Infant Metabolome in Relation to Prenatal DHA Supplementation and Maternal Single-Nucleotide Polymorphism rs174602: Secondary Analysis of a Randomized Controlled Trial in Mexico

ABSTRACT

Background

Although DHA (22:6n–3) is critical for fetal development, results from randomized controlled trials (RCTs) of prenatal DHA supplementation report inconsistent effects on offspring health. Variants in fatty acid desaturase (FADS) genes that regulate the conversion of n–3 and n–6 essential fatty acids into their biologically active derivatives may explain this heterogeneity.

Objectives

We investigated the effect of prenatal DHA supplementation on the offspring metabolome at age 3 mo and explored differences by maternal FADS single-nucleotide polymorphism (SNP) rs174602.

Methods

Data were obtained from a double-blind RCT in Mexico [POSGRAD (Prenatal Omega-3 Fatty Acid Supplementation and Child Growth and Development)] in which women (18–35 y old) received DHA (400 mg/d) or placebo from mid-gestation until delivery. Using high-resolution MS with LC, untargeted metabolomics was performed on 112 offspring plasma samples. Discriminatory metabolic features were selected via linear regression (P < 0.05) with false discovery rate (FDR) correction (q = 0.2). Interaction by SNP rs174602 was assessed using 2-factor ANOVA. Stratified analyses were performed, where the study population was grouped into carriers (TT, TC; n = 70) and noncarriers (CC; n = 42) of the minor allele. Pathway enrichment analysis was performed with Mummichog (P < 0.05).

Results

After FDR correction, there were no differences in metabolic features between infants whose mothers received prenatal DHA (n = 58) and those whose mothers received placebo (n = 54). However, we identified 343 differentially expressed features in the interaction analysis after FDR correction. DHA supplementation positively enriched amino acid and aminosugars metabolism pathways and decreased fatty acid metabolism pathways among offspring of minor allele carriers and decreased metabolites within the tricarboxylic acid cycle and galactose metabolism pathways among offspring of noncarriers.

Conclusions

Our findings demonstrate differences in infant metabolism in response to prenatal DHA supplementation by maternal SNP rs174602 and further support the need to incorporate genetic analysis of FADS polymorphisms into DHA supplementation trials.

This trial was registered at clinicaltrials.gov as NCT00646360.

Introduction

The n–3 long-chain polyunsaturated fatty acid (LC-PUFA) DHA (22:6n–3) accumulates during the second half of pregnancy to support optimal fetal neurodevelopment, visual acuity, and immune function (1, 2). Maternal DHA status is associated with infant DHA status, especially during the first 6 mo of postnatal life; therefore, ensuring adequate maternal DHA status during pregnancy may be critical for normal offspring growth and development (3). Given the increased nutritional requirements during pregnancy and lactation, it would be expected that pregnant women and their offspring will benefit from supplementation with preformed DHA. However, results from multiple well-designed randomized controlled trials (RCTs) of DHA supplementation during pregnancy report inconsistent effects on birth outcomes (4–6). This heterogeneity may be attributable to variants of fatty acid desaturase (FADS) genes that modulate the conversion of n–3 and n–6 essential fatty acids into their LC-PUFA forms (7, 8).

Tissue LC-PUFA concentrations are determined by both dietary intake of preformed LC-PUFAs and endogenous synthesis from their dietary precursors linoleic acid (LA; 18:2n–6) and α-linolenic acid (ALA; 18:3n–3). Conversion of dietary precursors to LC-PUFAs occurs through a series of consecutive desaturation and elongation steps, in which n–6 and n–3 fatty acids compete for conversion. The rate-limiting desaturase steps are mediated by Δ-6 and Δ-5 desaturase enzymes encoded in the FADS gene cluster (FADS1,FADS2,FADS3) (8). Single-nucleotide polymorphisms (SNPs) in FADS genes may explain up to nearly 30% of variability in PUFA and LC-PUFA concentrations in tissues (9). Multiple variants in FADS genes are associated with lower LC-PUFA concentrations, demonstrating slower conversion of dietary precursors (8, 10). Offspring of carriers of some FADS genetic variants may therefore uniquely benefit from prenatal DHA supplementation via an increased LC-PUFA supply.

Our group recently assessed the effect of prenatal supplementation with DHA on birth weight across 4 selected FADS SNPs and showed significant heterogeneity by maternal SNP rs174602, an intron variant located in the FADS2 gene, where carriers of the minor T allele in the treatment group had higher birth weights (11). Although these findings provide initial evidence that genetic variations in maternal fatty acid metabolism may contribute to differences in the effects of prenatal DHA supplementation, an understanding of the biological mechanisms by which this occurs remains limited.

In recent decades, the rise of systems biology approaches has facilitated powerful investigations of gene–environment interactions. High-resolution untargeted metabolomics offers a robust molecular measurement of phenotype by providing a quantitative global snapshot of small-molecule metabolites in biological samples (12). Untargeted metabolomics can be used to identify distinct metabolic changes that result from prenatal DHA supplementation as well as by maternal FADS genotype. The integration of genomic and metabolomics data provides a unique opportunity to bridge current knowledge gaps within fatty acid metabolism.

The objective of this exploratory study was to add biological context to previously observed heterogeneity in the POSGRAD (Prenatal Omega-3 Fatty Acid Supplementation and Child Growth and Development) trial by examining the influence of prenatal DHA supplementation and maternal FADS genotype on the offspring metabolome at 3 mo of age. Here, we report a metabolome-wide association study, coupled with pathway enrichment and module analysis, to assess metabolic changes in infants whose mothers participated in a prenatal DHA supplementation RCT in Mexico and explore differences by maternal SNP rs174602.

Methods

Study sample population

Data were obtained from POSGRAD, a double-blind RCT (NCT00646360) conducted in Cuernavaca, Mexico, which was originally designed to assess the effect of prenatal DHA supplementation on offspring growth and development (13). Briefly, from 2005 to 2007, pregnant women were recruited at 18–22 weeks of gestation and randomly assigned to receive 2 capsules containing either 200 mg algal DHA each (treatment) or a corn/soy blend (placebo) daily. Supplemental Table 1 gives the fatty acid profile of the treatment and placebo capsules. Eligible women were 18–35 y old, planned to deliver at the Mexican Institute for Social Security General Hospital in Cuernavaca, to breastfeed for ≥3 mo, and to continue living in the area for ≥2 y after delivery. Women and infants were followed up after birth. The subsample for untargeted metabolomics analysis included all mother–infant dyads selected from the sample of live singleton births with maternal FADS genotype information and infant plasma samples at 3 mo of age available for analysis (Figure 1). Analyses were completed blinded to randomization status.

FIGURE 1.

Flow of sample selection from the Prenatal Omega-3 Fatty Acid Supplementation and Child Growth and Development trial. SNP, single-nucleotide polymorphism.

The study was conducted according to the guidelines of the Declaration of Helsinki. The Emory University Institutional Review Board and the Mexican National Public Health Institute (INSP) ethics committee approved all procedures involving human subjects.

Blood collection and maternal genotyping

Fasting venous blood samples were obtained from all pregnant women at recruitment. Plasma, buffy coat, and RBCs were separated and stored at INSP laboratories at −80°C until transport to the Hemholtz Center, Munich, where the genetic analysis was carried out during 2012–2013 for those who provided genetic consent, using methods that have been previously described (9). The resulting data sets containing information on 15 FADS1,FADS2, and FADS3 SNPs, selected based on biological evidence of an effect on LC-PUFA metabolism, were sent to Emory University via encrypted files (11). For the exploratory stratified analyses by SNP rs174602, mothers who were homozygous or heterozygous for the minor T allele (TT, TC) were classified as carriers of the minor allele, whereas mothers homozygous for the major C allele (CC) were classified as noncarriers.

Determination of plasma fatty acids

Fatty acid concentrations in maternal plasma samples that were obtained at baseline and delivery were determined in a subsample as part of the original study at INSP laboratories (2006–2007), using methods that have been previously described (11). Total fatty acids were expressed as percentage by weight of total detected fatty acids.

Maternal dietary intake and infant feeding practices

Maternal dietary intake at baseline was assessed using a previously validated 110-item FFQ specifically designed for the population of interest to include important dietary sources of PUFAs (14, 15). Data on infant feeding practices at 3 mo of age, obtained by maternal interview at 3 mo of age, were used to categorize breastfeeding (BF) status as exclusive BF (EBF), predominantly BF (PreBF), partial BF (PaBF), and non-BF (NBF) according to the WHO classification (16, 17).

Plasma high-resolution metabolomics

High-resolution untargeted metabolomics was performed in 2014 on plasma samples obtained from offspring at 3 mo of age using previously standardized methods (18) in the Emory Clinical Biomarkers Laboratory. All samples were transported from INSP and stored at −80°C at Emory University until analysis. On the day of analysis, samples were thawed, mixed with 2 volumes of ice-cold acetonitrile containing a mixture of stable isotopic internal standards (18), allowed to stand on ice for 30 min, and centrifuged at 14,000 × g for 10 min at 4°C. Supernatants were transferred to autosampler vials and maintained in a refrigerated autosampler at 0–4°C until analysis by MS. Mass spectral data were collected in triplicate with a Thermo Q-Exactive mass spectrometer (ThermoFisher) set to collect data from m/z 85 to 1275 using C18 LC with a 10-min gradient (18) and positive electrospray ionization. Data were stored as .raw files and converted using Xcalibur file converter software (ThermoFisher) to .cdf files for further data processing. Raw MS data were processed using the R-based packages xMSanalyzer version 1.3.5 (19) with apLCMS version 5.9.4 (20) to perform peak detection, noise filtering, and feature alignment; samples with a mean correlation <0.70 were removed from the final feature table. The raw data were averaged and batch-effect correction was performed using ComBat (21), resulting in a feature table with 9533 features. For each analysis, data imputation was performed by replacing missings with one-half of the lowest signal intensity for the current feature. Only features with nonmissing values in >80% of all samples in either one of the treatment groups were retained. Data were log₂ transformed and quantile normalized to improve feature comparability.

Statistical analyses and bioinformatics

Baseline characteristics of the included subsample were compared with those of the rest of the birth cohort using Student's t tests (parametric) and Wilcoxon's rank-sum tests (nonparametric) for continuous variables and chi-square tests for categorical variables. Plasma fatty acid concentrations were compared between carriers (TT, TC) and noncarriers (CC) of the minor allele using t tests.

Discriminatory metabolic features were selected using linear regression (P < 0.05) and represented through Manhattan plots and heatmaps. False discovery rate (FDR) correction was applied using the Benjamini–Hochberg method (q = 0.2) (22). Covariates of interest included offspring sex. Sensitivity analyses were conducted using a subset of the data to assess the role of dietary fat intake outliers [including DHA, arachidonic acid (AA; 20:4n–6), ALA, LA, and total n–6 and n–3 fatty acids] on the infant metabolome.

Statistical interaction between DHA supplementation and maternal genotype for SNP rs174602 was assessed by performing feature selection using 2-factor ANOVA. Stratified analyses were also performed by maternal genotype for SNP rs174602, with the study population grouped into carriers (TT, TC) and noncarriers (CC) of the minor allele. All biomarker discovery analyses were conducted using the “xmsPANDA” package version 1.0.7.5. Metabolites that significantly differed between treatment groups (P < 0.05) were further analyzed using mummichog pathway enrichment and module analysis software (version 2.3.3), which bypasses metabolite identification, the bottleneck of untargeted metabolomics, by predicting biological activity directly from MS data (23). Empirical P values were estimated by permutation test (P < 0.05). Mummichog also performs module analysis, which produces modules, or subnetworks, of highly correlated metabolites that are unbiased by predefined metabolic pathways. Because the use of raw P thresholds for pathway enrichment has been shown to improve detection of biological effects for discovery purposes (24), no FDR correction was applied to the input data. Only significantly enriched metabolic pathways with ≥4 metabolites were included in the presented findings. A sensitivity analysis was performed to ensure that pathway enrichment results held across different feature selection methods (“limma” method in xmsPANDA). Metabolites within each enriched pathway were annotated with the R-based package xMSannotator (25) using the Human Metabolome Database (HMDB) and LipidMaps at 10 ppm tolerance. xMSannotator uses multiple criteria to assign a score-based annotation and corresponding confidence level. Confidence levels range from 0 (no confidence) to 3 (high confidence); only annotations with level 2 (medium) confidence or higher were reported. Amino acids, LA, and other metabolites have confirmed identification at Schymanski Level 1 (26) in other studies (27, 28) using these methods, but these were not confirmed in the present analyses so identifications are considered Schymanski Level 5. All statistical analyses were performed using R version 3.8 (R Foundation for Statistical Computing, Vienna, Austria.

Results

Table 1 presents baseline characteristics of the mother–infant dyads in the analytic sample. The mean age and BMI of the mothers were 26 y and 25.9 kg/m², respectively. Median dietary intakes of LA and AA were 20.2 g/d (IQR: 16.0–24.3 g/d) and 0.16 g/d (0.10–0.21 g/d), respectively, and median intakes of ALA, EPA (20:5n–3), and DHA were 1.6 g/d (1.2–2.3 g/d), 0.02 g/d (0.01–0.05 g/d), and 0.06 g/d (0.04–0.13 g/d), respectively. The median total dietary n–6:n–3 fatty acids ratio was 12:1. There were no differences in maternal characteristics, including maternal dietary intake and compliance with the intervention, by treatment group. At 3 mo of age, 74% of infants were consuming breast milk as part of their diet, but only 11% were exclusively breastfed. Dietary intakes of total fat, n–3, and n–6 fatty acids were higher in the analytic sample than in the rest of the birth cohort (P < 0.05) (Supplemental Table 2).

TABLE 1.

Maternal baseline characteristics and offspring characteristics at birth, stratified by treatment group and maternal fatty acid desaturase 2 SNP rs174602¹

	Placebo (n = 54)	DHA (n = 58)	Carriers (n = 70)	Noncarriers (n = 42)
Maternal characteristics
Age, y	26.3 ± 4.9	26.1 ± 5.4	26.3 ± 5.4	26.1 ± 4.8
Socioeconomic status score	−0.06 ± 1.17	−0.03 ± 0.98	0.10 ± 1.04	−0.30 ± 1.09
Schooling, y	11.5 ± 3.52	11.9 ± 3.47	12.0 ± 3.32	11.2 ± 3.74
Height, cm	154 ± 5.03	155 ± 5.19	155 ± 4.79	154 ± 5.61
BMI, kg/m2	26.2 ± 4.44	25.7 ± 4.12	25.8 ± 4.39	26.1 ± 4.11
First pregnancy	11 (20.4)	20 (34.5)	23 (32.9)	8 (19.0)
Dietary intake,2 g/d
n–3 Fatty acids	1.7 [1.2–2.4]	1.7 [1.3–2.4]	1.6 [1.2–2.4]	1.7 [1.3–2.4]
ALA	1.5 [1.1–2.7]	1.6 [1.2–2.3]	1.5 [1.2–2.3]	1.7 [1.2–2.3]
EPA	0.02 [0.01–0.04]	0.02 [0.01–0.05]	0.02 [0.01–0.05]	0.01 [0.01–0.03]
DHA	0.05 [0.03–0.10]	0.06 [0.04–0.13]	0.06 [0.04–0.13]	0.05 [0.03–0.08]
n–6 Fatty acids	19.9 [17.0–24.1]	20.4 [15.0–24.6]	20.4 [15.8–24.3]	20.3 [16.3–24.7]
LA	19.8 [16.8–24.0]	20.2 [14.9–24.4]	20.2 [15.7–24.2]	20.2 [16.2–24.5]
AA	0.15 [0.10–0.20]	0.16 [0.12–0.23]	0.16 [0.11–0.21]	0.15 [0.10–0.22]
n–6:n–3 ratio	12.4 [9.2–15.0]	12.0 [8.5–14.5]	12.0 [8.5–14.4]	12.4 [9.3–16.2]
Compliance to intervention,3 %	94.4 ± 5.0	95.9 ± 4.7	95.7 ± 4.5	94.3 ± 5.5
SNP rs174602
Carrier of minor T allele (TT, TC)	29 (53.7)	41 (70.7)
Noncarrier (CC)	25 (46.3)	17 (29.3)
Offspring characteristics, birth
Girls	30 (55.6)	35 (60.3)	41 (58.6)	24 (57.1)
Gestational age, wk	39.2 ± 1.31	39.1 ± 1.53	39.2 ± 1.57	39.0 ± 1.13
Length,4 cm	50.4 ± 1.80	50.4 ± 2.54	50.1 ± 2.28	51.0 ± 1.95
Weight,4 g	3255 ± 497	3267 ± 457	3173 ± 407	3408 ± 544
Head circumference,4 cm	34.1 ± 1.58	34.3 ± 1.51	34.0 ± 1.42	34.7 ± 1.62
BF status at 3 mo5
EBF	6 (11.5)	6 (10.7)	7 (10.6)	5 (11.9)
PreBF	7 (13.5)	3 (5.4)	5 (7.6)	5 (11.9)
PaBF	31 (59.6)	37 (66.1)	43 (65.2)	25 (59.5)
NBF	8 (15.4)	10 (17.9)	11 (16.7)	7 (16.7)

¹Values are mean ± SD, median [IQR], or n (%) unless otherwise indicated. Chi-square tests, Student's t tests, and Wilcoxon's rank-sum tests were used to test differences between groups (P < 0.05). There were no significant differences between placebo and treatment (DHA) groups for any of the characteristics included in the table. AA, arachidonic acid; ALA, α-linolenic acid; BF, breastfeeding; EBF, Exclusive Breastfeeding; LA, linoleic acid; NBF, No Breastfeeding; PaBF, Partial Breastfeeding; PreBF, Predominantly Breastfeeding; SNP, single-nucleotide polymorphism.

²Assessed by a 110-item FFQ designed for a Mexican population. Differences between groups were tested using Wilcoxon's rank-sum tests.

³Measured as percentage of consumed capsules.

⁴Carriers and noncarriers were significantly different (P < 0.05).

⁵EBF, defined as intake of breast milk only, allowing the consumption of drops, syrups, oral rehydration solution, and/or vitamins and minerals; PreBF, defined as intake of breast milk plus certain fluids such as water, water-based drinks, fruit juices, oral rehydration solution, and vitamins and/or minerals; PaBF, defined as breast milk plus any food or liquid, including nonhuman milk; NBF, defined as the intake of formula or other nonhuman milk and/or solid foods without BF.

Distribution of maternal FADS SNP rs174602

Within the analytic sample, 20 mothers (17.9%) were homozygous carriers of the minor T allele (TT), 50 (44.6%) were heterozygous carriers (TC), and 42 (37.5%) were homozygous noncarriers (CC). There were no violations of Hardy–Weinberg equilibrium (P = 0.49), and no significant differences in maternal genotype distribution for SNP rs174602 were observed across treatment groups. The maternal genotype distribution in the analytic sample also aligned with the distribution observed in the larger sample with complete genotype information (Supplemental Table 2). Offspring birth, length, weight, and head circumference, however, differed between carriers and noncarriers of the minor allele for SNP rs174602 (each P < 0.05). Carriers of the minor allele also tended to have higher mean ± SD plasma AA concentrations (% by wt of total fatty acids in maternal plasma) at both baseline (carriers: 4.50 ± 1.11; noncarriers: 4.02 ± 1.12; P = 0.232, n = 34) and delivery (carriers: 3.44 ± 1.00; noncarriers: 2.96 ± 0.46; P = 0.046, n = 40), whereas plasma DHA concentrations did not vary by genotype at either time point (Supplemental Table 3).

Differences by intervention on high-resolution metabolomics

We first examined whether the metabolome differed between infants whose mothers received DHA and infants whose mothers received placebo. Of the 9533 identified features, 7772 remained after data preprocessing; 279 metabolic features differed in infants whose mothers received prenatal DHA as opposed to placebo using a linear regression model (P < 0.05). After FDR correction, 0 significant features remained. However, pathway enrichment analysis using mummichog showed significant enrichment of the de novo fatty acid biosynthesis pathway among infants whose mothers received DHA as opposed to placebo. Annotated metabolites within the enriched pathway all belonged to the class of unsaturated fatty acyl CoAs (Supplemental Table 4).

Interaction between prenatal DHA supplementation and SNP rs174602

We identified 1684 differentially expressed features in the interaction analysis (P < 0.05); after correction for FDR (q = 0.2), 343 features remained (Figure 2A). Mummichog identified 5 metabolic pathways that were enriched by the interaction between DHA supplementation and SNP rs174602, including aspartate and asparagine metabolism, arginine and proline metabolism, carbon fixation, fatty acid metabolism, and drug metabolism (Figure 2B). Boxplots for select metabolic features, shown in Figure 2C, display differences in metabolite intensity levels across each treatment and genotype group. Briefly, metabolites within amino acid metabolism, including 5-oxoprolinate, 4-hydroxy-l-proline, and l-glutamine, were increased in offspring of carriers who received DHA and decreased in offspring of noncarriers who received DHA, relative to placebo. Metabolites within the carbon fixation pathway, such as glyceraldehyde 3-phosphate and oxalacetic acid, were decreased in noncarriers who received DHA and increased in carriers who received DHA, compared with placebo. Metabolites within fatty acid metabolism, including (2E)-dodecenoyl-CoA and 5-octadecynoic acid, were increased in noncarriers who received DHA and decreased in carriers who received DHA, compared with placebo. Supplemental Table 5 presents a complete list of the annotated metabolites contained in each enriched pathway. Mummichog identified 3 modules that may capture metabolite activities within and in between established metabolic pathways. The first 2 modules, which were related to lipid metabolism, consisted of 11 metabolites across de novo fatty acid biosynthesis, linoleate metabolism, and fatty acid activation and 30 metabolites across AA metabolism, de novo fatty acid biosynthesis, and glycerophospholipid metabolism. The third module was composed of 68 metabolites across the carnitine shuttle, aspartate and asparagine metabolism, and arginine and proline metabolism.

FIGURE 2.

Interaction between prenatal DHA supplementation and maternal SNP rs174602. (A) Manhattan plot as a function of m/z in interaction analysis (n = 112; P < 0.05: 1684 features; q < 0.2: 343 features). (B) Pathway enrichment analysis in Mummichog version 2.3.3 using input from metabolic features significantly associated with the interaction between treatment group and maternal SNP rs174602. The y axis represents differentially enriched pathways, whereas the x axis represents the negative log₁₀P value of each pathway. The radii of data points correspond to the number of significantly associated metabolic features within the pathway. Only pathways with ≥4 overlapping metabolites are shown. (C) Boxplots comparing the intensity of select metabolic features within significantly enriched pathways between groups. The box portion of the plot represents the interquartile range (IQR: 25th percentile, median, and 75th percentile) of the data. Error bars represent smallest and largest values within 1.5 times IQR. ¹Select metabolites from arginine and proline metabolism and aspartate and asparagine metabolism pathways. ²Select metabolite from carbon fixation pathway. ³Select metabolites from fatty acid metabolism pathway. C, carriers of minor allele for SNP rs174602; DHA, offspring whose mothers received prenatal DHA; HMDB, Human Metabolome Database; NC, noncarriers of minor allele for SNP rs174602; P, offspring whose mothers received placebo; rt, retention time; SNP, single-nucleotide polymorphism.

Stratified analysis by maternal FADS2 genotype

Among carriers, linear regression identified 615 metabolic features at P < 0.05 (Figure 3A), whereas among noncarriers, 666 metabolic features were selected (Figure 3B), with an overlap of 89 metabolic features across groups. After FDR correction, 1 metabolic feature remained significant among both subgroups [carriers: 2-methoxy-(S)-oleuropein (HMDB35445); noncarriers: 2-(methylthio)-3H-phenoxazin-3-one (HMDB35996)]. Two-way hierarchical cluster analysis showed clear differences in patterns of metabolite intensity when stratified by maternal FADS genotype (Figure 4).

FIGURE 3.

Stratified analysis by maternal fatty acid desaturase 2 single-nucleotide polymorphism rs174602. Manhattan plots as a function of m/z in (A) carriers (n = 70; P < 0.05: 615 features; q < 0.2: 1 feature) and (B) noncarriers (n = 42; P < 0.05: 666 features; q < 0.2: 1 feature).

FIGURE 4.

Two-way hierarchical cluster analysis of treatment group (DHA compared with placebo) and (A) 279 discriminatory features selected via linear regression (P < 0.05) from the complete analytic sample (n = 112), (B) 615 features identified from carriers of the minor T allele (n = 70), and (C) 666 features identified from noncarriers of the minor T allele (n = 42). Each column represents an individual's metabolic profile based on differentially expressed features.

Pathway enrichment analysis showed that, among carriers, 5 metabolic pathways were significantly enriched by DHA supplementation (Figure 5A). Metabolites within aspartate and asparagine metabolism and arginine and proline metabolism pathways, such as l-glutamic acid, 4-hydroxy-l-proline, and l-arginine, were increased in offspring whose mothers received DHA as opposed to placebo. Most metabolites within fatty acid metabolism and AA metabolism pathways were decreased, whereas metabolites within aminosugar metabolism were increased. Among noncarriers, 2 metabolic pathways were significantly enriched. Metabolites within the tricarboxylic acid (TCA) cycle (thiamin pyrophosphate, oxalacetic acid) and galactose metabolism (dihydroxyacetone phosphate, lactose-6-phosphate) were decreased in offspring whose mothers received DHA relative to placebo. In addition, we observed differences for dodecenoyl CoA, a metabolite noted for its role in fatty acid synthesis and oxidation, between carriers and noncarriers (29). In carriers, metabolite intensity was increased in infants whose mothers received DHA relative to placebo, whereas in noncarriers, it was decreased. Boxplots of select metabolic features for carriers and noncarriers are shown in Figure 5B and C, respectively, whereas complete annotations of the metabolites from the pathway enrichment analysis are available in Supplemental Tables 6 and 7.

FIGURE 5.

Pathway enrichment from stratified analysis by maternal fatty acid desaturase 2 single-nucleotide polymorphism rs174602. (A) Pathway enrichment analysis in Mummichog version 2.3.3 using input from metabolic features significantly associated with treatment group in the stratified analysis. The y axis represents differentially enriched pathways, whereas the x axis represents the negative log₁₀P value of each pathway. The radii of data points correspond to the number of significantly associated metabolic features within the pathway. Only pathways with ≥4 overlapping metabolites are shown. (B, C) Boxplots comparing the intensity of select metabolic features within significantly enriched pathways between infants whose mothers received DHA or placebo in (B) carriers and (C) noncarriers. The box portion of the plot represents the interquartile range (IQR: 25th percentile, median, and 75th percentile) of the data. Error bars represent smallest and largest values within 1.5 times IQR. ¹Select metabolites from arginine and proline metabolism and aspartate and asparagine metabolism pathways. ²Select metabolites from fatty acid metabolism and AA metabolism pathways. ³Select metabolites from TCA metabolism, galactose metabolism, and aminosugars metabolism pathways. AA, arachidonic acid; HMDB, Human Metabolome Database; LMFA, Lipid Maps Fatty Acyls; rt, retention time; TCA, tricarboxylic acid.

Discussion

Determination of metabolic phenotype via high-resolution metabolomics offers a powerful opportunity to investigate biological responses to interventions. In this population of Mexican women and their offspring, we used untargeted metabolomics to assess the offspring metabolome at 3 mo of age in response to prenatal DHA supplementation and explored differences by maternal FADS2 SNP rs174602. As opposed to focusing on individual metabolites, we primarily focused on pathway enrichment analysis in subsets of the study population, with the aim of facilitating better understanding of the specific effects of supplementation among SNP carriers.

The metabolome-wide association study comparing offspring whose mothers received DHA or placebo yielded no significant features after FDR correction, although pathway analysis using significant metabolic features at P < 0.05 showed significant enrichment of the de novo fatty acid biosynthesis pathway. Unsaturated fatty acyl CoAs, important coenzymes involved in fatty acid metabolism, were increased in offspring whose mothers received DHA compared with placebo. Within our subsample, plasma DHA concentrations were also significantly higher at delivery in supplemented mothers than in mothers who received placebo; this aligns with previously reported findings that plasma DHA concentrations at delivery were 26% higher in supplemented mothers in the larger sample (13).

However, when interaction between DHA supplementation and maternal FADS2 SNP rs1742602 was considered, differences in the offspring metabolome became more apparent. We identified 343 differentially expressed features across lipid metabolism and amino acid metabolism pathways. Moreover, we found that the relative intensity of multiple metabolites within fatty acid and amino acid metabolism pathways differed between offspring of carriers and noncarriers in response to DHA supplementation. Intensity of metabolites within amino acid and carbon fixation pathways was higher in offspring of carriers relative to noncarriers, whereas intensity of metabolites in fatty acid metabolism pathways was lower in offspring of carriers relative to noncarriers. These findings suggest that offspring of individuals with specific FADS2 genotypes may respond differently to prenatal DHA supplementation.

Results were consistent across both the interaction and stratified analyses by FADS2 SNP rs174602. Among offspring born to maternal carriers of the minor allele, amino acid metabolism and aminosugars metabolism pathways were positively enriched, whereas there were reductions in fatty acid metabolism, after DHA supplementation; among noncarriers, we observed decreases in metabolites from the TCA cycle and galactose metabolism pathways among infants whose mothers received DHA supplementation relative to placebo. Changes in nonessential amino acid metabolism and other pathways could reflect adaptive changes that occur as a consequence of supplementation. Changes in amino acid metabolism among carriers may also reflect the rapid growth and development observed in infancy. The first few months of life are characterized by complex changes, including rapid increases in weight and length and development of the immune system, nervous system, endocrine system, and metabolism (30). Within the carrier group, important metabolites for protein synthesis and collagen production, including glutamate, arginine, and 4-hydroxy-l-proline (31), were increased in offspring whose mothers received DHA compared with placebo. We previously showed that offspring of carriers of the T allele (TT, TC) in the treatment group were significantly heavier than those in the placebo group, with no differences in homozygote carriers of the C allele (11). Metabolomic differences observed at 3 mo suggest that differences in response to prenatal DHA supplementation by maternal genotype persist beyond birth; however, further confirmation of our findings is needed to determine any clinical relevance.

Results from the fatty acid analysis may provide further insight. In the larger sample, SNP rs174602 was positively associated with plasma AA concentrations but inversely associated with DHA concentrations (11). In our subsample, we observed a similar association for AA concentrations, but no differences in DHA concentrations by genotype. FADS2 encodes for the Δ-5 desaturase enzyme that regulates the conversion of EPA to DHA. Indeed, previous studies have shown that SNP rs174602 was associated with lower Δ-5 desaturase activity (32). This contributed to evidence suggesting that carriers of the minor allele in our sample would be at greater risk of DHA deficiency.

Our results align with previously reported findings that maternal genetic variants in the FADS gene cluster influence offspring health. Multiple observational studies have shown that various FADS1 and FADS2 SNPs influence breast-milk fatty acid composition (33, 34), child LC-PUFA status (35), the infant immune response at 6 mo of age (36), and child cognition at 14 mo (37). In a prenatal supplementation trial in the United States, pregnant women who were homozygous for the minor allele of FADS1 SNP rs174533 had lower DHA and AA concentrations at baseline; after daily supplementation with 600 mg DHA until delivery, the intervention increased DHA concentrations and decreased the AA:DHA ratio only among carriers (38). A recent study from Denmark used untargeted metabolomics to study the influence of prenatal n–3 LC-PUFA supplementation on the offspring metabolome at 6 mo of age and investigated differences by maternal FADS2 SNP rs1535. Findings showed that the minor allele for rs1535 was inversely associated with concentrations of all n–6 pathway metabolites in the placebo strata, whereas no association was observed in the treatment group (39). A key difference between this study and ours is the composition and dose of the supplement [2.4 g fish oil–derived n–3 LC-PUFAs (55% EPA, 37% DHA) and 400 mg algal DHA, respectively]. Although the investigated SNPs and observed results vary across each study, they collectively support the importance of incorporating FADS genetic information in prenatal DHA supplementation studies.

It is important to consider that findings may not be consistent or comparable across distinct populations, because there is racial/ethnic variation in the genotype distribution of FADS variants (40, 41). Most studies of FADS SNPs and LC-PUFA concentrations have been conducted in European populations. Whereas European populations predominantly have FADS alleles associated with more rapid conversion of precursor PUFAs, Native American and Mexican populations, including our study population, have a higher frequency of alleles associated with less efficient conversion of dietary precursors to their LC-PUFA derivatives, along with diets high in n–6 PUFAs and low in n–3 LC-PUFAs (8, 42). It is currently not well understood why there are such distinct variations in genetic distribution; however, genetic adaptations in response to changes in dietary intake after the agriculture transition are hypothesized to be major drivers (40, 43, 44). Additional evidence is needed across distinct populations that vary in genotype and PUFA intake to guide development of targeted supplementation recommendations based on genotype.

Several limitations should be considered when interpreting our findings. We used secondary data from a previously conducted RCT and there was limited overlap of participants with complete metabolomics, genetics, and fatty acid data. Consequently, sample sizes for stratified analyses were small, particularly among noncarriers, limiting our ability to detect significantly enriched pathways. Plasma samples for metabolomics analysis were stored for 8 y before analysis. Although multiple studies have demonstrated that long-term storage of biospecimens does not substantially influence metabolic composition, others report altered concentrations of lipids, amino acids, and hexoses (45, 46). However, because all samples were stored for the same duration under similar well-controlled conditions, any potential deterioration of samples should be randomly distributed across treatment and carrier groups. Dietary intakes of total fat, n–6 fatty acids, and n–3 fatty acids differed between the analytic sample and the rest of the birth cohort. Although these differences may be due to chance, selection bias may also be an issue. We performed a sensitivity analysis excluding participants with outliers of dietary fat intake from the metabolomics analyses, and there were no substantial differences in the results.

Another important consideration is that the original intervention was carried out only during pregnancy. Therefore, several factors besides the intervention and maternal genotype may have influenced the offspring metabolome in the first 3 mo of postnatal life. DHA is secreted in the breast milk; thus, infant feeding practices and breast milk composition are a possible mechanism by which DHA was differentially available to the infant postdelivery (47). We previously showed that DHA and ALA breast milk concentrations significantly differed between treatment groups at 1 mo postpartum (48). Owing to limited sample availability, we were unable to investigate the influence of maternal FADS genotype on breast milk concentrations in the present analysis and were underpowered to examine further by infant feeding practices. Therefore, we cannot determine whether observed differences in the offspring metabolome resulted from metabolic programming during pregnancy or differences in breast milk composition. There were no differences in infant feeding practices by treatment group or maternal genotype, and most children were predominantly or partially breastfed. Increasing evidence suggests that lipids (49) and metabolomic profiles (50, 51) differ between breastfed and formula-fed infants. Further, other environmental exposures could confound results because linoleate metabolism has been recently found to be altered by air pollution (52) and per- and polyfluoroalkyl substances [PFAS (53), dichloro-diphenyl-trichloroethane (DDT) (54)].

Although we have demonstrated the importance of maternal FADS genotype, the role of offspring genotype in fatty acid metabolism during infancy remains unclear. Thus, lack of information on offspring genotype is also a potential limitation of this work. Moreover, endogenous synthesis of n–3 and n–6 LC-PUFAs is mediated by both desaturase and elongase enzymes; therefore, any other polymorphisms in the FADS and elongation of very-long-chain fatty acids (ELOVL) families may potentially alter biosynthesis of LC-PUFAs (55, 56). Future studies should incorporate both offspring genetic information and additional FADS and ELOVL polymorphisms to fully elucidate this complex relation. Finally, metabolite annotations provided by Mummichog and xMSannotator are tentative and for discovery purposes only. Databases used for annotation (HMDB, LipidMaps) typically have limited coverage of experimentally measured metabolites in experimental databases (57). Additional validation via MS/MS and comparison to an authentic chemical standard is needed to confirm the structural identity of metabolites.

Several strengths of this exploratory study should be noted. We addressed this research question using data from a large double-blind RCT with high compliance to the intervention in a well-characterized study population of mothers and their offspring. Our study participants were representative of a population with low dietary intakes of preformed DHA, high dietary intakes of n–6 fatty acids, and a high prevalence of genotypes associated with slower conversion into LC-PUFAs. Data collection and laboratory assay protocols were standardized, validated, and conducted by trained personnel within a clinical setting. In addition, we performed pathway enrichment analyses using Mummichog, an algorithm that utilizes the collective power of metabolic networks and is designed on the premise that true metabolites should show local enrichment on metabolic networks whereas false positives are distributed randomly. Improved confidence in interpretation of pathway enrichment and module analysis is supported by the use of Mummichog version 2.3.3, which has a relatively conservative approach for metabolite matching with statistical criteria for pathway identification, along with stringent inclusion criteria for reported results (23).

In summary, our findings provide new insights into the potential biological responses associated with prenatal DHA supplementation and demonstrate that the infant metabolome at 3 mo differs in response to prenatal DHA supplementation by maternal FADS SNP rs174602. Our results align with previous epidemiologic evidence suggesting that differences in FADS genotype may explain inconsistent results observed across prenatal DHA supplementation trials and further support the need to incorporate genetic analysis of FADS polymorphisms. However, given the large variation in genotype distribution across populations and limitations of this exploratory work, these findings need to be reproduced in independent cohorts. Whether metabolomic differences persist beyond infancy and have long-term consequences on offspring health require additional investigation; future studies should determine whether these pathways mediate associations between prenatal DHA exposure and offspring health. Ultimately, this work may provide insight on targeting interventions toward vulnerable populations that would uniquely benefit from DHA supplementation during pregnancy.

Notes

Supported by NIH grants HD087606 (to UR) and HD040399 (to UR), March of Dimes (to UR), the Nutricia Foundation (to UR), and Consejo Nacional de Ciencia y Tecnología (CONACYT), Mexico. IG-C is supported by National Heart, Lung, and Blood Institute grant HL137338-S1. BK is supported financially in part by the Else Kröner Fresenius Foundation and the Ludwig-Maximilians-Universität Medical Faculty. AB-V is supported by CONACYT Mexico grant SALUD-2008-01-87121.

Author disclosures: The authors report no conflicts of interest.

ADS is an Editor on The Journal of Nutrition and played no role in the Journal's evaluation of the manuscript.

Supplemental Tables 1–7 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/.

Abbreviations used: AA, arachidonic acid; ALA, α-linolenic acid; BF, breastfeeding; ELOVL, elongation of very-long-chain fatty acids; FADS, fatty acid desaturase; FDR, false discovery rate; HMDB, Human Metabolome Database; INSP, National Public Health Institute; LA, linoleic acid; LC-PUFA, long-chain polyunsaturated fatty acid; POSGRAD, Prenatal Omega-3 Fatty Acid Supplementation and Child Growth and Development; RCT, randomized controlled trial; SNP, single-nucleotide polymorphism; TCA, tricarboxylic acid.