Early-mid gestation as a critical predictive window for infantile atopic dermatitis: A longitudinal multi-matrix...

Atopic dermatitis (AD) is a prevalent inflammatory skin disease with origins increasingly traced to the intrauterine environment. While maternal diet and immunity influence infant health, the optimal time window and specific metabolic signatures conferring protection against AD remain unclear. This study aims to characterize longitudinal perinatal metabolic profiles to identify stage-specific predictive biomarkers and critical windows for fetal metabolic exposure.

A longitudinal birth cohort study was conducted with 109 mother-infant pairs recruited between 2021 and 2023. Biological samples including maternal plasma (collected at early-mid gestation and pre-delivery), umbilical cord blood, and meconium were analyzed using untargeted LC-MS metabolomics. Infants were followed for 1 year to ascertain AD status (AD group n = 39; Healthy group n = 70). Logistic regression was performed to identify significant metabolites associated with health outcomes, and Random Forest classification was utilized to compare predictive performance across different developmental windows.

Distinct metabolic phenotypes were identified. In early-mid gestation maternal plasma, the Healthy group exhibited a significant enrichment of protective metabolites across 2 major classes: Polyunsaturated Fatty Acids (PUFAs), including 13S-HODE (OR = 0.11, P < 0.001) and Alpha-Linolenic Acid (OR = 0.51, P = 0.014); and Flavonoids, including Formononetin (OR = 0.33, P = 0.005), Equol (OR = 0.48, P = 0.012), and Quercetin (OR = 0.87, P = 0.027). Fewer metabolic alterations were observed in subsequent phases, such as ergocalciferol (OR = 0.56, P = 0.028) in late gestation and isovitexin (OR = 0.88, P = 0.037) in meconium. In cord blood, key predictive markers included phenyl acetate (OR = 0.28, P = 0.025), the antioxidant Ribose-1,5-bisphosphate (OR = 0.09, P = 0.017), and sterculic acid (OR = 1.23, P = 0.026). Notably, the latter 2 were consistently identified across both early-mid gestation maternal plasma and cord blood. Crucially, Random Forest modeling revealed that integrating early-mid gestation metabolites significantly improved AD prediction over the Baseline Clinical Model (AUC 0.831, 95% CI 0.73-0.93 vs. AUC 0.463, 95% CI 0.35-0.57), vastly outperforming late gestation metabolites (AUC 0.669, 95% CI 0.48-0.86). In neonates, the cord blood model (AUC 0.817, 95% CI 0.69-0.95) demonstrated superior predictive value compared to meconium (AUC 0.644, 95% CI 0.46-0.83). This predictive pattern remained highly robust in parsimonious models restricted to only 6 core metabolites.

This study highlights potential protective and risk perinatal metabolic profiles, suggesting that the early maternal metabolic milieu plays a critical role in 'priming' the infant immune system against AD and identifying early-mid gestation as an optimal window for early risk prediction. These predictive associations require confirmation in future mechanistic and interventional trials before informing targeted nutritional strategies.