Branched chain amino acid metabolism and microbiome in adolescents with obesity during weight loss therapy

Current knowledge about the physiology of obesity and treatment responses is dominated by studies conducted in adults. Since the 1960s, studies have established that elevation of circulating branched-chain amino acids (BCAA: valine, leucine, and isoleucine) and various products of their catabolism are associated with adult obesity, metabolic syndrome, and future development of T2D (1–3). The ketoacid metabolites of BCAA (BCKA) have been measured in fewer studies but are also reported to be elevated in adults with obesity (4, 5). Adults who lose weight via dietary and exercise interventions with or without bariatric surgery experience a reduction in BCAA and BCKA levels (6–8). Feeding of BCAA-restricted diets in rodents support a causal relationship of BCAA with obesity and type 2 diabetes (T2D) (9, 10). Circulating glutamate, a product of BCAA metabolism, is also associated with both insulin resistance and obesity in adults (11). Finally, some short and medium chain acylcarnitines, such as the C3 and C5 species generated from BCAA metabolism, are consistently found in metabolite principal components associated with adult T2D and insulin resistance (2, 12). Together, these results establish that adults with obesity have a metabolic profile consisting of elevated BCAA, BCKA, and associated acylcarnitines and amino acids.

Adolescence is a critical window during which time successful treatment of obesity dramatically reduces the risk for adult obesity, T2D, and cardiovascular mortality. Significant reduction of childhood obesity by late adolescence reduces risk to levels similar to children who never experienced obesity (13, 14). A similar magnitude of weight loss in adults with obesity does not lead to the same degree of risk reduction, suggesting that obesity in earlier life may be driven by physiological mechanisms that are still reversible before adulthood. Yet, little is known about how the metabolic pathways of obesity differ between adolescence and adulthood.

Limited prior work has indicated that obesity in youth aged 4–19 years is characterized by altered BCAA and acylcarnitine catabolism and changes in nucleotides, lysolipids, and inflammatory markers also seen in adults with obesity, as well as reduced fatty acid catabolism that may be unique to obesity in children (15). Recent work comparing male and female adolescents with obesity in the age range of 12–18 years who underwent a 6-month lifestyle intervention program suggests a sex difference in the baseline level of circulating BCKA, with males having higher levels than females despite similar BMI profiles (16). In addition, a recent study showed an elevation in urinary BCAA and BCKA levels in female adolescent participants with T2D compared to obese, nondiabetic, or lean controls, accompanied by diversion of tryptophan metabolism from the serotonin to the kynurenine pathway (17). These metabolic indices may be related to the higher risk of T2D in obese adolescent females than in adolescent males or relatively higher rates of anabolic metabolism in males. However, the similarities and differences in obesity physiology between adolescents and adults remain unclear, especially with respect to sex and age.

Several studies in adults have reported gut microbiome alterations associated with obesity and related diseases (18). Fecal microbiome transplant (FMT) from adult humans with and without obesity to germ-free (GF) mouse recipients has indicated that the microbiome can causally contribute to obesity and associated metabolic phenotypes. GF mice receiving FMT from adult donors with obesity gain more weight, gain more fat mass, and have distinct microbiota-linked changes to serum metabolite profiles than those receiving FMT from a healthy weight adult donor (19–22). However, it remains unknown if the microbiome causally contributes to obesity physiology during earlier adolescent stages.

We established the Pediatric Obesity Microbiome and Metabolism Study (POMMS) to test the hypothesis that adolescents with obesity have metabolic and microbiome characteristics at baseline and following weight loss intervention that distinguish them from adults with obesity (23). Our study cohort was drawn from the Duke Children’s Healthy Lifestyles program, which offers multicomponent behavioral, pharmacotherapeutic, and surgical treatment options as per the current American Academy of Pediatrics guidelines for youth aged 23 and younger with obesity (24). By studying the microbiome and metabolome in this diverse population, we sought to identify new prognostic and therapeutic targets for greater obesity-associated disease risk reduction and reversibility in the most affected populations. Further, adolescents tend to be treatment naive. Thus, their clinical and metabolic measures are less complicated by drug effects. The study design, data collection, and demographics of the POMMS cohort were previously described (23). Based on the lack of gut microbial diversity differences we observed in our preliminary cohort analysis (23), we hypothesized that adolescent obesity would not be directly correlated with gut dysbiosis to the same extent as observed in adults. Here, we report our findings from the complete metabolome and microbiome datasets at baseline and after intervention.

Baseline characteristics. The POMMS clinical cohort was recruited upon referral to the Duke Healthy Lifestyles Clinic in Durham, North Carolina, USA. Details concerning recruitment, treatment plans, and demographics of our study cohort have been previously reported (23) and are summarized in Figure 1, A and B, and Supplemental Data; supplemental material available online with this article; https://doi.org/10.1172/JCI196742DS1 Participants with obesity (OB, n = 220) and healthy weight controls (HWC, n = 71) had similar distributions of race and ethnicity according to self-reported data. However, the OB group was significantly more female identifying and younger than the HWC group (Supplemental Table 1). CRP, HbA1c, ALT, LDL cholesterol, triglyceride, insulin, systolic and diastolic blood pressures, and triglyceride/HDL ratio were higher in the OB group compared with the HWC group (Table 1). HDL cholesterol levels were lower in the OB group compared with the HWC group. Fasting glycerol levels were also significantly higher in the OB group using a targeted lab analysis (Supplemental Table 2), suggesting higher levels of lipolysis in adolescents with OB compared with HWCs (25). However, these 2 groups had no significant difference in fasting glucose levels, suggesting that even though several metrics associated with OB in adults were significantly higher in the OB cohort when compared with HWCs, high BMI alone is not enough to significantly confer a difference in fasting glucose commonly observed in adults with obesity (26).

Figure 1

Significant metabolic changes are associated with clinical measures of obesity in adolescents. (A) POMMS study overview depicting the inclusion criteria for both the participants in the healthy weight control (HWC) and obesity (OB) groups. (B). Flow diagram depicting sample sets available at every timepoint of the observational trial. (C–F). Linear regression analysis adjusted for age, race, and sex of participant. Values with a False Discovery Rate (FDR) of < 0.1 are labeled as triangles, those with FDR < 0.2 are labeled as squares. See Statistics description in Methods for details. (C) Targeted serum metabolites associated with %95th BMI, (D) Homeostatic Insulin Resistance score (HOMA-IR), (E) Hemoglobin A1c (HbA1c), or (F) C-Reactive Protein (CRP), at baseline. For C–F, metabolites above the dotted line are considered nominally significantly associated with the noted clinical measure. Higher levels of metabolites in the right upper quadrant are significantly associated with higher levels of the clinical measure. In comparison, higher levels of metabolites in the left upper quadrant are significantly associated with lower levels of the clinical measure. %95th BMI is defined as the percent of the BMI above the 95th percentile. Diagram in A was created in BioRender.

Table 1

Baseline clinical lab values in obesity versus healthy-weight control groups

Serum metabolites reveal distinct adolescent sex- and age-dependent associations with BMI, insulin resistance, and inflammation. In an adjusted analysis of serum metabolites (Supplemental Table 3) in OB and HWC adolescents at baseline, levels of BCAA valine (Val), the amino acids tyrosine (Tyr), glutamine/glutamate (Glx), and phenylalanine (Phe), and C3 (propionyl, a bioproduct of BCAA catabolism) and C8:1 acylcarnitines were all significantly and positively associated with BMI measured as a percent of the BMI 95th percentile (%95th BMI) (27). In contrast, higher serum levels of the amino acids glycine (Gly), serine (Ser), histidine (His), methionine (Met), and several even chain acylcarnitines and decarboxylated acylcarnitines were significantly negatively associated with %95th BMI (Figure 1C), as were the ketoacids of BCAAs, leucine (KIC), and isoleucine (KMV). This is in notable contrast with adults in which these ketoacids are typically positively associated with obesity and T2D (4, 7). Several of the metabolites associated with %95th BMI at baseline were also associated with HOMA-IR at baseline, including those negatively associated (Gly, Ser, and acylcarnitines C20 and C6-DC/C8-OH) and positively associated (Val, Tyr, Phe, Glx, C3) with HOMA-IR (Figure 1D). HOMA-IR was also significantly positively associated with levels of leucine/isoleucine (Leu/Ile) and proline (Pro) but was not associated with BCKA. In adults, elevated BMI and HOMA-IR have similarly been associated with higher BCAA, Phe, Tyr, and C3 acylcarnitine (2, 4, 5). However, adult BMI and HOMA-IR are also associated with high levels of BCKA as noted above, but we did not observe this in this adolescent cohort. These data suggest that adolescents with OB have metabolic features like adults with obesity, including higher BCAA and associated metabolites, but also display adaptations that differ from adults, including lower BCKA in association with high BMI.

In analyses of HbA1c, only the amino acids Leu/Ile, Phe, and Tyr were nominally positively associated with HbA1c (Figure 1E); the acylcarnitines C3-C5, and BCKA KIV were negatively (but not significantly) associated with higher HbA1c, again suggesting that metabolites associated with IR and OB in adults are not significantly associated with key clinical measures of metabolic syndrome in our adolescent cohort. Several even chain acylcarnitines were significantly negatively associated with HbA1c levels, suggesting that high HbA1c is associated with impaired or incomplete fatty acid oxidation in adolescents. It should be noted that HbA1c is a predictive measure of progression to T2D in adults, but its use as a measure of metabolic health during adolescence can sometimes be difficult to interpret, as high levels can occur in adolescents with normal weight and fasting glucose profiles (28).

CRP, a clinical marker of general inflammation, nominally correlated with fewer metabolites either positively (Val, Phe, Asx) or negatively (acylcarnitine C20, His, Gly, and alanine Ala) at baseline (Figure 1F). Some of these CRP-metabolite associations were similar in directionality to those with %95th BMI or HOMA-IR (Val, Phe, His, Gly, C20), while others were unique to CRP (Asx) or exhibited reversed directionality compared with HOMA-IR (Ala). Whereas only a subset of acylcarnitine species measured here were significantly associated with these 3 clinical measures, we observed that most acylcarnitines were generally positively correlated with CRP but negatively correlated with HOMA-IR and %95th BMI and HbA1c. Therefore, the general association trend between most serum acylcarnitines and CRP is inverse to that of %95th BMI, HbA1c, and HOMA-IR.

BCAA metabolism in adolescent obesity is sex and age dependent. Prior work suggests that there is an interaction between obesity status, pubertal status, sex, and age that influences BCAA and their related metabolite levels (16). To further assess these metabolic signatures in our adolescent cohort, we analyzed data stratified by OB versus HWC status and age on serum metabolite levels to identify factors that influence metabolite levels in our cohort. We found initially that signatures associated with OB versus HWC status were driven largely by metabolite levels in male participants in the HWC group (Figure 2). We also found that serum levels of the BCAA, along with several other amino acids and acylcarnitines involved in BCAA metabolism (Figure 2A) were significantly varied by OB versus HWC status and sex. Males generally had higher levels of most affected amino acids and their metabolites. When we analyzed raw data and adjusted only for multiple comparisons, males with HWC status had significantly higher levels of Leu/Ile, Glx, and the BCAA metabolite C3 than males in the OB group, as well as the ketoacids of Leu and Ile (KIC and KMV, respectively; Figure 2, B–J). Following adjustment for age and race, several of these relationships remained significant, including higher levels of Leu/Ile, C8:1-OH/C6:1DC, C18, C20-OH/C18-DC, Met, His, Orn, and Cit in male participants in the HWC group compared with male participants in the OB group. Because adults with OB tend to have higher levels of BCKA compared with HWC (4–8), we tested if the opposite trends in the adolescent data were still seen in older adolescents by assessing metabolites affected by the interaction of obesity, sex, age, and Tanner (pubertal) stage (29). We found that the amino acids Leu/Ile, His, Orn, Val, Met, Cit, and Phe, and the acylcarnitines C18, C20-OH/C18-DC, C16, C18:1, and C8:1-OH/C6:1DC were significantly affected by the overall interaction between HWC versus OB status and sex (Figure 3A). Second stage testing further delineated the role of status and sex for each indicated metabolite. Based on the interaction of Tanner stage and HWC versus OB status, KMV (ketoacid of Ile) was significantly associated with Tanner stages 4 and 5, suggesting that more mature teens had a greater difference in serum levels of this ketoacid (Figure 3B). To test whether BCKA are associated with sex and obesity at younger ages, we analyzed metabolite data from the Hearts and Parks (HP) Study, which included children enrolled as early as age 5 (30). In these analyses of children aged 5–9 years, we found that the ketoacid KMV was no longer associated with sex in the younger children. Still, the BCAA acylcarnitine metabolite C3 was associated with sex in the older age group after stratified analysis (Figure 3C). These results indicate that obesity-associated differences in BCAA and related metabolite levels are sex dependent and emerge later in adolescence.

Figure 2

Branched chain amino acids (BCAA) and related metabolite levels are sex dependent, significantly different in HWC v OB cohorts, and unique to adolescents with obesity. (A) BCAA related metabolism pathway. Valine (VAL), Leucine (LEU) and Isoleucine (ILE) are reversibly transaminated by the branched chain amino acid transferase (BCAT2) in the mitochondria to their respective ketoacids ketoisovalerate (KIV), ketoisocaproate (KIC) and ketomethylvalerate (KMV). The amino group removed from the BCAA is transferred to α-ketoglutarate (α-KG), yielding glutamate (GLU). The ketoacids can then be irreversibly hydrolyzed by the branched chain ketoacid dehydrogenase (BCKDH) where BCAA-derived carbons can then enter the TCA cycle or contribute to lipogenesis. When BCAA are elevated, as in obesity, excess BCAA increases BCAT2 activity, which could result in an increased nitrogen load, especially in muscle tissues. This load can be relieved, most likely via coordinated action of serine dehydratase, serine hydroxymethyltransferase, and glycine acyltransferase. Resulting increased metabolites, such as glutamine (GLN) and acylglycine, can be secreted from the affected tissues. ALA, alanine; PYR, pyruvate. (B–J) Levels of serum amino acids and keto acids associated with BCAA and alanine-pyruvate metabolism in HWC versus OB cohorts divided by sex. Box plots represent mean and SD. Statistics were performed on mean metabolite values following log₂ transformation, and P values result from Mann-Whitney tests and are corrected for multiple comparisons. **P < 0.01; ***P < 0.001; ****P < 0.0001.

Figure 3

Sex-dependent differences in serum ketoacids are less evident in younger children than in more mature adolescents. Log₂-transformed levels of serum metabolites tested for significance in an overall interaction and between (A) sex and HWC versus OB status; (B) Tanner (puberty) stage and HWC versus OB status; and (C) Young age (9 years or younger) versus older age metabolites and sex. Only significant metabolites are shown, P ≤ 0.05, after adjustment for race and multiple comparisons. Complete metabolite results are shown in Supplemental Table 3.

Microbiome α and β diversity were similar between OB and HWC adolescents, but taxa and microbial features were distinct between cohorts. We analyzed 16S rRNA gene amplicon sequence data from 786 fecal samples covering 54 HWC participants at baseline, and from participants with OB that contributed a combination of fecal and serum samples, along with clinical measures, at 0, 6, 12, 18, and 24 months of the POMMS study (Figure 1B). As in the interim analysis (23), fecal microbial communities did not display significant differences in either α or β diversity based on HWC versus OB status at baseline, sex, or age at entry (Supplemental Figure 1A and data not shown). When we tested whether specific bacterial lineages might be associated with %95th BMI at baseline, we found that relative abundance of several family-level taxa were positively correlated with %95th BMI, while the HWC group had significantly higher abundance in members of several phylogroups, including the Bifidobacteriaceae, Lachnospiraceae, and Bacteroiodaceae, compared with the OB group (Supplemental Figure 1B).

To gain further insight into these microbial communities, shotgun DNA sequencing data were obtained for 396 participant-derived fecal samples, including quality controls for sequencing fidelity and sampling replication. High quality sequence data were obtained for 140 OB at baseline and 6 months and 50 HWC participants at the baseline time point. Consistent with the 16S rRNA data, α and β diversity did not significantly differ between the OB and HWC cohorts (Supplemental Figure 2), and profiles of the most abundant family- and species-level taxa were generally similar between the 2 cohorts (Supplemental Figure 3). Further analysis determined, however, that participants from the OB cohort had a significantly higher Firmicutes-to-Bacteroidetes ratio than HWC (P = 0.001). ALDEx2 (31), ANCOM-BC (32), and corncob (33) analyses were employed to identify differential taxa between the 2 groups at baseline. The methods detected an increased relative abundance of genera Ligilactobacillus and reduced Alistipes among the OB participants. Of the Ligilactobacillus, the species L. ruminis was most prevalent (> 90% of samples), and of Alistipes, the species A. finegoldii was found in most samples (> 90%). Notably, Ligilactobacillus and Alistipes abundance remained positively and negatively correlated, respectively, with %95th BMI (Figure 4A). Notably, the taxa associated with HOMA-IR, %P95BMI and CRP shared little overlap, with Alistipes finegoldii as the only taxon negatively associated with all 3 measures (Figure 4B).

Figure 4

Several gut microbial taxa and predicted enzymatic functions are associated with clinical variables at baseline. (A) Abundance of microbial species determined by fecal metagenomic sequencing that were positively or negatively significantly associated with %95th BMI at baseline. (B) Venn diagram showing shared and distinct numbers of fecal microbial species associated with %95th BMI, insulin-resistance score (HOMA-IR) and C-reactive protein (CRP) levels at baseline. (C) Linear regression correlation following dimensionality reduction between covarying microbial factors and clinical variables at baseline. Factors are named for prominent taxa within each cluster, and the underlying taxa and weights in each cluster are listed in Supporting Data tables. (D) Top 20 KEGG pathways or orthologs in metagenomic sequences that predict OB versus HWC status at baseline in a random forests machine learning model. Predicted functions are labeled by symbols and remaining undefined predicted KEGG orthologs or pathways are described in Supporting Data.

To reduce data dimensionality, taxa and functional capacity of the gut microbiomes were clustered using hierarchical clustering, and the resulting factors were associated with clinical scores at baseline (Figure 4C and Supplemental Figure 4). Though microbial species diversity was similar between the 2 cohorts at baseline, we identified 2 taxonomic clusters significantly associated with either %95th BMI or CRP as continuous variables (Figure 4C, with factors defined in the Supporting Data file). Further clustering of serum metabolomics, KEGG orthology functions, and taxa revealed associations with clinical variables. We found several clusters of covarying taxa, genes, phages, and serum metabolites, indicating that there is interaction between metabolomic status and the microbiome (Supplemental Figure 4). Using a random forest machine learning model, we generated a model that predicts OB versus HWC status, which was significantly enriched for pathways involved in microbial amino acid synthesis, 2-component signaling, lipopolysaccharide biosynthesis, and sugar metabolism and transport (Figure 4D and Supplemental Figure 5, A and B). Notably, several genes required for microbial synthesis of BCAA and tryptophan were independently identified as significant contributors to this model, mirroring prior research that associated gut microbial BCAA synthesis genes with obesity and T2D in adults (22, 34) and children with obesity (35). No factors were significantly associated with HOMA-IR score at baseline (data not shown).

We next generated metagenome-associated genomes (MAGs) from our shotgun sequencing data and found that there was no difference in either α or β MAG diversity between HWC and OB groups at baseline (data not shown), and there was no single MAG taxon significantly associated with HWC versus OB status at baseline (Supplemental Table 4). We then tested whether any MAGs were associated with clustered serum metabolites and found that 3 amplicon sequence variant (ASV) of Faecalibacterium prausnitzii were negatively associated with a small cluster of acylcarnitines consisting of C10:2, C10:3, and C8:1, and 2 ASV of Anaerobutyricum hallii positively associated with ketoacid concentrations (Supplemental Table 5, with clusters defined in Supplemental Table 6).

Donor-status–independent metabolites and microbes were associated with weight gain and percent adipose in FMT-recipient mice. Previous FMT experiments demonstrated that fecal microbiome content can confer phenotypic and metabolomic traits of the donor to germ-free mouse (20–22) and human (36) recipients. However, all prior studies that examined the role of the microbiota in obesity used fecal samples from adult donors. We hypothesized that adolescent microbiota, influenced by host growth and development, are optimized for peak energy harvest, regardless of OB-HWC status, and that weight gain may not differ between mice receiving FMT from HWC or OB donors. To test this, we chose FMT samples of male and female donor adolescents from the HWC and OB groups. Germ-free male C57Bl/6J mice aged 5–7 weeks were weighed, gavaged with 150–200 mL of donor fecal slurry (Figure 5A), housed under gnotobiotic conditions, and fed a sterilized standard chow diet ad libitum with weight measured every 3–4 days (Figure 5B), and then euthanized at 2 weeks post gavage. While there was a nonsignificant trend towards more weight gained (Figure 5D) and heavier epididymal fat pads (Figure 5E) and adiposity (Figure 5F) in mice that received donor slurry from adolescents with OB, we did not observe significant changes in weight gain or fat pad mass associated with donor OB versus HWC status or with sex or race of donor (data not shown). The strongest predictor of weight gain across all recipient mice was the starting weight of the recipient (R² = 0.72, P < 0.0001). Comparing metabolite profiles across all recipient mice, levels of serum amino acids Gly, Tyr, Asx, and Arg, as well as acylcarnitines C16:1, C10:3, C8:1C4-DC/Ci4-DC were negatively associated with weight gained (Figure 5C).

Figure 5

Fecal microbial transplants from adolescent POMMS study participants did not significantly affect weight gain in recipient mice but did influence serum metabolites. (A) Experimental design. (B) Average weight and SD of each group of mice over the 2 weeks after gavage period. (C) Mouse serum metabolites associated with percent weight gain over the FMT incubation period. Dots above the dashed line represent metabolites significantly associated with less weight gain over time following linear regression analysis. (D) Mean percent weight gain of all mice grouped by donor (HWC versus OB) at baseline. (E) Mean epidydimal fat pad weight grouped by donor at baseline. (F) Mean percent epidydimal fat pad weight grouped by donor at baseline. (G) Fecal microbial taxa were identified by 16S rRNA gene sequencing. ASV significantly associated (P ≤ 0.05 after adjustment for multiple comparisons) with clustered mouse serum metabolites (Supplemental Figure 6) using ANCOM-BC (33) are shown on the heatmap. Clusters are described by the predominant metabolite species in each cluster. Each taxon in the heatmap represents a unique ASV identified to the lowest classification possible. BCAA, branched chain amino acids; AC, acylcarnitines.

We were intrigued to see that microbiome samples from some individual adolescent donors were able to confer greater weight gain over others, raising the possibility that some transplanted human gut microbes might be associated with weight gain. Indeed, we detected a microbial signature associated with weight gained, percent adipose, and serum metabolite features (Table 2). Fecal microbial α diversity was significantly higher in mice receiving donor microbiomes from HWC donors than OB donor recipients (Supplemental Figure 6C). We found Bray-Curtis dissimilarity of fecal microbial composition at 2 weeks after colonization to be significantly associated with percent weight gain and donor group (P values < 0.01) but not adipose percent change. We also saw significant engraftment of Akkermansia, Ruminoccocus, and Bacteroides genera in recipient fecal samples 2 weeks after gavage compared with the composition of the donor sample, and a relative decrease in members of the Blautia genus throughout engraftment (Supplemental Figure 6B). Only 2 taxa were significantly negatively associated with weight gain across all recipient mice – ASV 153 Barnesiella sp. (log₂ fold change –0.80, P_adj. 5.2^–13) and ASV 21 Collinsinella aerofaciens (log₂ fold change –0.19, P_adj. 0.02). We also found several microbes significantly associated with clusters of covarying metabolites (Supplemental Figure 6C). Many of these microbes are significantly associated with metabolite cluster 1, comprised of mostly even medium-to-long chain acylcarnitines. Cluster 2 was defined by the BCAA valine, BCKA KIC, KIV, and KMV, along with BCAA metabolite 3-HIB (Supplemental Figure 6F) (37). This metabolite cluster was both positively and negatively associated with several Blautia and Bacteroides ASVs (Figure 5G).

Table 2

Gut microbial taxa significantly associated with percent adipose tissue in mice 2 weeks after Fecal microbiome transplants

When we examined the level of human donor microbes detected in the mice following the 2-week colonization period, we found that several taxa were identified in mice that were not detected in the donor gavage sample, and vice versa (Supplemental Figure 6D). The levels of engraftment of human donor taxa in the mice fecal samples, averaged 27% (± 7% SD) at the ASV level and 45% (± 10% SD) at the genus level (Supplemental Figure 6E). These data suggest that any microbiome-transferrable phenotype (or lack thereof) may depend on the human-sourced consortia that is able to colonize the germ-free mouse gut.

Serum levels of aromatic and branched chain amino acids are positively associated with BMI and IR over the 6 month observational study period, while levels of even chain acylcarnitines are negatively correlated with change in BMI and IR. We next sought to determine whether metabolic shifts cooccur with improvements in health metrics during weight-loss intervention. Using linear regression models adjusted for age, race, and sex, we examined whether a change in metabolites correlated with a positive or negative change in BMI or HOMA-IR. We observed that, in general, changes in aromatic and branched chain amino acids were positively associated with change in BMI (Figure 6D) and HOMA-IR (Figure 6E) over the 6 month observational study period, while changes in several even-chain acylcarnitines and glycine were negatively associated with changes in BMI and HOMA-IR. While the majority of participants either maintained or reduced their %95th BMI score, HOMA-IR increased for most participants over the observation period (Figure 6, A–C, and Supplemental Figure 7, A and B), which is a common phenomenon during adolescence (38–40). No change in any metabolite was associated with change in CRP (Supplemental Table 7) or HbA1c after false discovery rate correction (data not shown).

Figure 6

Serum aromatic amino acids increase with BMI and HOMA-IR scores over the 6 month observational study period. (A) Absolute change in %95th BMI in OB cohort participants over the observational study period. (B) Percent change in HOMA-IR over the observational study period. (C) Graph of change in %95th BMI and percent change in HOMA-IR over the observational study period. (D and E) Estimated coefficients for metabolite levels associated with change in %95th BMI (D) or percent change in HOMA-IR (E), after adjusting for age, race, sex, and baseline measurement. For A–B, all participants with net reduction in the indicated measure are labeled by observational treatment type. For C, all patients with recorded change in %HOMA and change in %95th BMI data are labeled by observational treatment type. For D and E, acylcarnitines are labeled in black, amino acids are labeled in purple, and ketoacids are labeled in green. Correlations that met false discovery rate (FDR) cutoff of < 0.1 are labeled as triangles, and those that met an FDR cutoff of < 0.2 are squares.

Microbiome taxa and pathways at baseline predict change in health measures over the observational period, while distinct sets of taxa and features are associated with change in CRP. We analyzed the metagenomic data to determine if there were any microbial taxa or pathways at baseline that were significantly associated with change in BMI, HOMA-IR, CRP, or HbA1c. While there were no significant associations with change in HOMA-IR, we found that relative abundance of several microbial taxa at baseline were predictive of changes in BMI (Figure 7A), HbA1C, or CRP levels, with some taxa shared between changes in 2 measures (Figure 7B). For example, Bacteroides thetaiotaomicron, finegoldii, and uniformis as well as Parabacteroides distasonis and merdae were all associated with change in CRP and %95th BMI, but often in opposite directions. In addition, Anaerostipes hadrus, Blautia producta, and Escherichia virus Lambda_4A7 were all commonly associated with a change in HbA1c (Supplemental Figure 7C) and CRP (Supplemental Figure 7D). Notably, 93 taxa were significantly associated with changes in CRP, suggesting that the subset of microbes that potentially predict changes in this inflammatory marker are more numerous when compared with those associated strictly with BMI or other metabolic health indicators. Finally, we tested for potential associations between change in MAGs and change in clinical metrics. While the change in abundance of any MAG did not significantly associate with change in HOMA-IR or BMI in the OB cohort, we did find several taxa that significantly associated with CRP, including 3 Blautia wexlerae ASV, 2 Bifidobacterium adolescentis ASV, 1 B. bifidium ASV, and 1 Anaerobutyricum hallii ASV. These were all negatively associated with increase in CRP (Supplemental Table 8).

Figure 7

Microbial taxa associated with longitudinal outcomes. (A) Differential abundance of microbial taxa at baseline associated with change in %95th BMI over the 6-month observational period. (B) Venn diagram of microbial taxa shared between changes in clinical outcomes over the study period. See Statistics in Methods for details.

Based on studies of adult obesity indicating that the gut microbiome plays a substantial role in driving obesity-associated phenotypes, similar microbiome and metabolic coupling has been presumed in adolescent obesity. Here, we studied a cohort of adolescents with and without obesity to compare their metabolic and microbiome states, with the hypothesis that the obesity-associated adolescent metabolome and microbiome differ from that reported in adults with obesity. We further postulated that the adolescent metabolic rearrangement associated with obesity would not be associated with gut dysbiosis to the same extent as observed in adults, suggesting a decoupling of metabolic state and the gut microbiome in young individuals.

BCAA and BCKA have an inversely correlated signature in obesity that is unique to adolescents. In our study of adolescents with obesity, we found that serum levels of the BCAA and their direct metabolites, the BCKA, significantly differed between the OB and HWC cohorts. The BCKA skewed higher in the HWC group, whereas BCAA were higher among adolescents with obesity. Circulating levels of BCAA have been associated with obesity, IR, and T2D in adults since the late 1960s (2, 3, 41). At baseline, BCAA and aromatic amino levels are prognostic for the development of T2D in adults in long-term studies (1, 42), and this same metabolite cluster also predicts improvements in HOMA-IR in response to weight loss (43). While circulating amino acids are often measured in metabolite studies of adults, it is less common that those data sets also include BCKA measurements. When measured, BCKA levels tend to mirror BCAA levels in adults with obesity, with both being higher than in HWC (7) and both dropping significantly after effective treatments for obesity and T2D such as an exercise-based intervention (8) or bariatric surgery (44, 45). This may be related to obesity-driven changes in expression and activity of the kinase (branched-chain ketoacid dehydrogenase kinase, BCKDK) and phosphatase (protein phosphatase Mg²⁺/Mn²⁺ dependent 1 kinase, PPM1K) enzymes that control phosphorylation state and activity of branched-chain ketoacid dehydrogenase (BCKDH), the rate-limiting enzyme of BCAA catabolism (46, 47). Similarly, studies in adult rodent models, including ob/ob mice and Zucker-obese rats, show combined higher levels of BCAA and BCKA in the obese state (7, 46). Our finding of anticorrelated BCAA and BCKA levels in adolescents with obesity indicates a metabolic phenotype during these critical early stages that is distinct from that found in adults with obesity (Figure 8). We speculate this could be due to flexible regulation of PPM1K/BCKDK in adolescents, allowing for BCAA catabolism even with higher circulating BCAA, or patterns of organ-specific BCAA metabolism in adolescents that are distinct from what has been observed in adults. In agreement with our results, a recent study in adolescents suggested that high BCAA and low Gly should be included as part of a metabolic signature that could predict β cell failure (48).

Figure 8

Obesity in adolescence is characterized by unique transitional metabolomic and microbiome features. Lines indicate the direction of microbiome and metabolome features in individuals with obesity (OB) relative to age matched healthy weight (HW) individuals. Several of these features display changing relationships with OB status across the lifespan. Dashed lines indicate age ranges where there is currently missing data for that feature. We found that while high BMI is associated with high levels of serum BCAA, BCKA differences between OB and HW adolescents are sex dependent and pronounced in later stages of puberty. Further, serum measures of insulin resistance increased during adolescence for nearly all participants during our observational study, regardless of whether the participant lost weight, but there was no difference in fasting glucose between OB and HWC at baseline. We observed no significant difference in the ability of adolescent microbiome to transmit more weight gain to germ free mice, while evidence suggests that microbiome transplant from adult donors with OB leads to more weight gain in recipients. Evidence for each of these associations is cited in the text.

We found that the intersection of sex and obesity status plays a significant role in BCAA and BCKA levels. Our data suggest that these effects do not emerge until puberty onset, as we found that there were few sex-based differences in BCAA-BCKA metabolites in kids aged 5–9 years (30) (Figure 8). Our observation of sexual dimorphism in circulating levels of BCAA and their metabolism in adolescents is in accord with previous studies. Kobayashi and colleagues demonstrated that female sex hormones affected the diurnal regulation of the enzyme complex responsible for BCAA metabolism in rat liver, whereas gonadectomy of adult male rats did not affect circadian BCAA metabolism (6). Population-based human data examining the serum metabolomic differences of 1,756 adult participants (903 females and 853 males) suggested that BCAA metabolites were significantly higher in males and were significantly associated with male sex when taken as a group (49). In another study, exercising men had higher rates of leucine oxidation than women (50). Focusing on children and teens, Newbern and colleagues (51) published that, in a cohort of 82 teens with obesity, BCAA levels and byproducts of BCAA catabolism are higher in teen boys with obesity than in girls of comparable BMI. A subsequent study by the same group found that plasma levels of KMV, the ketoacid of isoleucine, went down in males but not in females when measured 6-months after intervention (16). The mechanisms underlying these emerging sex- and age-associated differences in BCAA metabolism remain unknown, but they could represent useful diagnostic and prognostic markers of obesity-associated pathologic progression.

Lower serum glycine is a common signature of obesity in adolescents and adults. We also found that serum glycine levels were significantly lower in adolescents with obesity when compared with the HWC group. This observation has been documented in adult obesity (52–54), and glycine levels increase after weight loss and improvement with insulin sensitivity (8, 44, 45, 55, 56). A recent study showed that elevations of BCAA in obesity activate glycine consumption to replenish pyruvate needed for the alanine transaminase reaction and nitrogen unloading in skeletal muscle (57) (Figure 2A). Treatment with a specific BCAT inhibitor that interferes with BCAA transamination increases muscle and plasma glycine levels in Zucker fatty rats (9). Our data suggest that the reciprocal relationship between BCAA catabolism and glycine levels exists in adolescents, despite their divergent BCKA signatures.

Acylcarnitine data from adult humans suggests that serum or plasma levels of C3-C5 species, which are derived from the metabolism of BCAA and other amino acids (58), often cluster in principal component metabolite factors along with the BCAA and aromatic amino acids, with higher levels in obesity, insulin resistance, and T2D (2, 43, 59). Low-grade chronic inflammation is a hallmark of T2D (60, 61) and obesity (62), and CRP levels drop with weight loss (63, 64). Herein, we identified an acylcarnitine signature that associated with CRP levels at baseline and that was distinct from the acylcarnitine species associated with baseline BMI and HOMA-IR. Fewer data sets incorporate metabolites when comparing HWC and OB in children and adolescents. One of the most comprehensive reports describes a cohort of 803 Hispanic children in the Viva la Familia study, ranging in age from 4–19 years and split between HWC and OB at a single timepoint (15). They examined 304 named serum metabolites and found that their OB participants had increased BCAA and acylcarnitine metabolites and increased inflammation markers. The Viva la Familia study did not include BCKA measurements, nor did it include longitudinal sampling or interventions. The data sets reported here therefore provide new insights into the relationships between BCAA, BCKA, and acylcarnitine species in the context of pediatric obesity and intervention outcomes.

Functional microbiome signatures are associated with obesity and metabolic signatures. Fecal microbiome transplants from adult lean donors and donors with obesity into germ-free mice demonstrate that the microbiome can contribute to weight gain and adiposity and affect the serum metabolomic profiles of the recipients (20, 22, 65). While we did not observe significant differences in α and β diversity between our adolescent case and control groups, the functional suite of genes represented in the microbiome associated with obesity in adolescence may still play a role in the serum availability of some obesity-related metabolic biomarkers. We found significant associations between host BMI status and presence of microbial genes required to synthesize several amino acids, including the BCAA and tryptophan (Supplemental Figure 5B). Research in pigs suggests that approximately 30% of dietary glycine is consumed by the small intestinal microbiota (66, 67), while other work in both conventionally raised and germ-free mice and rats suggests that the gut microbiome contributes on the order of grams per day of essential amino acids to the nutritional pool (65). Further, human FMT studies have revealed that adolescent recipients of adult donor stool retain specialized metabolic functions observed in healthy adolescents, including an elevated capacity for BCAA synthesis (68). In the future, longitudinal studies in preclinical models of diet-induced obesity could be used to clarify the role of age and sex in BCAA and acylcarnitine metabolism, to investigate mechanisms underlying the transition from high BCAA–low BCKA in youth to high BCAA–high BCKA in adult obesity, and the contributions of the microbiome to those metabolic shifts. Further, continued study of baseline metabolomic and microbiome associations with changes in health outcomes could be used to help predict successful treatment therapies in a personalized scenario.

Gut microbiome across HWC and OB adolescents is less differential than in adult cohorts. Our results provide important insights into the composition and function of the gut microbiome of pediatric obesity. Previous studies suggest that the gut microbiome undergoes changes between adolescence and adulthood in healthy individuals (69). Adults with obesity and healthy weight display substantial differences in gut microbiome diversity and composition (22, 70, 71), and those differences are distinct from adolescents with obesity or healthy weight (72). Our comparisons of adolescents with obesity and healthy weight detected no differences in diversity and limited differences in composition and functional potential. When we analyzed MAG abundance in association with clinical measures, we found that changes in CRP were significantly but negatively associated with abundance in several ASV from the Bifidobacterium, Blautia, and Anaerobutyricum genera. For Bifidobacterium, these results align with literature suggesting that Bifidobacteria are associated with both youth and health due to their high abundance in breast-milk fed infants, and with protection from high fat diet–induced obesity in rodents and negative association with visceral adiposity (73–75). Anaerobutyricum (formerly Eubacterium) (76) hallii is thought to be restricted to humans (77) and was notably absent from the mice following FMT even though it was detected in the gavage samples (Supplemental Figure 6D). This species is associated with improvement in insulin sensitivity in genetic mouse models of OB (78). Abundance of Blautia species has been both negatively and positively associated with OB or inflammation (79).

Previous studies examining the role of the microbiome in the development of obesity in germ-free animal models demonstrated that the microbiome from adult donors with OB contributes to more weight gain, higher adiposity, and higher circulating BCAA compared with a microbiome transplant from a lean adult donor (22, 80). However, there was no specific association between weight gain and donor BMI in our FMT studies here from adolescent donors. The caveat, however, for all such FMT experiments, is the potential for limited engraftment of the human microbiome in the animal (Supplemental Figure 6, D and E). We also only examined the relationship between donor microbiota and weight gain under a chow diet. These relationships could be further tested using a Western or high-fat diet to better replicate typical dietary intake in American adolescents (81, 82). However, these results raise the possibility that the gut microbiomes harbored by adolescents with obesity do not yet have sufficient compositional or functional changes to causally affect host metabolism, whereas microbiome changes and causal consequences increase progressively as individuals with obesity progress into adulthood (Figure 8). In the future, it will be important to determine if the ability of the gut microbiome to contribute to obesity phenotypes is due to microbiome functional differences (72) or differences in host growth and physiology during that critical stage. The resulting information could inform efforts to target the microbiome in treatment of obesity.

Recently, several medications were FDA approved for the treatment of obesity in adolescents age 12 and older (24). While these agents are remarkably effective in reducing body mass index (BMI), questions remain about the necessary duration of treatment. Some studies have suggested that a normalization of BMI prior to adulthood, in effect, “resets” the metabolic pathways and may offer hope for these teens to avoid the need for lifelong medication. In the future, it will be interesting to determine how these newly approved medications affect the distinct metabolic and microbial phenotypes described here.

One limitation of this study is that we did not have longitudinal sampling of any HWC adolescent to examine any changes in insulin resistance associated with typical adolescence to compare with adolescents with OB in our cohort (38–40). While none of the participants were treated with medications targeting weight loss at baseline, another limitation is that we were underpowered to understand the effect of any single medication on microbiome or metabolome in the cohort with obesity at the final timepoint. One objective of our study was to test whether microbiome transplant from teens with obesity would be sufficient to alter physiology in humanized mouse FMT experiments. These experiments were also limited by a relatively small sample set of donors and the analysis of a single timepoint following fecal colonization under normal dietary conditions. Future studies that build on this work could include Western or high-fat diets to better model the role of transplanted microbiomes in overnutrition settings.

Sex as a biological variable. Both male and female participants were included in this study. All relevant analyses were adjusted for sex, age, and race. Participants self identified their race and ethnicity based on investigator-defined options, with more than one option allowed for each category.

Supplemental methods. Detailed descriptions of serum metabolomics, fecal DNA preparation and sequencing, and metagenome associated genome assembly and analysis can all be found in the Supplemental Methods document associated with this article.

Fecal microbiome transplant experiments. Human fecal samples were prepared for animal gavage as described (84). Briefly, samples were initially frozen at –80°C and then thawed, opened in a flexible film anaerobic chamber (Coy Laboratory Products), homogenized with a sterile plastic loop, and distributed into 2 ml cryovials in ~200 mg aliquots without preservative. Samples were then stored at –80°C. Immediately before fecal transplant, samples were thawed, resuspended into 10% weight per volume sterile PBS +0.05% cysteine, vortexed repeatedly and allowed to settle at room temp for 5–10 minutes. Supernatant was then removed for gavage. Three to 5 male germ-free C57BL/6J mice between 5–7 weeks of age received between 150 and 200 ml of gavage slurry for each human sample administered. Mice were then housed with ALPHA-Dri (Shepherd Specialty Papers) bedding for 2 weeks in sterile filter top cages and fed autoclaved Teklad 2020SX diet (Envigo) and water ad libitum. Mice were weighed at gavage and every 3–4 days following fecal transplant. At 2 weeks, mice were euthanized by isoflurane overdose and immediately exsanguinated by cardiac puncture. Fecal samples and tissues were removed and snap frozen in liquid nitrogen.

Statistics. Clinical lab measures and targeted metabolites were log₂ transformed to ensure normality. Metabolites with missing values (i.e., below the lower limits of quantification) in greater than 25% of samples were not further analyzed. Patients that underwent bariatric surgery and for whom we had a complete data set (n = 2) were excluded from the longitudinal metabolomics analysis. For the remaining metabolites, values below the limit of quantification were imputed at half the minimum observed result (85). Logistic regression models were used to test association between each clinical lab measure with OB versus HWC groups, controlling for age, sex, and race/ethnicity. Serum levels of 65 targeted metabolites were investigated for associations with baseline %95th BMI, HOMA-IR score, HbA1c, and CRP using linear regression models adjusting for age, sex, and race/ethnicity. The Benjamini-Hochberg method was applied to control the false discovery rate (FDR) at 0.1 for multiple comparisons. The association between case-control status and each metabolite was assessed using linear regression models with the interaction between status and age, sex, or race in separate models. The association between change in each clinical outcome (%95th BMI, HOMA-IR score, HbA1C, and CRP) and change in each metabolite between baseline and 6 months were assessed using linear regression models adjusting for age, sex, race/ethnicity, and baseline clinical outcome. All analyses were conducted in R 4.1 (R Core Team, 2022).

Microbiome data was analyzed as follows. Spearman correlation was used to assess the association of α diversity metrics (Observed species, Shannon index, Faith phylogenetic diversity) with mouse or human physiologic measures, and Wilcoxon rank sum test was used to determine significant differences between diversity metrics. Principal coordinate analysis with Bray-Curtis dissimilarity was used to determine differences in overall microbial composition between gavage and fecal samples as well as a clear clustering of fecal samples by donor. Compositional differences were determined by PERMANOVA (R package vegan). Hierarchical clustering (R package hclust, default setting) was used to cluster metabolites as well as species in the metagenomic data set. Taxa were assembled, annotated, and functionally characterized using Kaiju (86) and HUMAnN (87) pipelines. Differential abundance analysis of ASVs with relative abundance above 0.1% in at least 20% of the samples (R package ANCOMBC) was performed for donor weight group, percent weight gain, and adipose percentage change, respectively, as well as for metabolite factors jointly. The Benjamini-Hochberg method was applied to control the false discovery rate (FDR) at 0.05 for multiple comparisons of ASVs. Random forests machine learning was (88) used to model associations between fecal metagenomic taxa or functional genes and clinical measures.

Study approval. Recruitment of participants and handling of clinical samples was approved as part of the Pediatric Obesity Microbiome and Metabolism Study (POMMS; Clinical trial identifier # NCT03139877) described in detail in ref. 23, and approved by the Duke Institutional Review Board in document Pro00074729. Written informed consent was received prior to participation and prior to inclusion of serum and fecal samples for further study in the POMMS biobank, which was governed under IRB protocol Pro00074546. Data, serum, and stool samples from the study participants are available as a repository (Clinical trial identifier #NCT02959034) for further research as described at https://sites.duke.edu/pomms Sample size estimates were based on likelihood of recruitment via the Duke Healthy Lifestyles clinic and estimated based on difference in BMI z-score by treatment group, with the exclusion of the bariatric surgery group, as this number was very small and considered exploratory. A sample size of n greater than 34 per group was estimated to provide greater than 80% power (2-sided α = 0.05, assuming SD of 0.20) to detect a mean treatment difference of –0.14. This effect size was achieved by a comparable intervention trial with youth and was found to be clinically significant (83). All animal experiments were performed either at the Duke University Gnotobiotic Core or in the National Gnotobiotic Rodent Research Center at the Center for Gastrointestinal Biology and Disease at the University of North Carolina Chapel Hill in AAALAC-accredited facilities and according to IACUC-approved protocols.

Data availability. Data collected as part of the POMMS project can be found in the Supplemental Data Tables file. Data used to generate figures is shown in the Supporting Data Values file. All raw sequencing data can be accessed via the NIH Sequence Read Archive at Accession number PRJNA1403442. Code used to generate figures is available at https://gitlab.oit.duke.edu/duke-microbiome/dmc-analysis/POMMS_JCI