Retrospective analysis: reproducibility of interblastomere differences of mRNA expression in 2-cell stage mouse...

What is the prevalence, reproducibility and biological significance of transcriptomic differences between sister blastomeres of the mouse 2-cell embryo?

Sister 2-cell stage blastomeres are distinguishable from each other by mRNA analysis, attesting to the fact that differentiation starts mostly early in the mouse embryo; however, the interblastomere differences are poorly reproducible and invoke the combinatorial effects of known and new mechanisms of blastomere diversification.

Transcriptomic datasets for single blastomeres in mice have been available for years but have never been systematically analysed together, although such an analysis may shed light onto some unclarified topics of early mammalian development. Two unknowns that remain are at which stage embryonic blastomeres start to diversify from each other and what is the molecular origin of that difference. At the earliest postzygotic stage, the 2-cell stage, opinions differ regarding the answer to these questions; one group claims that the first zygotic division yields two equal blastomeres capable of forming a full organism (totipotency) and another group claims evidence for interblastomere differences reminiscent of the prepatterning found in embryos of lower taxa. Regarding the molecular origin of interblastomere differences, there are four prevalent models which invoke (1) oocyte anisotropy, (2) sperm entry point, (3) partition errors of the transcript pool and (4) asynchronous embryonic genome activation in the two blastomeres.

Seven transcriptomic studies published between 2011 and 2017 were eligible for retrospective analysis, since both blastomeres of the mouse 2-cell embryo had been analysed individually regarding the original pair associations and since the datasets were made available in public repositories. Five of these studies, encompassing a total of 43 pairs of sister blastomeres, were selected for further analyses based on high interblastomere correlations of mRNA levels. A double cut-off was used to select mRNAs that had robust interblastomere differences both within and between embryos (hits). The hits of each study were compared and contrasted with the hits of the other studies using Venn diagrams. The hits shared by at least four of five studies were analysed further by bioinformatics.

PubMed was systematically examined for mRNA expression profiles of single 2-cell stage blastomeres in addition to publicly available microarray datasets (GEO, ArrayExpress). Based on the original normalizations, data from seven studies were screened for pairwise sample correlation at the gene level (Spearman), and the top five datasets with the highest correlation were subjected to hierarchical cluster analysis. Interblastomere differences of gene expression were expressed as a ratio of the higher to the lower mRNA level for each pair of blastomeres. A double cut-off was used to make the call of interblastomere difference, accepting genes with mRNA ratios above 2 when observed in at least 50% of the pairs, and discarding the other genes. The proportion of interblastomere differences common to at least four of the five datasets was calculated. Finally, the corresponding gene, pathway and enrichment analyses were performed utilizing PANTHER and GORILLA platforms.

An average of 17% of genes within the datasets are differently expressed between sister blastomeres, a proportion which falls to 1% when considering the differences that are common to at least four of the five studies. Housekeeping mRNAs were not included in the 17% and 1% gene lists, suggesting that the interblastomere differences do not occur simply by chance. The 1% of shared interblastomere differences comprise 100 genes, of which 35 are consistent with at least one of the four prevalent models of sister blastomere diversification. Bioinformatics analysis of the remaining 65 genes that are not consistent with the four models suggests that at least one more mechanism is at play, potentially related to the endomembrane system. Although there are many dimensions to the issue of reproducibility (biological, experimental, analytical), we consider that the sister blastomeres are poised to escape high interblastomere correlations of mRNA levels, because at least five sources of diversity superimpose on each other, accounting for at least 25 = 32 different states. As a result, interblastomere mRNA differences of a given 2-cell embryo are necessarily difficult to reproduce in another 2-cell embryo.

Data were as provided by the original studies (GSE21688, GSE22182, GSE27396, GSE45719, GSE57249, E-MTAB-3321, GSE94050).

The original studies present similarities (e.g. fertilization in vivo after ovarian stimulation) as well as differences (e.g. mouse strains, method and timing of blastomere separation). We identified robust mRNA differences between the sister blastomeres, but these differences are underestimated because our double cut-off method works with thresholds and affords more protection against false positives than false negatives. Regarding the false negatives, transcriptome analysis may have captured only part of the interblastomere differences due to: (1) the 2-fold cut-off not being sensitive enough to detect the remaining part of the interblastomere differences, (2) the detection limit of the transcriptomic methods not being sufficient, or (3) interblastomere differences being oblivious to transcriptomic identification because transcriptional changes are oscillatory or because differences are mediated non-transcriptionally or post-transcriptionally. Regarding the false positives, it seems unlikely that a difference was found just by chance for the same group of transcripts due to the same technical error, given that different laboratories produced the data.

It is clear that the sister blastomeres are distinguishable from each other by mRNA analysis even at the 2-cell stage; however, efforts to identify large stable patterns may be in vain. This elicits thoughts about the wisdom of adding new transcriptomic datasets to the ones that already exist; if all transcriptomic datasets produced so far show a reproducibility of 1%, then any future study would probably face the same issue again. Possibly, a solid identification of the 'large stable pattern that should be there but was not found' requires an even larger dataset than the sum of the seven datasets considered here. Conversely, small stable patterns may be easier to identify, but their biological relevance is less obvious. Alternatively, interblastomere differences may not be mediated by nucleic acids but by other cellular components.

This study was supported by the Deutsche Forschungsgemeinschaft (grant DFG BO 2540-4-3 to M.B. and grant NO 413/3-3 to V.N.). The authors declare that they have no competing financial interests.