Two prenatal chemical studies, opposite answers, and why both count
One found specific phenanthrene metabolites associated with fetal growth restriction and chased the mechanism. The other tested a 24-metabolite mixture against ADHD in 3,962 children and found nothing.
Environmental epidemiology has a publication problem that is easy to state and hard to fix: studies that find an association get written up as discoveries, and studies that find nothing tend not to get written up at all. Two papers appearing in Environment International three days apart are a useful corrective, because one of each landed.
The one that found something
The first is a nested case-control study within a prospective birth cohort in Guangzhou, China, comparing 100 cases of fetal growth restriction against 194 matched controls [s1].
Polycyclic aromatic hydrocarbons are ubiquitous environmental pollutants that can cross the placental barrier, but evidence linking specific PAH metabolites measured inside the body to fetal growth restriction has been scarce and mechanistically unclear [s1]. The researchers quantified maternal urinary hydroxylated and aminated PAH metabolites in first-trimester samples by high-performance liquid chromatography with tandem mass spectrometry, then used conditional logistic regression with machine learning to identify which metabolites carried the association and rank their contributions [s1].
Two came out on top: elevated 1+9-hydroxyphenanthrene and 9-aminophenanthrene were significantly associated with increased risk and were consistently identified as the strongest contributors [s1].
What distinguishes this paper from the standard exposure-outcome study is what happened next. Network toxicology analysis identified 172 genes overlapping between these phenanthrene metabolites and fetal growth restriction, converging on six core proteins — EGFR, STAT3, NFKB1, PPARG, ESR1 and PTGS2 — linked to oxidative stress, inflammation and cellular homeostasis, processes involved in placental development [s1]. Molecular docking predicted binding affinities of -5.8 kcal/mol or stronger between the metabolites and those targets [s1]. And in vitro work in trophoblast cells found that hydroxylated phenanthrene metabolites downregulated EGFR while 9-aminophenanthrene upregulated PTGS2 [s1].
That is a chain of evidence rather than a single correlation: an epidemiological association, a computational prediction of which proteins would be engaged, and a cell-based test showing that two of the predicted targets moved. Each link is weak on its own. Together they constitute a specific, falsifiable proposal about mechanism, which is more than most studies of this kind offer.
The limits remain real. A hundred cases is a small case-control study. First-trimester spot urine represents recent exposure to non-persistent compounds, not exposure across gestation. Trophoblast cells in culture are not a placenta. And a computational pipeline that starts from a hit list will generally return plausible-looking pathways, because inflammation and oxidative stress are involved in nearly everything.
The one that found nothing
The second paper is larger, older in provenance, and answers in the negative.
It used 3,962 mother-child pairs from the Environmental influences on Child Health Outcomes cohort, across 18 sites, with data spanning 2001 to 2021 [s2]. Maternal spot urine samples collected during pregnancy were analysed at the Wadsworth Center Human Health Exposure Analysis Resource laboratory using a single assay workflow — which matters, because inconsistent laboratory methods across sites are a standard source of noise in multi-site cohorts [s2].
Twenty-four urinary metabolites with detection frequency above 70% were retained and grouped into molar sums: high- and low-molecular-weight phthalates, polycyclic aromatic hydrocarbons, bisphenol S, organophosphate esters, and organophosphate and pyrethroid pesticides [s2]. Child ADHD symptoms came from the Child Behavior Checklist Attention-Deficit/Hyperactivity Problems subscale, in both the preschool and school-age versions, with scores at or above the 85th and 95th percentiles classified as borderline and clinical thresholds [s2]. Quantile g-computation estimated the change in outcome associated with a one-quartile increase across all mixture components at once, using negative binomial models for raw scores and binomial models for the thresholds [s2].
Prenatal exposure to the mixture was not associated with child ADHD symptoms in preschool-aged or school-aged children in adjusted models, for either continuous raw scores or the percentile thresholds [s2]. The authors state that their findings were not robust to full adjustment, and note that future work delineating windows of vulnerability and trajectories of ADHD-related symptoms may help clarify mixed findings in the literature [s2].
Why the null is the more useful paper
The temptation is to treat the second study as an absence of news. It is the opposite. A null in a sample of 3,962, using a harmonised single-laboratory assay and a mixture method designed for correlated exposures, constrains the size of any real effect far more tightly than a positive finding in 294 people establishes one.
The phrase to hold onto is "not robust to full adjustment" [s2]. It implies associations that appeared in less-adjusted models and disappeared as confounders were added — the exact pattern that produces a positive literature when analysts stop adjusting early, and the exact reason mixed findings accumulate in this field.
It is also worth being precise about what the null does not say. It concerns a specific mixture of non-persistent chemicals, measured in spot urine, against a parent-reported ADHD symptom scale [s2]. Non-persistent compounds have short biological half-lives, so a single spot sample is a noisy proxy for exposure across pregnancy — and measurement noise pushes results toward the null. A finding of nothing under a noisy exposure measure is weaker evidence of absence than a finding of nothing under a precise one.
What the pair demonstrates
Read together, the two papers show the two directions this field can go. One narrows from a broad class to two specific metabolites and then tests a mechanism [s1]. The other takes a broad mixture, measures it consistently, and reports that the expected association was not there [s2]. Both are more informative than another study reporting that a chemical class is associated with a developmental outcome.
Sources
- [s1] Prenatal urinary phenanthrene metabolites associated with fetal growth restriction: a nested case-control study with integrated computational toxicology. Environment International, 26 August
- [s2] Prenatal exposure to a mixture of non-persistent endocrine-disrupting chemicals and attention and hyperactivity symptoms in children in the ECHO cohort. Environment International, 29 August
Sources
- Prenatal urinary phenanthrene metabolites associated with fetal growth restriction: a nested case-control study with integrated computational toxicology — Environment International , August 26, 2026
- Prenatal exposure to a mixture of non-persistent endocrine-disrupting chemicals and attention and hyperactivity symptoms in children in the ECHO cohort — Environment International , August 29, 2026
More on
Stimulants beat behaviour therapy for ADHD at 14 months. By 8 years the gap was gone.
The MTA trial randomised 579 children and found medication management clearly superior short-term. Its own long-term follow-up found the treatment groups no longer differed by 6 to 8 years.
Athletes with ADHD face higher concussion odds and longer recovery, a review finds
The paper is a literature review, not a new trial — it synthesizes existing research rather than testing anything new. The authors argue return-to-play protocols may need to treat ADHD as a distinct variable.
FDA approves a new ADHD drug. It still needs the DEA's sign-off before anyone can buy it
Simtriyo works on a different neurotransmitter pathway than existing stimulants. But its own manufacturer classifies it as a CNS stimulant, so a scheduling review stands before pharmacy shelves.
Intellectual disability fell across South America. Autism diagnoses barely moved.
A single-author analysis of Global Burden of Disease estimates finds the region's neurodevelopmental picture split in two — and argues the flat autism curve reflects diagnostic capacity, not risk.