Analysis of Differentiation Protocols Defines a Common Pancreatic Progenitor Molecular Signature and Guides Refinement of Endocrine Differentiation

Abstract

Several distinct differentiation protocols for deriving pancreatic progenitors (PPs) from human pluripotent stem cells have been described, but it remains to be shown how similar the PPs are across protocols and how well they resemble their in vivo counterparts. Here, we evaluated three differentiation protocols, performed RNA and assay for transposase-accessible chromatin using sequencing on isolated PPs derived with these, and compared them with fetal human pancreas populations. This enabled us to define a shared transcriptional and epigenomic signature of the PPs, including several genes not previously implicated in pancreas development. Furthermore, we identified a significant and previously unappreciated cross-protocol variation of the PPs through multi-omics analysis and demonstrate how such information can be applied to refine differentiation protocols for derivation of insulin-producing beta-like cells. Together, our study highlights the importance of a detailed characterization of defined cell populations derived from distinct differentiation protocols and provides a valuable resource for exploring human pancreatic development.

Keywords: cell identity; directed differentiation; disease modeling; endocrine differentiation; multi-omics analysis; open chromatin; pancreatic endoderm; pancreatic progenitors; pluripotent stem cells; transcriptomics.

Figure 1

Derivation of PPs from hPSCs Using Multiple Differentiation Protocols (A) Schematic outline of the three PP protocols applied. (B) Representative examples of flow cytometry pseudo color dot plots of PPs from the three protocols stained for PDX1 and NKX6.1. Negative control is definitive endoderm cells. (C) Immunofluorescence images of PPs stained for PDX1 and NKX6.1. Scale bar, 100 μm. (D) Quantification of PDX1 and NKX6.1 co-expressing cells based on the flow cytometric analysis shown in (B). Graph shows a scatterplot of the mean ± SEM of five individual hPSC lines. Dots are color coded according to individual cell lines (details in Figure S1C). n = 10 independent experiments. (E) Percentage of PDX1 and NKX6.1 co-expressing cells from the SB AD3.1 hiPSC line differentiated with protocol C with or without 50 ng/mL Noggin included during stage two; n = 3 independent experiments; ^∗p < 0.05, paired t test. (F) Quantification of PDX1, NKX2.2, NEUROD1, and percentage NKX6.1+ cells co-expressing Ki67. Bars show means and dots represent individual differentiations. One-way ANOVA with the Tukey test for multiple comparisons, ^∗p < 0.05, ^∗∗p < 0.01, different from the two other groups. PDX1, n = 10 independent experiments, same hPSC lines as in (D). NKX2.2 and NEUROD1, n = 5 independent experiments, one for each of the following hPSC lines: SA121 hESC, SB NEO1.1 hiPSC, SB AD2.1 hiPSC, SB AD3.1 hiPSC, SB AD3.4 hiPSC. Ki67, n = 4 independent experiments, three using SB AD3.1 hiPSC and one using SA121 hESC.

Figure 2

Global Gene Expression and Chromatin Accessibility Analysis of FACS-Isolated PP Populations (A) Schematic showing the experimental setup. NKX6.1-GFP hiPSCs were differentiated side by side using all three protocols and GFP+ and GFP− cells as well as unsorted cells were collected following FACS for RNA and ATAC sequencing. Cells were collected from three independent differentiations of all three protocols. (B) Principal component analysis (PCA) of RNA-seq (left) and ATAC-seq data (right). Legend applies to both PCA plots.

Figure 3

Common Transcriptomic and Epigenomic PP Signatures across Three Differentiation Protocols (A) PCA of RNA-seq samples together with data collected at all stages of the hPSC differentiation toward beta-like cells generated with protocol A (Perez-Alcantara et al., 2018). (B) Venn diagram of the RNA-seq GFP+ PP signatures generated separately for each of the protocols by comparison with cells at the other differentiation stages. (C) Selected gene ontology enrichment of the common PP signature genes across the three protocols; shown separately for signatures derived with presort cells (in blue), GFP+ (in green), and GFP− (in red) cell populations. The length of the bar represents −log10 of the enrichment p value. (D) PCA of RNA-seq samples from this study, together with transcriptomes of fetal pancreas cell subpopulations (Ramond et al., 2018). (E) PCA of ATAC-seq samples from this study, together with data collected at all stages of the hPSC differentiation toward beta-like cells, generated with protocol A (Perez-Alcantara et al., 2018). (F) Venn diagram of the ATAC-seq GFP+ PP signatures generated separately for each of the protocols by comparison with cells at the other differentiation stages. (G) Enrichment of selected TFs within the common PP signature open chromatin peaks across the three protocols; shown separately for signatures derived with presort cells (in blue), GFP+ (in green), and GFP− (in red) cell populations. (H) Mean expression of CUX2 gene across all stages of hPSC differentiation toward beta-like cells. Shaded gray area indicates ± SEM (Perez-Alcantara et al., 2018). (I) Footprinting analysis of CUX2 binding motifs within open chromatin peaks of the GFP+ populations generated with the three differentiation protocols. BLC, beta-like cells; EN, endocrine cells; ; EP, endocrine progenitors; GT, gut tube; PE, pancreatic endoderm/progenitors; PF, posterior foregut.

Figure 4

Protocol-Specific Differences in Transcriptomic and Epigenomic Profiles of PPs Generated with Three Differentiation Protocols (A) Heatmap of RNA-seq co-expressed gene modules eigengenes across all RNA-seq samples. Higher red color intensities indicate higher eigengene values. (B) Heatmap of ATAC-seq co-open chromatin modules eigengenes across all ATAC-seq samples. Higher red color intensities indicate higher eigengene values. (C) Pairwise Pearson correlation heatmap of module eigengene values for RNA-seq and ATAC-seq modules. (D) Hypergeometric enrichment p values for gene signatures of PP signatures, selected developmental gene ontologies, and gene signatures of selected intestinal and hepatic tissue/cell types. Only modules with enrichment p values <0.01 for any of the selected categories in (D) and (E) are plotted. Full list is available in Table S1. (E) Hypergeometric enrichment p values for transcriptomic signatures of in vivo fetal pancreatic cell subpopulations from Ramond et al. (2018). Only modules with enrichment p values <0.01 for any of the selected categories in (D) and (E) are plotted. (F) Enrichment p values of selected known TF binding sites within modules of co-open chromatin. Only modules with enrichment p values <1e−20 for any of the selected TFs are plotted. (G) Selected highly correlated pairs of RNA-seq and ATAC-seq modules eigengenes. The height of the bars represents the eigengene values for each sample in the presented module. Bars are colored by each protocol/cell population, as outlined in figure legends for PCA plots in Figure 2B, and in other subsequent figures in this paper. I, protocol A GFP− module pair; II, protocol A GFP+ module pair; III, protocol B module pair; IV, protocol C module pair.

Figure 5

Reduction of Expression of the Intestinal Marker CDX2 in PPs (A) TPM for CDX2 in GFP− and GFP+ sorted cell populations from differentiation protocols A, B, and C. Graph shows scatterplot of mean with each dot representing individual differentiations. (B) Flow cytometry analysis of NKX6.1 and CDX2 of PPs from the three differentiation protocols. DE was used as negative control. Representative pseudo color dot plots of five individual differentiations. (C and D) Representative pseudo color dot plots of two individual differentiations. Ten modifications of protocol B were assessed for the ability to maintain PDX1 and NKX6.1 expression (C) while simultaneously reducing expression of CDX2 (D). Representative pseudo color dot plots of cells stained for NKX6.1 and PDX1 (C) or NKX6.1 and CDX2 (D). DE cells were used as negative controls. (E) Heatmap summarizing the percentage of PDX1/NKX6.1 and CDX2/NKX6.1 co-expressing cells. Average percentage of one differentiation each of SB AD3.1 and SB AD3.4 hiPSC lines (n = 2 independent differentiations). Conditions tested were (1) 50 ng/mL Noggin, ST2; (2) 50 ng/mL Noggin, ST2-3; (3) 50 ng/mL Noggin, ST2-4; (4) ST4 only 2 days; (5) 100 nM LDN, ST2; (6) 100 nM LDN, ST2-3; (7) 100 nM LDN, ST2-4; (8) 50 ng/mL Activin A, ST4.2; (9) without KGF, ST4.2; (10) reduced retinoic concentration (0.2 μM) second day of ST3.

Figure 6

Improved Endocrine Differentiation Following Reduction of CDX2 Expression in PPs (A) Flow cytometry-based quantifications of stage 5 endocrine progenitor differentiation of protocol B and the two modified conditions (condition 5 and 3). Scatterplots show percentage of NEUROD1 and NKX2.2-positive cells as means ± SEM, n = 5 independent experiments (three with SB AD3.1 hiPSC, one with SB AD3.4 hiPSC, and one with SA121 hESC). (B) Immunofluorescence microscopy images of stage 6 cells stained with NKX6.1 and C-peptide antibodies. DAPI is used to visualize the nuclei of all cells. Scale bar, 100 μm. (C) Representative pseudo color dot plots of stage 6 beta-like cells stained for C-peptide and NKX6.1. Numbers mark the percentage of cells in each quadrant. (D) Quantification of C-peptide/NKX6.1 double positive by flow cytometry as shown in (C) (conditions 5 and 3). Scatterplot shows percentage of C-peptide/NKX6.1 double-positive cells as means ± SEM, n = 6 independent experiments (four with SB AD3.1 hiPSC, two with SA121 hESC). (A and D) One-way ANOVA with Tukey test for multiple comparisons, ^∗∗∗p < 0.001, different from the two other groups.