasma samples and replicate samples created from a large pool of extensively characterized human plasma (MTRX3). Nontargeted metabolomic profiling was conducted by using the following 3 independent platforms: ultra-HPLC–tandem mass spectrometry optimized for basic species, ultra-HPLC–tandem mass spectrometry optimized for acidic species, and gas chromatography–mass spectrometry. Metabolomic analyses were performed as outlined previously (20).

The original data set consisted of 409 named metabolites. Of the original 409 named metabolites, we excluded 77 metabolites that were missing values for >40% of participants and 28 xenobiotics; hence, our final data set was comprised of 304 named metabolites.

Block normalization was performed on the original raw ion intensities to correct for variation that resulted from instrument interday tuning differences. Essentially, raw area counts for each compound in run-day blocks were rescaled to set medians to equal one. Any missing values were assumed to be below the limits of detection, and these values were imputed with the compound minimum (minimum value imputation).

To test for data reliability, a subset of 40 plasma samples were reanalyzed. Of the original 304 metabolites, 11.8% of metabolites in nonobese subjects and 8.9% of metabolites in obese subjects were shown to be significantly different between runs; however, the mean (±SD) intraclass correlation for these metabolites was 0.82 ± 0.17.

Statistical analysis

Numerical methods (i.e., skewness, kurtosis, and the Shapiro-Wilk test) and graphical methods (i.e., histograms and quantile-quantile plots) were used to test for normality. If the distribution of the metabolite was not normal, a log transformation was performed to meet the assumptions of a parametric test.

Mixed-effects linear regression models were used to test for differences in child characteristics and plasma metabolites between nonobese and obese children, with adjustment for sex, age, and Tanner stage, as needed, and by accounting for the nonindependence of related children within families. A Bonferroni correction was applied to the metabolomic data to counteract the problem of multiple comparisons; P < 1.151 × 10⁻⁴ was considered statistically significant.

A random forest analysis, which is a supervised classification technique that is based on an ensemble of decision trees, was used to classify nonobese and obese children. For a given decision tree, a random subset of data with identifying true class information was selected to build the tree (training set), and the remaining data were passed down the tree to obtain a class prediction for each sample repeatedly. The prediction accuracy was an unbiased estimate of how well the sample class in a new data set could be predicted. The mean decrease accuracy was used to determine which metabolites made the largest contribution to the classification. The mean decrease accuracy was determined by randomly permuting a variable, running observed values through the trees, and reassessing the prediction accuracy.

A principal components analysis (PCA), which is an unsupervised linear mixture model aimed at accounting for the variance within a data set by a smaller number of mutually uncorrelated PCs, was performed on the 304 log-transformed metabolites. In our application, the PCs were vectors of metabolite contributions. An orthogonal varimax rotation was used, and factors with eigenvalues >1.0 were retained. Component scores for each participant were calculated by using the standardized scoring coefficient. Metabolites with component loading values with magnitudes that were ≥0.50 were retained for a given component. The PCA was performed with the Proc Factor and Proc Score procedures in SAS software (version 9.4; SAS Institute Inc.) Mixed-effects linear regression models were used to test the effect of the 20 PCs on selected phenotypes, with adjustment for age, sex, Tanner staging, and obesity status. Statistical analyses were performed in the SAS software and STATA software (version13; StataCorp).

RESULTS

Global metabolomic profiling was performed in 803 Hispanic children (405 boys and 398 girls; mean ± SD age: 11.1 ± 3.9 y; age range: 4–19 y). This cohort was highly enriched for obesity with 56% of children classified as obese (BMI ≥95th percentile) according to CDC criteria (21). The remaining children were nonobese with 16% classified as overweight and 28% classified as normal weight. Fasting metabolites and hormones shown in Table 1 differed significantly between nonobese and obese children except for nonesterified fatty acids (NEFAs) and cortisol. Fasting insulin concentrations and HOMA-IR were elevated in obese children. When adjusted for age, sex, Tanner stage, FFM, and FM, 24-h total energy expenditure (TEE) did not differ by obesity status. When adjusted for age, sex, Tanner stage, and energy balance, the TEE RQ and net fat oxidation over the 24-h did not differ by obesity status; however, the sleep RQ was slightly lower in obese than in nonobese children.

TABLE 1.

Characterization of the nonobese and obese Hispanic children¹

	Nonobese	Obese	Test	Covariates	Adjusted P
n	353	450	—	—
Sex, M/F, n	163/190	242/208	Chi square	—	3.1 × 10−2
Age, y	11.09 ± 4.402	11.04 ± 3.52	t	—	8.6 × 10−1
Tanner stage	2.24 ± 1.44	2.15 ± 1.28	t	—	3.9 × 10−1
Anthropometric measures/body composition
Weight, kg	41.74 ± 18.78	67.30 ± 27.55	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
BMI, kg/m2	19.94 ± 3.78	30.07 ± 6.70	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
BMI z score	0.64 ± 0.76	2.31 ± 0.38	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
FFM, kg	30.43 ± 13.78	39.61 ± 14.61	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
FM, kg	11.36 ± 6.48	26.58 ± 11.43	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
FM, % of weight	0.26 ± 0.07	0.39 ± 0.06	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
Fasting biochemistry
Glucose, mg/dL	90.63 ± 9.57	94.38 ± 14.02	Mixed effects	Sex, age, Tanner stage	2.2 × 10−7
Insulin, μU/mL	13.11 ± 8.53	31.41 ± 21.26	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
HOMA-IR	3.01 ± 2.49	7.46 ± 5.67	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
Leptin, μg/L	8.98 ± 7.90	26.22 ± 14.89	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
Triacylglycerol, mg/dL	87.77 ± 46.62	124.27 ± 61.78	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
HDL cholesterol, mg/dL	50.85 ± 12.12	43.66 ± 9.70	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
LDL cholesterol, mg/dL	99.98 ± 28.65	106.74 ± 29.74	Mixed effects	Sex, age, Tanner stage	1.3 × 10−3
Uric acid, mg/dL	4.57 ± 1.36	5.92 ± 1.75	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
CRP, μg/L	651.53 ± 909.37	1761 ± 1680	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
ALT, U/L	16.27 ± 12.73	31.44 ± 30.25	Mixed effects	Sex, age, Tanner stage	1.0 × 10−32
NEFAs, mEq/L	0.51 ± 0.25	0.53 ± 0.18	Mixed effects	Sex, age, Tanner stage	1.0 × 10−4
Cortisol, μg/L	66.32 ± 30.04	63.33 ± 30.51	Mixed effects	Sex, age, Tanner stage	1.3 × 10−1
Calorimetry
TEE, kcal/d	1826 ± 380	2293 ± 495	Mixed effects	Sex, age, Tanner stage, FFM, FM	4.8 × 10−1
Fat oxidation, % of nonprotein TEE	40.50 ± 11.60	41.67 ± 10.51	Mixed effects	Sex, age, Tanner stage, FFM, FM	4.0 × 10−1
24-h RQ	0.87 ± 0.03	0.87 ± 0.03	Mixed effects	Sex, age, Tanner stage, energy balance	3.4 × 10−1
Sleep RQ	0.85 ± 0.03	0.84 ± 0.03	Mixed effects	Sex, age, Tanner stage, energy balance	4.1 × 10−3

¹For tests, “mixed effects” denotes mixed-effects linear regression models. ALT, alanine aminotransferase; CRP, C-reactive protein; FFM, fat-free mass; FM, fat mass; NEFA, nonesterified fatty acid; RQ, respiratory quotient; TEE, total energy expenditure.

²Mean ± SD (all such values).

A summary of the 38 AAs, peptides, and their metabolites that differed significantly between nonobese and obese children is presented in Table 2. When adjusted for sex, age, and Tanner stage, BCAAs (Leu, Ile, and Val) were significantly higher in obese children as were BCAA catabolites 2-methylbutyrylcarnitine, 3-methyl-2-oxobutyrate, and isovalerylcarnitine. In addition, propionylcarnitine and butyrylcarnitine, which are carnitine conjugates of propionyl-CoA and butyryl-CoA, were significantly elevated (Table 3). Additional AAs (alanine, glutamate, lysine, phenylalanine, and tyrosine), polyamines, several γ-glutamyl dipeptides, and polypeptides were also elevated in obese children. In contrast, levels of asparagine, aspartate, glycine, serine, and histidine were lower in obese children. Notably, there were significantly higher levels of α-hydroxybutyrate (AHB) and α-ketobutyrate (AKB) in obese children.

TABLE 2.

Global metabolomic profiling of amino acid metabolism of Hispanic children by obesity status¹

	Scaled, imputed values		Adjusted results2
Superpathway/subpathway and metabolites	Nonobese	Obese	t (reference nonobese)	P
Alanine and aspartate
Alanine	0.98 ± 0.29	1.07 ± 0.29	4.83	1.8 × 10−6
Asparagine	1.10 ± 0.30	0.95 ± 0.25	−6.69	5.8 × 10−11
Aspartate	1.06 ± 0.25	0.99 ± 0.21	−4.2	3.1 × 10−5
Creatine
Creatine	0.94 ± 0.28	1.02 ± 0.26	4.43	1.1 × 10−5
Glutamate
Glutamate	1.19 ± 1.29	1.52 ± 1.45	7.32	9.1 × 10−13
Pyroglutamine	1.45 ± 1.20	1.19 ± 0.81	−4.87	1.5 × 10−6
Glycine, serine, and threonine
Glycine	1.15 ± 0.27	0.94 ± 0.23	−10.86	5.6 × 10−25
N-Acetylglycine	1.49 ± 0.90	1.00 ± 0.60	−10.28	9.4 × 10−23
Serine	1.11 ± 0.25	0.95 ± 0.21	−8.82	1.7 × 10−17
Histidine
Histidine	1.03 ± 0.12	0.97 ± 0.14	−6.61	9.4 × 10−11
Leucine, isoleucine, and valine
C5	0.96 ± 0.25	1.08 ± 0.26	6.59	1.1 × 10−10
3-Methyl-2-oxobutyrate	0.97 ± 0.20	1.04 ± 0.19	4.93	1.1 × 10−6
α-Hydroxyisovalerate	1.00 ± 0.34	1.11 ± 0.40	4.39	1.4 × 10−5
Isoleucine	0.95 ± 0.16	1.04 ± 0.15	8.83	1.5 × 10−17
C5-OH	0.92 ± 0.32	1.16 ± 0.40	9.47	9.2 × 10−20
Leucine	0.96 ± 0.16	1.04 ± 0.14	8.78	2.2 × 10−17
Valine	0.95 ± 0.14	1.05 ± 0.13	11.24	1.8 × 10−26
Lysine
Lysine	0.96 ± 0.19	1.03 ± 0.18	6.38	3.8 × 10−10
Methionine, cysteine, SAM, and taurine
AHB	1.01 ± 0.46	1.10 ± 0.37	4.43	1.1 × 10−5
AKB	0.94 ± 0.56	1.24 ± 0.67	6.54	1.5 × 10−10
3-(4)-Hydroxyphenyl lactate	0.97 ± 0.30	1.09 ± 0.30	7.61	1.3 × 10−13
Phenylalanine and tyrosine
Phenylalanine	0.95 ± 0.12	1.04 ± 0.13	11.67	3.5 × 10−28
Tyrosine	0.90 ± 0.15	1.10 ± 0.18	17.7	1.0 × 10−32
Acisoga	0.96 ± 0.26	1.07 ± 0.27	5.26	2.1 × 10−7
Polyamine
C-glycosyltryptophan	0.97 ± 0.15	1.04 ± 0.15	7.65	9.6 × 10−14
Kynurenate	0.82 ± 0.37	0.98 ± 0.36	6.48	2.1 × 10−10
Kynurenine	0.95 ± 0.21	1.08 ± 0.23	8.96	5.5 × 10−18
Tryptophan	0.98 ± 0.16	1.03 ± 0.15	5.26	2.1 × 10−7
Citrulline	1.05 ± 0.24	0.97 ± 0.21	−4.49	8.6 × 10−6
Urea cycle; arginine and proline
Ornithine	0.99 ± 0.38	1.13 ± 0.45	5.37	1.2 × 10−7
Peptide
γ-Glutamylglutamate	1.03 ± 1.10	1.22 ± 1.04	5.01	7.3 × 10−7
γ-Glutamyl amino acid
γ-Glutamylleucine	1.09 ± 0.65	1.19 ± 0.65	3.86	1.3 × 10−4
γ-Glutamylphenylalanine	0.97 ± 0.24	1.09 ± 0.26	7.38	6.3 × 10−13
γ-Glutamyltyrosine	0.92 ± 0.23	1.12 ± 0.28	11.73	1.9 × 10−28
Bradykinin	1.17 ± 2.05	2.32 ± 5.85	6.67	6.5 × 10−11
Polypeptide
Bradykinin-des-Arg(9)	0.89 ± 1.40	2.05 ± 6.65	7.32	9.4 × 10−13
HWESASLLR	1.62 ± 3.04	2.53 ± 6.17	5.24	2.4 × 10−7
HWESASXX	1.04 ± 0.77	1.50 ± 1.81	5.72	1.7 × 10−8

¹All values are means ± SDs. Main effect of obesity status, P < 1.151 × 10⁻⁴. AHB, α-hydroxybutyrate; AKB, α-ketobutyrate; C5, 2-methylbutyrylcarnitine; C5-OH, isovalerylcarnitine; SAM, S-adenosylmethionine.

²Mixed-effects linear regression model adjusted for age, sex, and Tanner stage (ln metabolite).

TABLE 3.

Global metabolomic profiling of lipid metabolism of Hispanic children by obesity status¹

	Scaled, imputed values		Adjusted results2
Superpathway/subpathway and metabolites	Nonobese	Obese	t (reference nonobese)	P
Carnitine
Carnitine	0.94 ± 0.14	1.04 ± 0.13	10.71	2.3 × 10−24
Acylcarnitine
Propionylcarnitine (3)	0.92 ± 0.23	1.10 ± 0.25	10.78	1.2 × 10−24
Butyrylcarnitine (4)	1.03 ± 0.66	1.30 ± 0.71	8.2	1.8 × 10−15
Hexanoylcarnitine (6)	1.01 ± 0.48	1.10 ± 0.32	5.59	3.5 × 10−8
Stearoylcarnitine (18)	1.07 ± 0.43	0.87 ± 0.36	−6.37	4.2 × 10−10
Oleoylcarnitine (18:1)	1.20 ± 0.52	1.01 ± 0.39	−5.57	4.1 × 10−8
Fatty acid, amide
2-Aminoheptanoate	1.01 ± 0.40	1.11 ± 0.44	4.47	9.5 × 10−6
2-Aminooctanoate	1.29 ± 0.77	1.03 ± 0.66	−6.30	6.4 × 10−10
Fatty acid, dicarboxylate
Dodecanedioate	1.25 ± 2.24	0.92 ± 0.52	−4.23	2.7 × 10−5
Tetradecanedioate	1.60 ± 3.96	1.04 ± 0.85	−3.84	1.4 × 10−4
2-Hydroxydecanoate	1.14 ± 0.46	1.00 ± 0.34	−4.89	1.3 × 10−6
Ketone bodies
BHBA	3.42 ± 7.69	1.73 ± 3.17	−6.10	2.0 × 10−9
Long-chain fatty acid
10-Heptadecenoate (17:1n–7)	1.03 ± 0.55	1.12 ± 0.44	5.00	7.9 × 10−7
Arachidate (20:0)	1.19 ± 0.48	1.00 ± 0.28	−4.99	8.3 × 10−7
Myristoleate (14:1n–5)	1.04 ± 0.72	1.12 ± 0.55	5.61	3.3 × 10−8
Nonadecanoate (19:0)	1.15 ± 0.44	0.99 ± 0.28	−6.88	1.6 × 10−11
Palmitoleate (16:1n–7)	1.00 ± 0.60	1.13 ± 0.48	6.59	1.1 × 10−10
Lysolipid
1-Arachidonoylglycerophosphocholine (20:4n–6)	1.11 ± 0.34	0.98 ± 0.38	−6.38	3.9 × 10−10
1-Dihomo-linoleoylglycerophosphocholine (20:2n–6)	1.14 ± 0.76	0.94 ± 0.74	−4.53	7.2 × 10−6
1-Docosahexaenoylglycerophosphocholine (22:6n–3)	1.17 ± 0.50	1.01 ± 0.42	−5.25	2.2 × 10−7
1-Docosapentaenoylglycerophosphocholine (22:5n–3)	1.10 ± 0.44	0.99 ± 0.43	−3.7	2.4 × 10−4
1-Eicosenoylglycerophosphocholine (20:1n–9)	1.14 ± 0.92	0.85 ± 0.71	−6.17	1.4 × 10−9
1-Linolenoylglycerophosphocholine (18:3n–3)	1.22 ± 0.58	1.01 ± 0.48	−5.31	1.6 × 10−7
1-Linoleoylglycerophosphocholine (18:2n–6)	1.16 ± 0.28	0.92 ± 0.23	−12.62	1.0 × 10−32
1-Linoleoylglycerophosphoethanolamine	1.10 ± 0.37	0.98 ± 0.32	−4.97	8.9 × 10−7
1-Margaroylglycerophosphocholine (17:0)	1.32 ± 1.03	1.05 ± 1.02	−4.23	2.8 × 10−5
1-Oleoylglycerophosphocholine (18:1)	1.24 ± 0.55	0.97 ± 0.49	−8.33	6.7 × 10−16
1-Oleoylglycerophosphoethanolamine	1.19 ± 0.44	1.01 ± 0.36	−5.98	4.2 × 10−9
1-Palmitoylglycerophosphoinositol	1.01 ± 0.76	1.20 ± 0.68	6.18	1.3 × 10−9
1-Stearoylglycerophosphocholine (18:0)	1.35 ± 0.90	1.17 ± 0.89	−4.27	2.3 × 10−5
2-Linoleoylglycerophosphocholine	1.25 ± 0.44	0.93 ± 0.30	−12.18	3.0 × 10−30
2-Linoleoylglycerophosphoethanolamine	1.05 ± 0.44	0.86 ± 0.38	−5.82	1.0 × 10−8
2-Oleoylglycerophosphocholine	1.22 ± 0.51	0.96 ± 0.40	−8.49	2.1 × 10−16
2-Oleoylglycerophosphoethanolamine	1.13 ± 0.48	0.94 ± 0.41	−5.28	1.9 × 10−07
2-Palmitoylglycerophosphocholine	1.22 ± 0.59	1.09 ± 0.59	−4.05	5.9 × 10−05
2-Stearoylglycerophosphocholine	1.38 ± 1.00	1.21 ± 1.14	−3.71	2.3 × 10−04
Medium-chain fatty acid
Caprate (10:0)	1.16 ± 0.49	0.99 ± 0.29	−5.08	5.2 × 10−07
Laurate (12:0)	1.28 ± 0.81	1.05 ± 0.50	−4.15	3.8 × 10−05
Monoacylglycerol
1-Palmitoylglycerol (1-monopalmitin)	0.98 ± 0.29	1.08 ± 0.35	4.08	5.1 × 10−05
Phospholipid
Choline	0.96 ± 0.19	1.05 ± 0.19	6.90	1.4 × 10−11
PUFA
Dihomo-linolenate (20:3n–3 or n–6)	1.00 ± 0.43	1.14 ± 0.48	6.57	1.2 × 10−10
Secondary bile acid
Taurolithocholate 3-sulfate	1.22 ± 1.22	0.95 ± 0.96	−4.50	8.3 × 10−06
Sphingolipid
Stearoylsphingomyelin	0.96 ± 0.29	1.07 ± 0.33	5.11	4.5 × 10−7
Steroid
4-Androsten-3β-17β-diol disulfate 1	1.12 ± 1.19	1.51 ± 1.24	11.31	9.4 × 10−27
4-Androsten-3β-17β-diol disulfate 2	1.01 ± 0.71	1.17 ± 0.65	8.55	1.3 × 10−16
5α-Androstan-3β-17β-diol disulfate	1.08 ± 1.44	1.24 ± 1.30	7.18	2.3 × 10−12
Androsteroid monosulfate 2	1.23 ± 1.29	1.43 ± 1.27	6.60	1.0 × 10−10
Androsterone sulfate	1.40 ± 1.50	1.34 ± 1.09	−4.50	8.5 × 10−6
Cortisone	1.06 ± 0.20	0.96 ± 0.21	−6.64	7.6 × 10−11
DHEA-S	1.15 ± 1.04	1.28 ± 0.92	7.42	4.6 × 10−13
Epiandrosterone sulfate	1.34 ± 1.50	1.29 ± 1.08	−4.70	3.4 × 10−6
Sterol
Lathosterol	0.65 ± 0.42	1.16 ± 0.59	15.38	1.0 × 10−32

¹All values are means ± SDs. Main effect of obesity status, P < 1.151 × 10⁻⁴. BHBA, β-hydroxybutyrate; DHEA-S, dehydroisoandrosterone sulfate.

²Mixed-effects linear regression model adjusted for age, sex, and Tanner stage (ln metabolite).

Consistent with elevated C-reactive protein concentrations, several metabolites indicative of inflammation and the activation of an immune response were elevated in the obese children. Bradykinin (a potent vasodilator), bradykinin-des-Arg(9) (an active metabolite of bradykinin), and HWESASLLR and HWESASXX (peptides that map to complement C3 protein) were elevated in obese children, consistent with the activation of an immune response. Kynurenine and kynurenate, which are involved with the dilation of blood vessels during inflammation, were elevated in obese children. In contrast, 1-acyl-lysolipids and 2-acyl-lysolipids (Table 3), which are involved in innate immunity, were generally decreased.

A summary of the 52 lipid metabolites that differed significantly between nonobese and obese children is presented in Table 3. Carnitine and the short-chain acylcarnitines [propionylcarnitine (3), butyrylcarnitine (4), and hexanoylcarnitine (6)] were elevated in obese children, whereas a few of the long-chain acylcarnitines [stearoylcarnitine (18) and oleoylcarnitine (18:1)] were lower. There was a striking decrease in lysolipids (glycerophosphocholines and glycerophosphoethanolamines) and dicarboxylated fatty acids (dodecanedioate, tetradecanedioate, and 2-hydroxydecanoate) in obese children. The ketone body β-hydroxybutyrate was significantly lower in obese children than in nonobese children. With the exceptions of cortisone, androsterone sulfate, and epiandrosterone sulfate, steroid derivatives were markedly higher in obese children. Consistent with elevated LDL-cholesterol concentrations in obese children, the cholesterol precursor lathosterol was significantly higher in obese children.

A summary of carbohydrate and nucleotide metabolites that differed significantly between nonobese and obese children is presented in Table 4. Mannose and pyruvate were higher, but glycerate was lower, in obese children than in nonobese children. In the tricarboxylic acid (TCA) cycle, the only metabolite that reached significance was citrate, which was lower in obese children.

TABLE 4.

Global metabolomic profiling of carbohydrate and nucleotide metabolism of Hispanic children by obesity status¹

	Scaled, imputed values		Adjusted results2
Superpathway/subpathway and metabolites	Nonobese	Obese	t (reference nonobese)	P
Carbohydrate
Mannose	0.91 ± 0.22	1.11 ± 0.29	13.22	1.0 × 10−32
Glycolysis, gluconeogenesis, and pyruvate
Glycerate	1.07 ± 0.36	0.98 ± 0.36	−3.81	1.5 × 10−4
Pyruvate	0.95 ± 0.42	1.15 ± 0.49	5.02	7.2 × 10−7
Pentose metabolism
Bilirubin (E-E)	1.37 ± 0.99	1.08 ± 0.75	−4.35	1.6 × 10−5
Nicotinate
N1-Methyl-2-pyridone-5-carboxamide	0.87 ± 0.37	1.15 ± 0.48	8.37	5.0 × 10−16
Pantothenate and CoA
Pantothenate	0.96 ± 0.28	1.09 ± 0.27	7.85	2.3 × 10−14
Tocopherol
γ-Tocopherol	0.93 ± 0.35	1.13 ± 0.42	8.76	2.6 × 10−17
TCA3 cycle
Citrate	1.10 ± 0.29	0.96 ± 0.21	−8.21	1.7 × 10−15
Purine metabolism, (hypo)xanthine/inosine
Urate	0.92 ± 0.19	1.10 ± 0.24	13.24	1.0 × 10−32
Xanthine	0.97 ± 0.37	1.17 ± 0.48	5.94	5.2 × 10−9
Purine metabolism, adenine
N6-Carbamoylthreonyladenosine	0.93 ± 0.25	1.03 ± 0.23	5.48	6.6 × 10−8
N2-N2-Dimethylguanosine	0.93 ± 0.31	1.02 ± 0.29	5.27	1.9 × 10−7
Pyrimidine metabolism, uracil
5-Methyluridine (ribothymidine)	1.10 ± 0.69	0.96 ± 0.23	−6.84	2.1 × 10−11
N-Acetyl-β-alanine	1.00 ± 0.35	1.10 ± 0.35	5.03	6.8 × 10−7
Pseudouridine	0.97 ± 0.11	1.03 ± 0.12	7.39	5.7 × 10−13

¹All values are means ± SDs. Main effect of obesity status, P < 1.151 × 10⁻⁴.

²Mixed-effects linear regression model, adjusted for sex, age, and Tanner stage (ln metabolite).

³TCA, tricarboxylic acid.

Consistent with elevated uric acid, higher levels of purine and pyrimidine metabolites were seen in obese children. Levels of the purine catabolite xanthine were significantly elevated, which may have reflected an increased generation of H₂O₂ during the conversion of hypoxanthine to xanthine by xanthine oxidase as part of an inflammatory response. Xanthine oxidase also converts xanthine to urate, the levels of which were also higher in the obese-children group.

To assess the predictive accuracy of classifying children by obesity status, a random forest analysis was performed on the metabolites. Tyrosine was the highest-ranked metabolite on the basis of its contribution to the obesity classification. The top-ranked metabolites were from all the major pathways (Figure 1). The predictive accuracy for obesity classification was 81%.

FIGURE 1.

Random forest plot of top-ranked metabolites with the highest MDA for the classification of nonobese and obese Hispanic children.

A PCA was used to reduce the large number of metabolites to fewer clusters of highly correlated metabolites. The PCA consolidated the 304 metabolites into 65 PCs with eigenvalues >1 and factor loading values with magnitudes ≥0.50 These 65 PCs explained 75.6% of the variance in the data set. On the basis of a scree plot, the first 20 PCs with eigenvalues >2.5 were retained and explained 53.4% of the variance (Table 5). These 20 PCs were comprised of 161 of 304 metabolites.

TABLE 5.

PC analysis¹

PC	Pathway (metabolites, n)	Metabolites within PCs	Eigenvalue	Variance explained, %	Cumulative variance, %
1	Amino acid and lipid (37)	Amino acid: α-hydroxybutyrate	39.00	12.83	12.83
		Lipid: acetylcarnitine, 15-methylpalmitate, 17-methylstearate, hexadecanedioate, tetradecanedioate, 16-hydroxypalmitate, 3-hydroxydecanoate, 3-hydroxyoctanoate, glycerol, 3-hydroxybutyrate, 10-heptadecenoate, 10-nonadecenoate, arachidate, cis-vaccenate, eicosenoate, margarate, myristate, myristoleate, nonadecanoate, oleate, palmitate, palmitoleate, pentadecanoate, stearate, 5-dodecenoate, caprate, laurate, dihomo-linoleate, docosadienoate, docosahexaenoate, docosapentaenoate, docosapentaenoate, linoleate, linolenate, stearidonate
2	Amino acid and lipid (14)	Amino acid: glutamate, glutamine, 5-oxoproline	20.28	6.67	19.50
		Lipid: leukotriene B4, 13-HODE plus 9-HODE, glycerol 3-phosphate, 1-oleoylplasmenylethanolamine, 1-palmitoylplasmenylethanolamine, glycerophosphorylcholine, arachidonate, dihomo-linolenate, eicosapentaenoate, sphingosine, leucylleucine
3	Lysolipid: GPCs (19)	Lysolipid: 1-arachidonoylglycerophosphocholine, 1-dihomo-linoleoylglycerophosphocholine, 1-docosahexaenoylglycerophosphocholine, 1-docosapentaenoylglycerophosphocholine, 1-eicosatrienoylglycerophosphocholine, 1-eicosenoylglycerophosphocholine, 1-linolenoylglycerophosphocholine, 1-linoleoylglycerophosphocholine, 1-margaroylglycerophosphocholine, 1-myristoylglycerophosphocholine, 1-oleoylglycerophosphocholine, 1-palmitoleoylglycerophosphocholine, 1-palmitoylglycerophosphocholine, 1-pentadecanoylglycerophosphocholine, 1-stearoylglycerophosphocholine, 2-linoleoylglycerophosphocholine, 2-oleoylglycerophosphocholine, 2-palmitoylglycerophosphocholine, 2-stearoylglycerophosphocholine	15.90	5.23	24.73
4	Steroid (12)	Steroid: 21-hydroxypregnenolone disulfate, 4-androsten-3β-17β-diol disulfate 1, 4-androsten-3β-17β-diol disulfate 2, 5α-androstan-3β-17β-diol disulfate, 5α-pregnan-3β-20α-diol disulfate, andro steroid monosulfate 2, androsterone sulfate, dehydroisoandrosterone sulfate, epiandrosterone sulfate, pregn steroid monosulfate, pregnen-diol disulfate, pregnenolone sulfate	13.30	4.37	29.10
5	γ-Glutamyl amino acid (7)	γ-Glutamyl amino acid: γ-glutamylglutamate, γ-glutamylisoleucine, γ-glutamylleucine, γ-glutamylmethionine, γ-glutamylphenylalanine, γ-glutamylvaline; Polypeptide: HWESASLLR	10.51	3.46	32.56
6	Amino acid: BCAAs and aromatic AAs (7)	Amino acid: isoleucine, leucine, lysine, methionine, phenylalanine, tyrosine, γ-glutamyltyrosine	7.21	2.37	34.93
7	Amino acid (3)	Amino acid: 3-methyl-2-oxobutyrate, 3-methyl-2-oxovalerate, 4-methyl-2-oxopentanoate	6.35	2.09	37.02
8	Lysolipid: GPEs (7)	Lysolipid: 1-eicosatrienoylglycerophosphoethanolamine, 1-linoleoylglycerophosphoethanolamine, 1-oleoylglycerophosphoethanolamine, 1-palmitoylglycerophosphoethanolamine, 1-stearoylglycerophosphoethanolamine, 2-oleoylglycerophosphoethanolamine, 2-palmitoylglycerophosphoethanolamine	6.03	1.98	39.01
9	Amino acid and TCA (7)	Amino acid: aspartate, isoleucylglycine, valylglycine, dimethylarginine, pro-hydroxy-pro, trans-4-hydroxyproline; TCA cycle: citrate	5.06	1.67	40.67
10	Amino acid (3)	Amino acid: asparagine, glycine, serine	4.58	1.51	42.18
11	Vitamin and lipid (5)	Vitamin: α-Tocopherol, 1-stearoylglycerol, palmitoyl sphingomyelin, stearoyl sphingomyelin, cholesterol	4.47	1.47	43.65
12	Amino acid: shorter-chain acylcarnitines (5)	Amino acid–acylcarnitines: 2-methylbutyrylcarnitine, isovalerylcarnitine Lipid: carnitine, butyrylcarnitine, propionylcarnitine	4.31	1.42	45.07
13	Amino acid, glycogen, vitamin, and nucleotide (5)	Amino acid: serotonin; glycogen metabolism: maltose, maltotriose; vitamin: nicotinamide; nucleotide: adenosine 5-monophosphate	4.06	1.34	46.40
14	Lipid: cholates (4)	Lipid: glycochenodeoxycholate, glycocholate, taurochenodeoxycholate, taurocholate	3.49	1.15	47.55
15	Peptide (5)	Peptide: glycylleucine, leucylalanine, leucylglycine, serylleucine, threonylleucine	3.40	1.12	48.67
16	Lipid, longer-chain acylcarnitines (5)	Lipid: cis-4-decenoyl carnitine, decanoylcarnitine, hexanoylcarnitine, laurylcarnitine, octanoylcarnitine	3.17	1.04%	49.71%
17	Lipid (4)	Lipid: azelate, sebacate, caproate, heptanoate	3.03	1.00	50.71
18	Amino acid (4)	Amino acid: isobutyrylcarnitine, p-cresol sulfate, phenylacetylglutamine, 3-indoxyl sulfate	2.80	0.92	51.63
19	Lysolipid: inositols (5)	Lysolipid: 1-arachidonoylglycerophosphoinositol, 1-oleoylglycerophosphoinositol, 1-palmitoylglycerophosphoinositol, 1-stearoylglycerophosphoinositol, 2-stearoylglycerophosphoinositol	2.71	0.89	52.52
20	Amino acid (3)	Amino acid: α-hydroxyisovalerate, 3-(4-hydroxyphenyl)lactate, indolelactate	2.62	0.86	53.38

¹PC analysis was used to identify the top 20 PCs with eigenvalues >2.5 comprised of 161 metabolites with factor loading values with magnitudes ≥0.50. AA, amino acid; BCAA, branched-chain amino acid; GPC, glycerophosphocholine; GPE, glycerophosphoethanolamine; HODE, hydroxy-octadecadienoic acid; PC, principal component; TCA, tricarboxylic acid.

First, we tested which of the 20 PCs were associated with BMI and adiposity. Six PCs (PC3, PC4, PC6, PC10, PC12, and PC18) were associated with BMI and accounted for 46% of the variance in BMI, with PC6 (BCAA and aromatic AA) and PC10 (asparagine, glycine, and serine) making the largest contributions (Table 6). Eight PCs (PC3, PC4, PC6, PC9, PC10, PC12, PC18, and PC19) were associated with adiposity (percentage of FM) and accounted for 36% of the variance, with PC10 and PC12 (acylcarnitines) making the largest contributions.

TABLE 6.

Mixed-effects linear regression models predicting child characteristics from metabolite PC¹

	β ± SE	F	Standardized β	PCs, partial R2	All, full R2	P
Anthropometric measures and body composition
BMI2				0.46	0.58
PC3	−0.043 ± 0.007	−6.34	−0.16			4.8 × 10−10
PC4	0.044 ± 0.010	4.63	0.17			4.6 × 10−6
PC6	0.135 ± 0.013	10.30	0.31			8.5 × 10−23
PC10	−0.135 ± 0.012	−11.43	−0.26			3.5 × 10−27
PC12	0.068 ± 0.011	6.16	0.18			1.4 × 10−9
PC18	−0.040 ± 0.009	−4.64	−0.10			4.5 × 10−6
FFM2				0.45	0.84
PC3	−0.034 ± 0.006	−5.20	−0.08			2.8 × 10−7
PC4	0.071 ± 0.009	7.62	0.17			1.2 × 10−13
PC5	0.156 ± 0.013	12.32	0.23			1.0 × 10−32
PC6	−0.089 ± 0.011	−7.77	−0.10			4.4 × 10−14
PC7	0.034 ± 0.011	3.23	0.05			1.3 × 10−3
FM%2				0.36	0.42
PC3	−0.046 ± 0.009	−5.34	−0.16			1.4 × 10−7
PC4	0.048 ± 0.012	4.12	0.18			4.5 × 10−5
PC6	0.089 ± 0.016	5.39	0.20			1.1 × 10−7
PC9	−0.044 ± 0.012	−3.78	−0.14			1.8 × 10−4
PC10	−0.140 ± 0.015	−9.38	−0.27			2.2 × 10−19
PC12	0.095 ± 0.014	6.99	0.24			8.8 × 10−12
PC18	−0.048 ± 0.011	−4.55	−0.13			6.7 × 10−6
PC19	0.030 ± 0.009	3.31	0.10			1.0 × 10−3
Fasting biochemistries
HOMA-IR3				0.20	0.57
PC3	−0.088 ± 0.020	−4.55	−0.11			6.6 × 10−6
PC6	0.337 ± 0.039	10.24	0.31			1.4 × 10−22
PC9	0.089 ± 0.027	3.58	0.11			3.8 × 10−4
PC10	−0.303 ± 0.037	−8.72	−0.23			3.5 × 10−17
Leptin3				0.16	0.64
PC6	0.157 ± 0.040	3.91	0.11			1.1 × 10−4
PC9	−0.088 ± 0.027	−3.25	−0.09			1.2 × 10−3
PC10	−0.284 ± 0.037	−7.65	−0.17			9.2 × 10−14
PC11	0.081 ± 0.027	2.97	0.06			3.1 × 10−3
PC12	0.132 ± 0.032	4.08	0.12			5.3 × 10−5
PC19	0.079 ± 0.022	3.64	0.08			3.0 × 10−4
Triglycerides3				0.24	0.47
PC6	0.168 ± 0.023	7.27	0.24			1.3 × 10−12
PC8	0.155 ± 0.014	11.32	0.31			8.9 × 10−27
PC10	−0.213 ± 0.023	−9.12	−0.28			1.6 × 10−18
PC11	0.087 ± 0.017	5.04	0.14			6.4 × 10−7
Uric acid3				0.12	0.48
PC4	0.046 ± 0.012	3.84	0.17			1.4 × 10−4
PC6	0.089 ± 0.016	5.72	0.19			1.8 × 10−8
PC9	0.062 ± 0.011	5.49	0.17			6.3 × 10−8
PC10	−0.086 ± 0.015	−5.56	−0.17			4.3 × 10−8
CRP3				0.10	0.35
PC6	0.462 ± 0.073	6.34	0.22			5.0 × 10−10
PC9	−0.256 ± 0.054	−4.74	−0.18			2.7 × 10−6
PC10	−0.481 ± 0.074	−6.53	−0.20			1.5 × 10−10
ALT3				0.06	0.23
PC6	0.181 ± 0.038	4.73	0.17			2.9 × 10−6
PC14	0.108 ± 0.024	4.45	0.16			1.0 × 10−5
PC17	0.082 ± 0.024	3.38	0.13			7.7 × 10−4
NEFAs3				0.61	0.67
PC1	0.302 ± 0.011	28.02	0.73			1.0 × 10−32
PC3	0.033 ± 0.010	3.45	0.08			5.9 × 10−4
Calorimetry
TEE4				0.15	0.88
PC6	0.027 ± 0.007	4.01	0.07			7.4 × 10−5
PC9	0.029 ± 0.005	6.38	0.11			5.5 × 10−10
PC10	−0.036 ± 0.008	−4.68	−0.07			4.2 × 10−6
FATOX%5
PC8	−0.053 ± 0.011	−4.64	−0.18	0.14	0.14	4.8 × 10−6
PC16	0.081 ± 0.011	7.48	0.25			5.6 × 10−13
Sleep RQ5				0.15	0.22
PC4	−0.006 ± 0.002	−3.18	−0.15			1.6 × 10−3
PC8	0.009 ± 0.001	6.43	0.25			3.9 × 10−10
PC16	−0.010 ± 0.001	−7.38	−0.27			1.0 × 10−12
PC17	0.006 ± 0.002	3.81	0.20			1.6 × 10−4

¹ALT, alanine aminotransferase; CRP, C-reactive protein; FATOX%, fat oxidation as a percentage of nonprotein energy expenditure; FFM, fat-free mass; FM, fat mass; NEFA, nonesterified fatty acid; PC, principal component; RQ, respiratory quotient; TEE, total energy expenditure.

²Mixed-effects linear regression models, adjusted for age, sex, and Tanner stage.

³Mixed-effects linear regression models, adjusted for age, sex, Tanner stage, and obesity status.

⁴Mixed-effects linear regression models, adjusted for age, sex, Tanner stage, obesity status, FFM, and FM.

⁵Mixed-effects linear regression models, adjusted for age, sex, Tanner stage, obesity status, and energy balance.

Next, we tested which of the 20 PCs were associated with metabolic risk factors, energy expenditure, and fat oxidation. Metabolic risk factors for insulin resistance, hyperleptinemia, hypertriglyceridemia, hyperuricemia, and inflammation (C-reactive protein) were predicted by many of the same PCs as for BMI and the percentage of FM, namely PC6, PC9, and PC10. The TEE was also associated with PC6, PC9, and PC10 after adjustment for age, sex, Tanner stage, obesity status, FFM, and FM. Fasting NEFAs correlated highly with PC1 (AHB and lipids) and PC3 (glycerophosphocholines). Fat oxidation and sleep RQ were inversely related to PC8 (glycerophosphoethanolamines) and positively related to PC16 (acylcarnitines), independent of age, sex, Tanner staging, obesity status, and energy balance.

DISCUSSION

Global metabolomic profiling revealed different patterns of AAs, lipids, carbohydrate, and nucleotide intermediary metabolism between obese Hispanic children and their nonobese siblings indicative of insulin resistance, increased BCAA catabolism, mitochondrial dysfunction, reduced fatty acid β-oxidation, increased oxidative stress and inflammation, and altered steroid metabolism. The obesity-related increases in AAs, acylcarnitines, lysolipids, and nucleotides are largely concordant with those in adult studies; however, increased steroid metabolites may be unique to children with obesity. The use of calorimetry has not supported metabolic inflexibility in obese children, but there may be a metabolic cost to maintain homeostasis through the activation of alternative intermediary pathways. Insulin resistance, hyperleptinemia, hypertriglyceridemia, hyperuricemia, and oxidative stress and inflammation were evident in obese children.

The PCA linked correlated metabolites to clinical biomarkers of obesity and metabolic risk. Of the first 20 PCs, 8 PCs were associated with BMI or adiposity. PC6 (BCAAs and aromatic AAs), PC9 (aspartate, dipeptides, and citrate), and PC10 (asparagine, glycine, and serine) were associated with risk factors for insulin resistance, hyperleptinemia, hypertriglyceridemia, hyperuricemia, and inflammation. PC12 (shorter-chain acylcarnitines) made a large contribution to BMI, adiposity, and leptin but not to metabolic risk factors. PC1 (AHB and lipids) and PC3 (lysolipids- glycerophosphocholines) were associated with NEFAs. PC8 (lysolipids) was negatively related to fat oxidation, and PC16 (longer-chain acylcarnitines) was positively related to fat oxidation.

Insulin resistance

Along with elevated glucose and insulin concentrations, higher levels of AHB and lower levels of 1-linoleoylglycerophosphocholine and 2-linoleoylglycerophosphocholine were associated with insulin resistance in obese children in concordance with adult studies (22, 23). Mannose, which is an essential monosaccharide constituent of glycoproteins and glycolipids, was also elevated in obese children. In individuals with diabetes, plasma mannose levels are high, and levels of mannose and glucose are positively correlated (24).

Increased BCAA catabolism

Increased circulating BCAAs also reflected insulin resistance in obese Viva la Familia Study children. In our PCA, BCAAs and aromatic AAs formed a cluster (PC6), and the by-products of BCAA catabolism (C3 and C5 acylcarnitines) formed another cluster (PC12). Increased BCAA catabolism may be due to increased dietary consumption or alterations in skeletal muscle protein turnover, energy metabolism, and mitochondrial function. Elevated levels of circulating BCAAs, propionylcarnitine, and isovalerylcarnitine were proposed as markers of insulin resistance and diabetes risk in obese adults (2, 3), but the elevation of other AAs such as Phe, Tyr, and sulfur-containing AA were also consistently reported (25) as in our study.

Mitochondrial dysfunction

With the increased catabolism of BCAAs, AHB may be generated from AKB. Under conditions where propionyl-CoA cannot enter the TCA cycle or that result in the accumulation of NADH, the metabolism of AKB may shift to the production of AHB. In this study, AHB accumulation may have reflected a limited TCA capacity or demand relative to available substrate. This hypothesis was supported by significantly elevated levels of propionylcarnitine, which is a surrogate marker for propionyl-CoA levels as well as elevated levels of pyruvate, which suggested the reduced conversion of pyruvate to acetyl-CoA. In addition, the TCA-cycle intermediate citrate was significantly lower in obese children than in nonobese children. In the first step of the TCA cycle, oxaloacetate and acetyl-CoA combine to form citrate, which is shuttled out of the mitochondria to the cytosol where the acetyl-CoA moiety is converted to malonyl-CoA for fatty acid synthesis. The lower citrate perhaps reflects increased rates of fatty acid synthesis, resulting in the higher palmitate and triacylglycerol in obese children.

Oxidative stress also may contribute to elevated AHB levels in obese children because the increased synthesis of cysteine, which is a key component of the antioxidant glutathione, produces AKB as a byproduct. We previously showed higher plasma total cysteine and lower total glutathione in these obese children than in their nonobese siblings (26). Together with the described alterations in BCAA catabolism, these changes are indicative of mitochondrial dysfunction in obese children.

Reduced fatty acid β oxidation

Despite higher levels of long-chain fatty acids, levels of the long-chain acylcarnitines stearoylcarnitine, and oleoylcarnitine, which are generated during the import of long-chain fatty acids into the mitochondria for catabolism, were significantly lower in the obese-children group. Levels of lysolipids (glycerophosphocholines and glycerophosphoethanolamines) and dicarboxylate fatty acids, which ware intermediates generated during ω-oxidation, also were lower. Reduced lysophosphatidylcholines (18:1, 18:2, and 20:4) were reported previously in obese children (9). Levels of 3-hydroxy fatty acid (a readout of β oxidation) also tended to be lower in obese children. There was no evidence of ketogenesis because β-hydroxybutyrate was lower in the obese-children group. In contrast to findings in obese adults of overloading the mitochondria with lipid substrates leading to incomplete fatty acid oxidation (3), these changes in obese children do not suggest an excessive fatty acid supply or dysfunctional β-oxidation but, instead, represent a strong signature of the reduced catabolism of fatty acids.

Metabolic flexibility: appropriate fuel use

Respiratory calorimetry did not reveal any significant differences in the 24-h RQ or net fat use between obese and nonobese children (27). The sleep RQ, which is reflective of the fasting state, was actually lower in the obese-children group, which indicated slightly higher net fat use. Although the net fuel use did not differ, flux through metabolic pathways may have differed. Intermediary metabolite interconversions do not affect the calculation of nutrient use from respiratory gas measurements so long as there is no accumulation or depletion of metabolites during the study period (28). With the use of stable isotopes and calorimetry, obese adolescents were shown to adapt appropriately to changes in carbohydrate and fat intakes (29, 30). However, to maintain normal glucose and lipid concentrations, normal rates of glucose production and lipolysis, and appropriate fuel use, obese adolescents required a 2-fold increase in insulin secretion compared with that of their lean counterparts. As in the Viva la Familia Study, obese adolescents were highly insulin resistant both in the overnight fasted state and in response to a glucose challenge.

Inflammation accompanies obesity

Several metabolites [bradykinin and bradykinin-des-Arg(9), complement C3 protein, kynurenine, and kynurenate] that are indicative of inflammation and the activation of an immune response were elevated in obese children. Also, xanthine, which is a player in the inflammatory response, and urate, which is a scavenger of oxidative species (31, 32), increased in obese children.

Increased androgens in obese group

Elevated androgen derivatives, which are precursors of testosterone and estrogens, were observed in obese children. These anabolic steroids may induce premature adrenarche, which was linked to unfavorable metabolic features including hyperinsulinism, dyslipidemia, and later-appearing ovarian hyperandrogenism (33). Indeed, our findings replicated those reported in 6–10 y-old children in which an androgen pattern of higher levels of dehydroisoandrosterone sulfate and its derivatives were associated with parent-reported sexual maturation (10).

Our metabolomic profiling corroborated findings in some (9, 10) but not all (8, 11) pediatric publications. Enhanced mitochondrial function and adaptive metabolic plasticity in obese youth were inferred because of the lack of evidence of defective fatty acid or AA metabolism in contrast with our results (8). Increased plasma AA concentrations (BCAAs, Phe, Met, His, Arg, Ser, Gly, and acylcarnitines) were positively associated with β cell function relative to insulin sensitivity (11), again contrary to our results. Lower acyl-alkyl phosphatidylcholines and lysophosphatidylcholines were shown in obese children (9) in agreement with our results. Higher levels of BCAAs (Val, Leu, Ile, and related intermediates C3 and C5 acylcarnitines), androgens, and large neutral AAs (Phe, Tyr, and Trp) in obese children were in agreement with our findings (10). Discrepancies in studies may be attributed to differences in sample sizes, subject characteristics, and analytic platforms.

In conclusion, global metabolomics profiling in obese children recapitulated the increased BCAA and acylcarnitine catabolism and changes in nucleotides, lysolipids, and inflammation markers seen in obese adults; however, a strong signature of reduced fatty acid catabolism and increased steroid derivatives may be unique to obese children. Metabolic flexibility in fuel use observed in obese children may have occurred through the activation of alternative intermediary pathways. Insulin resistance, hyperleptinemia, hypertriglyceridemia, hyperuricemia, and oxidative stress and inflammation evident in obese children were associated with distinct metabolomic profiles.