Endotoxin Biomarkers Are Associated With Adiposity and Cardiometabolic Risk Across 6 Years of Follow-up in Youth

Abstract

Context

Metabolic endotoxemia may be a shared mechanism underlying childhood obesity and early-onset metabolic diseases (eg, type 2 diabetes, nonalcoholic fatty liver disease).

Objective

Examine prospective associations of serum endotoxin biomarkers lipopolysaccharide (LPS) and its binding protein, LPS binding protein (LBP), and anti-endotoxin core immunoglobulin G (EndoCab IgG) with adiposity and cardiometabolic risk in youth.

Design/setting

This prospective study included 393 youth in the Exploring Perinatal Outcomes Among Children cohort in Colorado. Participants were recruited from 2006 to 2009 at age 10 years (baseline) and followed for 6 years (follow-up). We examined associations of endotoxin biomarkers at baseline with adiposity [body mass index (BMI) z-score, visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), skinfolds, waist circumference] and cardiometabolic risk (insulin, glucose, adipokines, lipid profile, blood pressure) across both visits using mixed-effects regression, and with hepatic fat fraction (HFF) at follow-up using linear regression.

Results

Higher LPS and LBP predicted greater adiposity across follow-up. Each 1-unit log-transformed LPS corresponded with 0.23 (95% CI 0.03, 0.43) units BMI z-score, 5.66 (95% CI 1.99, 9.33) mm³ VAT, 30.7 (95% CI 8.0, 53.3) mm³ SAT, and 8.26 (95% CI 4.13, 12.40) mm skinfold sum. EndoCab IgG was associated with VAT only [3.03 (95% CI 0.34, 5.71) mm³]. LPS was associated with higher insulin [1.93 (95% CI 0.08, 3.70) µU/mL] and leptin [2.28 (95% CI 0.66, 3.90) ng/mL] and an adverse lipid profile. No association was observed with HFF. Accounting for pubertal status and lifestyle behaviors did not change findings. However, adjustment for prepregnancy BMI and gestational diabetes attenuated most associations.

Conclusions

Serum endotoxin may be a marker of pathophysiological processes underlying development of childhood obesity and cardiometabolic conditions associated with exposure to fetal overnutrition.

In parallel with the childhood obesity epidemic, chronic metabolic diseases like type 2 diabetes (T2D) and nonalcoholic fatty liver disease (NAFLD) are on the rise in youth (1, 2). While the behavioral and biological underpinnings are complex and multifaceted, a Western diet high in refined carbohydrates and unhealthy fats is recognized as a key behavioral risk factor (3). One biological pathway that may link a Western diet to obesity-related disease is metabolic endotoxemia, a condition caused by translocation of lipopolysaccharide (LPS), the major glycolipid component released from the cell wall of gram negative bacteria from the gut into circulation (4). This phenomenon is a consequence of gut dysbiosis and compromised integrity of the mucosal epithelial barrier (5) and typically co-occurs with translocation of LPS binding protein (LBP), an acute-phase protein that binds LPS to elicit an immune response to gram-negative bacterial infections (6). The presence of LPS and LBP in circulation results in multiple physiologic and metabolic disturbances including inflammation, hepatic and systemic insulin resistance, innate immune system activation, and disruptions to multiple aspects of lipid metabolism including circulating fatty acids, triglycerides, phospholipids, and bile acids (7, 8).

A relatively large literature links endotoxin exposure to development of T2D (9-11), cardiovascular disease (12), and NAFLD (13, 14) in adults. Only a handful of studies have examined these associations during childhood or adolescence, 2 sensitive life stages during which chronic disease precursors (eg, excess adiposity, hyperglycemia, elevated blood pressure, dyslipidemia) are established and subsequently track across the life course (15-18). Current literature in youth comprises small case-control studies (n < 100) that focus on LPS in relation to NAFLD and/or obesity. For example, Yuan et al (19) reported higher serum LPS among youth with nonalcoholic steatohepatitis or obesity as compared to normal weight controls in a study of 51 Chinese adolescents. Studies in German (n = 49) (20) and Italian (n = 49) (21, 22) children found higher serum LPS in youth with NAFLD or mild simple steatosis as compared to healthy controls. Similarly, studies in obese Hispanic adolescents observed higher LPS among participants with (n = 32) vs without (n = 11) NAFLD (23) and noted positive correlations between serum LPS, body mass index (BMI), triglycerides, and waist circumference (n = 32 obese; n = 32 normal weight) (24). While these findings implicate endotoxin exposure in the pathophysiology of obesity and related conditions, the cross-sectional nature of the previously discussed studies preclude inference on whether the elevated endotoxin is a cause or consequence of the health outcomes. Moreover, because LPS and LBP have short half-lives (ie, 12-24 hours) (25), they may not capture long-term endotoxin exposure. Markers of a persistent endotoxin load, such as the endotoxin-core immunoglobulin antibodies that have been linked to obesity (26) and NAFLD in adults (27), should be considered when studying the etiology of chronic metabolic diseases in youth.

Here, we capitalize on data from the Exploring Perinatal Outcomes among Children (EPOCH) cohort, a general-risk study of racially/ethnically diverse US youth, to explore prospective associations of acute (LPS and LBP) and chronic [anti-endotoxin core immunoglobulin G (EndoCab IgG)] endotoxin exposure with adiposity, a suite of cardiometabolic risk biomarkers, and hepatic fat from childhood through adolescence. Our overarching hypothesis was that higher serum levels of endotoxin at baseline would be associated with higher adiposity and cardiometabolic risk across childhood and adolescence. Of the 3 biomarkers, we anticipated the strongest and most consistent associations of EndoCab IgG with adiposity and cardiometabolic risk and that exposure to fetal overnutrition via maternal obesity and/or gestational diabetes would account for variation in the relationship between the endotoxin biomarkers and health outcomes.

Materials and Methods

Study Population

EPOCH is a prospective cohort of children whose mothers were members of the Kaiser Permanente of Colorado (KPCO) health plan at the time of their birth (1995-2002), initiated with the goal of characterizing long-term consequences of exposure to overnutrition in utero on cardiometabolic health across childhood and adolescence (28, 29). The baseline visit occurred when participants were median 10 years of age (n = 604; range 6-14 years) and with a second follow-up visit at median 16 years of age (n = 414; range 12-19 years). The present analysis is part of an ancillary study that seeks to identify determinants and risk factors of liver fat in youth (KL2-TR002534; PI: Perng) among the 395 participants for whom we measured hepatic fat fraction at follow-up. Of them, we excluded 2 participants with inadequate serum volumes at baseline for the endotoxin assays, yielding an analytic sample of 393. All mother provided informed consent, and offspring provided written assent. This study was approved by the Colorado Multiple Institutional Review Board.

Blood Collection

At baseline and follow-up, trained research assistants (RAs) collected an 8-hour fasting blood sample from the antecubital vein. All samples were refrigerated immediately, processed within 24 hours, and stored at −80°C until time of analysis. These samples were used for the endotoxin and cardiometabolic biomarker assays.

Endotoxin Biomarkers

LPS was analyzed in serum samples collected at baseline, diluted 10× using Limulus Amebocyte Lysate Chromogenic Endpoint Assay endotoxin detection kits (Hycult Biotech, Uden, The Netherlands; Antibody Registry ID: AB_10130891) according to the manufacturer’s instructions. LBP levels were determined in serum samples diluted 1500× using human LBP enzyme-linked immunosorbent assay (ELISA) kits following manufacturer’s instructions (Hycult; Antibody Registry ID: AB_10129813). In addition to LPS and LBP, we also measured EndoCab IgG, a marker of endotoxin core antibodies that reflect an integrated antibody response to a chronic and persistent endotoxin load (30). IgG endotoxin-core antibodies were analyzed in serum samples diluted 200× using EndoCab IgG ELISA kits following manufacturer’s instructions (Hycult; Antibody Registry ID: AB_10129811). Intra- and interassay coefficients of variation for these assays were between 5% and 18%.

Cardiometabolic Biomarkers

Using fasting blood collected at baseline and follow-up, we assayed fasting glucose enzymatically, fasting insulin (Millipore, Darmstadt, Germany; Antibody Registry ID: AB_2801577) and leptin via radioimmunoassay (Millipore; Antibody Registry ID: AB_2894698), and adiponectin via ELISA (Millipore; Antibody Registry ID: AB_2801457). We used fasting glucose and insulin values to calculate the homeostatic model assessment of insulin resistance as glucose mg/dL × insulin µIU/mL)/405. Total cholesterol, high-density lipoprotein (HDL), and triglycerides were assayed on the Olympus AU400 advanced chemistry analyzer system. We calculated low-density lipoprotein (LDL) as total cholesterol − HDL − (triglycerides/5) using the Friedewald method (31).

RAs measured participants’ blood pressure twice in the sitting position using an oscillometric monitor (Dinamap ProCare V100). We used the average of the 2 values in the analysis and focused on systolic rather than diastolic blood pressure because it is more accurately measured in children and is a stronger determinant of future health than diastolic blood pressure (32).

Anthropometric and Body Composition Assessment

At both visits, RAs measured the participants’ height (m) and weight (kg). We used these values to calculate BMI (kg/m²). We standardized BMI as an age- and sex-specific z-score using the World Health Organization growth reference for children 5 to 19 years (33). RAs also measured waist circumference (cm) using the National Health and Nutrition Examination Survey protocol (34) and the subscapular, triceps, and suprailiac skinfold thicknesses via Holtain calipers (mm), the sum of which was used in the analysis as a proxy for subcutaneous adiposity. A trained technician performed magnetic resonance imaging (MRI) of the abdominal region with a 3 T HDx Imager (General Electric, Waukesha, WI, USA) with the participant in the supine position. A series of T1-weighted coronal images were taken to locate the L4/L5 plane. One axial, 10-mm, T1-weighted image, at the umbilicus or L4/L5 vertebrae was analyzed to determine visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) depots. We assessed VAT and SAT separately, as well as the ratio of VAT:SAT as an indicator of central visceral adiposity.

At the follow-up visit only, we performed hepatic imaging on 395 subjects (35). In brief, MRIs were performed by trained technicians using a breath-hold 6-point MRI technique. Hepatic fat was calculated from the mean pixel signal intensity data for each flip angle acquisition using the Lipoquant plug-in from OsiriX (Pixmeo) (36). Given the low proportion of participants with clinical NAFLD [6.1% with hepatic fat fraction (HFF) > 5.5%], we focused on assessing HFF continuously following a natural log (ln) transformation for the main analysis.

Covariates

Maternal prepregnancy BMI was calculated using prepregnancy weight from KPCO medical records and measured height and categorized according to standard definitions for underweight (<18.5 kg/m²), normal weight (18.5-24.9 kg/m²), overweight (25.0-29.9 kg/m²), and obese (>30 kg/m²). All pregnant women at KPCO are routinely screened for gestational diabetes mellitus (GDM) at 24 to 28 weeks using the standard 2-step protocol (37). We abstracted information on birthweight (g) from medical records. For bivariate analyses, we categorized birthweight according to standard thresholds for low birthweight (<2500 g), normal (2500 to 4000 g), and macrosomic (>4000 g).

At the baseline, mothers self-reported total gestational weight gain for the index pregnancy; their education level, which we categorized as less than high school, high school or equivalent, or high school graduate; and smoking habits during pregnancy (prenatal smoking yes vs no) via a questionnaire (38). At both visits, we collected self-reported Tanner stage of pubic hair development in boys and breast and pubic hair development in girls (39). At these visits, we also inquired on lifestyle behaviors, including dietary habits via the Block Kids Food Frequency Questionnaire (40) and physical activity levels at T2 using the 3-Day Physical Activity Recall Questionnaire (41), which was subsequently used to derive average energy expenditure (metabolic equivalents) over a 3-day period.

Data Analysis

Prior to formal analysis, we assessed univariate distributions of LPS, LBP, and EndoCab IgG, as well as the adiposity and cardiometabolic outcomes. Due to nonnormal distributions, we ln-transformed all 3 endotoxin biomarkers and hepatic fat fraction prior to regression analysis. Next, we assessed bivariate associations of maternal/perinatal and participant characteristics at baseline with the endotoxin biomarkers to identify potential confounders to the relationships of interest. Here, we estimated mean ± SD for each endotoxin biomarker within categories of background characteristics and assessed the statistical significance of each relationship using the P-value from a Chi-squared test. This step, in conjunction with our a priori knowledge of determinants of cardiometabolic health in youth informed covariate selection for multivariable models. Covariates of particular interest were maternal prepregnancy BMI and GDM as these maternal conditions are markers of fetal overnutrition (42) that may contribute to offspring metabolic endotoxemia by influencing the infant gut microbiome (43, 44) and are also established risk factors for offspring obesity and cardiometabolic risk (45-47). Finally, to gain an understanding of how these biomarkers correlate with known indicators of metabolic risk, we assessed Spearman’s correlations among the endotoxin biomarkers with indicators of adiposity and cardiometabolic risk at baseline.

Next, we examined associations of the ln-transformed endotoxin biomarkers with adiposity and cardiometabolic risk from childhood (baseline) through adolescence (follow-up) using mixed-effects linear regression with the exception of HFF, which we analyzed using simple linear regression since we collected data on HFF at follow-up only.

In the mixed models, the explanatory variables were the endotoxin biomarker of interest, longitudinal age (indicator for time), a random effect for individual ID, and an unstructured covariance matrix. The dependent variable (outcome) was repeated measurements of each indicator of adiposity or cardiometabolic risk. This approach efficiently leverages longitudinal data as it do not require that all participants have the same number of outcome measurements or that the outcomes be assessed at exactly the same time. The estimate of interest is the beta for a given endotoxin biomarker, interpreted as the association of that endotoxin biomarker with average adiposity or cardiometabolic risk over the 6 year follow-up.

For the multivariable analysis, we adjusted for key covariates that may confound the relationship between endotoxin and adiposity/cardiometabolic risk. In Model 1, we accounted for race/ethnicity and maternal education at the time of birth. Model 2 further accounted for longitudinal assessments of Tanner stage, mean energy expenditure and total energy intake. Model 3 included Model 1 covariates plus maternal prepregnancy BMI and GDM as indicators of fetal overnutrition that are independent risk factors for offspring obesity and cardiometabolic risk. In models where HFF (measured at follow-up only) was the outcome, we implemented general linear regression using the same strategy for covariate adjustment but focusing on baseline covariates only. In reporting the results from multivariable models, we show the beta estimate and 95% CI as well as the P-value associated with the effect estimate. We considered an estimate to be statistically significant based on alpha = 0.05.

In all models, we tested for interactions of each endotoxin biomarker with longitudinal age (to evaluate for the possibility of differential rate of change in outcomes with across increasing levels of the endotoxin biomarkers), sex (to assess for sex-specific associations), race/ethnicity (given evidence of higher risk of cardiometabolic disease among racial/ethnic minorities), and baseline weight status [in light of evidence from adult studies of an interaction between endotoxin biomarkers and overweight/obesity status (48)]. We considered P-interaction < 0.05 as the threshold for a need to conduct stratified analysis.

Finally, we implemented some sensitivity analyses. First, we accounted for percentage of total energy intake from fat in Model 2. We also accounted for longitudinal BMI z-score when the cardiometabolic biomarkers were the outcomes of interest given that excess adiposity is a determinant of cardiometabolic risk, keeping in mind that this type of statistical adjustment is prone to collider stratification bias (49). Finally, given the established literature on endotoxins and NAFLD, but in light of the low prevalence of NAFLD in our study sample, which hampered assessment of this variable as an outcome in regression analysis, we compared endotoxin biomarker concentrations among youth with vs without NAFLD. For all models, we assessed jackknifed studentized residuals to confirm assumptions of normality.

We carried out all analyses using Statistical Analyses System software (version 9.3; SAS Institute Inc., Cary, NC, USA).

Results

At baseline, participants were 10.5 ± 1.5 years (range 6.0-13.9 years). At follow-up, average age was 16.7 ± 1.2 years (range 12.6-19.6 years). Approximately half of the participants were female (49.3%). Fifty-one percent (51.3%) were non-Hispanic White, 35.9% were Hispanic, and 12.2% identified as non-Hispanic other. Average BMI z-score at baseline was 0.23 ± 1.21, with 4.1% classified as underweight, 66.4% classified as normal weight, 22.4% classified as overweight, and 7.1% classified as obese.

Table 1 shows mean ± SD of adiposity indicators and cardiometabolic risk biomarkers at baseline and follow-up. In general, we observed an increase in adiposity and a worsening of cardiometabolic profile across the 6-year study period.

Table 1.

Mean ± SD for adiposity and cardiometabolic risk at baseline, follow-up, and change across follow-up among 393 youth in the EPOCH cohort

	Baseline	Follow-up	Change
Age, years	10.5 ± 1.5	16.7 ± 1.2	6.2 ± 0.9
BMI z-scorea	0.23 ± 1.21	0.40 ± 1.12	0.18 ± 0.71
VAT, mm2	21.3 ± 14.3	32.7 ± 21.4	11.4 ± 16.4
SAT, mm2	116.0 ± 102.9	201.2 ± 151.4	86.0 ± 92.8
Skinfold thickness sum, mm	43.5 ± 24.0	35.1 ± 21.6	-8.2 ± 19.9
Waist circumference, cm	65.1 ± 11.8	80.8 ± 13.1	15.8 ± 9.0
Fasting glucose, mg/dL	82.3 ± 14.4	90.8 ± 24.0	8.4 ± 23.8
Fasting insulin, µU/mL	10.8 ± 7.8	16.6 ± 11.3	5.9 ± 11.5
HOMA-IR	2.22 ± 1.60	3.96 ± 4.81	1.75 ± 4.53
Leptin, ng/mL	7.9 ± 8.1	13.0 ± 14.2	5.2 ± 12.0
Adiponectin, µg/mL	11.9 ± 4.8	9.8 ± 5.2	-2.2 ± 6.2
Total cholesterol, mg/dL	222.3 ± 102.1	145.1 ± 28.3	-75.8 ± 104.7
HDL, mg/dL	49.8 ± 11.6	46.1 ± 9.7	-3.7 ± 11.3
LDL, mg/dL	91.4 ± 26.5	81.5 ± 23.0	-10.0 ± 22.6
Triglycerides, mg/dL	87.6 ± 40.6	87.3 ± 47.5	0.0 ± 55.2
SBP, mmHg	102.8 ± 9.8	115.8 ± 10.7	13.0 ± 11.3

Abbreviations: BMI, body mass index; HDL, high-density lipoprotein; HOMA-IR, homeostatic model assessment of insulin resistance; LDL, low-density lipoprotein; SAT, subcutaneous adipose tissue; SBP, systolic blood pressure; VAT, visceral adipose tissue.

^aAge- and sex-specific BMI z-score ≥ 1, according to the World Health Organization growth reference for children 5 to 19 years.

Table 2 shows bivariate associations (mean ± SD) of LPS (0.94 ± 0.54 EU/mL), LBP (13.0 ± 6.3 µg/mL), and EndoCab IgG (66.3 ± 50.0 GMU/mL) across maternal, perinatal, and baseline characteristics. We observed positive associations of maternal prepregnancy weight status with all 3 endotoxin biomarkers. Maternal GDM (P = 0.03), higher maternal education (P-trend = 0.006), and non-Hispanic White race/ethnicity (type 3 P-value = 0.008) were each associated with higher LPS. LPS (P-trend = 0.01) and LBP (P-trend = 0.04) were positively associated with baseline BMI z-score. LPS and LBP did not differ by sex, but EndoCab IgG was higher in boys (74.3 ± 47.0 vs 58.1 ± 31.9 GMU/mL for boys vs girls, respectively). Age at baseline was not associated with any of the endotoxin biomarkers.

Table 2.

Bivariate associations of background characteristics (mean ± SD) with endotoxin biomarkers among 393 youth in the EPOCH cohort

	na	LPS, EU/mL	P b	LBP, µg/mL	P b	EndocabIgG, GMU/mL	P b
		0.94 ± 0.54		13.0 ± 6.3		66.3 ± 50.0
		Range: 1.5-45.4		Range: 1.5-45.4		Range: 14.6-291.3
Maternal perinatal characteristics
Maternal prepregnancy weight status			0.30		0.002		0.02
Underweight (<18.5 kg/m2)	10	0.86 ± 0.47		13.4 ± 6.3		62.6 ± 32.5
Normal weight (18.5-24.9 kg/m2)	134	0.86 ± 0.46		12.3 ± 5.1		61.4 ± 32.8
Overweight (25.0-29.9 kg/m2)	75	0.94 ± 0.54		13.8 ± 6.3		74.7 ± 51.5
Obese (>30 kg/m2)	60	0.93 ± 0.56		15.2 ± 7.1		72.5 ± 50.7
Gestational weight gain			0.07		0.19		0.98
Q1 (median: 17 lbs)	88	0.95 ± 0.60		13.3 ± 7.0		64.5 ± 48.0
Q2 (median: 25 lbs)	76	1.05 ± 0.55		13.2 ± 6.9		69.2 ± 33.2
Q3 (median: 30 lbs)	86	0.98 ± 0.54		12.5 ± 5.7		67.7 ± 44.6
Q4 (median: 47 lbs)	115	0.84 ± 0.49		12.7 ± 5.9		65.4 ± 39.6
Maternal gestational diabetes mellitus			0.03		0.16		0.88
Yes	69	1.06 ± 0.62		14.0 ± 7.2		66.9 ± 52.1
No	322	0.91 ± 0.52		12.8 ± 6.0		66.1 ± 38.4
Maternal education level			0.006		0.12		0.66
<High school	13	0.77 ± 0.42		12.3 ± 5.1		68.6 ± 39.2
High school or equivalent	56	0.77 ± 0.41		11.7 ± 5.5		62.1 ± 39.8
>High school	324	0.98 ± 0.56		13.3 ± 6.4		66.9 ± 41.3
Mother smoked during pregnancy			0.86		0.67		0.61
Yes	29	0.92 ± 0.50		12.6 ± 8.3		62.6 ± 44.9
No	364	0.94 ± 0.54		13.1 ± 6.1		66.6 ± 40.7
Offspring birthweight			0.21		0.78		0.34
	36	0.86 ± 0.42		15.0 ± 9.1		67.5 ± 45.6
2500-4000 g	326	0.94 ± 0.55		12.6 ± 5.6		65.0 ± 40.3
	29	1.02 ± 0.59		15.0 ± 7.7		78.6 ± 42.4
Child’s characteristics at baseline
Sex			0.36		0.59		<0.0001
Female	194	0.96 ± 0.57		12.9 ± 6.4		58.1 ± 31.9
Male	199	0.91 ± 0.51		13.2 ± 6.1		74.3 ± 47.0
Race/ethnicity			0.008		0.51		0.18
Non-Hispanic White	202	1.05 ± 0.59		12.7 ± 6.3		63.3 ± 42.4
Hispanic	141	0.82 ± 0.48		13.4 ± 6.3		67.6 ± 40.0
Non-Hispanic other	48	0.82 ± 0.38		13.4 ± 5.9		75.1 ± 36.9
Age			0.77		0.33		0.33
6 to < 10 y	148	0.93 ± 0.48		14.0 ± 6.7		71.2 ± 47.9
10 to < 12 y	184	0.94 ± 0.56		12.0 ± 5.6		61.4 ± 35.0
12 to < 14 y	61	0.96 ± 0.62		13.9 ± 6.5		69.2 ± 38.4
Weight statusc			0.01		0.04		0.27
Underweight (<−2 z)	16	1.00 ± 0.61		13.1 ± 6.0		75.3 ± 44.5
Normal weight, −2 to <1 z	261	0.88 ± 0.48		12.6 ± 6.3		63.6 ± 42.1
Overweight, ≥1 to <2 z	88	1.07 ± 0.63		13.8 ± 6.0		70.4 ± 37.4
Obese, ≥2 z	28	1.07 ± 0.72		14.8 ± 6.2		73.1 ± 39.1
Total energy intake, kcal/day			0.57		0.27		0.98
Q1, median 1232 kcal/day	96	0.90 ± 0.57		12.6 ± 5.9		64.4 ± 42.9
Q2, median 1486 kcal/day	96	1.03 ± 0.54		12.7 ± 6.3		67.6 ± 42.4
Q3, median 1862 kcal/day	96	0.96 ± 0.58		13.5 ± 6.0		69.3 ± 45.6
Q4, median 2379 kcal/day	99	0.88 ± 0.47		13.2 ± 6.8		64.1 ± 32.9
Pubertal statusd			0.007		0.44		0.49
Tanner stage 1	174	0.86 ± 0.51		12.9 ± 6.5		68.2 ± 46.1
Tanner stage 2	135	0.99 ± 0.52		12.9 ± 5.7		65.1 ± 37.5
Tanner stage 3 or 4	83	1.03 ± 0.60		13.6 ± 6.7		64.9 ± 35.0

Abbreviations: EndoCab IgG, antibodies to endotoxin core immunoglobulins; LBP, lipopolysaccharide-binding protein; LPS, lipopolysaccharide.

^aTotals may not add up to 393 due to missing values.

^bFrom a P-for-linear-trend for ordinal variables; from a type 3 test for a difference for categorical variables.

^cAccording to the World Health Organization growth reference for children 5 to 19 years of age.

^dBased on pubic hair development.

We noted weak positive correlations of the 3 endotoxin biomarkers with one another (Spearman rho 0.05-0.12) and moderate to high correlations of the endotoxin biomarkers with indicators of adiposity and cardiometabolic risk (Table 3). The correlation coefficients were, on average, stronger among females.

Table 3.

Spearman correlations (rho) among endotoxin biomarkers, adiposity, and metabolic biomarkers at baseline (age ~10 years) in the EPOCH cohort

	LPS	LBP	Endocab	BMIz	SAT	VAT	SKF	WC	Glucose	Insulin	HOMA	LEP	APN	TC	HDL	LDL	TRG	SBP
Boys (n = 199)
LPS	1.00
LBP	0.11	1.00
Endocab IgG	0.05	0.12	1.00
BMI z-score	0.05	0.10	0.08	1.00
SAT	0.07	0.11	0.12	0.89	1.00
VAT	0.17	0.11	0.13	0.72	0.83	1.00
Skinfold thickness	0.16	0.16	0.16	0.87	0.93	0.81	1.00
Waist circumference	0.04	0.05	0.06	0.87	0.89	0.74	0.83	1.00
Fasting glucose	0.24	0.02	0.23	−0.05	−0.02	0.08	0.08	−0.05	1.00
Fasting insulin	0.15	−0.04	0.05	0.52	0.52	0.41	0.52	0.57	0.11	1.00
HOMA-IR	0.16	−0.02	0.09	0.51	0.51	0.42	0.52	0.56	0.28	0.98	1.00
Leptin	0.04	0.11	0.08	0.83	0.89	0.75	0.87	0.79	−0.08	0.56	0.54	1.00
Adiponectin	0.05	−0.01	−0.10	−0.20	−0.18	−0.12	−0.13	−0.23	0.03	−0.18	−0.17	−0.14	1.00
Total cholesterol	−0.07	−0.04	−0.09	0.07	0.05	0.05	0.01	0.05	−0.05	−0.02	−0.02	0.04	−0.04	1.00
HDL	−0.24	−0.09	−0.02	−0.15	−0.18	−0.18	−0.22	−0.21	−0.15	−0.14	−0.17	−0.08	−0.04	0.10	1.00
LDL	0.14	−0.08	0.05	0.04	0.05	0.13	0.14	0.03	0.20	−0.06	−0.03	0.10	0.09	0.04	−0.06	1.00
Triglycerides	0.28	0.11	−0.11	0.30	0.35	0.30	0.34	0.35	−0.01	0.38	0.36	0.35	−0.04	−0.10	−0.28	−0.01	1.00
SBP	−0.02	0.03	0.11	0.40	0.37	0.28	0.31	0.43	−0.10	0.25	0.22	0.36	−0.03	0.03	−0.03	−0.04	0.28	1.00
Girls (n = 194)
LPS	1.00
LBP	0.16	1.00
Endocab IgG	0.01	0.12	1.00
BMI z-score	0.15	0.19	0.12	1.00
SAT	0.16	0.19	0.10	0.87	1.00
VAT	0.20	0.20	0.11	0.70	0.81	1.00
Skinfold thickness	0.28	0.20	0.11	0.83	0.92	0.81	1.00
Waist circumference	0.10	0.17	0.04	0.82	0.89	0.72	0.81	1.00
Fasting glucose	0.24	−0.05	−0.06	0.03	0.05	0.23	0.19	−0.01	1.00
Fasting insulin	0.24	0.00	−0.04	0.47	0.56	0.51	0.53	0.60	0.25	1.00
HOMA-IR	0.26	−0.01	−0.06	0.45	0.54	0.53	0.54	0.57	0.39	0.98	1.00
Leptin	0.18	0.16	0.09	0.78	0.87	0.74	0.85	0.77	0.10	0.60	0.59	1.00
Adiponectin	−0.04	0.09	−0.08	−0.13	−0.11	−0.06	−0.13	−0.16	−0.01	−0.18	−0.17	−0.12	1.00
Total cholesterol	−0.07	0.07	0.14	0.06	−0.03	−0.01	−0.01	−0.03	−0.09	−0.13	−0.13	0.00	0.00	1.00
HDL	−0.29	0.03	−0.05	−0.21	−0.28	−0.24	−0.32	−0.21	−0.17	−0.27	−0.29	−0.23	0.04	0.04	1.00
LDL	0.14	0.03	0.02	−0.05	−0.02	0.06	0.05	−0.16	0.21	−0.15	−0.11	0.07	0.21	0.05	−0.07	1.00
Triglycerides	0.41	0.16	−0.02	0.31	0.36	0.34	0.38	0.35	0.07	0.43	0.42	0.44	−0.03	−0.07	−0.32	0.07	1.00
SBP	0.00	0.09	0.02	0.37	0.35	0.29	0.28	0.42	0.06	0.32	0.31	0.33	−0.01	−0.01	−0.01	−0.08	0.16	1.00

Bolded values indicate statistical significance at alpha = 0.05.

Abbreviations: APN, adiponectin; LEP, leptin; LPS, lipopolysaccharide, LBP, lipopolysaccharide-binding protein; EndoCab IgG, antibodies to endotoxin core immunoglobulins; BMI, body mass index; SAT, subcutaneous adipose tissue; SKT, skinfold thickness; TC, total cholesterol; TRG, triglycerides; VAT: visceral adipose tissue; HDL, high-density lipoprotein; LDL, low-density lipoprotein; SBP, systolic blood pressure.

In multivariable models that accounted for age, sex, race/ethnicity, and maternal education level (Table 4, Model 1), LPS and LBP were each associated with higher adiposity across the follow-up. For instance, each 1-unit increment in ln-transformed LBP corresponded with 0.23 (95% CI 0.03, 0.43) units BMI z-score, 5.66 (95% CI 1.99, 9.33) mm³ VAT, 30.7 (95% CI 8.0, 53.3) mm³ SAT, 8.26 (95% CI 4.13, 12.40) mm skinfold sum, and 3.01 (95% CI 0.82, 5.21) cm waist circumference. The estimates for LBP were similar to LPS but of slightly smaller magnitude (Table 4). Accounting for pubertal status, total energy intake, and physical activity level in Model 2 did not materially change results for LPS or LBP. On the other hand, the only adiposity indicator associated with EndoCab IgG was VAT: each 1-unit of ln-transformed EndoCab IgG corresponded with 3.03 (95% CI 0.34, 5.71) mm³ VAT in Model 1 and 2.97 (95% CI 0.22, 5.72) mm³ in Model 2. After accounting for maternal prepregnancy BMI and GDM in Model 3, the only relationship that remained significant was between LPS and skinfold sum [5.32 (95% CI 0.77, 9.86) mm per 1-unit ln-transformed LPS]. None of the endotoxin biomarkers at baseline were associated with HFF at follow-up.

Table 4.

Prospective associations [β (95% CI)] of endotoxin biomarkers at baseline (age 10 years) with adiposity across 6 years of follow-up among 393 youth in the EPOCH cohort

	LPS		LBP		EndoCab IgG
Model 1a	n = 391	P	n = 393	P	n = 393	P
BMI z-score	0.23 (0.03, 0.43)	0.0245	0.28 (0.06, 0.50)	0.0137	0.11 (−0.08, 0.30)	0.2577
VAT, mm3	5.66 (1.99, 9.33)	0.0026	3.69 (0.50, 6.87)	0.0233	3.03 (0.34, 5.71)	0.0271
SAT, mm3	30.7 (8.0, 53.3)	0.0081	27.8 (5.8, 49.7)	0.0134	10.8 (−8.5, 30.0)	0.2723
Skinfold sum, mm	8.26 (4.13, 12.40)	0.0001	4.65 (0.71, 8.59)	0.0207	2.69 (−0.86, 6.25)	0.1369
Waist circumference, cm	3.01 (0.82, 5.21)	0.0073	2.64 (0.52, 4.74)	0.0147	0.79 (−1.13, 2.72)	0.4189
Hepatic fat fraction, %b	0.04 (−0.51, 0.58)	0.9001	0.18 (−0.41, 0.77)	0.5544	−0.10 (−0.67, 0.46)	0.7222
Model 2 c	n = 391	P	n = 393	P	n = 393	P
BMI z-score	0.24 (0.03, 0.44)	0.0219	0.27 (0.04, 0.49)	0.0191	0.09 (−0.10, 0.28)	0.3284
VAT, mm3	5.96 (2.30, 9.63)	0.0015	3.81 (0.54, 7.07)	0.0225	2.97 (0.22, 5.72)	0.0341
SAT, mm3	32.3 (10.57, 54.0)	0.0037	27.5 (5.5, 49.4)	0.0143	8.3 (−10.7, 27.3)	0.3911
Skinfold sum, mm	8.56 (4.44, 12.68)	<0.0001	4.69 (0.66, 8.72)	0.0226	2.38 (−1.19, 5.95)	0.1908
Waist circumference, cm	3.10 (0.93, 5.27)	0.0052	2.54 (0.42, 4.66)	0.0191	0.60 (−1.32, 2.51)	0.5401
Hepatic fat fraction, %a	−0.01 (−0.57, 0.56)	0.9801	0.19 (−0.41, 0.80)	0.5278	−0.11 (−0.68, 0.47)	0.7155
Model 3 d	n = 278	P	n = 279	P	n = 279	P
BMI z-score	0.12 (−0.09, 0.33)	0.2580	0.11 (−0.17, 0.39)	0.4472	0.08 (−0.13, 0.29)	0.4790
VAT, mm3	3.61 (−0.38, 7.59)	0.0755	2.74 (−0.73, 6.20)	0.1208	2.26 (−0.67, 5.18)	0.1296
SAT, mm3	15.8 (−8.2, 39.7)	0.1964	8.6 (−16.1, 33.2)	0.4954	8.54 (−11.2, 28.4)	0.3941
Skinfold sum, mm	5.32 (0.77, 9.86)	0.0221	1.61 (−3.29, 6.51)	0.5178	1.55 (−2.41, 5.51)	0.4417
Waist circumference, cm	1.84 (−0.49, 4.18)	0.1215	0.94 (−1.62, 3.50)	0.9411	0.64 (−1.43, 2.71)	0.5455
Hepatic fat fraction, %b	−0.16 (−0.78, 0.45)	0.6006	−0.16 (−0.92, 0.59)	0.6706	−0.08 (−0.71, 0.56)	0.8161

Abbreviations: BMI, body mass index; EndoCab IgG, antibodies to endotoxin core immunoglobulins; LBP, lipopolysaccharide-binding protein; LPS, lipopolysaccharide, SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue.

^aModel 1: From a linear mixed model with the endotoxin biomarker, age in decimal years as an indicator of time, random effect for individual ID, and an unstructured correlation matrix that accounts for sex, race/ethnicity, and maternal education level.

^bEstimates are from a linear regression model where each endotoxin biomarker is the exposure and hepatic fat fraction at follow-up is the outcome.

^cModel 2: Model 1 + longitudinal pubertal status, total energy intake, and physical activity levels.

^dModel 3: Model 1 + maternal prepregnancy BMI, and gestational diabetes mellitus.

Table 5 shows associations of endotoxin with indicators of cardiometabolic risk. Higher LPS was related to higher fasting insulin [1.93 (95% CI 0.08, 3.79) mg/dL], leptin [2.28 (95% CI 0.66, 3.90) ng/mL], total cholesterol [0.19 (95% CI 0.06, 0.31) mg/dL], LDL [0.17 (95% CI 0.07, 0.28) mg/dL], and triglycerides [0.19 (95% CI 0.12, 0.26) mg/dL] and lower HDL [−0.07 (95% CI −0.11, −0.03) mg/dL] across follow-up (Model 1). Adjustment for pubertal status and lifestyle behaviors did not change estimates (Model 2). Accounting for maternal prepregnancy BMI and GDM attenuated associations with fasting insulin, leptin, and HDL, but the estimates for total cholesterol [0.22 (95% CI 0.08, 0.37) mg/dL], LDL [0.18 (95% CI 0.06, 0.31) mg/dL], and triglycerides [0.15 (95% CI 0.08, 0.22) mg/dL] remained significant. Correlations with VAT:SAT ratio with indicators of adiposity and cardiometabolic risk were nearly null, as were associations from regression analyses; therefore, we do not show results for this variable (available upon request).

Table 5.

Prospective associations [β (95% CI)] of endotoxin biomarkers at baseline (age 10 years) with biomarkers of cardiometabolic risk across 6 years of follow-up among 393 youth in the EPOCH cohort

	LPS		LBP		EndoCab IgG
Model 1a	n = 391	P	n = 393	P	n = 393	P
Fasting glucose, mg/dL	−1.82 (−8.12,0.47)	0.5695	0.94 (−2.39, 4.26)	0.5811	3.36 (−1.09, 7.80)	0.1388
Fasting insulin, µU/mL	1.93 (0.08, 3.79)	0.0414	0.24 (−1.24, 1.71)	0.7535	0.34 (−1.03, 1.72)	0.6236
HOMA-IR	−0.02 (−1.19, 1.07)	0.9743	0.05 (−0.46, 0.55)	0.8582	0.25 (−0.17, 0.68)	0.2602
Leptin, ng/mL	2.28 (0.66, 3.90)	0.0060	1.63 (−0.16, 3.43)	0.0747	0.93 (−0.62, 2.48)	0.2404
Adiponectin, µg/mL	−0.39 (−1.18, 0.40)	0.3367	0.29 (−0.73, 1.31)	0.5730	−0.67 (−1.44, 0.09)	0.0854
Total cholesterol, mg/dL	0.19 (0.06, 0.31)	0.0033	0.05 (−0.08, 0.17)	0.4657	0.00 (−0.10, 0.10)	0.9744
HDL, mg/dL	−0.07 (−0.11, −0.03)	0.0016	0.01 (−0.03, 0.06)	0.5681	−0.01 (−0.05, 0.03)	0.5995
LDL, mg/dL	0.17 (0.07, 0.28)	0.0016	0.01 (−0.10, 0.12)	0.9095	0.03 (−0.07, 0.12)	0.5783
Triglycerides, mg/dL	0.19 (0.12, 0.26)	<0.0001	0.06 (−0.01, 0.12)	0.0655	−0.02 (−0.10, 0.05)	0.5270
SBP, mmHg	1.31 (−0.16, 2.77)	0.0788	1.20 (−0.34, 2.75)	0.1271	0.74 (−0.68, 2.16)	0.3083
Model 2 b	n = 391	P	n = 393	P	n = 393	P
Fasting glucose, mg/dL	−1.75 (−8.11, 4.61)	0.5885	1.04 (−2.34, 4.42)	0.5459	3.30 (−1.13, 7.73)	0.1441
Fasting insulin, µU/mL	2.09 (0.24, 3.94)	0.0266	0.17 (−1.28, 1.61)	0.8208	0.07 (−1.29, 1.40)	0.9336
HOMA-IR	0.03 (−1.03, 1.09)	0.9576	0.05 (−0.45, 0.55)	0.8517	0.21 (−0.22, 0.64)	0.3378
Leptin, ng/mL	2.45 (0.88, 4.02)	0.0023	1.63 (−0.20, 3.46)	0.0804	0.68 (−0.86, 2.23)	0.3857
Adiponectin, µg/mL	−0.33 (−1.15, 0.50)	0.4368	0.47 (−0.58, 1.51)	0.3800	−0.72 (−1.49, 0.05)	0.0681
Total cholesterol, mg/dL	0.19 (0.07, 0.32)	0.0024	0.06 (−0.06, 0.18)	0.3390	0.01 (−0.09, 0.11)	0.8589
HDL, mg/dL	−0.07 (−0.11, −0.03)	0.0011	0.02 (−0.03, 0.07)	0.4814	−0.01 (−0.05, 0.04)	0.7832
LDL, mg/dL	0.18 (0.07, 0.28)	0.0012	0.02 (−0.10, 0.13)	0.7810	0.03 (−0.06, 0.13)	0.4693
Triglycerides, mg/dL	0.20 (0.13, 0.27)	<0.0001	0.06 (0.00, 0.13)	0.0625	−0.03 (−0.11, 0.04)	0.3761
SBP, mmHg	1.34 (−0.11, 2.79)	0.0704	1.11 (−0.43, 2.65)	0.1564	0.72 (−0.69, 2.14)	0.3163
Model 3 c	n = 278	P	n = 279	P	n = 279	P
Fasting glucose, mg/dL	−4.60 (−12.59, 3.39)	0.2580	0.54 (−3.90, 4.97)	0.8118	3.79 (−1.76, 9.34)	0.1804
Fasting insulin, µU/mL	1.28 (−0.94, 3.51)	0.2572	−0.69 (2.58, 1.21)	0.4767	0.10 (−1.54, 1.74)	0.9058
HOMA-IR	−0.43 (−1.85, 0.98)	0.5468	−0.19 (−0.90, 0.52)	0.6036	0.21 (−0.31, 0.73)	0.4322
Leptin, ng/mL	1.21 (−0.41, 2.83)	0.1430	0.59 (−1.47, 2.64)	0.5747	1.00 (−0.57, 2.57)	0.2114
Adiponectin, µg/mL	−0.83 (−1.72, 0.05)	0.0638	0.85 (−0.37, 2.06)	0.1717	−0.60 (−1.53, 0.32)	0.2024
Total cholesterol, mg/dL	0.22 (0.08, 0.37)	0.0024	−0.01 (−0.18, 0.17)	0.9397	0.00 (−0.13, 0.12)	0.9355
HDL, mg/dL	−0.03 (−0.08, 0.02)	0.2099	0.00 (−0.06, 0.06)	0.9424	0.00 (−0.05, 0.04)	0.8314
LDL, mg/dL	0.18 (0.06, 0.31)	0.0040	−0.01 (−0.16, 0.15)	0.9123	0.02 (−0.08, 0.13)	0.6525
Triglycerides, mg/dL	0.15 (0.08, 0.22)	<0.0001	0.01 (−0.08, 0.09)	0.8401	−0.05 (−0.13, 0.03)	0.2393
SBP, mmHg	−0.01 (−1.67, 1.66)	0.9949	0.26 (−1.65, 2.17)	0.7873	0.00 (−1.66, 1.66)	0.9985

Abbreviations: EndoCab IgG, antibodies to endotoxin core immunoglobulins; HDL, high-density lipoprotein; HOMA-IR, homeostatic model assessment of insulin resistance; LBP, lipopolysaccharide-binding protein; LDL, low-density lipoprotein; LPS, lipopolysaccharide; SBP, systolic blood pressure.

^aModel 1: From a linear mixed model with the endotoxin biomarker, age in decimal years as an indicator of time, random effect for individual ID, and an unstructured correlation matrix that accounts for sex, race/ethnicity, and maternal education level.

^bModel 2: Model 1 + longitudinal pubertal status, total energy intake, and physical activity levels.

^cModel 3: Model 1 + maternal prepregnancy body mass index and gestational diabetes mellitus.

There was no statistical interaction between any of the endotoxin biomarkers and longitudinal age, so the estimates from the mixed models represent average levels of the outcomes across the 6-year follow-up period. Similarly, we did not observe evidence of effect modification by sex, race/ethnicity, or baseline weight status; therefore, no stratified analyses were conducted with respect to these variables.

Sensitivity Analyses

Adjusting for percentage of energy intake due to total fat or carbohydrate intake—both of which are known determinants of obesity-related chronic conditions including fatty liver disease (50, 51)—in lieu of total energy intake did not change estimates from Model 2 at all. Therefore, we use total energy intake given that this variable is a broad indicator of intake of energy-providing foods/nutrients that is associated with chronic disease (52). In addition, we included longitudinal BMI z-score in models where indicators of cardiometabolic risk were the outcomes to assess for an effect of the endotoxin biomarkers on cardiometabolic profile that is independent of weight status. As expected, doing so attenuated significant estimates for the relationship between LPS and the cardiometabolic markers. For example, the estimate for fasting insulin went from 1.93 (95% CI 0.08, 3.79) as reported in Table 5, Model 1, to 1.16 (95% CI −0.50, 2.82). Similarly, the estimate for leptin went from 2.28 (95% CI 0.66, 3.90) to 0.90 (95% CI −0.17, 1.98). However, the relationship of LPS with lipid profile were only slightly attenuated (ie, by <10%) in comparison to values reported in Table 5, Model 1, and remained statistically significant: 0.17 (95% CI 0.05, 0.30) for total cholesterol, 0.06 (95% CI −0.10, −0.02) for HDL, 0.16 (95% CI 0.06, 0.27) for LDL, and 0.17 (95% CI 0.10, 0.23) for triglycerides.

A comparison of endotoxin biomarkers among youth with vs without NAFLD indicated no difference in any of the biomarkers. Mean ± SD biomarker concentrations for youth with vs without NAFLD were 0.80 ± 0.39 vs 0.95 ± 0.55 EU/mL (P = 0.20) for LPS, 12.4 ± 4.2 vs 13.0 ± 6.3 µg/mL (P = 0.96) for LBP, and 64.0 ± 32.5 vs 66.0 ± 41.2 GMU/mL (P = 0.95) for EndoCab IgG.

Discussion

Summary

In this study of 393 children and adolescents, we examined associations of 3 endotoxin biomarkers (LPS, LBP, EndoCab IgG) at age 10 years with adiposity and cardiometabolic risk during 6 years of follow-up. LPS and LBP were each associated with higher total, subcutaneous, and visceral adiposity across childhood and adolescence, whereas higher EndoCab IgG was only associated with visceral adiposity. LPS predicted several indicators of cardiometabolic risk including higher fasting insulin, leptin, total cholesterol, LDL, and triglycerides and lower HDL. Adjustment for maternal prepregnancy BMI and GDM attenuated nearly all associations, except for the relationship of LPS with skinfold thickness and non-HDL lipids, suggesting that these common maternal conditions are shared common causes of metabolic endotoxemia and cardiometabolic disease risk in offspring.

Endotoxin and Adiposity

Our finding that serum LPS and LBP are associated with higher adiposity across childhood and adolescence corroborates a large literature in adults (53-55) and aligns with findings from small case-control studies comparing LPS and LBP among youth with vs without obesity (19, 24). Mechanistic experiments showing that endotoxin exposure initiates development of adiposity through alterations in the gut microbiome, inflammation, and/or activation of the immune response [see review (4)] support the mechanistic plausibility of results from observational studies.

Interestingly, the relationship between endotoxin and adiposity became attenuated after accounting for maternal prepregnancy BMI and GDM, suggesting that maternal adiposity and hyperglycemia are upstream causes of metabolic endotoxemia and obesity. There is a biological basis for this scenario given that the maternal metabolic milieu can influence offspring gut microbiome through the in utero programming (43) and/or via vertical transmission (56, 57) and has a substantial impact on offspring obesity and cardiometabolic risk (42).

Endotoxin and Cardiometabolic Risk

Higher LPS at baseline was associated with higher fasting insulin, leptin, total cholesterol, LDL, and triglycerides and lower HDL across follow-up. These findings align with those of Lassenius et al (48) who found that serum LPS predicted an increasing number of metabolic syndrome components (waist circumference, triglycerides, HDL, blood pressure, and fasting glucose) in a dose-response fashion among 1347 patients with type 1 diabetes. Similarly, a meta-analysis of 7178 individuals in 3 Finnish cohorts (FINRISK97, Paragene, and FinnTwin16) reported that serum LPS corresponded with higher levels of very low-density lipoprotein, intermediate-density lipoprotein, and LDL and lower HDL (58). While the specific mechanisms linking endotoxin exposure to an unfavorable cardiometabolic profile remain uncertain, it is possible that the previously described associations transpire from an initially stimulatory effect of endotoxin on pancreatic insulin secretion (59) that gives way to insulin resistance and other metabolic perturbations over time.

Adjusting for the child’s concurrent BMI attenuated the relationship of LPS with insulin and leptin to the null. This likely reflects the strong interrelations between adiposity, glucose-insulin homeostasis (fasting insulin), and adipocyte-derived hormones (leptin) (60). However, the associations of LPS with total cholesterol, HDL, LDL, and triglycerides were not attenuated, suggesting an effect of LPS on lipid profile across the adolescent transition that is independent of concurrent body size and adiposity.

Similarly, adjustment for maternal prepregnancy BMI and GDM—2 known determinants of offspring adiposity and perturbed glucose-insulin homeostasis (42, 61)—attenuated the effect of LPS on insulin and leptin but not for total cholesterol, triglycerides, and LDL. This suggests that the association of LPS with lipid profile is independent of the in utero metabolic milieu and that maternal BMI and GDM may primarily influence fetal programming of offspring adiposity and glucose-insulin homeostasis but not lipid profile. One biological pathway through which LPS may directly induce hyperlipidemia is through activation of the toll-like receptor 4 pathway, which induces lipolysis and stimulates the release of free fatty acids from adipocytes into circulation (62). An efflux of free fatty acids can interfere with normal lipid metabolism and result in dyslipidemia (63). Additionally, prior studies in both humans and animals have shown that LPS engagement of toll-like receptor 4 in the liver drives increases in circulating cholesterol (64, 65).

Additional Considerations

Counter to our expectations and findings from adult studies of NAFLD patients (14, 66), EndoCab IgG was not associated with any outcomes other than VAT. One explanation for the largely null associations for EndoCab IgG is that in this healthy population of youth, few individuals have had long enough exposure to endotoxin to detect differences in physiology and metabolism with respect to this biomarker. Along these lines, average EndoCab IgG in EPOCH (66 ± 50 GMU/mL) is markedly lower than values reported in adults [~100 GMU/mL in healthy adult volunteers in the United States and the United Kingdom (67) and 188 ± 137 GMU/mL among adults with NAFLD (27)]. We also acknowledge the possibility that EndoCab IgG is not a suitable marker of chronic endotoxemia given that some studies in adults that have also reported null associations between EndoCab IgG and histological features of NAFLD (68). This notion warrants additional investigations in other youth populations.

Despite the generally null results for EndoCab IgG, the singular association with VAT supports a link between endotoxin exposure and metabolically active visceral fat (69) that may transpire from close proximity of VAT depots to the site of bacterial translocation in the gastrointestinal tract (70). Indeed, a recent cross-sectional study of 244 Finnish type 1 diabetic adults observed strong associations of metabolic endotoxemia with VAT, independent of traditional metabolic risk factors including age, sex, kidney function, C-reactive protein, and insulin sensitivity (71).

We also found null associations between endotoxin biomarkers and HFF. However, as noted earlier, HFF is relatively low in this population (2.4 ± 3.2%) and only 24 youth (6% prevalence) had NAFLD. It is possible that the relationship between endotoxin and HFF are not yet detectable and/or the present study was underpowered to detect smaller effect sizes.

Strengths and Limitations

A major strength of this study is assessment of 3 endotoxin biomarkers, including 2 markers of acute exposure (LPS and LBP) and a marker of chronic exposure (EndoCab IgG), thereby capturing multiple facets of metabolic endotoxemia. We also had research grade data on multiple indicators of adiposity, including MRI-assessed VAT, SAT and hepatic fat, as well as a suite of cardiometabolic risk factors that mark shared pathogenic mechanisms relevant to development of multiple metabolic diseases. Finally, EPOCH comprises general risk youth (as opposed to the high-risk youth who already have metabolic disease in the case-control studies mentioned earlier) for whom adiposity levels and cardiometabolic biomarkers are relevant to present-day US youth, thereby enhancing generalizability of our findings.

This study also has some limitations. Although the prospective design strengthens causal inference regarding the relationships explored, we acknowledge the possibility of reverse causation given the preexisting association between LPS and LPB with weight status at baseline. Additionally, low prevalence of NAFLD in this population may have hindered our ability to detect significant associations between the endotoxin biomarkers and hepatic fat fraction. Finally, due to the relatively small sample size, we may have had limited statistical power to detect smaller but biologically relevant associations, especially in the multivariable analysis.

Conclusions

Our findings suggest that serum endotoxin may be a marker of pathophysiological processes underlying childhood obesity and adverse cardiometabolic outcomes on the rise in youth. The fact that accounting for maternal prepregnancy BMI and GDM attenuated the relationship of endotoxin biomarkers with most adiposity indicators and some cardiometabolic biomarkers (insulin, leptin, and HDL) points toward fetal overnutrition as an important upstream pathway through which the relationship between metabolic endotoxemia, obesity, and cardiometabolic risk transpires and, therefore, a potential target for preventive intervention.

Financial Support

Supported by the National Institutes of Health (NIH), National Institute of Diabetes, Digestive, and Kidney Diseases (R01 DK068001), and the National Institute of General Medical Sciences (R01GM121081-08). J.E.F. and R.J. are supported by NIH-R01DK128416. W.P. is supported by the Center for Clinical and Translational Sciences Institute KL2-TR002534. The funders had no role in the design, conduct, or reporting of this work.

Disclosure

None of the authors have any conflicts of interest.

Data Availability

Data are available upon request.