Figure 4.

(A–B) PCA plots of the tissue frame of reference (TFOR) and an cell frame of reference (CFOR) shape spaces of the pLLP. Each point represents a cell and each color represents a different primordium. Selected example cells are shown as point clouds, illustrating that meaningful properties are encoded in PCs, namely cell orientations (A) and cell sphericity and surface smoothness (B). (C) Explained variance ratios of principal components. (D) t-SNE embedding of the shape space showing the absence of obvious clusters as already seen with PCA in (A–B). Colors indicate different primordia as in (A–B). (E–F) Bigraph visualizations of correlations between principal components of the embedded space (bottom nodes) and a set of engineered features (top nodes). Any edge between two nodes indicates a correlation with Pearson's r > abs(0.3) and stronger edges indicate stronger correlations. A blue hue implies a positive and a red hue a negative correlation. These correlations together with manual inspection as shown in (A–B) allow the biological meaning of embedded features to be determined. (G) Consensus tissue maps of shape space PCs. The contour map represents the local average of PC values across all registered primordia. The small circles show the centroid positions of cells from a single example tissue to aid orientation. The gray bars indicate, from left to right, the deposition zone, follower zone, transition zone, and leader zone. pLLP shape features show varied patterns, including orientation along the D-V axis (TFOR-PC1), characteristic differences between leader and follower cells (TFOR-PC3, CFOR-PC2), and complicated patterns likely arising from the superimposition of different processes (TFOR-PC1).