Fig. 5: Comparative analysis reveals conserved and divergent features across macaque and mouse cochlear epithelium.

a Schematic of label transfer-based integration of macaque and mouse clusters by scVI transfer learning. b UMAP plots of macaque cochlear epithelium snRNA-seq data anchored onto the mouse cochlear epithelium snRNA-seq data by label transfer embedding. Left: Colors represent cell type annotations determined from individual clustering prior to anchoring. Middle: Colors represent cell type annotations predicted from mouse identity. Right: Colors represent species mix. c Confusion matrix plot showing the percentage of macaque cells mapping onto the reference mouse clusters (cell types). d Heatmap showing the consistency of the DEGs within the same cell type across macaque and mouse, which was calculated by Jaccard similarity index. Hair cells showcase the highest similarity among cell type groups. e Heatmap depicting the classification performance upon AUROC scores for each cell type by MetaNeighbour analysis. Each column represents the performance of a single training group across the two snRNA-seq datasets. Hair cells, Deiter’s cells, pillar cells, and tympanic border cells have higher transcriptomic similarity across two species. f Cumulative distribution plot illustrating the number of integrative cochlear epithelium cell types expressing groups of genes in the following families: homeodomain TFs, zinc finger TFs, all TF families, GPCRs, ion channels, proteoglycans, cell adhesion molecules, and ribosomal genes. g Bar plot of TFs shared between species. h Dot plots showing the variance-scaled, log-transformed expression of TFs (x-axis) that were conserved within the same cell type (y-axis) across macaque and mouse snRNA-seq data. i Representative images of immunofluorescence demonstrating IKZF2, GATA3, NR2F2 and PBX3 localization in 5-year-old macaque cochlea. n = 3 independent experiments with similar results. j Scatter plots illustrating the average classification performance (chance = 0.5) of gene sets to classify cochlear epithelium types within and between species by applying the MetaNeighbor analysis. Linear regression fits are shown with cyan lines (slope values are shown at the top left). Colors indicate gene set types.