Cord Blood Metabolomics: Association With Newborn Anthropometrics and C-Peptide Across Ancestries

Abstract

Context

Newborn adiposity is associated with childhood obesity. Cord blood metabolomics is one approach that can be used to understand early-life contributors to adiposity and insulin resistance.

Objective

To determine the association of cord blood metabolites with newborn adiposity and hyperinsulinemia in a multiethnic cohort of newborns.

Design

Cross-sectional, observational study.

Setting

Hyperglycemia and Adverse Pregnancy Outcome study.

Participants

One thousand six hundred multiethnic mother–newborn pairs.

Main Outcome Measure

Cord blood C-peptide, birthweight, and newborn sum of skinfolds.

Results

Meta-analyses across four ancestry groups (Afro-Caribbean, Northern European, Thai, and Mexican American) demonstrated significant associations of cord blood metabolites with cord blood C-peptide, birthweight, and newborn sum of skinfolds. Several metabolites, including branched-chain amino acids (BCAAs), medium- and long-chain acylcarnitines, nonesterified fatty acids, and triglycerides were negatively associated with cord C-peptide but positively associated with birthweight and/or sum of skinfolds. 1,5-Anhydroglucitol, an inverse marker of recent maternal glycemia, was significantly inversely associated with birthweight and sum of skinfolds. Network analyses revealed groups of interrelated amino acid, acylcarnitine, and fatty acid metabolites associated with all three newborn outcomes.

Conclusions

Cord blood metabolites are associated with newborn size and cord blood C-peptide levels after adjustment for maternal body mass index and glucose during pregnancy. Negative associations of metabolites with C-peptide at birth were observed. 1,5-Anhydroglucitol appears to be a marker of adiposity in newborns. BCAAs were individually associated with birthweight and demonstrated possible associations with newborn adiposity in network analyses.

Cord blood metabolites are associated with newborn size and C-peptide.

The intrauterine environment has long been known to impact fetal development, newborn phenotype, and later childhood health, as classically described by the developmental origins of health and disease or DOHaD hypothesis (1–3). Developmental plasticity in utero allows the fetus to adapt to its environmental cues, modulating its phenotype at birth and chronic disease risk over time. Greater newborn adiposity, or high birthweight (BW), an anthropometric measure that is impacted by both genetics and in utero metabolic exposures, is also linked to a higher risk of adolescent and adult obesity and metabolic diseases such as type 2 diabetes mellitus (4–7). The profound impact of childhood and adolescent obesity is a public health concern, with a current prevalence of 18.5% or 13.7 million children and adolescents in the United States (8). Once present, obesity is difficult to treat and reverse; therefore, it is imperative to identify at-risk children early in life and to implement preventive interventions before obesity and its comorbidities develop.

Metabolomics, the systematic study of metabolites present in blood and other tissues, offers an approach to identifying and understanding metabolic signatures involved in early-life risk of obesity and related metabolic disorders, such as type 2 diabetes. The metabolome is impacted by several factors, including genetics, nutrition, and the environment; therefore, using a metabolomics approach integrates multiple factors affecting disease development and allows for a more comprehensive understanding of a pathophysiologic state. Metabolite signatures are emerging as biomarkers of disease states, and recent literature has shown that an adverse metabolic state contributes to the development of mitochondrial “metabolic gridlock,” best described as a condition in which mitochondria are overburdened with nutritional cues, leading to ineffective fuel switching between fed and fasted states. This leads to metabolic inflexibility and mitochondrial dysfunction, which contribute to an individual’s inability to maintain energy homeostasis and risk for subsequent metabolic diseases (9, 10). Thus, the use of metabolic signatures as biomarkers of metabolic disease offers an entry point for developing clinical interventions to combat metabolic and mitochondrial inflexibility early in life when developmental plasticity is present.

Studies of metabolomics as they relate to obesity and insulin resistance in newborns and young children are limited. In adolescents and adults, branched-chain amino acids (BCAAs), two of the four aromatic amino acids (AAAs; tyrosine and phenylalanine), short-chain acylcarnitines (AC C2, AC C3, AC C5, AC C6, and AC C8, with the short-hand C notation indicating the number of carbons in the acyl moiety of the conjugate), and the carnitine ester of the “ketone body,” 3-hydroxybutyrate (AC C4-OH), have all been associated with obesity, insulin resistance, higher glycated hemoglobin, and/or postprandial glucose levels (11–14). Although some studies have demonstrated that high BCAA and AAA levels can be present prior to the onset of insulin resistance or type 2 diabetes (15, 16), the timing of these changes is not known. Additional metabolites, such as phospholipids and fatty acids, have also been implicated in adolescent obesity (17).

A few studies have examined the association between cord blood metabolite levels and newborn anthropometrics (18–21); however, these studies were small, used indirect measures of adiposity, or did not account for known confounders of newborn size, such as maternal body mass index (BMI) or glucose. In the current study of 1600 well-characterized, multiethnic maternal–newborn pairs, we aimed to evaluate the association of targeted and nontargeted cord blood metabolites with measures of newborn adiposity and hyperinsulinemia to determine whether a unique cord blood metabolic signature might emerge as a proxy for the metabolic environment of the fetus and provide mechanistic insight into fetal fat deposition, insulin sensitivity, and future obesity and metabolic disease risk.

Materials and Methods

Data and sample collection

Maternal blood samples were obtained during a 75-g oral glucose tolerance test (OGTT) between 24 and 32 weeks gestation as part of the multicenter Hyperglycemia and Adverse Pregnancy Outcome (HAPO) study (22). Glucose levels at fasting, and after 1 and 2 hours, were measured as described (22). Details of cord blood sample collection, processing, and standardization across study sites were previously published (23). In brief, cord blood was obtained by needle aspiration of the umbilical vein or free drainage within 5 minutes of delivery and before delivery of the placenta, and cord blood C-peptide was later measured as described (22). Cord blood plasma samples were stored at −70°C prior to metabolomics assays. Metabolic profiling was performed on cord blood plasma samples for offspring of 400 mothers from each of four ancestry groups: Afro-Caribbean, Mexican American, Northern European, and Thai.

Trained personnel obtained maternal anthropometric measures using a standardized protocol at the time of the OGTT. Gestational age was calculated based on the date of the last menstrual period. When the date was uncertain, gestational age was based on the expected date of delivery as estimated by ultrasonography performed between 6 and 24 weeks of gestation. Demographic data were ascertained via questionnaire. Newborn anthropometric measurements, including BW and sum of skinfolds (SSF) at three sites (flank, subscapular, and triceps), were measured within 72 hours of birth using calibrated equipment and standardized methods across field centers. The protocol was approved by the Institutional Review Board at all participating sites, and each mother provided written informed consent.

Conventional metabolites and targeted metabolomics assays

Sixty-five conventional and targeted metabolites were assayed as previously described (24). Briefly, conventional metabolite levels [lactate, triglycerides, 3-hydroxybutyrate, glycerol, nonesterified fatty acids (NEFAs)] were measured on a Beckman Coulter Unicel DxC 600 clinical analyzer (Beckman Coulter, Brea, CA). Targeted metabolomics assays for acylcarnitines and amino acids were performed by tandem mass spectrometry with addition of known quantities of stable isotope-labeled internal standards on an Acquity triple-quadruple system (Waters Corporation, Milford, MA).

Nontargeted analyses

Nontargeted gas chromatography (GC)/mass spectrometry (MS) assays were performed to analyze the full range of metabolites in plasma, as previously described (24). Methanol, the extraction solvent, was spiked with a retention-time-locking internal standard of perdeuterated myristic acid. Extracts were prepared for GC/MS by methoximation and trimethylsilylation (25, 26). Plasma from Northern European and Thai ancestries were run on a 6890N GC-5975B inert mass spectrometer (Agilent Technologies, Santa Clara, CA). Afro-Caribbean and Mexican American samples were run on a 7890B GC-5977B inert mass spectrometer (Agilent Technologies).

Samples for GC/MS were run in batches of equal size during 50 days for each ancestry group and balanced by field center, maternal phenotypes, and newborn outcomes. Quality control pools, constructed using small volumes from all cord blood samples and prepared the same as above were injected as the first, middle, and last samples of each daily GC/MS batch. Data from these quality control samples were used to control technical variability attributable to batch and run order, as applied using the metabomxtr R package (24, 27). Peaks were deconvoluted with AMDIS freeware (28) and annotated against a retention-time-locking spectral library built upon that of Fiehn and colleagues (26) with additions from our laboratory. Manual curation included grouping metabolite peaks across samples according to similar GC retention times and mass spectra (25). Detected peak areas were log₂ transformed for analysis. In total, 73 GC/MS metabolites that had not been assayed using targeted approaches were used for subsequent data analysis.

Statistical analysis

Per-metabolite analysis

Acylcarnitine and 3-hydroxybutyrate levels were log transformed to improve normality. Log₂-transformed peak areas were used to quantify abundance for nontargeted metabolites. Outlying metabolite values, defined as values >5 SDs from the mean, were excluded prior to statistical analysis.

Associations among newborn phenotypes and metabolites were identified using linear regression within ancestry groups. Analyses treated newborn metabolites as predictors and newborn phenotypes as outcomes, and they were adjusted for field center, maternal age, and mean arterial pressure at HAPO study visit, neonatal sex, sample storage time, mode of delivery (vaginal delivery, elective cesarean section, emergent cesarean section), and gestational age at delivery (model 1). Two additional models adjusted for the model 1 covariates plus maternal BMI and maternal fasting glucose at OGTT (model 2) and maternal BMI, fasting glucose at OGTT, and cord blood C-peptide (model 3) were used.. Maternal fasting glucose was chosen as the glucose covariate in place of 1-hour or 2-hour glucose levels because women were fasting at the time of labor.

Analyses were limited to observed metabolite values. Fewer than 10% of data were missing for analyzed targeted metabolites. Nontargeted metabolites were included in analyses when the metabolite was detectable in at least 50% of samples. As is common for GC/MS data, values were “missing” either due to absence from the sample or presence below detectability. Rather than impute an arbitrary constant or use multiple imputation models that could errantly allow values reflecting high compound abundance, we chose to limit analyses only to observed values and interpret results as pertinent only to metabolite levels in the detectable range.

Meta-analysis

β Values from per-metabolite analyses were combined across ancestry groups using random effects meta-analysis with inverse variance weights and restricted maximum likelihood estimation (29) using the metafor R package (http://www.metafor-project.org). Heterogeneity of effects among ancestry groups was assessed descriptively with I² statistics and formally tested via Cochran Q tests. False discovery rate (FDR) correction was applied to meta-analysis and Cochran Q test P values, and FDR-adjusted P values <0.05 were considered statistically significant. Heat maps with hierarchical clustering were used for visual display of per-metabolite results, with the color scale determined by the association test statistic for each phenotype. All statistical analyses were conducted using R (version 3.4.3, http://www.r-project.org).

Network analyses

Graphical lasso techniques were used to identify networks including metabolites associated with phenotype (30). Each node represents an individual metabolite, and edges depict dependence of metabolite pairs conditional on all other metabolites. First, we identified the subset of metabolites associated with a phenotype at nominal P < 0.05. Next, we identified all other metabolites demonstrating partial correlation with magnitude >0.25 with at least one significant metabolite, after adjusting for model parameters as described previously. We calculated residuals from linear models for these metabolites, including all model covariates, and used these residuals as inputs into the graphical lasso analysis. This approach was used to accommodate adjustment for potential confounders when estimating networks. Graphical lasso uses a coordinate descent algorithm and a penalty term to provide a sparse estimate of the inverse covariance matrix for a given set of input features. Zero covariance indicates independence for pairs of features conditional on all others, whereas nonzero covariance indicates dependence, conditional on all other features. In network parlance, when nodes represent input features (i.e., metabolites), nonzero covariance estimates correspond to edges between pairs of nodes that represent their association with each other, independent of all other graph features. The penalty term trims likely false-positive edges by gauging the strength of estimated edges.

Graphical lasso was applied within each ancestry group and in meta-analysis, using meta-analytic estimates of partial correlations based on a Fisher z transformation. To further elucidate network structure, spin-glass clusters (31) were estimated on graphical lasso networks to identify communities of nodes that are more tightly connected to each other than the other nodes in the network.

Results

Descriptive characteristics of participating newborns are shown in Table 1. All infants were born full term, and newborn sex was roughly equally distributed. Most newborns were born via vaginal delivery. Thai newborns had the lowest BW, whereas Thai and Afro-Caribbean newborns had the lowest SSF. Cord C-peptide was similar across ethnic groups. Afro-Caribbean mothers were the youngest and all mothers were nonsmokers. Maternal glucose levels and BMI spanned the range observed in the HAPO study. Mexican American mothers had the highest fasting glucose levels, and Thai mothers had the highest 1-hour glucose levels. Thai mothers were the leanest and Mexican American mothers were heaviest, according to BMI.

Table 1.

Maternal and Newborn Characteristics

	Afro-Caribbean (n = 400)	Northern European (n = 400)	Mexican American (n = 400)	Thai (n = 400)
Newborn characteristics, mean (SD)
Birth weight, g	3251 (440)	3668 (485)	3555 (444)	3148 (373)
Length, cm	49.5 (2.4)	51.0 (2.2)	50.6 (1.8)	49.5 (1.5)
Sum of skinfolds, mm	11.6 (1.8)	12.8 (2.7)	14.2 (3.1)	11.6 (2.3)
Cord C-peptide	1 (0.6)	1.1 (0.6)	1.1 (0.5)	1 (0.5)
Sex, N (%)
Male	208 (52)	209 (52)	180 (45)	201 (50)
Female	192 (48)	191 (48)	220 (55)	199 (50)
Mode of delivery
Vaginal	365 (91)	294 (74)	319 (80)	287 (72)
Planned cesarean section	22 (6)	64 (16)	58 (15)	70 (16)
Emergency cesarean section	13 (3)	42 (11)	23 (6)	43 (11)
Maternal characteristics, mean (SD)
Age	25.7 (5.6)	29.4 (5.1)	29.0 (5.3)	28.0 (5.8)
Height, cm	164.0 (6.9)	164.5 (6.3)	159.7 (5.8)	154.0 (5.4)
BMI, kg/m2	28.4 (6.2)	28.0 (5.1)	29.7 (5.3)	26.1 (4.2)
Gestational age at delivery, wk	39.8 (1.2)	40.2 (1.1)	39.7 (1.1)	39.4 (1.3)
Fasting plasma glucose, mg/dL	80.6 (7)	82.1 (6.1)	83.7 (7.2)	80.1 (6.5)
One-hour plasma glucose, mg/dL	123.2 (29)	131.7 (27.2)	135.6 (33.9)	148 (30.9)
Two-hour plasma glucose, mg/dL	109.4 (22.5)	111 (20.3)	110.5 (23)	118.5 (23.9)
Mean arterial pressure, mm Hg	78.9 (7.5)	83.1 (6.9)	82.6 (7.3)	79.5 (7.1)
Parity, N (%)
No prior pregnancy	189 (47)	203 (51)	85 (21)	211 (53)
Prior pregnancy	211 (53)	197 (49)	315 (79)	189 (47)
Smoking status
Nonsmoker	400 (100)	400 (100)	400 (100)	400 (100)

Individual metabolites associated with newborn outcomes

In the meta-analysis (Fig. 1; Table 2) and ancestry-specific analyses [Fig. 2; (32)], we observed that cord metabolites were largely negatively associated with cord C-peptide levels and, in contrast, positively associated with newborn BW and SSF.

Figure 1.

Heat map displaying positive (red) and negative (blue) associations of cord blood metabolites with newborn outcomes across ancestries in the meta-analysis, with hierarchical clustering of metabolites according to test statistics reflecting strength of association. Analyses were adjusted for field center, mean arterial pressure, maternal age, newborn sex, sample storage time, mode of delivery, and gestational age at delivery (model 1). Model 2 includes model 1 covariates, maternal BMI, and maternal fasting glucose. Model 3 includes model 2 covariates and cord blood C-peptide.

Table 2.

Meta-Analysis: Associations of Cord Blood Metabolites With Newborn Outcomes

	C-Peptide				BW				SSF
	Model 1		Model 2		Model 1		Model 3		Model 1		Model 3
	β	P	β	P	β	P	β	P	β	P	β	P
Amino acids and related metabolites
Arginine	−5.00 × 10−4	0.48	−4.00 × 10−4	0.52	1.87	0.014	1.802	0.011	0.0033	0.44	0.0038	0.35
Asparagine/aspartic acid	−0.0053	0.029	−0.0057	0.013	−0.63	0.42	0.32	0.443	−0.0066	0.54	−0.0031	0.56
Citrulline	−0.0043	0.34	−0.0022	0.47	6.62	0.092	8.02	0.039	−0.018	0.44	−0.0077	0.55
Glutamine/glutamic acid	−0.0015	0.0030	−0.0014	0.013	0.75	0.021	1.09	8.93 × 10 −4	0.0012	0.47	0.0028	0.15
Glycine	−0.0021	3.32 × 10−5	−0.0020	5.78 × 10 −1	1.2	0.021	1.56	0.0050	0.0045	0.094	0.0064	0.010
Histidine	0.0010	0.28	0.0010	0.32	4.46	1.11 × 10 −9	4.19	2.70 × 10 −9	0.019	6.80 × 10 −6	0.018	2.14 × 10 −5
Leucine/isoleucine	−0.0019	0.0090	−0.0024	0.0050	1.29	0.0020	1.17	0.003	0.0051	0.12	0.0042	0.10
AC C4/Ci4	0.059	0.18	0.033	0.36	−44.25	0.16	−81.4	0.014	−0.18	0.473	−0.36	0.16
AC C5	0.068	0.024	0.052	0.069	−5.67	0.45	−27.19	0.134	−0.062	0.58	−0.16	0.20
Methionine	−0.0037	0.26	−0.0036	0.22	5.15	0.060	5.00	0.034	0.017	0.42	0.016	0.25
Ornithine	−0.0033	0.020	−0.0032	0.043	1.97	0.023	2.19	0.011	0.012	0.024	0.013	0.011
Proline	2.00 × 10−4	0.48	1.00 × 10−4	0.54	1.92	1.23 × 10 −4	1.63	8.93 × 10 −4	0.0082	0.009	0.0068	0.022
Serine	−0.0043	9.17 × 10 −7	−0.0044	1.99 × 10 −7	2.52	6.58 × 10 −4	2.96	0.001	0.014	7.57 × 10 −4	0.016	2.14 × 10 −5
Tyrosine	−0.0036	0.009	−0.0036	0.011	1.52	0.11	1.9	0.045	0.0023	0.61	0.0047	0.37
Valine	−4.00 × 10−4	0.28	−7.00 × 10−4	0.085	0.81	0.014	0.58	0.051	0.0033	0.13	0.0021	0.27
Lysine	−0.0151	0.76	−0.019	0.73	26.55	0.067	23.88	0.036	0.11	0.25	0.095	0.27
Threonine	−0.0059	0.81	−0.0087	0.80	51.88	2.13 × 10 −5	50.44	1.48 × 10 −5	0.19	0.014	0.19	0.033
3-Indolelactic acid	−0.016	0.77	−0.014	0.80	−35.6	0.0050	−33.79	0.006	−0.18	0.021	−0.17	0.039
NM/2AA/NE	−0.0030	0.91	−0.0092	0.86	25.51	0.097	23.98	0.047	0.045	0.65	0.032	0.61
Aminomalonic acid	−0.0065	0.79	−0.0044	0.88	22.78	0.088	26.07	0.033	0.065	0.61	0.086	0.41
Acylcarnitines and related metabolites
AC C8:1	0.018	0.40	0.0047	0.53	58.47	0.021	40.27	0.13	0.24	0.21	0.13	0.44
AC C8:1-DC	0.076	0.052	0.076	0.050	95.85	0.001	84.19	0.0030	0.41	0.024	0.34	0.062
AC C8:1-OH/C6:1-DC	0.069	0.062	0.061	0.094	72.12	0.014	53.34	0.045	0.20	0.32	0.11	0.43
AC C6-DC/C8-OH	0.0069	0.49	0.0143	0.46	63.38	0.0040	67.76	0.0010	0.10	0.43	0.13	0.27
AC C10:2	0.046	0.23	0.049	0.13	53.54	0.029	44.61	0.057	0.20	0.39	0.17	0.37
AC C10:3	−0.0015	0.57	−0.0047	0.53	45.06	0.021	40.7	0.042	0.18	0.39	0.15	0.38
AC C10-OH/C8-DC	0.018	0.46	0.022	0.46	88.03	1.20 × 10 −4	87.7	5.60 × 10 −5	0.32	0.019	0.33	0.011
AC C12	−0.15	0.013	−0.14	0.025	−15.99	0.41	11.51	0.41	−0.021	0.72	0.12	0.42
AC C12-OH/C10-DC	−0.022	0.43	−0.02	0.47	41.15	0.024	48.64	0.011	0.17	0.17	0.21	0.072
AC C14	−0.12	0.0070	−0.12	0.006	11.27	0.42	27.33	0.21	0.18	0.39	0.26	0.14
AC C14:1	−0.14	0.027	−0.14	0.026	−31.71	0.24	−12.00	0.35	−0.062	0.6	0.043	0.55
AC C14:2	−0.14	0.040	−0.14	0.043	−25.11	0.38	−5.39	0.45	−0.010	0.73	0.059	0.55
AC C16:1	−0.17	1.03 × 10 −5	−0.17	5.78 × 10 −6	−8.8	0.46	9.21	0.44	−0.0023	0.74	0.099	0.49
AC C18:1	−0.11	0.043	−0.11	0.037	43.99	0.19	52.77	0.077	0.25	0.43	0.28	0.18
AC C18:2	−0.11	0.021	−0.11	0.017	−32.93	0.24	−20.94	0.32	−0.16	0.43	−0.10	0.46
Fatty acids and related metabolites
NEFAs	−1.22	6.08 × 10 −4	−1.13	0.0020	−173.64	0.19	94.21	0.32	−1.27	0.39	0.072	0.65
Palmitoleic acid	−0.091	1.26 × 10 −5	−0.085	1.19 × 10 −5	−7.18	0.51	10.55	0.36	−0.042	0.82	0.054	0.60
Methyl palmitate	0.0079	0.79	0.0068	0.85	26.09	0.088	27.96	0.036	0.042	0.72	0.055	0.56
Methyl stearate	−2.00 × 10−4	0.93	0.0012	0.92	24.96	0.10	25.24	0.036	0.098	0.38	0.11	0.24
Methyl eicosatrienoate	0.015	0.77	0.014	0.80	28.97	0.088	29.32	0.036	0.075	0.50	0.075	0.45
Lauric acid	−0.052	0.13	−0.041	0.40	14.69	0.26	34.31	0.016	−0.019	0.87	0.096	0.38
Lipids and related metabolites
AC C4-OH	−0.045	0.037	−0.052	0.034	59.18	1.20 × 10 −4	61.73	2.56 × 10 −5	0.24	0.009	0.26	0.002
3-Hydroxybutyrate	−0.062	2.71 × 10 −4	−0.070	2.70 × 10 −4	55.36	1.99 × 10 −5	64.95	2.56 × 10 −5	0.23	8.01 × 10 −4	0.27	6.60 × 10 −5
Triglycerides	−0.0054	0.0030	−0.0053	5.43 × 10 −4	−4.29	1.35 × 10 −4	−3.63	0.014	−0.026	0.024	−0.023	0.072
Cholesterol	−0.0070	0.79	−8.00 × 10−4	0.92	26.23	0.097	31.63	0.033	0.032	0.85	0.059	0.60
Glycerol 1-phosphate	0.011	0.77	0.011	0.80	37.52	0.0050	37.21	0.0040	0.11	0.25	0.10	0.24
Carbohydrates and related metabolites
Aldopentoses	−0.052	0.003	−0.049	0.007	−2.45	0.64	5.01	0.52	2 × 10−4	0.94	0.039	0.60
Fructose or similar ketohexose	0.027	0.66	0.024	0.64	−23.44	0.26	−27.56	0.2	−0.15	0.25	−0.18	0.042
1,5-Anhydroglucitol	−0.041	0.66	−0.034	0.77	−33.55	0.088	−25.55	0.036	−0.19	0.014	−0.15	0.042
Purine/pyrimidines and related metabolites
Uric acid	−0.0033	0.89	−0.0022	0.92	25.67	0.076	23.20	0.045	0.079	0.47	0.076	0.38
Hypoxanthine	−0.020	0.66	−0.014	0.80	35.33	0.067	37.82	0.0060	0.15	0.10	0.17	0.042
Pseudouridine	0.022	0.73	0.018	0.80	38.04	0.005	31.79	0.016	0.20	0.014	0.16	0.042
Glycolysis/TCA cycle intermediates and related metabolites
Lactate	−0.040	8.70 × 10 −5	−0.038	2.06 × 10 −4	−9.21	0.24	−1.93	0.44	−0.016	0.68	0.024	0.54
Glyceric acid	−0.048	0.18	−0.041	0.34	16.67	0.25	28.72	0.033	−0.014	0.89	0.040	0.60
Organic acids and related metabolites
CMPF	0.0044	0.89	0.013	0.80	−28.02	0.094	−26.25	0.033	−0.19	0.25	−0.16	0.042
Other metabolites
Creatinine	−0.023	0.58	−0.019	0.77	24.24	0.14	28.31	0.033	0.11	0.47	0.14	0.14

Boldface indicates significant P values.

Abbreviations: NM/2AA/NE, N-methylalanine/2-aminobutanoic acid/N-ethylglycine; CMPF, 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid; TCA, tricarboxylic acid cycle.

Figure 2.

Heat map displaying ancestry-specific positive (red) and negative (blue) associations of cord blood metabolites with newborn outcomes in the fully adjusted models, with hierarchical clustering of metabolites according to test statistics reflecting strength of association. Analyses were adjusted for field center, mean arterial pressure, maternal age, newborn sex, sample storage time, mode of delivery, gestational age at delivery, maternal BMI, and maternal fasting glucose (model 2). Model 3 includes model 2 covariates and cord blood C-peptide. AC, Afro-Caribbean; EU, Northern European; MA, Mexican American; TH, Thai.

Cord C-peptide levels

Tyrosine, leucine/isoleucine, glycine, the ketone body 3-hydroxybutryate and its carnitine adduct (AC C4-OH), NEFAs, triglycerides, and several medium- and long-chain acylcarnitines were all negatively associated with C-peptide, whereas the acylcarnitine AC C5, a BCAA metabolite, was positively associated with cord blood C-peptide (Table 2). Adjusting for maternal BMI and fasting maternal glucose at OGTT (model 2) did not alter the direction of the findings and only attenuated the association with AC C5.

When evaluating metabolite associations in individual ancestries, the overall inverse association with cord C-peptide persisted, and additional associations were also evident [Fig. 2; (32)]. In Afro-Caribbean newborns, 1,5-anhydroglucitol and 3-hydroxybutyrate were inversely associated with C-peptide at baseline (model 1), and 3-hydroxybutyrate was also inversely associated in model 2. Several medium- and long-chain acylcarnitines were negatively associated with C-peptide among Afro-Caribbean and Northern European newborns in all models. In Afro-Caribbean, Thai, and Northern European newborns, palmitoleic acid was inversely associated with C-peptide in all models.

Birthweight

The BCAAs (leucine/isoleucine and valine), glutamate/glutamine, and several medium-chain acylcarnitines were among the metabolites positively associated with BW (model 1, Table 2). Triglycerides and 3-indoleacetic acid, a metabolite of the AAA tryptophan, were negatively associated.

Adjustment for maternal BMI, maternal fasting glucose at OGTT, and cord blood C-peptide (model 3) attenuated the association with valine, whereas inverse associations emerged between 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid (CMPF), 1,5-anhydroglucitol, and the BCAA metabolite, AC C4/Ci4, and BW. Cholesterol, fatty acid metabolites, and additional amino acids, including tyrosine, were also positively associated with BW in model 3.

In Afro-Caribbean and Northern European ancestry newborns, 3-hydroxybutryate was associated with BW both at baseline (model 1) and in the adjusted model (model 3). The ketone metabolite AC C4-OH was positively associated with BW in Afro-Caribbean neonates only. Histidine was positively associated with BW across all four ancestry groups at baseline (model 1) and in Afro-Caribbeans, Northern Europeans, and Thai newborns after adjustment for maternal BMI, maternal fasting glucose, and cord C-peptide (model 3). Serine and glycine were positively associated with BW in the adjusted model for Afro-Caribbean and Thai newborns (32).

Sum of skinfolds

As with BW, metabolite associations with SSF were also predominantly positive (Table 2). At baseline (model 1), the ketone body, 3-hydroxybutyrate, and its metabolite AC C4-OH, medium-chain ACs, the amino acids serine, ornithine, proline, and threonine, and pseudouridine, a nucleic acid derivative, were all positively associated with SSF. 1,5-Anhydroglucitol, triglycerides, and the tryptophan metabolite, 3-indoleacetic acid, were negatively associated. After adjustment for maternal fasting glucose, maternal BMI, and cord C-peptide (model 3), negative associations emerged with CMPF and fructose, and positive associations emerged with glycine and hypoxanthine, and the association with triglycerides was attenuated.

3-Hydroxybutyrate was positively associated with SSF in model 3 in Afro-Caribbean and Northern European newborns (32). Afro-Caribbean neonates also had positive associations between histidine, glycine, AC C10:3, and AC C8:1 and SSF, both at baseline (model 1) and after adjustment (model 3). The BCAA metabolite, AC C4/Ci4, was negatively associated with SSF in Afro-Caribbean newborns in model 3. There were no ancestry-specific associations in the Mexican American group. Thai newborns demonstrated negative associations of triglycerides and AC C20-OH/C18-DC, both at baseline (model 1) and in the adjusted model (model 3).

Heat maps

Subsequent analyses examined differences in metabolite–phenotype associations across different models in the meta-analysis and in ancestry specific analyses (Figs. 1 and 2) and identified groups of metabolites through hierarchical clustering that were similarly related with newborn outcomes at FDR adjusted P < 0.05. Broadly, cord blood metabolites were negatively associated with cord C-peptide and positively associated with BW and SSF across ancestries (Fig. 1). One group of metabolites that included leucine/isoleucine, glutamate/glutamine, and tyrosine was negatively associated with C-peptide and positively associated with BW and SSF. Another group of metabolites comprised largely of medium- and long-chain acylcarnitines (AC C12, AC C14:1, AC C14:2, AC C16:1, AC C18:2) was similarly negatively associated with C-peptide. Interestingly, the longer-chain acylcarnitines were negatively associated with BW and SSF in model 1, but the association was attenuated or trended in the positive direction after adjustment for maternal BMI and maternal fasting glucose. A third group of metabolites that included the tryptophan metabolite 3-indoleacetic acid, 1,5-anhydroglucitol, triglycerides, and CMPF, was similarly negatively associated with all three outcomes, although the association was stronger for BW and SSF.

In the fully adjusted models for ancestry-specific analyses (Fig. 2), several medium- and long-chain acylcarnitines were negatively associated with C-peptide in Northern Europeans, and to a lesser extent Afro-Caribbeans and Thais, but they were positively associated with C-peptide in Mexican Americans. Several metabolites including cholesterol, as well as AC C8:1, AC C10:1, AC C10:2, and AC C10:3, were positively associated with BW in Afro-Caribbeans but were negatively or not associated with the same outcomes in other ancestries. Additional metabolites, including 3-hydroxybutyrate, AC C4-OH, NEFAs, and various medium- and long-chain acylcarnitines, were positively associated with SSF in Afro-Caribbeans but were negatively or not associated with the same outcomes in other ancestries. These results highlight that metabolomics profiles vary among ancestry groups and are likely impacted by nutritional, genetic, and environmental cues.

Network and pathway analyses

Recognizing that dependencies exist among metabolites, we conducted network analyses to better characterize joint associations of cord blood metabolites with newborn outcomes. Network analyses allowed for the visualization of individual, nominally significant metabolite–phenotype associations in the context of a larger metabolite environment. The metabolite network associated with SSF (Fig. 3) included medium- and long-chain acylcarnitines, AAAs, BCAAs, fatty acid metabolites, and acetylcarnitine in the fully adjusted model. Metabolite networks associated with cord blood BW and C-peptide (33, 34) also included groups of medium- and long-chain acylcarnitines, AAAs, fatty acid metabolites, and acetylcarnitine, with the addition of BCAA metabolites as well in comparison with the network associated with SSF.

Figure 3.

Subnetworks including cord blood metabolites associated with newborn sum of skinfolds. The subnetwork including cord blood metabolites associated with newborn sum of skinfolds is shown in model 3. Covariates for the model include field center, mean arterial pressure, maternal age, neonatal sex, sample storage time, mode of delivery, gestational age at delivery, maternal BMI, maternal fasting glucose at OGTT, and cord blood C-peptide. Blue shading denotes spin-glass communities within the estimated networks. Edges represent dependence among metabolite pairs conditional on all other metabolites in the network according to graphical lasso. Solid edges represent dependencies for metabolites in the same spin-glass cluster, and red dashed lines represent dependencies for metabolites in different spin-glass clusters. Large nodes represent metabolites that are individually significant with newborn sum of skinfolds, whereas small nodes are correlated with an individually significant metabolite. Nodes are colored by metabolite class (AC, acylcarnitine; AA, amino acid; CHO, carbohydrate; FA, fatty acid; GC/TCA, glycolysis/tricarboxylic acid cycle; MISC, miscellaneous; OA, organic acid; PUR/PYR, purine or pyrimidine). Abbreviated metabolites are as follows: 3-OHB, 3-hydroxybutyrate; Asn/Asx, asparagine/aspartic acid; Glu/Glx, glutamine/glutamic acid; Leu/Ile, leucine/isoleucine; NM/2AA/NE, N-methylamine/2-aminobutanoic acid/N-ethylglycine; OHPro, hydroxyproline.

Pathway analyses were conducted to determine whether metabolites with common pathway memberships were associated with newborn outcomes (35). Metabolites in the ammonia recycling pathway were consistently associated with cord C-peptide whereas metabolites in the glycine and serine metabolism pathways were associated with BW. No pathways were associated with SSF. For the significant pathways, a limited number of members of each pathway were represented in our data set and available for analysis.

Adjustment for maternal gestational diabetes mellitus status

We performed secondary analyses to evaluate whether metabolite–phenotype associations differed in infants born to mothers without gestational diabetes mellitus (GDM, n = 1277) compared with the cohort as a whole (n = 1600) (36). GDM was retrospectively diagnosed using International Association of the Diabetes and Pregnancy Study Group criteria (37). Some metabolites that were significantly associated with newborn outcomes in the whole cohort were not significant when studying infants of non-GDM mothers alone, suggesting that these associations may be influenced by the complex metabolic state present in GDM. These metabolites include 1,5-anhydroglucitol, fatty acid metabolites such as methyl stearate and palmitoleic acid, lipid metabolites such as AC C4-OH, and amino acids and related metabolites such as tyrosine and 3-indoleacetic acid. Insufficient power precluded performing analyses that included only women with GDM.

Association of fatty acid desaturation index to newborn outcomes

Stearoyl–coenzyme A (CoA) desaturase-1 is upregulated in insulin resistance and obesity (38). Its activity leads to increased conversion of saturated fatty acids to monounsaturated fatty acids, which can then be incorporated into triglycerides. The palmitoleic acid/palmitic acid and oleic acid/stearic acid ratios reflect stearoyl-CoA desaturase-1 activity and have been associated with BW, waist circumference, and/or glucose levels in newborns (38). We evaluated associations of these ratios with newborn outcomes (39). In the meta-analysis, the palmitoleate/palmitate and oleic acid/stearic acid ratios were negatively associated with C-peptide, suggestive of greater insulin sensitivity and consistent with the other inverse associations of metabolites with C-peptide that we report in this study. There was no association of these ratios with BW or SSF in the meta-analysis. Ancestry-specific analysis confirmed the negative association between the palmitoleate/palmitate ratio and C-peptide in Afro-Caribbeans, Northern Europeans, and Thais as well as the association between the oleic acid/stearic acid ratio and C-peptide in Afro-Caribbeans and Thais only. There were no associations of either ratio with BW. The palmitoleate/palmitate ratio was positively associated with SSF in Afro-Caribbeans only.

Discussion

This study demonstrates associations of cord blood metabolites with newborn measures of adiposity and hyperinsulinemia across four ancestry groups in the HAPO cohort. We report novel associations of cord metabolites with newborn outcomes while also demonstrating that some of the known associations of metabolites with body composition in adolescents and adults might already be present at birth. The current study expands on previous studies in the HAPO cohort (40) by including nontargeted metabolites in the analysis, revealing additional novel metabolite–phenotype associations, as well as network analyses that demonstrate the interrelationships among metabolites with the different newborn outcomes.

BCAAs, AAAs, and their metabolites are associated with insulin resistance and type 2 diabetes in adults (12, 41) and obesity and insulin resistance in adolescents (42, 43); however, their temporal relationship to the adiposity, insulin resistance, and/or type 2 diabetes phenotype is not clear. Some studies in adolescents and adults show that higher BCAA and AAA levels are associated with future insulin resistance and diabetes in adolescents and adults (42, 44–46), although more recent studies suggest that higher levels occur subsequent to the development of adiposity and insulin resistance but prior to the development of type 2 diabetes mellitus (15, 47). Although we identified a positive association between BCAAs and their metabolic byproduct, glutamate/glutamine, with BW, the tryptophan metabolite, 3-indoleacetic acid, was negatively associated with both BW and SSF. BCAAs were associated with SSF only in the network analyses. Thus, our results suggest that, although BCAAs appear to be related to lean mass at birth, their relationship to fat mass is less clear. We also found that both BCAAs and AAAs were paradoxically negatively associated with cord C-peptide in newborns. Cord C-peptide is solely a measure of fetal insulin secretion. The lack of association may reflect incomplete development of fuel-stimulated insulin secretion in the fetus, which matures during postnatal life. Although BCAAs, AAAs, and their metabolites relate to birth size and insulin sensitivity, the relationships are not in consistent directions as seen in later childhood and adulthood.

Additional amino acid relationships of interest emerged in our study. Cord blood glycine was associated with both BW and SSF, and cord blood serine was associated with SSF. Fetal glycine is largely derived from fetal conversion of serine (48). Serine plays a role in phospholipid and sphingolipid biosynthesis, and glycine is critical for protein synthesis, especially collagen (49). Although serine and glycine have not been studied extensively in relationship to adiposity in newborns, a recent study of Hispanic children with obesity reported a positive association of glycine and serine with BMI (43), consistent with our reported findings. However, our finding of a negative association of glycine with C-peptide at birth is consistent with the reported inverse relationship of glycine with insulin resistance and type 2 diabetes in adults (50).

The “ketone body” 3-hydroxybutyrate crosses the placenta and serves as a substrate for fetal lipogenesis (51). 3-Hydroxybutryate and its metabolite, AC C4-OH, were both associated with newborn SSF independent of maternal BMI and glucose, and with cord C-peptide; this finding is consistent with a possible contributory role of these metabolites to newborn fat mass. Higher levels of 3-hydroxybutyrate could also reflect greater fatty acid oxidation or reduced insulin-mediated suppression of ketogenesis.

Higher levels of medium- and long-chain acylcarnitines have been described in adults with obesity and/or type 2 diabetes (11, 52), where their accumulation in tissues may interfere with cellular insulin signaling (52). Positive associations of cord blood medium- and long-chain acylcarnitines with BW, and, to a lesser extent, SSF in this study are consistent with previously described relationships in adults, although contrary to a negative association between medium-chain acylcarnitines and BW previously reported in newborns (21). The negative association of this metabolite class with cord C-peptide is consistent with a prior report in newborns but is opposite the relationship in adults, but the same result has previously been reported in a different cohort of newborns (53).

1,5-Anhydroglucitol is a dietary glucose analog that competes with glucose for renal reabsorption via SGLT2 and serves as a short-term marker of glucose control (54). In hyperglycemic states, serum levels of 1,5-anhydroglucitol decline secondary to increased renal excretion. Negative associations of cord blood 1,5-anhydroglucitol with BW and SSF, even after adjustment for maternal BMI, glucose, and cord C-peptide in this study, parallel the linear association of maternal glucose levels and newborn SSF across the spectrum of maternal glucose levels (55). An association of maternal 1,5-anhydroglucitol during pregnancy with BW in mothers with type 1 diabetes, gestational diabetes, and HAPO mothers has previously been reported (24, 56, 57), but, to our knowledge, the current study is the first to demonstrate an association of cord blood 1,5-anhydroglucitol with SSF, a direct measure of adiposity, in a cohort of newborns born to mothers without preexisting diabetes even after adjustment of maternal glucose levels during pregnancy. Whether 1,5-anhydrogluctol crosses the placenta or cord blood levels of this metabolite are completely fetal in origin is not known.

CMPF is a furoic acid that has been implicated in β-cell dysfunction and impaired insulin granule maturation and secretion. The origin of circulating CMPF is incompletely understood, but CMPF is thought to be generated by gut microbes (58). CMPF levels have been positively associated with maternal glycemia and gestational diabetes and rise over time in nonpregnant individuals that develop type 2 diabetes mellitus (59–61). However, we found a negative association of cord blood CMPF with BW and SSF. Interestingly, we previously identified a similar negative relationship between maternal CMPF and newborn size in the same cohort (unpublished observation), and others have described an association of lower cord blood CMPF levels in children who had rapid postnatal weight gain and were overweight or obese in mid-childhood compared with normal weight children (62).

Cord blood triglycerides were consistently negatively associated with cord C-peptide, BW, and SSF, whereas NEFAs and palmitoleic acid were negatively associated with cord C-peptide but not associated with BW or SSF. Interestingly, we and others have previously reported positive associations of maternal triglycerides and NEFAs during pregnancy with fetal growth (24, 63). The inverse association between cord blood levels of these same metabolites and measures of newborn size and hyperinsulinemia are consistent with prior studies (18, 20, 64, 65). Maternal triglycerides do not cross the placenta, but fatty acids do and provide a substrate for uptake into adipose tissue, triglyceride synthesis, and fetal energy production (63). This could lead to lower measurable levels in larger infants with greater utilization, and a negative association with newborn size.

Network analyses performed in this study highlight the interconnectedness of cord blood metabolites as potential contributors to newborn body composition and hyperinsulinemia. Networks associated with all three outcomes included similar metabolites, notably acetylcarnitine (AC C2), medium- and long-chain acylcarnitines, BCAAs, AAAs, ketone bodies and related metabolites, and fatty acid related metabolites. The BCAA metabolites, AC C3 and C5, were only present in networks associated with cord C-peptide and BW. Acetylcarnitine (AC C2), typically the most abundant AC in human sera, is a short-chain acylcarnitine that is a cognate analyte for acetyl CoA and is elevated in insulin-resistant states (66, 67). AC C2 may also be a marker of maternal metabolic inflexibility, because high levels of acetyl CoA lead to greater production of malonyl CoA, an inhibitor of fatty acid oxidation that can impact the ability of mitochondria to switch between glucose and fatty acid metabolism during a fed or fasted state (9). When fuel switching does not occur effectively, mitochondrial metabolic gridlock (9) may occur, and we speculate that this may lead to acylcarnitine accumulation in tissues (and secondarily in plasma), which, in turn, might affect newborn size and insulin action.

Existing studies of cord blood metabolites as they relate to newborn anthropometrics have revealed similarities and differences to the current study. Analogous to our results, Patel et al. (18) found that cord blood NEFAs and triglycerides were negatively associated with BW, SSF, and middle upper-arm circumference in 343 newborns born to women with obesity. Another study of 126 newborns used principal component analyses to show that tricarboxylic acid cycle intermediates, purine/pyrimidine metabolites, and selected lipid metabolites were associated with higher BW and odds of being born large for gestational age (19); we similarly identified that hypoxanthine and pseudouridine, two purine/pyrimidine metabolites, were positively associated with BW and SSF in a fully adjusted model (model 3). Two additional studies identified an association between cord blood lysophosphatidylcholines and BW (20, 21); these metabolites were not evaluated in the current study. All of these prior studies were smaller in scale and in some cases did not use direct measures of adiposity or account for known confounders of newborn size, such as maternal BMI and glucose.

Our finding of largely negative associations between various cord blood metabolites and newborn C-peptide levels was unexpected, particularly as some of these metabolite classes (BCAAs, AAAs, ketones, and medium- and long-chain acylcarnitines) are associated with insulin resistance in adolescents and adults. Our findings suggest that the positive association of metabolites with C-peptide does not begin at birth, and perhaps the inverse relationship at birth changes over time as fuel-stimulated insulin secretion matures and a child is exposed to external factors such as diet, environment, and level of physical activity. A parallel to this reversal in direction of association can be drawn from the literature on adiponectin, an adipokine involved in insulin sensitivity (68). Cord blood adiponectin levels are positively correlated with fat in newborns and predict adiposity at age 3 (69), but serum levels are negatively associated with BMI and measures of fat mass in older children and adults with obesity (70, 71). Alternatively, as cord blood C-peptide is directly related to maternal glucose levels, it may not fully and accurately represent newborn insulin sensitivity (72).

Strengths of this study include its multiethnic cohort and large size in comparison with previously published cord blood metabolomics studies, as well as the adjustment for both maternal BMI and maternal glucose levels (18–20, 62, 64). Maternal BMI and glucose are two exogenous factors that impact newborn size; accounting for these factors in the adjusted analysis allowed examination of the relationship of cord blood metabolites to newborn size in their absence. We also used SSF as a measure of newborn fat mass, which has been shown to be a more accurate predictor of fat mass compared with BW, weight for length, and BMI, all of which do not discriminate fat mass from fat-free mass (73).

Because this is a cross-sectional, observational study, we are limited in our ability to establish causality. Missing values for low abundance metabolites, particularly from nontargeted assays, also preclude generalizability of findings to low abundance compounds. Further longitudinal analysis in this cohort is necessary to evaluate how the relationship between cord blood metabolites and newborn size translates to obesity and insulin resistance in later life. Lastly, there are likely many additional contributors to the cord blood metabolome including but not limited to the placenta, maternal diet, maternal and fetal genetics, and the maternal microbiome. We are unable to explore many of these factors; however, our results span four ethnic groups that represent diverse socioeconomic, nutritional, and environmental backgrounds, and therefore some contributors may be accounted for on a high level.

In summary, cord blood metabolites are associated with newborn size and hyperinsulinemia, even when accounting for maternal BMI and glucose during pregnancy. 1,5-Anhydroglucitol, a marker of glycemia, may be an emerging marker of newborn adiposity. The known associations of medium-chain acylcarnitines with obesity in adolescents and adults appear to be present at birth although the relationship of BCAAs to lean vs fat mass in the newborn period is less clear. Further study is needed to elucidate the mechanisms underlying these associations to better understand how the complex relationship between metabolite signatures of adiposity and hyperinsulinemia in newborns relates to obesity and type 2 diabetes phenotypes in adolescents and adults.

Glossary

Abbreviations:

AAA

aromatic amino acid

AC C4-OH

carnetine ester of 3-hydroxybutyrate

BCAA

branched-chain amino acid

BMI

body mass index

BW

birthweight

CMPF

3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid

CoA

coenzyme A

GC

gas chromatography

FDR

false discovery rate

GDM

gestational diabetes mellitus

HAPO

Hyperglycemia and Adverse Pregnancy Outcome

MS

mass spectrometry

NEFA

nonesterified fatty acid

OGTT

oral glucose tolerance test

SSF

sum of skinfolds

Additional Information

Disclosure Summary: The authors have nothing to disclose.

Data Availability:

The datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.