(64) Volumetric segmentation data set and deep learning analysis of differentiation marker expression in blastomeres of at different stages of blastocyst development.

Lyda Yuliana Parra-Forero¹, Romana A. Nowak¹

¹Department of Animal Sciences, College of Agricultural, Consumer and Environmental Sciences, University of Illinois, Urbana, IL, USA

Abstract Text:

The expression of differentiation markers, such as GATA6, OCT4, and CDX2, in blastocysts, has been studied to assess the potential harmful effects of various substances on early embryonic development. These proteins are crucial for the formation of the epiblast, hypoblast, and trophectoderm, which ultimately give rise to tissues and the placenta. The blastomere, as the functional unit of the blastocyst, undergoes significant biochemical, anatomical, and genetic changes during maturation, influencing its specialization. Artificial intelligence models have recently been integrated into embryo selection and evaluation, with potential applications in clinical reproduction, pharmacology, and toxicology. This study presents preliminary results from developing a non-invasive, volumetric graphical model for evaluating the expression of differentiation markers (GATA6, OCT4, and CDX2) in blastomeres within mouse blastocysts. Using surface segmentation, we generated a volumetric spatial representation, enabling the precise identification of blastomere boundaries and the in situ expression of these proteins through deep learning analysis with Imaris 10.1 software. Blastocysts were obtained from an ex vivo model via oviductal lavage at day 4.5 of post-mating development. Immunofluorescence was used to assess protein expression levels while maintaining the morphological integrity of the blastocysts. The blastocysts were categorized based on the number of blastomeres into three groups: Group 1 (64-100 blastomeres; n= 412), Group 2 (more than 100 blastomeres with zona pellucida; n= 478), and Group 3 (hatching, no zona pellucida; n= 538). This approach allowed the collection of 30 quantitative variables per blastomere, including morphometric measures such as area (μm²), diameter (μm), ellipticity (oblate/prolate), sphericity, and volume (μm³), as well as intensity-based protein expression measurements for GATA6, OCT4, and CDX2, including max intensity, mean intensity, median intensity, min intensity, standard deviation, and intensity sum. Our deep learning-based method revealed significant differences in the intracellular assessment of blastomeres among the three groups, both in morphological parameters and in the expression of cell differentiation markers (GATA6, OCT4, and CDX2). These markers exhibited distinct patterns of co-expression at various stages of blastocyst development, suggesting a potential specialization based on the blastomere's location within the inner cell mass or the trophectoderm. These changes were accompanied by significant morphological differences in parameters such as area, volume, and sphericity. These data offer valuable insights into protein expression at different stages of blastomere development within a physiologically relevant environment. By preserving anatomical information, our measurements correlate closely with cellular compaction, excision, and differentiation processes, providing a more accurate reflection of these biological events. Given the large amount of numerical data generated, bioinformatics approaches are essential to develop a model with high sensitivity and specificity. These results lay the foundation for future advances in these technologies.