The study, published in the journal Metabolomics, drew on 65 mother–neonate pairs recruited between August 2014 and September 2017 from the labor and delivery unit of the University of Cincinnati Medical Center. All participants delivered full-term, single infants. At delivery, the team collected 15 milliliters of maternal venous blood and 10 milliliters of venous cord blood from each pair. Cord blood is a remarkably useful proxy for the fetal circulating environment: it is drawn within minutes of birth, and because the umbilical vein carries blood to the baby from the placenta, its chemical contents reflect the fetal metabolic state just before and during the delivery process, rather than after the profound physiological transition that follows separation from the placenta.

The analytical engine of the study was untargeted high-resolution metabolomics performed by liquid chromatography coupled to a Q-Exactive HF Orbitrap mass spectrometer. Each serum sample was extracted with acetonitrile spiked with stable isotope standards, and aliquots were run in triplicate under four complementary conditions: forward-phase hydrophilic interaction chromatography in positive electrospray ionization mode, and reverse-phase C18 chromatography in negative ionization mode. The instrument operated in full-scan mode at 120,000 resolution across a mass-to-charge range of 85 to 1,275. Across all 130 samples, the team detected 17,064 molecular features — ions characterized by their mass, retention time, and abundance. After stringent filtering for reliability, including exclusion of features with coefficients of variation at or above 30 percent and missingness exceeding 20 percent, 3,641 negative-mode and 5,244 positive-mode features entered the statistical analyses.

Quality control received careful attention. Samples were randomized into four analytical batches, with maternal and neonate samples deliberately not paired within batches to prevent technical artifacts from masquerading as biological differences. Standard reference materials from the National Institute of Standards and Technology, pooled plasma quality controls, and isotopically labeled internal standards were interspersed at the beginning, middle, and end of every batch, and the ComBat algorithm was applied to remove residual batch effects. These safeguards matter enormously in untargeted metabolomics, where the sheer dimensionality of the data — thousands of features measured in dozens of samples — creates ample opportunity for false positives if analytical drift goes uncontrolled.

The first statistical test was unsupervised. Principal component analysis of the log2-transformed feature data revealed clear separation between mothers and neonates in both ionization modes. In negative mode, the first two principal components explained 15.8 percent and 12.0 percent of the variance respectively, and the two groups formed distinct clusters with 95 percent confidence ellipses that barely touched. In positive mode the separation was present but less crisp, with the first two components explaining 16.0 percent and 9.5 percent of the variance and some overlap between groups — a pattern the authors attribute in part to the high dimensionality of the data and to a small set of outliers for whom no analytical error or batch bias could be identified. Crucially, the clustering arose despite the fact that the analytical batches mixed samples from both groups, arguing that the separation reflects genuine biology rather than laboratory artifacts.

To pin down which individual features drove this separation, the team then ran a metabolome-wide association study using a linear mixed-effects model. Each feature’s intensity was modeled as the outcome, with sample type — mother or neonate — as a fixed effect and a random intercept for each mother–neonate pair, thereby exploiting the paired design to control for dyad-level factors. The models were adjusted for five pre-specified covariates: maternal age, smoking status during pregnancy, cesarean delivery, neonatal sex, and number of prior pregnancies. False discovery was controlled with the Benjamini–Hochberg procedure at a permissive 20 percent threshold, a choice the authors justify by the exploratory nature of the analysis and the modest sample size relative to the number of correlated features. The result was striking: 3,720 features — 41.9 percent of those tested — differed significantly between mothers and neonates. In negative mode, 48.5 percent of features were significant, with the majority higher in neonates; in positive mode, 37.3 percent were significant, and a remarkable 85.4 percent of those were higher in the babies.

The pathway-level analysis is where the study delivers its most interesting insight. The significant features were mapped using the mummichog algorithm against the Homo sapiens MFN pathway library, with pathway significance assessed by permutation testing across 10,000 iterations. Across all significant features, 16 unique pathways were implicated, spanning five major classes of metabolism: amino acids and nucleotides, vitamins, lipids and fatty acids, energy metabolism, and a residual category. Among the amino acid and nucleotide pathways, the enriched set included the degradation of the branched-chain amino acids valine, leucine, and isoleucine, along with tyrosine, tryptophan, lysine, beta-alanine, arginine and proline, and glycine–serine–alanine–threonine metabolism. Energy pathways including propanoate and butanoate metabolism, vitamin pathways for D3, B9, and B3, de novo fatty acid synthesis, and the urea cycle also came out significant.

The asymmetry between the two generations is the headline finding. When the team restricted the pathway analysis to features that were significantly higher in mothers alone, seven pathways emerged as enriched — prominently including three lipid-related pathways: glycerophospholipid metabolism, fatty acid activation, and de novo fatty acid biosynthesis, alongside glycosphingolipid metabolism, butanoate metabolism, and two amino acid pathways. These lipid signals are consistent with well-documented physiological adaptations of pregnancy, in which the mother ramps up glucose production and develops insulin resistance to keep a steady fuel supply flowing to the fetus, while mobilizing and synthesizing fatty acids for fetal tissue construction. But when the analysis was restricted to features higher in neonates, no pathway reached statistical significance at all. The neonatal metabolic difference, in other words, is not concentrated in any single biochemical route; it is distributed diffusely across the global metabolic profile — a pattern the authors suggest is consistent with the rapid, system-wide metabolic reprogramming the newborn must execute at birth, when the continuous placental glucose supply ends and independent energy regulation begins.

Study co-author Katherine E. Manz of the University of Michigan and colleagues frame the work as a baseline map rather than a clinical verdict. The evidence supports the conclusion that maternal metabolomes are not adequate proxies for fetal metabolomes — a finding with practical consequences. Epidemiologists studying how environmental chemicals, nutrition, and maternal health shape fetal development frequently measure metabolites in maternal blood and infer fetal exposures; this study suggests those inferences carry real uncertainty. The team’s related work in the same cohort reinforces the point: in studies of polycyclic aromatic hydrocarbons, 43 of 59 exposure-associated metabolic pathways were distinct between mothers and neonates, and endocrine-disrupting chemicals likewise produced divergent metabolic signatures in the two generations, with only partial overlap.

The authors are candid about limitations. The cohort of 65 pairs is modest, limiting statistical power, and the 20 percent false discovery threshold, while defensible for exploratory work, means some reported features may be spurious. Unmeasured confounders — maternal diet in particular, along with unknown environmental exposures — could explain some of the dyad-level variation, and because fetal blood cannot be sampled in utero, the study cannot determine when during gestation the maternal–fetal metabolic divergence first appears. What it can establish, and does, is that at the moment of birth, before any postnatal adaptation begins, mother and baby are already biochemically distinct in measurable, pathway-informative ways. As untargeted metabolomics becomes cheaper and more widespread, the practice of treating maternal blood as a window onto fetal chemistry will need recalibration — and the distinct metabolic logic of the newborn, still largely unorganized into discrete pathways at birth, represents a frontier for understanding the critical transition from fetal to independent life.

Subject of Research: Comparative untargeted metabolomics of paired maternal and umbilical cord blood samples at delivery to determine whether maternal metabolic profiles can proxy fetal metabolic profiles.

Article Title: Comparative metabolomics of paired maternal–cord samples identifies pathway-level divergence

Article References: Van Winckel, A., Gupta, M., Klein, R., Puvvula, J., Braun, J. M., Pennell, K. D., DeFranco, E. A., Ho, S.-M., Leung, Y.-K., Vuong, A. M., Percy, Z., Bhashyam, P., Lee, R., Jones, D. P., Tran, V., Kim, D. V., Huang, S., Chen, A., & Manz, K. E. (2026). Comparative metabolomics of paired maternal–cord samples identifies pathway-level divergence. Metabolomics, 22(5), Article 157. https://doi.org/10.1007/s11306-026-02528-z

Image Credits: AI Generated

DOI: 10.1007/s11306-026-02528-z

Keywords: metabolomics, pregnancy, umbilical cord blood, neonate, maternal metabolism, mass spectrometry, metabolic pathways, fetal development, lipid metabolism, principal component analysis, MWAS, perinatal health