Linking the metals to metabolism in recurrent pregnancy loss through untargeted metabolomics and machine learning

The association between recurrent pregnancy loss (RPL) and environmental exposure has attracted increasing attention. However, associations between RPL and metal exposure in northwestern China remained unclear.

This case-control study (318 RPL women, 326 controls) investigated associations between serum metal concentrations and RPL. Five machine learning algorithms identified significant variables. Bayesian kernel machine regression (BKMR) and quartile g-computation (Qgcomp) models assessed the combined effects of metal mixtures on RPL risk. Untargeted metabolomics integrated with metal exposure data explored potential mechanisms underlying metal-induced disruption.

Compared to controls, RPL women exhibited higher BMI (P<0.001) and elevated serum Ti, Cu, and Se levels (P<0.05), while controls had higher Li, V, Cr, Sr, Pb, Ni, Zn, and Fe (P<0.05). Machine learning algorithms (AUC = 0.99-1.0) identified V, Li, Cr, Ti, and Ni as top five discriminative metals. Mixture analyses (BKMR/Qgcomp) revealed a significantly increased RPL risk with mixed metals (β=0.37, 95% CI: 0.31-0.42). Ti contributed positively to this risk, whereas V contributed negatively after adjusted for con-founders. Metabolomic analysis in a subset (n=100) linked these metals primarily to perturbations in purine metabolism, pantothenate and CoA biosynthesis, retinol metabolism, and ubiquinone/terpenoid-quinone biosynthesis.

Our study provides valuable insights into the metabolic and environmental factors associated with RPL.