Targeted metabolic signatures and diet in associations with obesity and insulin resistance in young adults

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1

Targeted metabolic signatures and diet in associations with
obesity and insulin resistance in young adults

By

Chenyu Qiu

A Thesis Presented to the
FACULTY OF THE KECK SCHOOL OF MEDICINE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
BIOSTATISTICS

May 2020

Copyright 2020 Chenyu Qiu
ii
TABLE OF CONTENTS

List of Tables ................................................................................................................................. iii
List of Figures ................................................................................................................................ iv
Abstract ........................................................................................................................................... v
1 Introduction .............................................................................................................................. 1
2 Materials and methods .............................................................................................................. 4
2.1 Study recruitment and study design ..................................................................................... 4
2.2 Adiposity and cardiometabolic outcomes ............................................................................ 6
2.2.1 Adiposity outcomes ....................................................................................................... 6
2.2.2 Glucose metabolism ...................................................................................................... 6
2.2.3 Lipid metabolism ........................................................................................................... 7
2.2.4 Inflammatory cytokines, chemokines and cardiovascular markers ............................... 7
2.2.5 Laboratory assays .......................................................................................................... 8
2.3 Metabolomic signatures ....................................................................................................... 8
2.4 Statistical methods ................................................................................................................ 9
3 Results .................................................................................................................................... 12
3.1 Association between metabolites and cardiometabolic traits ............................................. 12
3.2 Classification of metabolite PCs and outcome PCs ........................................................... 21
3.3 The association of metabolite PCs and outcome PCs ........................................................ 26
3.4 The association of dietary intake and outcome PC score, and effect modifications by
metabolite PC scores ................................................................................................................. 28

4 Discussion .............................................................................................................................. 33
References ..................................................................................................................................... 37

iii
List of Tables
Table 1. Sociodemographic characteristics of 107 Children's Health Study (CHS) adolescents
and young adults. .......................................................................................................................... 14
Table 2. Characteristics table for metabolites variables of 107 Children's Health Study (CHS)
adolescents and young adults. ....................................................................................................... 15
Table 3. Characteristics table for outcome variables of 107 Children's Health Study (CHS)
adolescents and young adults. ....................................................................................................... 16
Table 4. Loadings for the first four principal components identified from 23 targeted metabolites
concentrations among 107 Children's Health Study (CHS) adolescents and young adults. ......... 24
Table 5. Loadings for the first three principal components identified from 14 cardiometabolic
traits of 107 Children's Health Study (CHS) adolescents and young adults. ................................ 25
Table 6. Characteristics table for 24 hours dietary intake of 107 Children's Health Study (CHS)
adolescents and young adults. ....................................................................................................... 30
Table 7. Association between dietary intake and lipids-related outcome PC score stratified by
dichotomous BCAA-related metabolomic PC score. ................................................................... 32
Table 8. Association between dietary intake and lipids-related outcome PC score stratified by
dichotomous BCAA-related metabolomic PC score, additionally adjusting for body fat percent.
....................................................................................................................................................... 32

iv
List of Figures
Figure 1. Heatmap of similarity scores for the 1
st
and 2
nd
component identified from PLS
regression ...................................................................................................................................... 20
Figure 2. Scree and variance plots for principal components identified from 23 metabolites
among 107 CHS young adults. ..................................................................................................... 22
Figure 3. Scree and variance plots for principal components identified from 14 outcomes among
107 CHS young adults. ................................................................................................................. 22
Figure 4. Heatmap of loadings for the first four principal components identified from 23 targeted
metabolites among 107 young adults. ........................................................................................... 23
Figure 5. Heatmap of loadings for the first three principal components identified from 14
cardiometabolic outcomes among 107 young adults. ................................................................... 23
Figure 6. Association of metabolite PCs and metabolism trait PCs in 107 participants. ............. 27
Figure 7. Association of dietary intake and lipids-related PC scores, and effect modifications by
BCAA-related metabolomic PC scores in 107 participants. ......................................................... 31

v

Abstract
Background: The effects of metabolites on altering cardiometabolic traits including adiposity,
inflammation, glucose and lipid metabolism have been shown in several adults and animal studies.
However, only a few studies have conducted investigations on adolescents and young adults. The
biological mechanism of metabolic dysfunction, as well as the interaction between dietary intake
and metabolites in the association of key metabolic pathway remain unclear in young adults.
Objective: The purpose of this research was to investigate the association of targeted metabolic
signatures and diet with obesity and insulin resistance in young adults.
Methods: The 107 young adults, aged 18-22 years, were a subset of the Meta-air study recruited
from the Children’s Health Study in 2014-2018. Outcomes included adiposity measures, glucose
metabolism traits, lipids measures, inflammatory markers, and cardiovascular markers. The
concentration of the 64 targeted metabolites profiles were measured in fasting serum samples
from participants. Dietary intake variables, including daily total calorie intake, grams of
macronutrients intake, dietary glycemic index and percent calories from macronutrients, were
estimated by 24-hour dietary recall. Partial least squares regression was used to identify the
association between 64 targeted metabolites and 42 outcomes. Principal components analysis was
used to identify clusters that represent major metabolic pathways. Linear regression models were
built to assess the association between metabolite PCs and cardiometabolic trait PCs. Last,
multiple linear regression models were used to analyze the association of diet and outcome PC
score, and effect modification by metabolite PC scores.
Results: The aromatic amino acid, L-Glutamic acid & L-Glutamate, and L-Alanine related PC
score was positively associated with all outcome PC scores (p = 0.048 for insulin-resistance

vi
related PC score, p < 0.001 for adiposity-related PC score, and p = 0.01 for VLDL- and total
triglycerides-related PC score). BCAA-related metabolite PC score was positively associated with
the VLDL- and total triglycerides-related PC score (p = 0.01). The inverse association of percent
calories from fat and three polyunsaturated fatty acid intakes – total polyunsaturated fatty acid
intake, omega-6 fatty acid intake, and linoleic acid intake (LA, 18:2) – with VLDL- and
triglycerides-related PC score were significantly modified by BCAA related metabolomic PC
score. For example, by stratifying our participants by the median of the BCAA-related PC score,
one standard deviation (8.3g) increase in linoleic acid intake was associated with 0.25 unit
decrease in mean VLDL- and total triglycerides-related PC score among participants with low
BCAA-related metabolomic PC score (p = 0.60). On the other hand, among participants with high
BCAA-related metabolomic PC score, the association estimate was 1.65 unit decrease in mean
VLDL- and triglycerides-related PC score by one standard deviation increase in linoleic acid
intake (p = 0.01).
Conclusions: L-Glutamic acid & L-Glutamate, L-Alanine, aromatic amino acid (L-Phenylalanine
and L-Tyrosine), BCAA-related metabolites (L-Leucine/L-Isoleucine, C3 and C5 acylcarnitines)
were associated with alterations of lipids metabolism in young adults. Additionally, the estimated
inverse effects of linoleic acid intake and percent calories from fat on serum VLDL and total
triglycerides levels were greater for people who had high concentrations of BCAA related
metabolites, than that for people who had low concentrations of BCAA related metabolites.

1
1 Introduction
The crude prevalence of obesity (people with BMI greater than or equal to 25 kg/m2) was
39.8% in adults and 18.5% in youth during 2015-2016 (Hales et al., 2017). According to a report
from CDC (The National Diabetes Statistics Report, 2020), 34.2 million people in the United
States (10.5% of the nation’s population) have diabetes and 90%-95% of those have type 2
diabetes (T2D). The estimated number of individuals aged 18 years or older with prediabetes is
88 million (27.0% of the nation’s population). Several studies suggested that genetics, unhealthy
diet, obesity, and inadequate physical activity were associated with insulin resistance and
impaired glucose tolerance (prediabetes) (Weiss et al., 2008; Hannon et al., 2005). However, the
biological mechanism of metabolic dysfunction in young adults is still not clear. Cardio-
metabolic diseases have a common characteristic where the diseases might have developed for
a long time before a firm diagnosis can be made by abnormal clinical measures. Exploring the
association between metabolites and metabolic diseases in young adults might provide
meaningful understanding or new insight into the mechanism of these diseases. Such research
might also provide fundamental theoretical support for disease prevention strategies and risk
prediction.
The development of cardio-metabolic disease could be driven by altering numerous
metabolic pathways that involve many small molecules. The ability to identify and measure
small molecule chemicals, known as endogenous metabolites, from biological samples makes
metabolomics a powerful approach to reveal the underlying mechanism of cardio-metabolic
diseases. In metabolomics, mass spectrometry has been widely used with the development of
high-resolution and high-throughput analytical platforms. Both untargeted metabolomics and
targeted metabolomics could be detected by mass spectrometers with high sensitivity (Newgard,

2
2017). Untargeted metabolomics is used to discover all measurable chemical features in a sample
instead of detecting the quantity of known metabolic compounds. Therefore, untargeted
metabolomics could be a useful technique to explore novel metabolic pathway rather than
obtaining the exact concentration of chemicals. By contrast, targeted metabolomics is mainly
applied to identifying and quantitatively analyzing metabolites with known chemical identities,
which could provide significant information on the direction of regulation of metabolic pathways
(Roberts et al., 2012). Also, targeted metabolites profiling reflects the phenotype and biomarkers
that may be useful in disease prediction and progression assessment (Newgard, 2017).
Taking advantage of targeted metabolomics, several studies have confirmed that the risk of
type 2 diabetes is significantly associated with serum metabolites, including sugar metabolites,
amino acids, and choline-containing phospholipids (Floegel et al., 2013; Newgard et al., 2009;
Mai et al., 2013). Furthermore, serum medium-chain acylcarnitines, particularly
hexadenylcarnitine and octadenylcarnitine, have been found to be higher in gestational diabetes
mellitus (GDM) females and post-GDM females who have developed new-onset T2D than that
in non-GDM pregnant controls (Batchuluun et al.,2018). However, only a few studies have been
conducted to investigate the effects of metabolites on altering cardiometabolic traits including
adiposity, glucose and lipid metabolism, as well as inflammation in young adults (Chen et al.,
2019; Butte et al., 2015). Therefore, this study mainly focused on investigating the association
of metabolites with three major classes of metabolism traits – adiposity measures, glucose
metabolism traits, and lipids concentration in young adults. Additionally, dietary intake, as an
exogenous source of metabolites, is associated with obesity and insulin resistance. The
interaction between dietary intake and metabolites in the association of key metabolic pathways

3
and cardio-metabolic traits were also analyzed to provide a broad scope of possible mechanisms
of obesity and insulin resistance in young adults.

4
2 Materials and methods
2.1 Study recruitment and study design
As described in the publications of Chen et al. (2019) and Kim et al. (2019), a total of 107
participants were selected from the Meta-AIR study, which was part of a larger Southern
California Children’s Health cohort study (CHS) (McConnell et al., 2015). In the CHS, during
the last follow-up in 2011-2012, there were 1,154 participants with higher risk of metabolic
disorder (BMI greater than or equal to 85th percentile of the CDC 2000 age- and sex-specific
BMI growth chart (Kuczmarski et al., 2002)). Probability weighted sampling method was used
to sample 200 participants with lower and higher air pollution exposure within each CHS
community (Chen et al., 2019). 175 participants (87.5% of the 200 targeted participants) were
enrolled in the Meta-AIR study. Among all Meta-AIR participants, there were 126 participants
(72.0% of the 175 participants) who had complete oral glucose tolerance test (OGTT) data.
Among the 126 participants, 4 young adults were excluded due to incomplete diet data, and 15
participants were excluded due to incomplete data of adiposity measures or glucose metabolism
measures (excluding OGTT). Ultimately, the remaining 107 subjects (84.9% of the 126
participants) were included in this project and were free from type 1 diabetes, type 2 diabetes.
Also, going into the study, the 107 subjects were not taking any medication that may affect body
composition, insulin action, or glucose metabolism.
Participants completed their study visits at the Diabetes and Obesity Research Institute
(DORI) or Clinical Trials Unit (CTU) at the University of Southern California in 2014-2018
through clinical measurements and several questionnaires. The questionnaire contents included
detailed sociodemographic characteristics, parental education, history of e-cigarette usage,
attendance of exercise classes, self-reported physical activity status and 24-hour dietary intake

5
recalls. The self-reported physical activity status was evaluated by asking whether participants
were taking exercise classes (yes or no) and through a physical activity scale (less active (0-40),
moderately active (50-60), more active (70-100)). The Nutrition Data System for Research
(NDSR) software (Version 2014) (NDSR, 2014) was used to calculate the 24-hour diet recalls,
based on Nutrition Coordinating Center (NCC) Food and Nutrient Database. The daily total
calorie intake, grams of macronutrients intake, dietary glycemic index and percent calories from
macronutrients were estimated. Macronutrients intake included total carbohydrate intake, total
fat intake, total protein intake, total cholesterol intake, total saturated fatty acids intake, total
monounsaturated fatty acids intake, total polyunsaturated fatty acids intake, fructose intake,
galactose intake, glucose intake, lactose intake, maltose intake, sucrose intake, total sugars
intake, added sugars intake by available carbohydrate, added sugars intake by total sugars, total
fiber intake, soluble fiber intake, insoluble fiber intake, omega-3 fatty acids intake, PUFA 20:4
arachidonic acid intake, PUFA 18:2 linoleic acid intake, and Omega-6 fatty acids intake. These
five variables were calculated as percent calories from macronutrients, including percent calories
from fat, carbohydratea, proteins, total sugars, and added sugars. Dietary glycemic index
contained glycemic index glucose reference, glycemic index bread reference, glycemic load
glucose reference, and glycemic load bread reference. Omega-6 fatty acids intake was calculated
as the sum of arachidonic acid intake and linoleic acid intake.
Informed consent and assents were gained from all participants. The Institutional Review
Board at the University of Southern California approved this research.

6
2.2 Adiposity and cardiometabolic outcomes
2.2.1 Adiposity outcomes
Height (in nearest centimeters) and weight (in nearest 0.1kg without shoes) were measured
by trained technicians. Body mass index (BMI) was calculated by weight/height
2
(kg/m
2
). Body
fat percent was detected by dual-energy X-ray absorptiometry scan. Subcutaneous abdominal
adipose tissue (SAAT), visceral adipose tissue (VAT), and hepatic fata fraction (HFF) were
measured by abdominal scan with 3T magnetic resonance imaging (MRI).
2.2.2 Glucose metabolism
After fasting for at least 10 hours, fasting glucose and fasting insulin samples were collected
from all participants. A dose of 1.75g anhydrous glucose per kilogram of body weight (max dose
of 75g) was given to each participant to process the oral glucose tolerance test. Afterwards, blood
samples were taken at 30- and 120-min for glucose and insulin testing. Glucose area under the
curve (Glucose AUC) and insulin area under the curve (insulin AUC) were generated by all OGTT
time-points using the trapezoidal method. Homeostatic model assessment for insulin resistance
(HOMA-IR), serving as an insulin resistance index, was given by the formula (1) (Matthews,
1985).
HOMA-IR = fasting glucose(mg/dL) *fasting insulin(µU/mL) /405 (1)
Homeostatic model assessment for β-cell function (HOMA-β) was calculated by using the
formula (2) (Wallace et al., 2004).
HOMA-β = (20*fasting insulin (µU/mL)) /(fasting glucose(mg/dL)*18.02 - 3.5) (2)
The Matsuda index, which estimates an approximation of insulin sensitivity, was generated by
using the formula (3).
Matsuda Index =
1000
√𝑓𝑎𝑠𝑡𝑖𝑛𝑔 𝑔𝑙𝑢𝑐𝑜𝑠𝑒 (𝑚𝑔 /𝑑𝐿 )∗𝑓𝑎𝑠𝑡𝑖𝑛𝑔 𝑖𝑛𝑠𝑢𝑙𝑖𝑛 ((µU/mL)∗a∗b
,

7
where a =
fasting glucose ∗ 15 + 30−min glucose ∗ 30 + 60−min glucose ∗ 30 + 90−min glucose ∗ 30 + 120−min glucose ∗ 15
120

and b =
fasting insulin ∗ 15 + 30−min insulin ∗ 30 + 60−min insulin ∗ 30 + 90−min insulin ∗ 30 + 120−min insulin ∗ 15
120

(3)
Changed insulin level was defined by 30-minutes insulin after OGTT(µU/mL) - fasting insulin
(µU/mL). In summary, glucose metabolism traits include fasting glucose, fasting insulin, 30- and
120- minutes glucose after OGTT, 30- and 120- minutes insulin after OGTT, glucose AUC,
insulin AUC, HOMA- β, HOMA-IR, Matsuda index, HbA1c (mmol/mL), amylin (pg/mL), C-
peptide (µg/mL), GLP-1 (µg/mL), glucagon (pg/mL), and HMW adiponectin (ng/mL). All
insulin-related outcomes were log-transformed. Their means were calculated by 𝑒 ∑ 𝑙𝑛 𝑥 𝑖 𝑛 𝑖 =1
𝑛 , and
the standard deviations were computed by the delta method with the formula: 𝑒 ∑ 𝑙𝑛 𝑥 𝑖 𝑛 𝑖 =1
𝑛 *SD,
where SD =
√
∑ (𝑙𝑛 𝑥 𝑖 −
1
𝑛 ∗∑ 𝑙𝑛 𝑥 𝑖 )
𝑛 𝑖 =1
𝑛 𝑖 =1
2
𝑛 − 1
.
2.2.3 Lipid metabolism
Fasting samples were used to analyze lipid profiles, consisting of HDL-cholesterol (mg/dL),
LDL-cholesterol (mg/dL), VLDL-cholesterol (mg/dL), total cholesterol(mg/dL), triglycerides
(mg/dL), and LBP (ng/dL).
2.2.4 Inflammatory cytokines, chemokines and cardiovascular markers
Concentrations of interleukin 10 (IL-10) (pg/mL), interleukin 13 (IL-13) (pg/mL),
interleukin 17a (IL-17a) (pg/mL), interleukin 1b (IL-1b) (pg/mL), interleukin 4 (IL-4) (pg/mL),
interleukin 5 (IL-5) (pg/mL), interleukin 6 (IL-6) (pg/mL), sTNFRII (ng/mL), sCD14 (ng/mL),
and leptin (µg/mL) were also analyzed from fasting serum samples. Among them, leptin was
analyzed by human leptin ELISA kit, and the remaining were analyzed by Millipore Multiplex

8
assay kits. Three anthropometric measures were collected as cardiovascular markers: (1)
diastolic blood pressure(mmHg), (2) systolic blood pressure (mmHg), (3) pulse (bpm).
2.2.5 Laboratory assays
Fasting samples were collected in serum separator tube and centrifuged at 2000 RPM for 10
minutes after clot retracting completed. Serum samples were stored at -80°C for measurements
at a later time. Plasma was used to measure post OGTT glucose and insulin concentration. Blood
samples were collected in potassium oxalate, sodium fluoride tubes and centrifuged at 1500 RCF
for 15 minutes for glucose assays. After plasma sample were obtained, glucose concentrations
were measured by Roche Cobas C501 with hexokinase mediated reaction. For the detection of
insulin, blood samples were collected in sodium heparin tubes and then centrifuged at 2500 RPM
for 10 minutes. Plasma sample were frozen at -80°C and measured by Human Insulin ELISA
Kit.
2.3 Metabolomic signatures
Fasting serum samples were used to measure concentrations of 64 targeted metabolites. As
introduced in the publication of Chen et al. (2019), all serum samples were kept in -80°C and
transferred to laboratory on dry ice to secure the stability of serum metabolites (Breier et al.,
2014). 63 targeted metabolites were separated in to three groups: (1) 15 amino acids, (2) 45
acylcarnities, (3) non-esterified free fatty acids (NEFA), lactate, and beta-hydroxybutyrate.
Samples were block randomized by gender in analysis plate and measured by three batches. The
leftover samples were applied to detect the concentration of glycerol with a different batch.
Therefore, our research included 64 targeted metabolites for total. Acylcarnitines and amino
acids in the serum were detected by flow injection-tandem mass spectrometry (MS/MS) (Chen
et al., 2019; Newgard et al., 2009; Ferrara et al., 2008). Methanol was used to remove proteins

9
in the preprocess step. Hot acidic methanol and n-butanol were adopted to esterify dried
supernatants for acylcarnitines and amino acids, respectively. Inclusion of table isotope-labeled
internal standards were applied to obtain quantitative and reproducible measurements for
acylcarnitines and amino acids.
Non-esterified free fatty acids (NEFA), lactate, and beta-hydroxybutyrate were analyzed by
Beckman Unicel DxC600 autoanalyzer. Non-esterified free fatty acids (NEFA) and beta-
hydroxybutyrate were measured with kits from Roche Diagnostics (Indianapolis, IN) and lactate
with kits from Wako (Richmond, VA).
C7-DC had > 20% zero values with concentrations below the quantification limits. For data
processing, we used the raw metabolites data. To avoid the occurrence of infinite negative value
after log transformation and the violation of the normality assumption in later analysis, we added
1 to the concentration of all metabolites and made log transformation before analysis.
2.4 Statistical methods
To deal with multivariate data having high degrees of collinearity, we adopted partial least
squares (PLS) regression to explore the associations between 64 targeted metabolites and 42
outcomes. PLS aims to look for the largest covariance between the linear combinations of the
independent variable set and the dependent variable set, as well as generate mutually
uncorrelated components sequentially (Abdi, et al. 2010). To control potential confounders for
both metabolites and outcomes data in the PLS regression, we built multiple linear regression
models to obtain the residuals for each metabolite treating each metabolite as the dependent
variable and using the following covariates as independent variables: sex, age, race/ethnicity,
parental education, seasons of the research visit, self-reported physical activity scale, whether
participants smoked during the last seven days, whether participants ever used e-cigarettes, and

10
whether participants took exercise classes in the last year. Similarly, we also built multiple linear
regression models to obtain the residuals for each cardiometabolic trait treating each
cardiometabolic trait as the dependent variable and using the same covariates mentioned before
as independent variables. The residuals represent metabolites and cardiometabolic traits adjusted
for covariates. Q
2
total was calculated as a result of leave-one-out cross validation to assess the
model consistency and to identify the number of components for PLS regression. After building
a PLS model with two components, we then generated the heatmap of similarity scores which
stand for the approximations of correlation between metabolites and cardiometabolic traits
(Eisen et al., 1998; Weinstein et al., 1997; Gonzalez et al., 2012). Multiple metabolic pathways
were suggested by each metabolite-outcome cluster. Based on metabolite-outcome correlations
in each PLS components, we further performed principal components analysis (PCA) on
metabolites and cardiometabolic outcomes that were clustered with high correlations (similarity
score > 1.5). The PCA helped to dissect metabolic pathways from the big metabolites-outcomes
cluster. Also, the PCA created varimax orthogonal rotations for metabolites and outcomes with
reduced dimension and reduced collinearity. Scree plots were used to select the number of PCs
for metabolites and cardiometabolic traits. Metabolites and outcomes with an absolute value of
factor loading ≥ 0.4 were treated as the representative components of a given PC. Linear
regression was conducted to analyze the associations between metabolite PC scores and outcome
PC scores. To analyze whether diets interact with specific metabolic pathway represented by
various metabolite PC scores, we obtained the residuals of 34 diet variables from linear
regression, adjusting for covariates: sex, age, race, parental education, the season of the research
visit, the self-reported physical activity scale, whether the participants smoked during the last
seven days, whether the participants ever used e-cigarettes, and whether the participants took

11
exercise classes. Then, we used these residuals to estimate main effects for diets and metabolite
PCs on cardiometabolic traits, and interactions between diets and metabolite PCs in their
associations with outcome PCs by linear regression. In order to interpret the interaction
relationship across different metabolites levels, we dichotomized the PC scores of metabolites if
the interaction relationship tests were significant. Furthermore, we also analyzed the association
between dietary intake and outcome PC score across the stratification of dichotomous metabolite
PC score by multiple linear regression.
A two-sided p-value < 0.5 was treated as statistically significant for all independent tests.
SAS version 9.4 (SAS Institute, Cary, NC) and R version 3.6.1 were used to conduct data
analysis.

12
3 Results
This research included 56 male and 51 female young adults (107 in total) who had complete
metabolomic signature, cardiometabolic trait, and diet data. The range of the 107participants was
from 17.7 to 22.3 years old with mean of 19.6 and SD of 1.1. Among all participants, 58.8% of
participants were Hispanic White, 20.6% were non-Hispanic White, and the remaining 20.6%
were African American, Asian / Pacific Islander, other Non-White race, mixed races, and
unknown race. About half of participants’ parents completed high school, college or college
beyond education. 71.0% of participants did not have any e-cigarette smoking history and 94.4%
of them did not smoke in the past 7 days prior to the research. Approximately 31% of participants
had physical exercise classes during last 12 months and 75% of them had a moderate physical
activity or above prior to the research (Table 1).
3.1 Association between metabolites and cardiometabolic traits
Means and SDs of blood concentrations of all metabolites and cardiometabolic traits are
presented in Table 2 and Table 3, respectively. Among the 107 participants, 38 (35.5%) of them
were obese with BMI ≥ 30kg/m
2
, and 69 (64.5%) of them were normal weight or overweight
with BMI ≤ 30kg/m
2
.The average concentrations of all adiposity-related outcomes (BMI, body
fat percent, SAAT, VAT, and HFF) were significantly higher for obese people than for those
non-obese (all p’s < 0.001). Among glucose metabolism traits, an elevated mean concentration
of fasting insulin, 30-min insulin after OGTT, 2-hour insulin after OGTT, amylin, and C-peptide
were found to be higher in obese subjects compared with non-obese subjects (all p’s < 0.05); the
mean OGTT insulin area under the curve in obese people was significantly higher than that in
non-obese people (p < 0.001); the differences in mean HOMA-β, HOMA-IR, and Matsuda index
were statistically significant between non-obese people and obese people (all p’s <0.001). For

13
lipids measures, the mean concentrations of VLDL, triglycerides and LBP were higher in obese
subjects than in non-obese subjects (p = 0.004, p = 0.004, and p < 0.001); HDL in non-obese
subjects were significantly higher than in obese subjects (p = 0.01). According to our tests, the
mean concentrations of interleukin markers (Interleukin 10, Interleukin 13, Interleukin 17a,
Interleukin 1b, Interleukin 4, Interleukin 5, and Interleukin 6) were similar in obese subjects and
non-obese subjects (p > 0.05). However, higher levels of sTNFRII, sCD14 and leptin were seen
in obese subjects than in non-obese subjects (p = 0.02, p = 0.01, and p < 0.001). Finally, for
cardiovascular markers, only diastolic blood pressure showed difference in mean across the two
groups (p = 0.01). Details of the measures are presented in Table 3.

14

Table 1. Sociodemographic characteristics of 107 Children's Health Study (CHS) adolescents
and young adults.

Entire sample (N=107)
N
%
Sex

Male
56
52.3
Female
51
47.7
Race/Ethnicity

Non-Hispanic White
22
20.6
Hispanic White
63
58.8
Other *
22
20.6
Parental Education

Less than high school
36
33.6
Completed high school
31
29.0
Some college or beyond
26
24.3
Unknown
14
13.1
Ever used e-cigarette

Yes
31
29.0
No
76
71.0
Cigarette smoke in the last week

Yes
6
5.6
No
101
94.4
Exercise Class †

Yes
33
30.8
No
74
69.2
Physical activity

Less active
27
25.2
Moderately active
40
37.4
More active
40
37.4
Season

Spring
24
22.4
Summer
25
23.4
Fall
31
29.0
Winter
27
25.2
* Other race/ethnicity includes African American, Asian/Pacific Islander, mixed three or more races, other Non-White
race/ethnicity and unknown race/ethnicity.
†

Exercise Class refers to taking any exercise classes, lessons, or special programs during the past 12 months (outside of
school only).

15
Table 2. Characteristics table for metabolites variables of 107 Children's Health Study (CHS)
adolescents and young adults.
Metabolites N (%) Above Detection
Limits
Mean Standard Deviation
Amino acids (µmol/L)

L-Glycine 107 (100) 265.3 47.8
L-Alanine 107 (100) 366.5 69.0
L-Serine 107 (100) 120.5 18.1
L-Proline 107 (100) 176.3 44.8
L-Valine 107 (100) 235.5 40.5
L-Leucine or L-Isoleucine 107 (100) 162.9 26.8
L-Methionine 107 (100) 26.9 4.1
L-Histidine 107 (100) 80.1 8.7
L-Phenylalanine 107 (100) 63.6 9.7
L-Tyrosine 107 (100) 62.0 12.5
L-Aspartic acid and L-Asparagine 107 (100) 25.2 6.8
L-Glutamic acid and L-Glutamate 107 (100) 106.1 26.2
L-Ornithine 107 (100) 76.5 17.9
L-Citrulline 107 (100) 24.7 4.6
L-Arginine 107 (100) 85.5 14.9
Acylcarnitines (µmol/L)

C2 107 (100) 6.45 1.98
C3 107 (100) 0.28 0.09
C4/Ci4 107 (100) 0.15 0.06
C5:1 107 (100) 0.19 0.04
C5's 107 (100) 0.13 0.05
C4-OH 107 (100) 0.035 0.018
C6 86 (80.4) 0.058 0.015
C5-OH/C3-DC 106 (99.1) 0.033 0.012
C4-DC/Ci4-DC 107 (100) 0.035 0.056
C8:1 107 (100) 0.226 0.085
C8 107 (100) 0.105 0.055
C5-DC 107 (100) 0.043 0.012
C8:1-OH/C6:1-DC 107 (100) 0.02 0.008
C6-DC/C8-OH 107 (100) 0.06 0.027
C10:3 107 (100) 0.07 0.025
C10:2 107 (100) 0.022 0.008
C10:1 107 (100) 0.129 0.054
C10 107 (100) 0.182 0.098
C7-DC 24 (22.4) 0.021 0.013
C8:1-DC 107 (100) 0.019 0.007
C10-OH/C8-DC 107 (100) 0.033 0.018
C12:1 107 (100) 0.077 0.033
C12 107 (100) 0.063 0.030
C12-OH/C10-DC 107 (100) 0.007 0.003
C14:2 107 (100) 0.042 0.024
C14:1 107 (100) 0.062 0.031
C14 107 (100) 0.022 0.007
C14:1-OH 107 (100) 0.014 0.005
C14-OH/C12-DC 106 (99.1) 0.008 0.003
C16:2 107 (100) 0.008 0.004
C16:1 107 (100) 0.020 0.008
C16 107 (100) 0.080 0.019
C16:1-OH/C14:1-DC 107 (100) 0.007 0.002
C16-OH/C14-DC 107 (100) 0.004 0.002
C18:2 107 (100) 0.066 0.020
C18:1 107 (100) 0.104 0.030
C18 107 (100) 0.032 0.009
C18:2-OH 99 (92.5) 0.006 0.002
C18:1-OH/C16:1-DC 106 (99.1) 0.005 0.002
C18-OH/C16-DC 107 (100) 0.006 0.002
C20:4 106 (99.1) 0.007 0.002
C20 107 (100) 0.004 0.001
C18:1-DC 107 (100) 0.006 0.003
C20-OH/C18-DC 107 (100) 0.007 0.003
C22 107 (100) 0.003 0.003
Ketones and free fatty acids

3-Hydroxybutyrate (µmol/L) 107 (100) 103.2 100.4
Lactate (mmol/L) 107 (100) 2.07 0.58
Non-esterified fatty acids (mmol/L) 107 (100) 0.61 0.22
Glycerol (mg/dL) 107 (100) 1.90 1.07

16
Table 3. Characteristics table for outcome variables of 107 Children's Health Study (CHS)
adolescents and young adults.
Outcomes
Total (N=107) Non-obese (N=69) Obese (N=38)
p-value* Mean (SD) Mean (SD) Mean (SD)
Adiposity

Body mass index, BMI (kg/m2) 29.8 (5.0) 26.8 (2.0) 35.3 (4.1) <0.001
Body fat percent (%) 31.8 (9.0) 28.5 (8.4) 37.8 (6.7) <0.001
Subcutaneous adipose tissue, SAAT (L) 5.1 (2.6) 3.9 (1.6) 7.4 (2.5) <0.001
Visceral adipose tissue, VAT (L) † 1 (0.7) 0.8 (0.5) 1.8 (0.9) <0.001
Hepatic fat fraction, HFF (%) † 3.7 (2.1) 3 (1.3) 5.6 (2.8) <0.001
Glucose Metabolism Traits

Fasting glucose (mg/dL) 90 (7.0) 89.1 (6.3) 91.6 (7.8) 0.08
30-min glucose after OGTT (mg/dl) 147.4 (22.5) 146.1 (21.3) 149.8 (24.7) 0.42
2-hour glucose after OGTT (mg/dL) 121.5 (25.7) 118.6 (27.6) 126.8 (21.3) 0.12
OGTT glucose area under the curve 15971.6 (2546.0) 15631.5 (2551.4) 16589.2 (2448.0) 0.06
Fasting insulin (µU/mL) † 6.1 (6.6) 4.4 (4.5) 11 (10.1) <0.001
30-min insulin after OGTT (µU/mL) † 99.2 (77.3) 88.6 (70.6) 121.6 (87) 0.04
2-hour insulin after OGTT (µU/mL) † 71.6 (72.5) 57.9 (60.5) 105.2 (87.8) 0.003
OGTT insulin area under the curve † 10726.4 (6963.4) 9176.5 (5899.6) 14239.9 (8052.6) <0.001
Insulin change (µU/mL) †‡ 89.3 (73.8) 81.2 (69.2) 106 (80.7) 0.13
HOMA-β †§ 83.7 (88.8) 62.1 (65.8) 143.9 (119.8) <0.001
HOMA-IR †§ 1.3 (1.5) 1 (1) 2.5 (2.4) <0.001
Matsuda index † 4.0 (3.3) 5.1 (4) 2.5 (1.8) <0.001
Hemoglobin A1c, HbA1c (mmol/mol) 5.2 (0.3) 5.2 (0.3) 5.3 (0.3) 0.10
Amylin (pg/ml) † 31.6 (14.1) 27.6 (11.3) 40.3 (16.6) <0.001
C-peptide (μg/mL) † 1149.7 (473.0) 1006.7 (389.2) 1463.4 (501.6) <0.001
GLP-1 (μg/mL) †§ 166.1 (35.6) 161.5 (34.8) 174.8 (35.9) 0.07
Glucagon (pg/ml) † 65.9 (30.7) 63.2 (30.0) 70.9 (31.7) 0.23
HMW adiponectin (ng/mL) † 3032.2 (2139.2) 3319.3 (2111.4) 2572.9 (2057.1) 0.07
Lipids

HDL (mg/dL) § 40.2 (10.3) 42.0 (10.4) 36.9 (9.4) 0.01
LDL (mg/dL) § 103.5 (32.7) 103.4 (33.6) 103.5 (31.5) 0.99
VLDL (mg/dL) †§ 15.5 (8.2) 13.9 (7.3) 18.9 (9.1) 0.004
Triglycerides (mg/dL) † 77.6 (41.1) 69.6 (36.5) 94.6 (45.7) 0.004
Total Cholesterol (mg/dL) 161.5 (38.5) 161.4 (39.9) 161.5 (36.5) 0.99
LBP (ng/mL) †§ 21466.3 (6686.3) 19872.7 (5671.7) 24693.6 (7692.1) <0.001
Inflammatory Markers

Interleukin 10 (pg/ml) † 19.2 (11.7) 19.5 (11.4) 18.7 (12.3) 0.73
Interleukin 13 (pg/ml) † 14.9 (12.6) 15.4 (13.1) 14.0 (12) 0.59
Interleukin 17a (pg/ml) † 15.8 (7.4) 16.5 (7.7) 14.6 (6.7) 0.18
Interleukin 1b (pg/ml) † 3.5 (1.1) 3.6 (1.1) 3.3 (1.1) 0.29
Interleukin 4 (pg/ml) † 87.4 (42.0) 88.3 (40.9) 85.8 (44.3) 0.77
Interleukin 5 (pg/ml) † 5.1 (2.4) 5.2 (2.5) 5 (2.3) 0.58
Interleukin 6 (pg/ml) † 7.6 (5.9) 7.7 (6.0) 7.3 (5.8) 0.72
sTNFRII (ng/mL) †§ 4963.8 (1198.2) 4772.4 (1204.1) 5331 (1090.2) 0.02
sCD14 (ng/mL) †§ 1629.4 (339.2) 1566.5 (321.8) 1750.4 (343.1) 0.01
Leptin (μg/mL) † 6707 (7185.6) 4413.4 (4470.8) 14341.3 (9961.5) <0.001
Cardiovascular markers

Diastolic blood pressure (mmHg) 69.9 (6.7) 68.8 (7.1) 72.1 (5.3) 0.01
Systolic blood pressure (mmHg) 114.8 (10.4) 113.5 (10.4) 117.3 (10.1) 0.07
Pulse (bpm) 64.6 (9.6) 63.4 (9.4) 66.6 (9.6) 0.10

17
* Two sample t-test was used to investigate the mean value of each variable across non-obese and obese subjects.
† Log transformation was applied to specific outcome variable to approximate normal distribution.
‡ Insulin change = 30-min insulin after OGTT- Fasting insulin.
§ HOMA-β = homeostatic model assessment-β-cell function;
HOMA-IR = homeostatic model assessment-insulin resistance;
MCP-1 = monocyte chemoattractant protein-1;
HDL = high-density lipoprotein;
LDL = low-density lipoprotein;
VLDL = very low-density lipoprotein;
LBP = lipopolysaccharide-binding protein;
sTNFRII = soluble tumor necrosis factor receptor type II;
sCD14 = soluble cluster of differentiation 14.

18
Partial least square (PLS) regression identified 2 components with Q
2
total > 0 (the Q
2
total
of the 1
st
component is 0.05 and the Q
2
total of the 2
nd
component is 0.04). Afterwards, our PLS
regression model was refitted with 2 components. The percentage of explained variance by
component 1 was 23.4% for metabolites and 26.0% for cardiometabolic traits. Additionally, the
percentage of explained variance by component 2 was 7.2% for metabolites and 19.0% for
cardiometabolic traits. The clustered correlation heatmaps of component 1 and component 2 are
presented in Fig 1 A) and B) respectively. The main cluster on component 1 showed inverse
correlations between medium chain acylcarnitines and body fat measures, which has been well-
studied previously (Chen et al., 2019). Therefore, we mainly focused on the cluster of
metabolomic signatures and cardiometabolic traits on component 2. Based on the hierarchical
clustering, it is obvious that several metabolites and outcomes were clustered within a big block
of strong positive correlations in component 2. This cluster almost consisted all of the main
phenotypes – adiposity measures, glucose metabolism traits, lipids levels, and inflammatory
markers, including BMI, body fat percent, VAT, SAAT, HFF, fasting insulin, insulin AUC, C-
peptide, HOMA-β, HOMA-IR, VLDL, triglycerides, sTNFRII, and leptin. The metabolites
clustered in this big block were C4-DC/Ci4-DC, C5's, C6, C8:1, C10:2, C10:3, C14, C14:1,
C14:2, C16, C16:1, C16:2, C18:2, C18:1, L-Tyrosine, L-Phenylalanine, L-Glutamic acid & L-
Glutamate, L-Leucine/L-Isoleucine, L-Alanine, Lactate, NEFA, and Glycerol. Though C3 was
not assigned to the cluster of branched chain-amino acids (BCAA) in the PLS component 2, it
was known to be a main downstream catabolism product of BCAA. Therefore, we included C3
with other BCAAs in the following PCA analysis. In addition, we did not find any obvious cluster
involved interleukin markers (Interleukin 10, Interleukin 13, Interleukin 17a, Interleukin 1b,

19
Interleukin 4, Interleukin 5, and Interleukin 6), sCD14, diastolic blood pressure, systolic blood
pressure, and pulse in the PLS component 1 and component2.

20

A) Component 1

B) Component 2

Figure 1. Heatmap of similarity scores for the 1
st
and 2
nd
component identified from PLS
regression. The color and the dendrogram illustrated the value of similarity score and the proximity between correlated
variables, respectively.

21
3.2 Classification of metabolite PCs and outcome PCs
The PCs were identified from 23 metabolites and 14 cardiometabolic traits extracted from
the PLS component 2 cluster as described before. There were 5 orthogonal factors with
eigenvalue > 1.0 for metabolites data and 3 orthogonal factors with eigenvalue > 1.0 for
outcomes data (Fig 2 and Fig 3, respectively). We focused on the top 4 PCs of metabolites and
top 3 PCs of the outcomes according to the scree plots. The loading heatmaps for each PC were
shown in Fig 4 for metabolite PCs and Fig 5 for cardiometabolic trait PCs. Table 4 and Table 5
were the detailed loadings for metabolites and cardiometabolic traits, respectively. Among 4 PCs
of metabolites data, PC1 primarily depicts the group of long-chain fatty acids and NEFA, PC2
represents the branched-chain amino acid (BCAA, L-Leucine/L-Isoleucine) and related
metabolic signature (C3 and C5 acylcarnitines), PC3 is the medium-chain acylcarnitines, and
PC4 is characterized by the L-glutamic acid and L-glutamate acid and L-Alanine, as well as
aromatic amino acids (L-Phenylalanine and L-Tyrosine) and lactate. As shown in the figure,
among the 3 PCs of cardiometabolic outcomes, PC1 stands for insulin resistance (HOMA-IR,
HOMA-β and fasting insulin), PC2 mainly represents adiposity measures, and PC3 stands for
specific lipids measures (i.e. VLDL and total triglycerides).

22

Figure 2. Scree and variance plots for principal components identified from 23 metabolites
among 107 CHS young adults.

Figure 3. Scree and variance plots for principal components identified from 14 outcomes
among 107 CHS young adults.

23

Figure 4. Heatmap of loadings for the first four principal components identified from 23
targeted metabolites among 107 young adults.

Figure 5. Heatmap of loadings for the first three principal components identified from 14
cardiometabolic outcomes among 107 young adults.

24
Table 4. Loadings for the first four principal components* identified from 23 targeted
metabolites concentrations among 107 Children's Health Study (CHS) adolescents and young
adults.
Metabolomic profiling PC1 ‡ PC2 ‡ PC3 ‡ PC4 ‡
Amino acids

L-Alanine
-0.18 0.32 0.03 0.57
L-Leucine or L-Isoleucine
-0.01 0.71 -0.21 0.24
L-Phenylalanine
0.19 0.53 -0.07 0.62
L-Tyrosine
-0.01 0.50 -0.07 0.55
L-Glutamic acid and L-Glutamate
-0.17 0.00 0.03 0.77
Acylcarnitines

C3
-0.21 0.76 0.05 -0.05
C4/Ci4
0.01 0.57 0.21 0.04
C5's
0.06 0.71 -0.05 0.16
C6
0.47 0.18 0.24 -0.23
C8:1
0.18 -0.13 0.81 0.17
C10:3
0.21 -0.09 0.83 0.20
C10:2
0.28 0.30 0.73 -0.13
C14:2
0.68 0.03 0.52 -0.31
C14:1
0.76 -0.04 0.41 -0.36
C14
0.74 0.10 0.20 -0.30
C16:2
0.74 0.00 0.37 -0.23
C16:1
0.87 -0.01 0.24 -0.12
C16
0.75 0.07 -0.10 0.13
C18:2
0.72 -0.07 0.35 0.03
C18:1
0.82 -0.14 0.07 -0.02
Ketones and free fatty acids

Lactate
-0.13 0.04 0.20 0.58
Non-esterified fatty acids
0.74 -0.23 0.01 0.00
Glycerol
0.41 0.05 -0.08 0.32
Variance explained (%) † 7.2 3.5 1.9 1.6
* Loadings of each metabolite for specific principal component (PC) factor are presented in the table. Loadings with an absolute
value > 0.4 are bolded and metabolites with bolded loadings in each PC are representative for that PC factor.
† Percent of variance of entire metabolomic profiling explained by each PC factor is presented.
‡ PC1 primarily depicts the group of long-chain fatty acids and NEFA;
PC2 represents the branched-chain amino acid (BCAA, L-Leucine/L-Isoleucine) and related metabolic signature (C3 and C5
acylcarnitines);
PC3 represents the medium chain acylcarnitines;
PC4 is characterized by the L-glutamic acid and L-glutamate acid and L-Alanine, as well as aromatic amino acids (L-
Phenylalanine and L-Tyrosine) and lactate.

25
Table 5. Loadings for the first three principal components* identified from 14 cardiometabolic
traits of 107 Children's Health Study (CHS) adolescents and young adults.
Outcomes PC1 † PC2 † PC3 †
Adiposity

Body mass index, BMI
0.30 0.82 0.06
Body fat percent
0.08 0.90 0.00
Subcutaneous adipose tissue, SAAT
0.32 0.88 0.09
Visceral adipose tissue, VAT (L)
0.31 0.72 0.24
Hepatic fat fraction, HFF
0.32 0.54 0.48
Glucose Metabolism Traits

Fasting insulin
0.96 0.18 0.10
OGTT insulin area under the curve
0.66 0.37 0.14
HOMA-β‡
0.96 0.11 0.10
HOMA-IR‡
0.96 0.20 0.10
C-peptide
0.62 0.52 0.32
Lipids

Very low-density lipoprotein, VLDL (mg/dL)
0.12 0.09 0.98
Triglycerides (mg/dL)
0.12 0.09 0.98
Inflammatory Markers

sTNFRII (ng/mL) ‡
0.00 0.43 0.05
Leptin (μg/mL)
0.25 0.78 0.10
Variance explained (%)§ 7.0 2.0 1.8
* Loadings of each outcomes for specific principal component (PC) factor are presented in the table. Loadings with an absolute
value > 0.4 are bolded and outcomes with bolded loadings in each PC are representative for that PC factor.
† PC1 represents insulin resistance (HOMA-IR, HOMA-β and fasting insulin);
PC2 represents adiposity measures;
PC3 stands for lipids measures.
‡ HOMA-β = homeostatic model assessment-β-cell function;
HOMA-IR = homeostatic model assessment-insulin resistance;
sTNFRII = soluble tumor necrosis factor receptor type II.
§ Percent of variance of entire cardiometabolic traits explained by each PC factor is presented.

26
3.3 The association of metabolite PCs and outcome PCs
We found that metabolite PC4, which represents the concentration of aromatic amino acid
(L-Phenylalanine and L-Tyrosine), L-Glutamic acid & L-Glutamate, L-Alanine, and lactate, was
significantly associated with all outcome PCs (Fig 6, p = 0.048 for insulin resistance related PC
score, p < 0.001 for adiposity-related PC score, and p = 0.01 for lipids-related PC score). One unit
increase in the metabolite PC4 score was associated with 0.19, 0.54 and 0.24 unit increase in
mean insulin-related outcome PC score, adiposity-related PC score, and lipids-related PC score,
respectively. Meanwhile, metabolite PC2, which is characterized for BCAA-related metabolic
signature, was significantly associated with the lipids-related PC score (p = 0.01). One unit
increase in the BCAA-related metabolic signature PC scores was found to be associated with 0.25
unit increase in the average of lipids-related PC score.

27

Figure 6. Association† of metabolite PCs and metabolism trait PCs in 107 participants.
* Glx represents L-Glutamic acid & L-Glutamate acid;
Ala represents L-Alanine.
† Linear regression model was used to estimate the association of metabolite PCs and metabolism trait PCs. The association
estimated by β with its 95% confident interval.
‡ Insulin resistance = insulin resistance related outcome PC score;
Adiposity = adiposity related outcome PC score;
Lipids = Lipids related outcome PC score.
x

28
3.4 The association of dietary intake and outcome PC score, and effect modifications
by metabolite PC scores
The mean and SD of 24-hour recall dietary intake are shown in Table 6. Mean fructose and
glucose intakes were significantly higher for obese people than for non-obese people (both p’s
= 0.02). For other diets, there was no significant differences in mean intake between non-obese
people and obese people (all p’s > 0.05).
Percent calories from fat and three polyunsaturated fatty acids intakes – total
polyunsaturated fatty acids intake, omega-6 fatty acid intake, and linoleic acid intake (LA, 18:2)
– were significantly associated with the lipid-related PC score (p = 0.046, p = 0.04, p = 0.03, and
p = 0.03, respectively), and effect modification by BCAA-related metabolomic PC score was
also significant (all p’s = 0.02). There was no other significant effect medication found by
multiple linear regression model (p < 0.05).
Further, the BCAA-related metabolomic PC score were dichotomized by its median (median
= 0), and then we examined the associations between 4 dietary variables (percent calories from
fat, total polyunsaturated fatty acids intake, omega-6 fatty acid intake, and linoleic acid intake)
and lipids-related outcome PC score stratified by low vs. high BCAA-related metabolomic PC
score (Fig 7 and Table 7). The results for total polyunsaturated fatty acids intake, omega-6 fatty
acids intake, and linoleic acid intake (LA, 18:2) were found to be similar. Our results suggested
that one standard deviation (8.3g) increase in linoleic acid intake was associated with 0.25 unit
decrease in mean lipids-related PC score among participants with low BCAA-related
metabolomic PC score and 1.65 unit decrease for participants with high BCAA-related
metabolomic PC score (p = 0.60 and p = 0.01, respectively); one standard deviation (8.3g)
increase in omega-6 fatty acids intake was associated with 0.28 unit decrease in mean lipids-

29
related PC score among participants with low BCAA-related metabolomic PC score and 1.65
unit decrease for participants with high BCAA-related metabolomic PC score (p = 0.58 and p =
0.009, respectively); one standard deviation (9.2g) increase in total polyunsaturated fatty acids
intake was associated with 0.27 unit decrease in mean lipids-related PC score among participants
with low BCAA-related metabolomic PC score and 1.73 unit decrease for participants with high
BCAA-related metabolomic PC score (p = 0.61 and p = 0.011, respectively). Additionally, one
standard deviation (7.6%) increase in percent calories from fat was associated with 0.07 unit
decrease in the average of lipids-related PC score for participants with low BCAA-related
metabolomic PC score and 0.46 unit decrease for participants with high BCAA-related
metabolomic PC score (p = 0.59 for BCAA-related metabolomic PC score and p = 0.01 for high
BCAA-related metabolomic PC score). Similar estimated effects were identified using the
datasets further adjusted for body fat percent and processed by the analysis done before (Table
8).

30
Table 6. Characteristics table for 24 hours dietary intake of 107 Children's Health Study (CHS)
adolescents and young adults.
Total (N=107) Non-obese
(N=69)
Obese (N=38) p-value*
Mean (SD) Mean (SD) Mean (SD)
Total calorie intake (kcal) 2014.2 (597.7) 2011.8 (629.4) 2018.4 (543.5) 0.96
Macro-nutrients
Total carbohydrate intake (g) 245.5 (73.4) 239.9 (74.0) 255.8 (72.1) 0.28
Total fat intake (g) 81.2 (32.9) 83.4 (35.6) 77.3 (27.2) 0.37
Total protein intake (g) 81.8 (33.2) 82.1 (34.5) 81.4 (30.9) 0.92
Total cholesterol intake (mg) † 227.9 (160) 232.7 (156.6) 219.4 (166.7) 0.68
Total saturated fatty acids intake (g) 26.1 (11.5) 27.1 (12.5) 24.3 (9.3) 0.23
MUFA (g) ‡ 28.6 (11.8) 29.4 (12.5) 27.2 (10.4) 0.35
PUFA (g) †‡ 16.8 (9.2) 16.8 (9.1) 16.7 (9.5) 0.95
Fructose intake (g) 21.6 (15.4) 18.7 (13.2) 26.9 (17.7) 0.02
Galactose intake (g) † 0.1 (0.2) 0.1 (0.2) 0.1 (0.2) 0.90
Glucose intake (g) 22.1 (13.9) 19.7 (12.5) 26.3 (15.3) 0.02
Lactose intake (g) † 6.1 (8.8) 7.0 (9.4) 4.7 (7.5) 0.17
Maltose intake (g) † 2.2 (1.7) 2.3 (1.7) 2.1 (1.8) 0.65
Sucrose intake (g) 42.1 (24.3) 41.1 (23.4) 43.9 (26.2) 0.57
Total sugars intake (g) 99.5 (44.0) 94.8 (41.0) 108.2 (48.4) 0.13
ASUG (g) ‡ 69.2 (40.2) 64.4 (36.1) 77.9 (45.9) 0.10
ASUTS (g) †‡ 62.6 (36.7) 57.9 (32.8) 71.2 (42.1) 0.07
Total fiber intake (g) 17.9 (6.1) 17.9 (6.5) 18 (5.2) 0.97
Soluble fiber intake (g) 5.8 (2.2) 5.7 (2.3) 5.9 (1.9) 0.74
Insoluble fiber intake (g) 12.0 (4.6) 12 (4.8) 11.9 (4.2) 0.94
Omega-3 fatty acids intake (g) 1.9 (1.0) 1.9 (1.0) 2 (1.0) 0.62
AA (g) ‡ 0.2 (0.1) 0.1 (0.1) 0.2 (0.1) 0.59
LA (g) †‡ 14.6 (8.3) 14.7 (8.1) 14.4 (8.6) 0.86
Omega-6 fatty acids intake (g) † 14.8 (8.3) 14.9 (8.2) 14.6 (8.7) 0.86
Percent calories from specific
nutrients

Percent calories from fat 34.9 (7.6) 35.7 (7.7) 33.4 (7.4) 0.13
Percent calories from carbohydrate 48.7 (9) 47.8 (9.1) 50.3 (8.7) 0.18
Percent calories from protein† 15.7 (4.3) 15.7 (4.4) 15.6 (4.3) 0.90
Percent calories from total sugars 20.4 (9) 19.5 (7.9) 22.0 (10.5) 0.20
PASUG †‡ 13.9 (7.4) 13.2 (6.8) 15.2 (8.3) 0.17
PASUTS‡ 12.6 (6.9) 11.8 (6.3) 14.0 (7.8) 0.13
Glycemic index

Glycemic index glucose reference 138.2 (47.1) 134.2 (46.1) 145.5 (48.7) 0.24
Glycemic index bread reference 197.5 (67.3) 191.8 (65.9) 208.0 (69.6) 0.24
Glycemic load glucose reference 60.4 (4.9) 60.2 (4.9) 60.9 (5) 0.48
Glycemic load bread reference 86.4 (7) 86.1 (7) 87.1 (7.1) 0.48
* Two sample t-test was used to investigate the mean value of each variable across non-obese and obese subjects.
† Log transformation was applied to specific outcome variable to approximate normal distribution.
‡ MUFA = total monounsaturated fatty acids intake;
PUFA = total polyunsaturated fatty acids intake;
ASUG = Added sugars intake by available carbohydrate;
ASUTS = Added sugars intake by total sugars;
AA = PUFA 20:4 arachidonic acid intake;
LA = PUFA 18:2 linoleic acid intake;
PASUG = Percent calories from added sugars by available carbohydrate;
PASUTS = Percent calories from added sugars by total sugars.

31

Figure 7. Association† of dietary intake and lipids-related PC scores, and effect
modifications by BCAA-related metabolomic PC scores in 107 participants.
* PUFA = total polyunsaturated fatty acids intake.
† Linear regression was used to investigate the association between dietary intake and lipids-related metabolism trait PC
score stratified by dichotomous BCAA-related metabolomic PC score, after adjusting for gender, age, race, parental
education, the season of the research visit, the self-reported physical activity scale, whether the participants smoked during
the last seven days, whether the participants ever used e-cigarettes, and whether the participants took exercise classes.

32
Table 7. Association* between dietary intake and lipids-related outcome PC score stratified by
dichotomous BCAA-related metabolomic PC score.
Dietary intake
Low BCAA PC score High BCAA PC score
β p-value β p-value
Percent calories from fat

-0.01 0.59

-0.06 0.007
Total polyunsaturated fatty acids intake†

-0.12 0.61

-0.78 0.011
Linoleic acid intake (LA, 18:2) †

-0.12 0.60

-0.78 0.009
Omega-6 fatty acids intake† -0.13 0.58 -0.78 0.009
* Linear regression was used to investigate the association between dietary intake and lipids-related metabolism trait PC score
stratified by dichotomous BCAA-related metabolomic PC score, after adjusting for gender, age, race, parental education, the
season of the research visit, the self-reported physical activity scale, whether the participants smoked during the last seven days,
whether the participants ever used e-cigarettes, and whether the participants took exercise classes.
† Log transformation was applied to specific dietary intake variable to approximate normal distribution.

Table 8. Association* between dietary intake and lipids-related outcome PC score stratified by
dichotomous BCAA-related metabolomic PC score, additionally adjusting for body fat percent.
Dietary intake
Low BCAA PC score High BCAA PC score
β p-value β p-value
Percent calories from fat

-0.01 0.58

-0.06 0.007
Total polyunsaturated fatty acids intake†

-0.13 0.60

-0.80 0.011
Linoleic acid intake (LA, 18:2) †

-0.13 0.57

-0.79 0.009
Omega-6 fatty acids intake† -0.13 0.57 -0.79 0.009
* Linear regression was used to investigate the association between dietary intake and lipids-related metabolism trait PC score
stratified by dichotomous BCAA-related metabolomic PC score, after adjusting for gender, age, race, parental education, the
season of the study visit, the self-reported physical activity scale, whether the participants smoked during the last seven days,
whether the participants ever used e-cigarettes, whether the participants took exercise classes and body fat percent.
† Log transformation was applied to specific dietary intake variable to approximate normal distribution.

33
4 Discussion
In this research, we investigated associations between 64 metabolites and 42 adiposity- and
cardiometabolic-related outcomes. We found an amino acid metabolomic PC score (representing
L-Glutamic acid & L-Glutamate acid, L-Alanine, and aromatic amino acid) was significantly
associated with insulin-, adiposity-, and lipids-related outcome PC score. The same associations
have been shown previous publications (Walford et al., 2013; Stanley et al., 2009; Newgard et
al., 2009). Additionally, a cohort study showed higher serum medium-chain acylcarnitines
occured in GDM females and post-GDM females who developed new-onset T2D compared to
that in non-GDM pregnant controls (Batchuluun et al.,2018). However, our PLS regression results
suggested medium chain acylcarnitines were inversely associated with adiposity measures in
young adults, such as BMI, body fat percent, HFF, VAT, and SAAT.
Surprisingly, we did not find any statistically significant association between BCAA-related
metabolomic PC score and insulin-resistance related outcome PC score, though other evidence
suggests that HOMA-IR is positively associated with BCAA-related metabolic signatures in other
studies (Wallace et al., 2004; Newgard et al., 2009; Walford et al., 2013; Rousseau et al., 2019).
Nevertheless, we discovered significant associations between the BCAA-related metabolite PC
score and the lipid-related outcome PC score. Lee et al. (1997) found that the occurrence of β-cell
dysfunction was preceded by a six-fold increase in triglycerides and by high levels of lipids
intermediate products in an animal model of obesity and type 2 diabetes. Higher lipogenesis and
raised fatty acid concentrations could be risk factors for developing diabetes.
Our research also investigated the association of dietary intake with adiposity- and
cardiometabolic-related outcomes. We found that the association of total polyunsaturated fatty
acids, omega-6 fatty acids, and linoleic acid intake with lipids-related outcome PC score were

34
similar when we stratified by dichotomous BCAA-related metabolomic PC score. This might be
because the total polyunsaturated fatty acids variable was constituted by omega-6 fatty acids and
omega-3 fatty acids intake, while omega-6 fatty acids intake was constituted by linoleic acid
intake and arachidonic acid intake. There was no evidence showing omega-3 fatty acids intake or
arachidonic acid intake were significantly associated with lipids related outcome PC score.
Therefore, the estimated effects for total polyunsaturated fatty acids and omega-6 fatty acids were
similar to linoleic acid. Additionally, these effects might be brought by the linoleic acid intake
from diet. Our research indicated that intake of linoleic acid was inversely associated with lipids-
related PC score. Linoleic acid, the most common polyunsaturated fatty acid in lipid rich dietary
intake, has high oxidation rate and can reduce lipogenesis by limiting lipogenic enzymes activity
(Wang et al., 2004). Compared with oleic acid feeding animals, taking oxidized linoleic acid could
lower plasma total triglyceride by the interchangeable binding of two apolipoproteins (Apo A5
and Apo CIII) with very low-density lipoprotein (VLDL) (Garelnabi et al., 2007). Another study
also found that dietary linoleic acid intake was associated with reduced coronary heart disease
(CHD) risk (Farvid, 2014). On the other hand, some studies indicated that the risk of chronic
inflammatory disease, cardiovascular disease and obesity might be related to a high ratio of
omega-6: omega-3 PUFA (Patterson et al., 2012). Furthermore, in an animal experimental study,
taking dietary supplements long-term with oxidized linoleic acid was found potentially cause
atheropathogenesis (Garelnabi et al., 2017). Additionally, our research also found the association
between linoleic acid and the lipid-related PC score was modified by BCAA-related metabolite
PC score, suggesting that the effect of unsaturated fatty acid intake on the lipids concentrations
varies across different BCAA levels. Linoleic acid intake had a stronger influence on lowering
lipid (total triglycerides and VLDL) levels in participants with high BCAA-related metabolites

35
concentration than in participants with low BCAA-related metabolites concentration. A possible
explanation for this might be that branched-chain amino acids catabolism and linoleic acid
catabolism appear to be competing for some common aspects in the enzyme system. The
mechanism of how dietary intake affects BCAA and lipid metabolism is still not clear. Yet,
several studies indicated that the risk of obesity, T2D, and cardiovascular disease might be
affected by the overall dietary pattern that could possibly be modulated through increasing BCAA
level (Merz et al., 2018; Rousseau et al., 2019; Newgard et al., 2009).
The results of this research fill the gaps of current knowledge regarding the association
among targeted metabolites and dietary intake on obesity and metabolic dysfunction in young
adults. Our research population is unique with detailed sociodemographic characteristics. This
research was conducted after adjusting for major covariates to control potential confounders that
are known risk factors for obesity and type 2 diabetes, such as smoking and physical exercise.
Additionally, the measurements comprehensively involved carbohydrates, lipids, and protein
catabolism. This provided us with a broad range of possible pathways for metabolic dysfunction
in adolescents and young adults who are at the prediabetes stage. Finally, the metabolites that
correlated with adiposity- and cardiometabolic-related outcomes might serve as biomarkers for
identifying the metabolic pathways for individuals, which could contribute to the disease
prevention, therapy strategies and prognosis prediction.
Nevertheless, this research has several limitations. First, given the small sample size, these
results need to be verified by future studies with a larger sample size, especially for the
stratification analysis. Second, since this was a cross-sectional study, we cannot infer any causal
relationships between metabolites and cardiometabolic outcomes. Last, the nutrition data was
generated by 24-hour diet recalls. It is important to have caution to interpret the results for the

36
association between dietary intake and outcomes. There could be recall bias in our dietary intake
data which may be differentially misclassified by obesity status. However, obesity status did not
significantly influence our main findings about unsaturated fatty acid and lipids metabolism
(Table 6).
In conclusion, this cross-sectional study using PLS regression, principal components
analysis, and multiple linear regression identified that L-Glutamic acid & L-Glutamate acid, L-
Alanine, lactate, aromatic amino acid (L-Phenylalanine and L-Tyrosine), BCAA-related
metabolites (L-Leucine/L-Isoleucine, C3 and C5 acylcarnitines) were positively associated with
altering on lipids metabolism. Additionally, we found that the relationship between linoleic acid
intake and lipids-related outcomes was modified by BCAA related metabolites. Future studies are
needed to explore the biological mechanism and study causality among metabolites, diet, and
cardiometabolic traits.

37
References
Abdi, H. (2010). Partial least squares regression and projection on latent structure regression
(PLS regression). Wiley Interdisciplinary Reviews: Computational Statistics, 2(1), 97-106.
Bain, J. R. (2013). Targeted metabolomics finds its mark in diabetes research. Diabetes, 62(2),
349-351.
Batchuluun, B., Al Rijjal, D., Prentice, K. J., Eversley, J. A., Burdett, E., Mohan, H., . . .
Wheeler, M. B. (2018). Elevated medium-chain acylcarnitines are associated with
gestational diabetes mellitus and early progression to type 2 diabetes and induce pancreatic
β-cell dysfunction. Diabetes, 67(5), 885-897.
Berg, A. H., & Scherer, P. E. (2005). Adipose tissue, inflammation, and cardiovascular disease.
Circulation Research, 96(9), 939-949.
Breier, M., Wahl, S., Prehn, C., Fugmann, M., Ferrari, U., Weise, M., . . . Lechner, A. (2014).
Targeted metabolomics identifies reliable and stable metabolites in human serum and
plasma samples. PloS One, 9(2), e89728.
Butte, N. F., Liu, Y., Zakeri, I. F., Mohney, R. P., Mehta, N., Voruganti, V. S., . . . Comuzzie, A.
G. (2015). Global metabolomic profiling targeting childhood obesity in the hispanic
population12 American Society for Nutrition.
Centers for Disease Control and Prevention. National Diabetes Statistics Report, 2020. Atlanta,
GA: Centers for Disease Control and Prevention, U.S. Dept of Health and Human Services;
2020
Chen, Z., Newgard, C. B., Kim, J. S., IIkayeva, O., Alderete, T. L., Thomas, D. C., . . . Gilliland,
F. D. (2019). Near-roadway air pollution exposure and altered fatty acid oxidation among

38
adolescents and young adults – the interplay with obesity. Environment International, 130,
104935.
Eisen, M. B., Spellman, P. T., Brown, P. O., & Botstein, D. (1998). Cluster analysis and display
of genome-wide expression patterns. Proceedings of the National Academy of Sciences of
the United States of America, 95(25), 14863.
Farvid, M. S., Ding, M., Pan, A., Sun, Q., Chiuve, S. E., Steffen, L. M., . . . Hu, F. B. (2014).
Dietary linoleic acid and risk of coronary heart disease: A systematic review and meta-
analysis of prospective cohort studies. Circulation, 130(18), 1568-1578.
Floegel, A., Stefan, N., Yu, Z., Mühlenbruch, K., Drogan, D., Joost, H., . . . Pischon, T. (2013).
Identification of serum metabolites associated with risk of type 2 diabetes using a targeted
metabolomic approach. Diabetes, 62(2), 639-648.
Garelnabi, M., Ainsworth, G., Mahini, H., Jamil, N., & Ochin, C. (2017). Dietary oxidized
linoleic acid modulates plasma lipids beyond triglycerides metabolism. Journal of Lipids,
2017, 1645828-7.
Garelnabi, M., Selvarajan, K., Litvinov, D., Santanam, N., & Parthasarathy, S. (2007). Dietary
oxidized linoleic acid lowers triglycerides via APOA5/APOClll dependent mechanisms.
Atherosclerosis, 199(2), 304-309.
González, I., Cao, K. L., Davis, M. J., & Déjean, S. (2012). Visualising associations between
paired ‘omics’ data sets. BioData Mining, 5(1), 19.
Hales, C. M., Carroll, M. D., Fryar, C. D., & Ogden, C. L. (2017). Prevalence of obesity among
adults and youth: United states, 2015-2016. NCHS Data Brief, (288), 1.
Hannon, T. S., Rao, G., & Arslanian, S. A. (2005). Childhood obesity and type 2 diabetes
mellitus. Pediatrics, 116(2), 473-480.

39
Kim, J. S., Chen, Z., Alderete, T. L., Toledo-Corral, C., Lurmann, F., Berhane, K., & Gilliland,
F. D. (2019). Associations of air pollution, obesity and cardiometabolic health in young
adults: The meta-AIR study. Environment International, 133(Pt A), 105180.
Mai, M., Tönjes, A., Kovacs, P., Stumvoll, M., Fiedler, G. M., & Leichtle, A. B. (2013). Serum
levels of acylcarnitines are altered in prediabetic conditions. PloS One, 8(12), e82459.
Merz, B., Frommherz, L., Rist, M. J., Kulling, S. E., Bub, A., & Watzl, B. (2018). Dietary
pattern and plasma BCAA-variations in healthy men and Women—Results from the
KarMeN study. Nutrients, 10(5), 623.
Newgard, C. (2012). Interplay between lipids and branched-chain amino acids in development of
insulin resistance. Cell Metabolism, 15(5), 606-614.
Newgard, C. B. (2017). Metabolomics and metabolic diseases: Where do we stand? Cell
Metabolism, 25(1), 43-56.
Newgard, C. B., An, J., Bain, J. R., Muehlbauer, M. J., Stevens, R. D., Lien, L. F., . . . Svetkey,
L. P. (2009). A branched-chain amino acid-related metabolic signature that differentiates
obese and lean humans and contributes to insulin resistance. Cell Metabolism, 9(4), 311-
326.
Patterson, E., Wall, R., Fitzgerald, G. F., Ross, R. P., & Stanton, C. (2012). Health implications
of high dietary omega-6 polyunsaturated fatty acids. Journal of Nutrition and Metabolism,
2012, 539426-16.
Roberts, L. D., Souza, A. L., Gerszten, R. E., & Clish, C. B. (2012). Targeted metabolomics.
Current Protocols in Molecular Biology, 98(1), 30.2.1-30.2.24.

40
Rousseau, M., Guénard, F., Garneau, V., Allam-Ndoul, B., Lemieux, S., Pérusse, L., & Vohl, M.
(2019). Associations between dietary protein sources, plasma BCAA and short-chain
acylcarnitine levels in adults. Nutrients, 11(1), 173.
Stanley, C. A. (2009). Regulation of glutamate metabolism and insulin secretion by glutamate
dehydrogenase in hypoglycemic children1234 American Society for Nutrition.
Tara M. Wallace, Jonathan C. Levy, & David R. Matthews. (2004). Use and abuse of HOMA
modeling. Diabetes Care, 27(6), 1487-1495.
Walford, G. A., Davis, J., Warner, A. S., Ackerman, R. J., Billings, L. K., Chamarthi, B., . . .
Florez, J. C. (2013). Branched chain and aromatic amino acids change acutely following
two medical therapies for type 2 diabetes mellitus. Metabolism, 62(12), 1772-1778.
Weinstein, J. N., Myers, T. G., O'Connor, P. M., Friend, S. H., Fornace A.J., J., Kohn, K. W., . . .
Paull, K. D. (1997). An information-intensive approach to the molecular pharmacology of
cancer. Science (Washington), 275(5298), 343-349.
Weiss, R., & Kaufman, F. R. (2008). Metabolic complications of childhood obesity: Identifying
and mitigating the risk. Diabetes Care, 31 Suppl 2(2), S310-S316.
Yanxin Wang, Brynn Jones Voy, Sumithra Urs, & Suyeon Kim. (2004). The human fatty acid
synthase gene and de novo lipogenesis are coordinately regulated in human adipose Tissue1.
The Journal of Nutrition, 134(5), 1032.
Young Lee, Hiroshi Hirose, Yan-Ting Zhou, Victoria Esser, J Denis McGarry, & Roger H
Unger. (1997). Increased lipogenic capacity of the islets of obese rats. Diabetes, 46(3), 408.

Asset Metadata

Creator Qiu, Chenyu (author)

Core Title Targeted metabolic signatures and diet in associations with obesity and insulin resistance in young adults

Contributor Electronically uploaded by the author (provenance)

School Keck School of Medicine

Degree Master of Science

Degree Program Biostatistics

Publication Date 04/27/2020

Defense Date 04/01/2020

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag diet,insulin resistance,OAI-PMH Harvest,obesity,targeted metabolites,Young adults

Language English

Advisor Chen, Zhanghua (committee chair), Gilliland, Frank (committee member), Thomas, Duncan (committee member)

Creator Email chenyuq@usc.edu,qiuchenyu@hotmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-291811

Unique identifier UC11663766

Identifier etd-QiuChenyu-8359.pdf (filename),usctheses-c89-291811 (legacy record id)

Legacy Identifier etd-QiuChenyu-8359.pdf

Dmrecord 291811

Document Type Thesis

Rights Qiu, Chenyu

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Abstract (if available)

Abstract Background: The effects of metabolites on altering cardiometabolic traits including adiposity, inflammation, glucose and lipid metabolism have been shown in several adults and animal studies. However, only a few studies have conducted investigations on adolescents and young adults. The biological mechanism of metabolic dysfunction, as well as the interaction between dietary intake and metabolites in the association of key metabolic pathway remain unclear in young adults. ❧ Objective: The purpose of this research was to investigate the association of targeted metabolic signatures and diet with obesity and insulin resistance in young adults. ❧ Methods: The 107 young adults, aged 18-22 years, were a subset of the Meta-air study recruited from the Children’s Health Study in 2014-2018. Outcomes included adiposity measures, glucose metabolism traits, lipids measures, inflammatory markers, and cardiovascular markers. The concentration of the 64 targeted metabolites profiles were measured in fasting serum samples from participants. Dietary intake variables, including daily total calorie intake, grams of macronutrients intake, dietary glycemic index and percent calories from macronutrients, were estimated by 24-hour dietary recall. Partial least squares regression was used to identify the association between 64 targeted metabolites and 42 outcomes. Principal components analysis was used to identify clusters that represent major metabolic pathways. Linear regression models were built to assess the association between metabolite PCs and cardiometabolic trait PCs. Last, multiple linear regression models were used to analyze the association of diet and outcome PC score, and effect modification by metabolite PC scores. ❧ Results: The aromatic amino acid, L-Glutamic acid & L-Glutamate, and L-Alanine related PC score was positively associated with all outcome PC scores (p = 0.048 for insulin-resistance related PC score, p < 0.001 for adiposity-related PC score, and p = 0.01 for VLDL- and total triglycerides-related PC score). BCAA-related metabolite PC score was positively associated with the VLDL- and total triglycerides-related PC score (p = 0.01). The inverse association of percent calories from fat and three polyunsaturated fatty acid intakes—total polyunsaturated fatty acid intake, omega-6 fatty acid intake, and linoleic acid intake (LA, 18:2)—with VLDL- and triglycerides-related PC score were significantly modified by BCAA related metabolomic PC score. For example, by stratifying our participants by the median of the BCAA-related PC score, one standard deviation (8.3g) increase in linoleic acid intake was associated with 0.25 unit decrease in mean VLDL- and total triglycerides-related PC score among participants with low BCAA-related metabolomic PC score (p = 0.60). On the other hand, among participants with high BCAA-related metabolomic PC score, the association estimate was 1.65 unit decrease in mean VLDL- and triglycerides-related PC score by one standard deviation increase in linoleic acid intake (p = 0.01). ❧ Conclusions: L-Glutamic acid & L-Glutamate, L-Alanine, aromatic amino acid (L-Phenylalanine and L-Tyrosine), BCAA-related metabolites (L-Leucine/L-Isoleucine, C3 and C5 acylcarnitines) were associated with alterations of lipids metabolism in young adults. Additionally, the estimated inverse effects of linoleic acid intake and percent calories from fat on serum VLDL and total triglycerides levels were greater for people who had high concentrations of BCAA related metabolites, than that for people who had low concentrations of BCAA related metabolites.