Design on What Was Actually Done

 Quick Analysis Plan Copied

 Long Analysis Plan

1. Cohort architecture (adapted from VDAART precedent)

The strongest available template is the VDAART ancillary study: 657 mother–child pairs, maternal third-trimester stool (n=120), infant stools at 3–6 months (n=265), 1 year (n=436), 3 years (n=506), with covariate adjustment for sex, race/ethnicity, site, and maternal age Longitudinal sampling schedule and covariates. For the dye cohort: recruit n≥650 pregnant women prenatally; collect maternal urine + plasma at each trimester (dye biomarkers: parent compounds and conjugates via LC-MS/MS); collect infant urine/plasma at birth (meconium/cord blood), 3–6 mo, 1 y, 2 y, 3 y, 5 y, 8 y; collect stool for shotgun metagenomics (not 16S — the VDAART authors note genus-level resolution was a limitation) and untargeted metabolomics at every visit.

2. Biomarker and metabolomics platform

Dye biomarkers in urine require targeted LC-MS/MS quantification of amino acid, catecholamine, lipid, and steroid-class metabolites with attention to hydration-status normalization — a documented challenge for urinary metabolomics Targeted urinary LC-MS/MS review. Plasma metabolomics should be interpreted against a consensus serum metabolome reference — 69,677 neutral mass registries from >100,000 acquisitions, with open code at github.com/shuzhao-li-lab/consensus_serum_metabolome — to improve annotation and avoid database gaps Consensus serum metabolome reference. Pre-register dye-specific MRM transitions, creatinine/specific-gravity correction for urine, and pooled-QC batch correction.

3. Neurodevelopmental testing schedule

Repeated observation-based behavioral assessment is preferred over purely parent-report instruments; the prenatal cortisol study showed observation-based protocol at 4 months discriminated infant reactivity (High Reactive OR≈1.6) but suffered ~attrition bias — participants lost to follow-up had higher cortisol/distress Observation-based infant reactivity protocol. Plan: Bayley-III at 6/12/24 mo, WPPSI/NEPSY at 4–5 y, WISC-V + executive function battery at 8 y; blinded certified testers; retention incentives and tracking to counter attrition bias.

4. Mediation analysis pipeline

Pre-specify the mediation model: dye exposure → gut microbiome/metabolome (e.g., Bacteroides depletion, sphingolipid/sialic acid metabolites) → neurodevelopmental score. Use SNF clustering of joint microbiome+metabolome data, then covariate-adjusted causal mediation (indirect-effect bootstrap p-values), as VDAART reported 33% mediation via SNF-M and 52–55% via Bacteroides for delivery mode, and breastfeeding-mediated fractions of 23/38/29% for early/active/any outcomes Reported mediation effect sizes.

5. Statistical and confounding controls

• Sex-stratified analyses mandatory: acute-exercise metabolomics showed 41.7% of fitness-correlated metabolites had opposite directions by sex, and lipid classes dominate sex differences Sex-dimorphic serum metabolomics.
• Power for maternal subsample: VDAART's n=120 maternal stools limited maternal-effect detection — oversample maternal biospecimens (target n≥400).
• Pre-register hypothesis, FDR control, and falsification criteria (e.g., mediation fails if indirect-effect CI crosses null after breastfeeding/antibiotic adjustment).
• Report null results; 16S→shotgun upgrade; avoid HARKing by locking the DAG before unblinding.

Confidence: high that this design follows the best available longitudinal precedent; the dye-specific exposure-outcome link itself is not established by the supplied evidence and requires the proposed cohort to test it.