Fig. 5
Quantification of spatial patterning during RA differentiation. a The computationally generated pattern class structures used to train the principle component analysis, derived from33, were applied to 120 experimental colony structures. b The seven selected metrics were calculated from each of the training set pattern classes (8 classes x 120 colony structures) and transformed into latent variable space through principal component analysis. PC1 represents extent of differentiation (temporal), and PC2 and PC3 represent organization/stochasticity and spatial locale, respectively (spatial characteristics). c The same metrics were calculated from experimental images of 0- (n = 24), 24- (n = 113), 48- (n = 139), and 72-h (n = 22) RA-treated colonies and transformed into latent variable space. The average simulation trajectory was capable of capturing the spatiotemporal trajectory of the experimental data (see Supplementary Figure 7 for simulation data points). At 24 h there is a steep transition along both spatial axes, indicating that there is a gain in random differentiation and that it propagates along the edges of colonies. By 48 h, the majority of differentiated cells are connected within a single, asymmetrical cluster
