A systematic review of metabolomic studies of childhood obesity: State of the evidence for metabolic determinants and consequences

Childhood obesity has become a global epidemic and carries significant long‐term consequences to physical and mental health. Metabolomics, the global profiling of small molecules or metabolites, may reveal the mechanisms of development of childhood obesity and clarify links between obesity and metabolic disease. A systematic review of metabolomic studies of childhood obesity was conducted, following Preferred Reporting Items for Systematic Reviews (PRISMA) guidelines, searching across Scopus, Ovid, Web of Science and PubMed databases for articles published from January 1, 2005 to July 8, 2020, retrieving 1271 different records and retaining 41 articles for qualitative synthesis. Study quality was assessed using a modified Newcastle–Ottawa Scale. Thirty‐three studies were conducted on blood, six on urine, three on umbilical cord blood, and one on saliva. Thirty studies were primarily cross‐sectional, five studies were primarily longitudinal, and seven studies examined effects of weight‐loss following a life‐style intervention. A consistent metabolic profile of childhood obesity was observed including amino acids (particularly branched chain and aromatic), carnitines, lipids, and steroids. Although the use of metabolomics in childhood obesity research is still developing, the identified metabolites have provided additional insight into the pathogenesis of many obesity‐related diseases. Further longitudinal research is needed into the role of metabolic profiles and child obesity risk.


| INTRODUCTION
Childhood obesity has become a global epidemic in developed as well as in developing countries. 1 Increased body mass index (BMI) and adiposity during childhood carries significant long-term consequences including increased risk of later development of chronic disease such as type 2 diabetes (T2D) and cardiovascular disease (CVD) and worse psychological health, social and economic outcomes. [2][3][4] Furthermore, gaining weight in childhood is likely to lead to lifetime overweight and obesity. 5 Behavioral dimensions such as diet and physical activity, and an "obesogenic environment" that shapes those behaviors, have contributed to the spread of childhood obesity. 6,7 Elucidating the mechanisms of development of childhood obesity at a molecular level may contribute to identifying potential targeted intervention approaches to prevent childhood obesity and clarify the links between obesity and metabolic disease.
Metabolomics, the study of the set of small molecules or metabolites (<1500 KDa) in a biological sample, can improve understanding of biological responses due to changes at the genetic, epigenetic or protein level and also due to environmental exposures such as diet, physical activity, microbiome, and toxins. Assessment of the metabolome has typically been conducted through two analytical chemistry techniques: nuclear magnetic resonance spectroscopy (NMR) or mass-spectrometry (MS) coupled to various chromatographic separations such as liquid or gas chromatography (LC or GC). Furthermore, analyses may be untargeted if they aim to assess a comprehensive range of metabolite classes or targeted if the chemical analysis is optimized to focus on particular classes of molecules, which can provide gains in precision, quantification and identification. While the field is relatively young and rapidly evolving, metabolomics may help to define molecular phenotypes and better characterize the metabolic alterations associated with obesity, such as processes related to insulin resistance (IR) 8 and inflammation. 9 While the literature regarding application of metabolomics to obesity in adults is relatively mature, fewer studies have been conducted specifically in child populations. 10 Metabolic signatures of obesity in children may differ from a signature observed in adults for reasons including a relatively shorter duration of obesity, ongoing linear growth, and pubertal hormones. Furthermore, metabolic alterations early in life may affect child propensity to overweight and obesity. We therefore conducted a systematic review of the literature related to obesity, BMI or other measures of adiposity and metabolomics in children.

This systematic review was accomplished based on the Preferred
Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statements. 11 A systematic search of Scopus, Ovid, Web of Science (WoS) and PubMed was conducted to identify the available evidence on metabolic signatures of childhood obesity. Different keywords were combined to retrieve metabolomics papers in the outcomes of interest, "adiposity or BMI" in "child." The search was carried out using free-text search terms (Table S1) with truncations to allow for different spellings as well as using Boolean operators. We included filters related to main language, the type of document and publication year.
The search strategy is provided in the supporting information (Tables S2-S5). Covidence software 12 was used for importing articles from Scopus, Ovid, WoS and PubMed removing duplicates and screening the articles.
A two-stage screening process were followed: First, literature search and study selection based on titles and abstracts were carried out independently by two researchers (EC and CHL). Then, a full text screening was contacted on first stage eligible papers. Any disagreements in selection process were resolved by the involvement of a third reviewer (OR). Articles titles were screened using the "Participants," "Exposure," "Comparator," and "Outcomes" (PECO) statement components. A study was included if: it was conducted in human children (age ≤ 18 years) (P); analyzed the association between metabolomics (including metabonomics and metabolic profile, defined as studies that apply NMR or MS, coupled to various types of chromatography, of urine, serum, saliva or plasma samples, to measure at least 10 molecules) (E) and childhood obesity/ overweight (or any of modification in body sizes related to obesity/ overweight including BMI, weight, waist circumference, adiposity, fat mass, waist-to-hip ratio) (O); had as control group with children without obesity/overweight (or compared continuous/categorical variation of body size measurements) (C). Additionally, eligibility criteria were 1 : full-text is available, 2 the paper is written in English, 3 the paper describes an observational study (cross-sectional studies and longitudinal studies including prospective and retrospective cohort studies) excluding controlled experiments conducted in manipulated rather than naturalistic settings (clinical trial involving administration of drugs) 4 the paper is peer-reviewed 5 and the paper is not a letter, editorial, study/review protocol, or review article.
Finally, we considered only studies published from Jan 1st 2005 to July 8, 2020.
Two authors independently extracted information on a predefined spreadsheet about study authors and year of study publishing, study population (country, size, and age), main outcomes, biofluid matrix, analytical platform and metabolite coverage, statistical analysis, covariate adjustment, and main findings. Metabolites extracted from these manuscripts were systematically annotated and stored for enhancing a synthetic data interpretation.
The risk of bias of included studies was assessed with a modified Newcastle-Ottawa Scale, 13 with additional fields related to metabolome coverage and metabolite level of identification as proposed by Sumner et al. 14 (Tables S7 and S8). We considered high quality articles as those that scored more than six stars.
The systematic review is registered with the International Prospective Register of Systematic Reviews (PROSPERO) database with registration number CRD42020208836.

| Overview of studies
In the primary search, 1916 records were identified from four databases. After removing duplicates, 1271 publications were screened for abstracts and title. 1169 publications were then excluded after review of the abstracts for not meeting the inclusion/exclusion criteria. Among the 101 remaining articles, 59 were excluded because: 24 were not observational, 17 did not have full text manuscripts, 18 did not analyze the defined outcome, and 1 had the wrong population. Finally, the remaining 41 papers were subjected to systematic review and are summarized in Tables S8-S11. A PRISMA flow chart of the paper selection process is shown in Figure S1. Sixteen studies included less than 100 participants while eight studies included more than 500 participants. Six studies used NMR analysis and 36 studies applied MS based assays. Seven MS studies used the Biocrates (Austria) kit that allows semi-quantitative measurement of up to 200 molecules from several classes, while six of the untargeted MS studies used the metabolic profiling service provided by the Metabolon company (USA) that uses a range of GC-MS and LC-MS assays to profile many hundreds of molecules. Seventeen were conducted on plasma, 15 were conducted on serum, six were conducted on urine (two of which also analyzed serum or plasma), three were conducted on umbilical cord blood, one study was conducted on dried blood spots, and one was conducted on saliva. Regarding quality of the studies, 24 were regarded as high quality (score >6) based on the adapted scoring assessment (Table S7).
Three categories of articles were identified: Cross-sectional studies that assessed metabolic profiles and adiposity at the same timepoint and formed most studies; longitudinal studies where metabolomic assessment was conducted in infancy prior to adiposity assessment; and intervention studies where metabolic profiles were assessed in relation to a lifestyle intervention designed to reduce BMI.
The three categories are presented in separate sections below.
3.2 | Cross-sectional studies: Describing the metabolic signature of obesity The 30 included cross-sectional studies are presented in Table S8, with extracted metabolite associations given in Table S9. Metabolites reported by at least two studies to be increased or decreased in blood with the adiposity measure are presented in Figure 1A. 22 studies analyzed BMI as categories of normal weight, overweight and/or obesity, seven studies used continuous BMI as the primary outcome, while one study used visceral fat as the primary outcome.
Three studies applied NMR analyses to blood. Bertram et al. 15 found no apparent effects of BMI on NMR profiles among 75 Danish adolescents, however the vast majority of participants were of normal weight. Bervoets et al. 16  BCAAs were predictive of IR and metabolic syndrome score at followup 2 years later. Wahl et al. 25 applied Biocrates analysis in serum of 80 German children with obesity and 40 with normal weight between 6 and 15 years of age. Concentrations of two acylcarnitines (C12:1 and C16:1) were significantly increased in obesity compared to normal-weight group, while concentrations of glutamine, methionine, proline, and nine phospholipids were significantly decreased in children with obesity. They also found differences in 69 metabolite ratios including ratios between saturated and unsaturated LPCs, between saturated LPCs and PCs and between SMs and PCs, which were all increased in children with obesity. Lau et al. 26  Most of the other targeted analyses have specifically targeted amino acids and carnitines. Hirschel et al. 29 analyzed dried blood sorts collected from 2,191 German children aged between 3 months and 18 years. They report positive association with BMI z score with leucine, isoleucine, tyrosine, valine, free carnitine, alanine, proline, and hydroxyproline and negative associations with citrulline. Sex differences were also observed: In boys only increases with BMI were observed for sarcosine, propionylcarnitine (C3), and acetylcarnitine (C2) and in girls only citrate and glycine were decreased with BMI.
Mihalik et al. 30 applied a targeted LC-MS analysis of short-chain acylcarnitine and amino acids in a study comparing children of normal weight, with obesity, and with obesity and T2D among 120 American children (aged 12-18 years). Although few Bonferroni corrected significant differences (for histidine and arginine) were observed between the normal and children with obesity only, significantly lower levels of acylcarnitines and amino acids were observed among children with obesity and T2D compared to normal weight controls, with generally intermediate levels among the obesity without T2D group.
There was also a negative correlation for these metabolites with BMI analyses as a continuous score. The associations reported with the short chain acylcarnitines, BCAAs and tyrosine contrast with all other studies in this review. Although associations appear to be driven mainly by the T2D group, which was slightly older than the other groups, there are not obvious sources of bias in this study and T2D is widely reported to be associated with higher levels of these metabolites. 31 Short et al. 32 investigated serum profiles of amino acids and related metabolites among 94 Indian Americans (aged 11-18) measured at baseline before participating in an exercise-based intervention trial. Higher levels of glutamate, phosphoethanolamine, aspartate, cystathionine, tyrosine, alloisoleucine, phenylalanine, leucine, alanine, valine, β-alanine, ornithine,2-aminoadipic acid, proline, histidine, and lysine and lower levels of glutamine, β-aminoisobutyric acid, cysteine, asparagine, homocysteine, γ-amino-n-butyric acid were observed among children with obesity compared to normal weight children.
Moran-Ramos et al. 33  were among the AAs more prevalently involved in the obesityderanged pathways, but they did not appear to accurately reflect specific hepatic or metabolic involvement. Two saturated fatty acids, palmitic, acid and myristic acid were also higher in the group with obesity.

| Prospective studies: Predicting obesity risk
Three studies analyzed cord blood and two studies analyzed plasma collected during infancy to predict child weight status in later life (Table S10) After adjustment for feeding group, six metabolites (asparagine, lysine, methionine, phenylalanine, tryptophan, and tyrosine) were positively associated with change in weight-for-age z score and one metabolite (tyrosine) was positively associated with change in BMI-for-age z score between 1 and 4 months of age. No metabolites predicted anthropometry at 4 years.

| Intervention studies: Is the metabolic signature of obesity reversible?
Seven studies examined changes in metabolites measured before and after a weight loss intervention program (Table S10) (Table S9). Despite wide differences in sample processing, metabolome coverage and analytical technique, 64 metabolites were reported by more than one study ( Figure 1A). The most widely reported and consistent associations were for BCAAs and for the aromatic AAs tyrosine and phenylalanine, followed by many other AAs. However, other groups of molecules were consistently reported by at least one study including acylcarnitines (particularly those of shorter chain length), steroid hormones, glycerophospholipids, sphingolipids, polyamines, peptides, purines and single metabolites from other classes. Figure 2 summarizes our main conclusions.
Of the BCAAs, eight studies report an increase with BMI for isoleucine 19,18,24,23,29,28,33,42 10 studies reported an increase for valine, 19,18,20,24,27,29,28,32,42,34 and 11 studies reported an increase for leucine. 19,18,24,26,29,28,32,33,42,34 An increase in tyrosine was reported by 11 studies 19,17,18,20,24,27,29,[32][33][34] and phenylalanine by seven studies. 19,17,18,24,26,32,33,.34 Only one study reported decreases in leucine, valine and tyrosine 30 and two studies reported a decrease in phenylalanine 23.30 Although the number of studies reporting associations with these metabolites partly results from the ability of most analytical techniques in the included studies to assess these molecules, it should be noted that these were among the highest ranked metabolites to be associated with BMI in studies that assessed a broad range of metabolite classes. 19,18,27 Associations between serum BCAAs levels and obesity and IR were first reported half a century ago; 57 however, the advantage of the metabolomic approach is putting these changes in the context of concomitant changes in other metabolites. Many studies included here also linked BCAAs to an insulin resistance or T2D. 16,18,27,33,42 Tyrosine also was among the strongest associates with IR among generally healthy European children. 27 Potential mechanisms for increased levels of AAs include increasing protein degradation, impairment of efficient oxidative metabolism in some tissues, 58 or even reduced de novo synthesis by the gut microbiome as indicated by one study in this review. 34 3-methyl-2-oxovalerate, produced from the incomplete breakdown of branched-chain amino acids was also reported decreased by two studies. 18,42 Whether BCAAs may cause IR (for instance through activation of the mammalian target of rapamycin complex 1 (mTORC1)) or more likely reflect metabolic changes related to the IR state is still unclear. 58 Acylcarnitines play a crucial role in transport of fatty acids to the mitochondria for β-oxidation and plasma levels have been linked to IR 59 and CVD. 60 Seventeen different acylcarnitines were reported as associated with BMI. In particular, increases in short chain acylcarnitines were commonly reported including free carnitine, 19,20,24,26,29 acetylcarnitine (C2), 29,33 propionylcarnitine (C3), 19,18,24,29,26 butyrylcarnitine (C4), 19,20 valerylcarnitine (C5), 24,26 and 2-methylbutyrylcarnitine (C5). 19,18 Most studies reported rises in acylcarnitines alongside BCAAs, and the increases likely represents increased availability of acyl-CoAs from BCAA catabolism. Increases were generally not reported for the longer-chained acylcarnitines and a decrease in oleoylcarnitine (C18:1) was reported by two studies 19,26 likely reflecting reduced fatty acid catabolism. 19 Tryptophan was reported to be increased by two studies 19,18 as were related polyamines, kynurenate 19,20 and kynurenine. 19,26 These compounds may reflect immune activation or low-grade systemic inflammation, and increases can result from upregulation of indoleamine 2,3-dioxygenase activity. The enzyme has been closely related to obesity, potentially resulting from reduced serotonin production and mood disturbances, depression, and impaired satiety, finally leading to increased caloric uptake. 61 Interestingly, enrichment of tryptophan and serotonin pathways were observed in cord blood of neonates who went on the become overweight in childhood. 46 Lysine was reported increased by three studies 19 19,32 arginine was increased in one study 33 and decreased in another, 30 while citrulline was reported to be decreased in four studies. 19,29,28,30 The reasons for decreased citrulline are unclear but it may reflect hepatic steatosis that is often present in children with obesity. 10 The sterol lathosterol was increased in two studies 19,39 and may also reflect hepatic and gastroenteric involvement. 62 Glutamate was reported to be increased by three studies 19,16,29 while glutamine was reported decreased by three studies. 19,23,30 Cysteine 16,32 and methionine 25,30 were reported decreased by two studies each. Glutamine and methionine were also reported to be decreased following weight loss. 55 These changes would be consistent with increased glutathione demand due to increased oxidative stress. Serine was decreased in three studies, 19,23,30 while glycine was decreased in three studies 19,29,30 and increased in one other. 26 Reduced levels of glycine and serine may indicate increased gluconeogenesis which is observed with insulin resistance. The organic acid α-hydroxybutyrate was increased in two studies 19,42 and is produced as a result of the conversion of cystathionine to cysteine and is produced downstream from glycine and serine. 63 α-hydroxybutyrate has been associated with increased glutathione demand and disrupted mitochondrial metabolism and shown to derive from hepatic glutathione stress. 64 Citrate, reported as decreased in three studies 19,16,29 also indicates altered energy metabolism in the mitochondria.
Purine metabolites, urate, 19,20 and xanthine 19,42 were increased in two studies each. It is possible that hyperuricemia may be causative of obesity through increase of hepatic and peripheral lipogenesis 65 although as urate is a scavenger of oxidative species, increases may reflect oxidative stress. Xantine is also involved in the inflammatory response. 19 Peptides γ-glutamyltyrosine 19,20 and bradykin 19,22 were increased in two studies each. Bradykin too, is indicative of inflammation and the activation of an immune response. 19 Increases in steroid hormones, in particular androgens, were reported by multiple studies, including 4-Androsten-3b-17b-diol disulfate, 19,18 5a-Androstan-3b-17b-diol disulfate, 19,20 androsteroid monosulfate, 19,18 and DHEA-S, 19,18,20 although one study also reported a decrease in DHEA-S. 40 DHEA-S was also reported to be decreased in a study of girls following weight-loss. 56 Obesity has been associated with puberty timing, particularly for girls, 66 70 We found that lower levels of BCAAs valine and leucine to be predictive of overweight in childhood, replicating the association with leucine reported by Isganaitis et al. 46 This supports the notion that metabolic profiles can identify determinative factors and improve identification of children at risk of developing obesity, supporting further longitudinal studies in this area.
While higher blood levels of BCAAs result from physiological changes associated with obesity, BCAAs are also reflective of nutritional quality, particularly protein intake, 71 and paradoxically higher intake can have positive effects on satiety and regulation of body weight. 58 Metabolomic analysis can simultaneously profile diet, including breast milk constituents, 72 products of microbial metabolism and physiological changes. Future metabolomic studies of child obesity, particularly prospective studies, should carefully consider these factors, considering their close relationship to child obesity. 73 We found only four studies analyzing metabolic profiles in urine, with 71 metabolite associations reported ( Figure 1B). Only p-cresol sulfate was reported by more than one study with contrasting direction of associations. 44,26 This may be expected as p-cresol sulfate is a microbially produced metabolite of tyrosine, so reflects both increased tyrosine levels and an altered gut microbiome. Generally, many associations reported in urine were also consistently reported in blood, although as indicted in the two studies that measured both matrices, 42,26 the biological interpretation of metabolites present in blood and urine may differ. Many more diet-specific and microbial metabolites were also detected in urine than blood. Increased use of urinary metabolomics, in conjunction with analysis in blood, may help clarify the role of these factors in obesity-related profiles. Diet is difficult to accurately assess, particularly in overweight populations where reporting bias may be greater and was not accounted for in many of the included studies. Urinary metabolomics is increasingly being used to provide more objective dietary assessments 74 and can also provide a more practical solution for microbiome analysis than incorporation of metagenomic analysis of faecal samples. 34 Another research gap is the integration of metabolomics with "omic" assessment at other biological layers. Epigenetics has attracted great interest in child obesity research due to in role in foetal programming, sensitivity to environmental factors including potential transgenerational effects, and may influence metabolism. 75 Metabolomics can clarify the role of observed epigenetic factors and their integration can provide a more complete picture of mechanistic pathways. 76 Reviewing metabolomics studies presents challenges: Structural annotation in metabolomics remains an issue and many included studies did not report identification levels according to current community standards, 14 so misclassification of reported metabolites is possible.
The breadth of metabolome coverage and also measurement precision and quantification differed widely between studies which makes assessing consistency of associations and quantitative synthesis challenging. Also, compared to the genome, the metabolome is much more highly correlated, and its size is not known and can vary across samples, which makes accounting for multiple testing difficult, particularly in untargeted studies where many features may represent analytical noise. Permutation based procedures 77 may be considered the goldstandard in addressing the multiplicity problem. We did not formally test publication bias as this is not readily applicable to omics studies where many features are tested. Only one study did not report any associations, but it should be noted that a large proportion of studies, particularly earlier studies, did not apply any multiple testing correction, increasing the likelihood of having associations to report. There are currently over 100,000 metabolites in the Human Metabolome Database, 78 while the highest number of molecules assayed by studies in this review was just over 1,000. Future untargeted studies will need to both increase metabolite coverage, through technological development and combining analytical platforms, alongside improvements in structural annotation 78 and appropriate statistical methodology, to provide comprehensive assessment and generate new hypotheses. In parallel, further targeted studies are required to further explore classes of molecules and test hypotheses. Both steroids and lipids appear from this review to be promising avenues.
In conclusion, a consistent metabolic profile of childhood obesity was observed including amino acids (particularly BCAAs and AAAs), carnitines, lipids and steroids. These signatures appear largely concordant with those in adult studies. 10