Extended Data Fig. 5: Differentiation of the prenatal hair follicle mesenchyme.

(a) Circos plot showing selected significant (adjusted p-value<0.05, significance calculated in CellphoneDB using empirical shuffling and FDR-adjusted) predicted interactions between pre-dermal condensate and ILC3 and LTi cells in prenatal skin. Arrows represent directionality of interactions (ligand to receptor); connection width is proportional to the CellphoneDB mean value for each ligand-receptor pair. (b) Schematic representation of mesenchymal-epithelial signalling and cellular processes during hair formation. (c) Heatmap showing significant (adjusted p-value <0.05, significance calculated in CellphoneDB using empirical shuffling and FDR-adjusted) predicted interactions between hair mesenchymal cells and epithelial cells (early: Immature basal; late: DPYSL2⁺ basal, POSTN⁺ basal, placode, matrix, ORS, CL, IRS, Cuticle/cortex) in SkO. Top 10 interactions per cell pair are shown. Colour scale represents the mean expression values of each ligand-receptor pair in corresponding cell pairs. (d) Heatmap on the left shows interesting trends of distributional distances in the expression of selected differentially expressed TFs across pseudotime between prenatal skin (reference) and SkO. The distributional distance is a Shannon information measure of dissimilarity (unit: nits), and the heatmap visualises these distances across time for each TF after log-transformation and smoothening using a Gaussian kernel (σ = 2) for highlighting their trends. Heatmaps in the middle and right show the interpolated and z-normalised mean expression of those selected TFs across pseudotime in prenatal skin and SkO respectively. (e) Gene expression plots for representative genes in prenatal skin (green) and SkO (blue) across pseudotime. Left column: the interpolated log1p transformed expression (y-axis) against pseudotime (x-axis). The lines represent mean expression trends; the faded data points are 50 random samples from the estimated expression distribution at each time point. Right two columns: actual log1p transformed expression (y-axis) against pseudotime (x-axis) where each point represents a cell. (f) Dot plot showing variance-scaled, mean expression (dot colour) and percent of expressing cells (dot size) of known genes involved in hair formation²⁶. (g) UMAP co-embedding of human prenatal (left) and mouse embryonic (E12.5, E13.5 and E14.5) (right) skin coloured by broad cell cluster annotations. (h) Heatmap showing prediction probabilities from a logistic regression model trained on human prenatal skin (x-axis), projected onto mouse embryonic skin (y-axis) for broad cell groupings. Colour scale indicates median prediction probabilities. (i) Heatmap showing prediction probabilities from a logistic model trained on human prenatal skin (x-axis), projected onto mouse embryonic skin (y-axis) for refined cell clusters (only fibroblast sub-populations shown from all cell types). (j) Dot plot showing variance-scaled, mean expression (dot colour) and percent of expressing cells (dot size) of fibroblast marker genes. DC, dendritic cells; Fib, fibroblast; LTi, lymphoid tissue inducer cells; Vessel BECs, Vessel blood endothelial cells; Vessel LECs, Vessel lymphatic endothelial cells. The images in b were created using BioRender (https://biorender.com).