Background:
Assisted reproductive technology (ART) relies on culture media to support preimplantation development. L-glutamine (Gln) is an essential component supporting embryonic metabolism, but its instability can lead to ammonia accumulation, prompting the use of the dipeptide structure, L-alanine-L-glutamine (GlutaMAX™). Increased Gln concentration has been associated with altered embryonic metabolism and adverse metabolic outcomes in offspring. The developmental origins of health and diseases (DOHaD) theory discusses how pre-implantation exposures induce developmental reprogramming that persists later in life. However, whether these changes produce detectable signatures that can identify the embryonic nutritional environment remains unknown. This study investigated whether machine learning (ML) models could distinguish culture formulations differing in Gln structure and concentration based on body weight trajectories.
Methods:
ICR mouse embryos were cultured in media supplemented with Gln or GlutaMAX™ at 1 mM (1Gln and 1MAX) or 2 mM (2Gln and 2MAX) before embryo transfer. A total of 190 mice, including naturally conceived controls, were recorded over 16 weeks. Following preprocessing, ML models were trained using longitudinal body weight trajectories and evaluated using accuracy, precision, recall, F1-score, confusion matrices, and multiclass receiver operating characteristic area under the curve (ROC-AUC).
Results:
Preimplantation glutamine exposure produced distinct postnatal growth trajectories that persisted throughout the 16-week observation period. ML models accurately classified offspring according to embryo culture conditions, with the highest accuracy model, GradientBoosting, achieving 86.4%. Cross-validation analysis revealed RandomForest demonstrated the best predictive performance. Feature importance analysis identified body weight at week 4 as one of the most influential predictors across most ML models.
Conclusion:
Postnatal growth trajectories retain signatures that distinguish the preimplantation nutritional environment, accurately identifying individual components. This integrative approach offers new opportunities to evaluate the long-term safety of ART culture systems and optimise embryo culture formulations to improve offspring health outcomes and may serve as a non-invasive phenotypic marker.