A fundamental process in embryonic development is the first cell fate decision, when cells take on distinct lineage identities for the first time. In mammals, this separates embryonic inner cell mass (ICM) from extra-embryonic trophectoderm (TE) during pre- implantation development. In the mouse, the process is classically attributed to the consequences of apical-basal polarity, which forms at the 8-cell stage: those cells which retain the apical domain after cell divisions are specified as TE, and the rest as ICM. However, more recent research has shown that early molecular heterogeneities between cells before polarisation can also bias cell fate. The existence of different models calls into question the role of polarisation in the first cell fate decision.

The first part of this study focuses on reconciling the polarity and heterogeneity models. It was previously thought that polarisation occurs only at the late 8-cell stage. By studying its timing in detail, it is possible to split cells into ‘normal polarising’ (NP) cells which polarise at the late 8-cell stage, and ‘early polarising’ (EP) cells at the early 8-cell stage, the latter found in approximately 20% of embryos. Although EP cells follow the canonical polarity pathway — involving the critical factors Tfap2c, Tead4 and RhoA — they have molecular and morphological differences from NP cells and are biased towards symmetric divisions and TE fate. Blastomeres with low activity of the arginine methyltransferase CARM1 prior to polarisation are known to be biased towards TE, and inhibition of CARM1, or overexpression of its substrate BAF155, increases the frequency of early polarisation. Thus, this study proposes that early heterogeneities influence cell fate by altering the timing of polarisation.

The final part of this study addresses the ability to detect polarity. Tracking polarisation over time currently requires invasive fluorescence imaging. Here, artificial intelligence is used to detect whether an embryo is polarised from unstained images, after training based on bright-field movies annotated using the corresponding fluorescence channel. The resulting model has an accuracy of 85% for detecting polarisation, significantly outperforming human volunteers trained on the same data (61% accuracy). Taken together, this study advances our understanding of polarisation and its role in the first cell fate decision, while also providing a tool for further investigation.