Nuclear segmentation and quantification
Stardist 3D employs a neuronal network to separate densely packed nuclei in 3D image stacks33. The model was trained with 5 manually annotated image stacks, and thereafter used to segment placental nuclei.
Quantitative analysis of predicted label maps was performed in FIJI with 3D ROI manager28. Nuclear density was defined as the number of nuclei relative to the villous volume (#nuclei/villous volume in mm3).
Segmentation of syncytial knots was performed with KnotMiner , available at https://github.com/stegmaierj/KnotMiner.KnotMiner was built as an extension package for the MATLAB toolbox SciXMiner34. The volume of knots was normalised to the volume of the corresponding villus to determine knot fraction. Knot shape was described by elongation index (EI=intermediate principal axis/ longest principal axis) and flatness index (FI=Short principal axis/intermediate principal axis).

Statistical analysis

Data are given as mean ±SD. GraphPad prism 5.0 was used for statistical analysis and graph representation. ANOVA with Bonferroni post-test was used to assess significant differences in parameters that are normally distributed. In case data was assumed not normally distributed Kruskal Wallis with Dunn’s post-test was employed. If normality could not be statistically determined (by Shapiro-Wilk test) because of too small sample size, the assumption of normality was made. P values of <0.05 were considered statistically significant.