Generation of primate embryoid bodies.

A) Overview about the EB differentiation workflow of the four primate species human (Homo sapiens), orangutan (Pongo abelii), cynomolgus (Macaca fascicularis) and rhesus (Macaca mulatta), including their phylogenetic relationship. Scale bar represents 500 µm. B) Immunofluorescence staining of day 16 EBs using α-fetoprotein (AFP), β-III-tubulin and α-smooth muscle actin (α-SMA). Scale bar represents 100 µm. C) Schematic overview of the sampling and processing steps prior to 10x scRNA-seq. D) UMAP representation of the whole scRNA-seq dataset, integrated across all four species with Harmony. Single cells are colored by the expression of known marker genes for the three germ layers and undifferentiated cells. E) UMAP representation, colored by assigned germ layers, split by species. Panels A-C created with BioRender.com.

Assignment of orthologous cell types across species.

A) Schematic overview of the pipeline to match clusters between species and assign orthologous cell types. B) Sankey plot visualizing the intermediate steps of the cell type assignment pipeline. Each line represents a cell which are colored by their species of origin on the left and by their current cell type assignment during the annotation procedure on the right. An initial set of 118 high resolution clusters (HRCs), 25-35 per species, was combined into 26 orthologous cell type clusters (OCCs). Similar cell type clusters were merged and after further manual refinement provided the basis for final orthologous cell type assignments. C) Fraction of annotated cell types per species. D) UMAPs for each species colored by cell type. E) To validate our cell type assignments, we selected three marker genes per cell type that exhibit a similar expression pattern across all four species and have been reported to be specific for this cell type in both human and mouse (Supplementary Table S1). The heatmap depicts the fraction of cells of a cell type in which the respective gene was detected for cell types present in at least three species.

Effect of cell type specificity on expression conservation.

A) UMAP visualizations depicting expression patterns of selected example genes: SOX10 (conserved cell type-specific expression in neural crest cells), ESRG (species-specific and cell type-specific expression in human iPSCs), and RPL22 (conserved, broad expression). B) For each gene, expression was summarized per species and cell type as the expression fraction and binarized into “not expressed”/”expressed” (black frame) based on cell type-specific thresholds. The same example genes as in A) are shown here. iPSCs: induced pluripotent stem cells, EE: early ectoderm, NC: neural crest, SMC: smooth muscle cells, CFib: cardiac fibroblasts, EC: epithelial cells, Hepa: hepatocytes. c) Boxplot of expression conservation of genes with different levels of cell type specificity in human. D) Boxplot of the fraction of coding sequence sites that were found to evolve under constraint based on a 43 primate phylogeny [34], stratified by human cell type specificity.

Evaluation of marker gene conservation.

A) UpSet plot illustrating the overlap between species for the top 100 marker genes per cell type. B) Heatmap showing the expression fractions of marker genes: on the left, markers shared among all species, and on the right, markers unique to the human ranking. For each cell type, one representative gene is labeled and further detailed in Supplementary Figure S8. iPSCs: induced pluripotent stem cells, EE: early ectoderm, NC: neural crest, SMC: smooth muscle cells, CFib: cardiac fibroblasts, EC: epithelial cells, Hepa: hepatocytes. C) Rank-biased overlap (RBO) analysis comparing the concordance of gene rankings per cell type for lncRNAs, protein-coding genes and transcription factors. D) Average F1-score for a kNN-classifier trained in the human clone 29B5 to predict cell type identity based on the expression of 1-30 marker genes. Each line represents the performance in a different clone, with shaded areas indicating 95% bootstrap confidence intervals.

Comparison of EB differentiation protocols using flow cytometry.

A) Antibody combination to analyze iPSCs and cells of the three primary germ layers in a single sample. Created with BioRender.com. B) Flow cytometry gating overview using human EBs at day 7 of differentiation. 1. Gating of cell population. 2. Gating of single cell population. 3. Gating of live cell population. 4.-6. Gating of cells belonging to pluripotent or germ layer populations based on the antibody combination shown in S1A). C) Phase contrast images of orangutan EBs on day 6 of differentiation in 4 different culture conditions. Scale bar represents 250 µm. D) Barplot of pluripotency and germ layer proportions of day 7 EBs from human, orangutan, cynomolgus and rhesus in the 4 different culture conditions.

Total number of recovered cells.

A) Barplot of cell numbers per species and experimental batch and 10x lane. B) Barplot of cell numbers per species and day of differentiation. C) Barplot of cell numbers per clone. D) Barplot of cell numbers per clone and day of differentiation.

Reference based cell type classification.

A) UMAP representations colored by labels from a classification with a reference dataset of day 21 human embryoid bodies [18]. B) Single cell clusters in integrated data from all 4 species. C) Stacked bar plot of the proportions of predicted labels across clusters obtained in the integrated dataset.

Replicability of cell types across species measured by reciprocal classification.

A) Heatmap illustrating ‘all vs all’ similarities of cell types from all four species. For each cell type pair the similarity represents the average classification fraction obtained through reciprocal classification between each species pair. B) Average classification fractions for cell types that are shared among each species pair. AP: astrocyte progenitor, CFib: cardiac fibroblasts, CEndo: cardiac endothelial cells, CPC: cardiac progenitor cells, EEC: early epithelial cells, EE: early ectoderm, EC: epithelial cells, Fib: fibroblasts, GPC: granule precursor cells, Hepa: hepatocytes, NCI: neural crest I, NCII: neural crest II, Neu: neurons, SMC: smooth muscle cells.

Replicability of cell types across species measured with MetaNeighbor.

A) Heatmap illustrating ‘all vs all’ similarities of cell types from all four species. For each cell type pair the similarity represents area under the receiver operator characteristic curve (AUROC) scores obtained with MetaNeighbor [6] in unsupervised mode. B) AUROC scores for cell types that are shared among each species pair. AP: astrocyte progenitor, CFib: cardiac fibroblasts, CEndo: cardiac endothelial cells, CPC: cardiac progenitor cells, EEC: early epithelial cells, EE: early ectoderm, EC: epithelial cells, Fib: fibroblasts, GPC: granule precursor cells, Hepa: hepatocytes, NCI: neural crest I, NCII: neural crest II, Neu: neurons, SMC: smooth muscle cells.

Cell type annotation.

A) Barplot of cell type fractions per species and clone. B) Barplot of cell type fractions per experimental batch and 10x lane. C) Barplot of cell type fractions per day of differentiation.

Characteristics of genes with different levels of cell type-specific expression.

A) Stacked bar plot of the number of genes per cell type specificity level for different species. B) Boxplot of expression conservation of genes with different levels of cell type specificity in orangutan, cynomolgus and rhesus. C) Boxplot of gene-level constraint based on primate phastCons scores [34] for protein-coding genes. D) Boxplot of mean expression per cell type for genes with different levels of cell type specificity. E) Boxplot of mean expression per cell type for a subset of 236 genes per cell type specificity and species that were sampled to have a similar distribution of mean expression. F) Boxplot of expression conservation of the same subsampled genesets as in E).

Expression patterns of shared and human specific marker genes.

A) UMAP representation per species filtered for the 7 cell types that are present in all 4 species. B) UMAP representations colored by the log-normalized expression of 7 representative marker genes that are shared among the top100 marker genes per cell type in all 4 species. C) UMAP representations colored by the log-normalized expression of 7 representative marker genes that are only present in the human top100 marker gene ranking per cell type.

kNN classification performance per cell type.

F1-score per cell type for a kNN-classifier trained in the human clone 29B5 to predict cell type identity based on the expression of 1-30 protein-coding marker genes. Each line represents the performance in a different clone, colored by species identity.

kNN classification performance for transcription factors and protein coding marker genes.

A) Average F1-score for a kNN-classifier trained in the human clone 29B5 to predict cell type identity in the other clones. The classifier is trained on the expression of the top 1-30 protein coding markers (solid lines) or transcription factor markers (dashed lines). B) Comparison of the maximum average F1-score between transcription factors and protein coding markers for the classifications depicted in A).

Marker genes.

Literature review for marker genes used in human and mouse / rodents to determine a specific cell type.