Unsupervised Approaches to Placental Protein Clustering: Which Best Captures Signals Linked to Childhood Metabolic Health?

Background Placental signaling pathways regulate nutrient transport and fetal growth, with potential long-term consequences for offspring metabolic health. Most prior human studies have focused on individual placental markers, limiting insight into the role of coordinated activity across multiple pathways in relation to offspring outcomes. Objective To compare three unsupervised data reduction techniques for characterizing placental signaling patterns across multiple pathways and assess their associations with neonatal and early childhood adiposity and metabolic biomarkers. Design: Among 108 mother-child pairs from the Healthy Start cohort, we quantified 33 placental signaling proteins and their phosphorylated-to-total protein ratios involved in nutrient sensing, insulin/growth factor signaling, stress/inflammation, and mitochondrial biogenesis using Simple Western assays of term placental villus tissue. We applied consensus clustering, weighted gene correlation network analysis (WGCNA), and principal component analysis (PCA) to derive signaling scores. Model performance (AIC, R², and RMSE) was compared, and associations with offspring outcomes at age 4 years (%fat mass; fasting adiponectin, leptin, insulin, glucose, and lipids) were estimated using multivariable linear regression adjusted for offspring age, race and ethnicity, and maternal pre-pregnancy BMI. Results Consensus clustering outperformed PCA and WGCNA based on model fit statistics. The mTOR/AMPK cluster, characterized by activation of mTOR complex 1 and energy sensing (e.g., phosphorylated 4E-BP1, RPS6, AMPK), was inversely associated with childhood %fat mass (β: − 2.51%, 95% CI: − 4.44, − 0.58). The IGF/Mitochondrial Biogenesis cluster was positively associated with childhood triglyceride levels (17.90 [6.14, 29.60] mg/dL). Conclusion Consensus clustering provided superior model fit compared to WGCNA and PCA. Placental signaling clusters were associated with childhood adiposity and metabolic markers, supporting the relevance of coordinated placental activity to early metabolic programming in a healthy pregnancy cohort. These findings highlight the utility of unsupervised analytic approaches in placental biology and the potential of early-life placental markers to inform pediatric metabolic disease risk. However, which approach is best for summarizing complex protein data is likely dependent on the data structure, dimensionality, and covariance.