Prenatal exposure to environmental contaminants and cord serum metabolite profiles in future immune-mediated diseases

Exposure to environmental contaminants contributes to the global burden of many chronic diseases. Over recent decades, autoimmune diseases have increased in prevalence and often manifest in early childhood and during pregnancy. Their etiology involves genetic predisposition, environmental and maternal factors, and their interactions. Prenatal exposure occurs during sensitive developmental windows; PFAS and other contaminants have been linked with abnormal metabolism and later autoimmune conditions (e.g., T1D, CD, IBD). While autoimmune diseases share some pathogenic mechanisms and genetic risk (e.g., class II HLA overlaps for T1D and CD), underlying mechanisms remain poorly understood, and additional perinatal factors (birthweight, gut microbiome, maternal diet) may contribute. Given the potential impact of contaminants and maternal factors, the study hypothesized that prenatal exposure to environmental contaminants alters the cord serum metabolome and may contribute to future autoimmune diseases in the ABIS cohort. The work aimed to: (i) examine exposure levels and differences between controls and future cases, (ii) assess associations of contaminant exposure with cord serum metabolic profiles, and (iii) evaluate the impact of exposure on cord serum metabolites. The rationale also recognizes strong individual variability and gene–environment interactions, as well as shared risk factors across autoimmune diseases and the linkage between T-cell functionality and cellular metabolic programs.

Design and cohort: Case-control study nested within All Babies in Southeast Sweden (ABIS), a general population cohort of ~17,000 children born 1997–1999, prospectively followed. Included N=62 children who later developed one or more autoimmune/inflammatory diseases (CD n=28, IBD n=7, HT n=6, JIA n=9, T1D n=12) and matched controls (N=268) matched for sex and age at diagnosis. Average ages at diagnosis: CD 11.5 y, IBD 16 y, JIA 15 y, HT 16 y, T1D 13.5 y. Cord blood collected at birth was used for analysis. Demographics included birthweight, maternal age, gestational age, BMI; birthweight-for-gestational-age (BWGA) Z-score was computed.

Analytical platforms: Integrated exposomics and metabolomics using UHPLC-QTOFMS. Two extractions applied: (1) lipids; (2) polar/semi-polar metabolites. Total of 545 lipids and 3417 polar metabolites measured. Identification followed MSI levels 1–2 using in-house MS/MS libraries and authentic standards when available. Quantification via calibration curves; unknowns reported as normalized peak areas. QC included blanks, authentic standards, extracted standards, pooled QC, and NIST SRM 1950 reference plasma; features with >30% RSD in pooled QC or high in blanks (sample:blank <5) were excluded.

Lipidomics: Serum (10 µL) extracted with CHCl3:MeOH (2:1) containing internal standards; UHPLC-QTOFMS on Waters ACQUITY UPLC BEH C18 (2.1×100 mm, 1.7 µm) with eluent A (10 mM NH4Ac in H2O + 0.1% FA) and B (10 mM NH4Ac in ACN:IPA 1:1 + 0.1% FA); gradient 35–100% B; flow 0.4 mL/min. MZmine 2.53 processing with specified parameters (mass detection, chromatogram building, deconvolution, isotopic grouping, alignment, gap filling, database search). Quantification used 7-point internal calibration (0.1–5 µg/mL) with class-specific standards; for unknown lipids, normalized peak areas used.

Polar metabolites and contaminants: Serum (40 µL) precipitated with MeOH/H2O (1:1) containing labeled internal standards; analysis on Agilent 1290/6545 QTOFMS with dual ESI, BEH C18 column; mobile phases A (2 mM NH4Ac in H2O:MeOH 7:3) and B (2 mM NH4Ac in MeOH); gradient to 100% B; negative ion mode m/z 100–1700. Processing mirrored lipidomics. Quantitation via 6-point calibrations (e.g., PFOA 3.75–120 ng/mL; bile acids 20–640 ng/mL; polar metabolites 0.1–80 µg/mL). A panel of bile acids and polar metabolites used for quantification.

Exposure assessment and clustering: Detected 20 contaminants (including PFAS, bisphenol S, mycotoxins). Contaminant data were log2-transformed and autoscaled. Model-based clustering (mclust) grouped contaminants into four clusters (CC1–CC4): CC1 (Bisphenol S, Deoxynivalenol, Monobutyl phthalate, α-Zearalanol), CC2 (Ethyl-, Methyl-, Propylparaben), CC3 (PFHxS and PFHxS branched), CC4 (seven PFAS and fragments). Metabolite features were previously clustered into 8 lipid clusters (LCs) and 12 polar metabolite clusters (PCs).

Statistics: Analyses in R 4.1.2. Pairwise Spearman correlations among contaminants, LCs, PCs, and covariates (Z-score, maternal age, BMI) visualized with corrplot; separate matrices for controls and cases. Partial correlations estimated using Debiased Sparse Partial Correlation (DSPC) and visualized as chord diagrams (circlize) with |r| ≥ 0.14 across contaminants, LCs, PCs, covariates. Univariate analyses: Subjects stratified into quartiles (Q1–Q4) by exposure; two-way ANOVA (factors: quartile and group) with post-hoc Tukey to compare metabolite clusters and lipids across exposure levels. Classification: Logistic ridge regression (LRR; glmnet) adjusted for Z-score, maternal age, BMI, with 10-fold CV for λmin; performance via bootstrapped AUC (10,000 resamples; 80/20 split), downsampling to address class imbalance, and stepwise recursive feature elimination. Regression: Linear ridge regression (LR) modeling contaminants as predictors and metabolite clusters/individual metabolites as responses; 10-fold CV for λmin; performance by mean R2 over 10,000 resamples. Predictor ranks based on normalized ridge coefficients (LR) or unit absolute odds ratio differences (LRR). Pathway analysis: MetaboAnalyst 5.0 Functional Analysis (MS Peaks) on negative-mode LC-HRMS features (10 ppm tolerance) for DON-associated changes, adjusted for Z-score, maternal age, BMI; enrichment via Mummichog against MFN model and KEGG; pathway impact via MetPA.

• Twenty contaminants were detected in cord blood, including several PFAS, Bisphenol S (BPS), and mycotoxins such as Deoxynivalenol (DON).
• Significant differences (p<0.05) in concentrations between cases and controls were observed for Perfluorooctanoic acid Branched 2, Environmental Contaminant 1 (putatively mOPFLCA n=2; identity unconfirmed), Perfluorooctanoic acid Linear 1, Perfluorooctanoic acid Linear 2, and Methylparaben.
• Logistic ridge regression using contaminant concentrations modestly distinguished cases from controls: mean AUC 0.65 (95% CI 0.63–0.67) with all predictors and 0.67 (95% CI 0.66–0.68) with recursive feature elimination.
• Spearman and partial correlation analyses showed multiple associations among exposures, metabolite clusters, and covariates. In partial correlation networks, controls exhibited stronger direct associations between contaminants and metabolic profiles than cases, while covariates (Z-score, maternal age, BMI) showed stronger associations with exposure and metabolic profiles in cases.
• Quartile-based analyses indicated greater and more frequent differences for polar metabolite clusters than for lipid clusters between highest (Q4) and lowest (Q1) exposure levels. For CC1, significant Q4–Q1 differences were observed for LC3, LC4, LC7; for CC3 (PFHxS, PFHxSBr), significant differences for LC5 and LC6.
• Linear ridge regression demonstrated that polar metabolite clusters were more impacted by exposure than lipid clusters: PC2 (R2=0.72), PC6 (R2=0.53), PC4 (R2=0.52), PC1 (R2=0.48), PC10 (R2=0.48), PC11 (R2=0.32). At the metabolite level, amino acids (e.g., tryptophan) and other features (e.g., a PC11 feature annotated as 3-Chlorothieno[2,3-b]thiophene-2-carbonyl chloride) were significantly associated with exposure. Lipid cluster LC6 (mainly triacylglycerols with MUFA/PUFA) showed a weak association (R2=0.04).
• DON and BPS emerged as top linear predictors for shifts in cord serum metabolite clusters (PC2, PC11) and specific metabolites, with additional roles for specific PFAS.
• Pathway enrichment for DON-associated changes revealed both shared and group-specific pathways: MFN pathways in controls included Tyrosine and Tryptophan metabolism; in cases included Glutathione metabolism, Alanine and Aspartate metabolism, and Glycerophospholipid metabolism. KEGG showed common Aminoacyl-tRNA biosynthesis and Glycine/Serine/Threonine metabolism, with several pathways specific to each group.
• Overall, higher prenatal contaminant exposures were associated with altered amino acid and free fatty acid profiles in cord serum, suggesting metabolic perturbations linked to future immune-mediated diseases.

The study addressed the hypothesis that prenatal exposure to environmental contaminants alters the neonatal cord serum metabolome in ways relevant to future immune-mediated diseases. Using integrated exposomics–metabolomics, the study detected multiple contaminants at birth and identified statistically significant but modest exposure differences between children who later developed autoimmune/inflammatory diseases and controls. While exposures alone had limited discriminatory power (AUC ~0.65–0.67), multivariable associations revealed clear links between exposures (notably DON, BPS, and PFAS) and cord serum metabolic profiles. Polar metabolites were more strongly affected than lipids, with amino acid metabolism (e.g., tryptophan) particularly impacted. DON-associated pathway differences between cases and controls highlighted metabolic processes tied to immune function, oxidative stress, and lipid remodeling (e.g., glutathione, amino acid pathways, and glycerophospholipid metabolism). The findings align with existing evidence that PFAS and endocrine disruptors influence lipid and amino acid metabolism and extend this by implicating DON and BPS as key exposures associated with neonatal metabolic shifts relevant to immune dysregulation. These results support a model in which prenatal environmental exposures, interacting with maternal and genetic factors (e.g., HLA risk), perturb neonatal metabolism in pathways integral to T-cell function and immune programming, potentially contributing to subsequent autoimmune disease risk.

Prenatal exposure to specific environmental contaminants is associated with distinct alterations in cord serum metabolomes that may predispose to immune-mediated diseases. In the ABIS cohort, DON, BPS, and selected PFAS were top predictors of metabolic shifts, particularly in polar metabolite clusters and amino acid pathways, while lipid associations were weaker. Pathway analyses for DON suggested both shared and group-specific metabolic perturbations in controls and future cases, implicating immune-relevant pathways (e.g., tryptophan and glutathione metabolism, glycerophospholipids). Although exposure differences alone modestly discriminated cases from controls, the integrated exposomics–metabolomics approach revealed biologically meaningful associations supporting the role of prenatal exposures in immune programming. Future work should include larger disease-specific samples, longitudinal exposure assessments (maternal and postnatal), mechanistic studies to elucidate causal pathways, and integration with genetic and microbiome data to dissect gene–environment interactions.

• Small sample sizes within each individual disease group limited disease-specific analyses.
• Lack of maternal exposure measurements and longitudinal exposure data between birth and disease onset constrained interpretation of exposure timing and dynamics.
• Modest effect sizes for exposure differences suggest exposures are contributory but not sole risk factors; multifactorial influences (genetics, environment, lifestyle) likely interact.
• One putative contaminant (Environmental Contaminant 1, suspected mOPFLCA n=2) lacked confirmation by authentic standard, limiting definitive identification.
• Median ages of diagnosis were relatively high compared with some high-risk cohorts, which may affect comparability.
• Potential residual confounding despite adjustment for BWGA Z-score, maternal age, and BMI.