Charting cellular identity during human in vitro β-cell differentiation

Abstract

In vitro differentiation of human stem cells can produce pancreatic β-cells; the loss of this insulin-secreting cell type underlies type 1 diabetes. Here, as a step towards understanding this differentiation process, we report the transcriptional profiling of more than 100,000 human cells undergoing in vitro β-cell differentiation, and describe the cells that emerged. We resolve populations that correspond to β-cells, α-like poly-hormonal cells, non-endocrine cells that resemble pancreatic exocrine cells and a previously unreported population that resembles enterochromaffin cells. We show that endocrine cells maintain their identity in culture in the absence of exogenous growth factors, and that changes in gene expression associated with in vivo β-cell maturation are recapitulated in vitro. We implement a scalable re-aggregation technique to deplete non-endocrine cells and identify CD49a (also known as ITGA1) as a surface marker of the β-cell population, which allows magnetic sorting to a purity of 80%. Finally, we use a high-resolution sequencing time course to characterize gene-expression dynamics during the induction of human pancreatic endocrine cells, from which we develop a lineage model of in vitro β-cell differentiation. This study provides a perspective on human stem-cell differentiation, and will guide future endeavours that focus on the differentiation of pancreatic islet cells, and their applications in regenerative medicine.

Extended Data Fig. 1 |. Comparison of two SC-β-cell protocol variants and resulting cell types.

a–c, Immunofluorescence imaging of differentiated (v8, stage 6, day 13) SC-islets showing staining of relevant markers. a, SC-β cells, which are typically positioned in the periphery, are positive for both NKX6.1 and C-peptide (fragment of proinsulin). b, SC-EC cells are positive for SLC18A1, an enterochromaffin cell marker. These cells are also present in the periphery. c, Non-endocrine cells, which are marked by SOX9, are most commonly found near the centre of SC-islets. Scale bars, 100 μm. d, e, Summary of changes in stages 3 and 4 in protocols x1 (a) and x2 (b) (see Extended Data Table 1 for protocol summaries), and representative flow cytometry results at the end of stages 4 and 6. f–i, t-SNE projection of cells sampled from the ends of stages 3–6 of protocol x2. Cells in f–i are coloured according to their assigned cluster. Horizontal bars indicate cell-type proportions. Related to Fig. 1d–g. j, Comparison of cell populations from protocols x1 and x2. Correlation is computed using the z-scores of mean TPM values (for each cluster) of 2,000 high-variance genes. Rows and columns are ordered using hierarchical clustering. Cells are labelled as in f–i and Fig. 1d–g. k, l, t-SNE projection of stage 6 from three differentiations, coloured by cell type (k) and by differentiation (l). m, Correlation of cell populations derived from HUES8 (embryonic stem cells, v4 and x3) and iPS1016/31 (induced pluripotent stem cells, v4). Colours are the same as in k. Correlation is computed as in j.

Extended Data Fig. 2 |. Functional assay of GSIS during stage-6 time course.

a, Design for sequential GSIS assay. b, Complete data for three independent flasks, assayed across several weeks. Circles are individual technical triplicates and bars show mean of those triplicates. c, Complete data for cadaveric human islets (seven donors), run alongside samples from b.

Extended Data Fig. 3 |. Stage-6 SC-β cells express characteristic β-cell markers.

a, b, t-SNE projection of stage-6 time-course data shaded by sampling time (a) and by representative marker genes (b). Expression is normalized relative to maximum value and smoothed over neighbouring cells. c, Expression profiles for key genes necessary for β-cell function. Shading displays mean expression (TPM, log-scaled) and diameter denotes fractional expression. d, e, Comparison of global expression between human islet β cells and in vitro progenitors (d) and SC-β cells (e). Note the shift in gene expression from progenitors to SC-β cells. All genes shown in all panels from c are circled in red. f, Results from GSEA show that gene sets from c are significantly upregulated during differentiation. Value plotted is -log₁₀ of the GSEA-reported FDR q-value (capped at 10), with sign (positive or negative) showing the direction of the effect (that is, purple positive values are upregulated in SC-β cells compared to NKX6.1 progenitors).

Extended Data Fig. 4 |. Comparison of SC-β and SC-α cells to each other, and their islet counterparts.

a, Insulin and glucagon expression in SC-β (purple distributions) and SC-α cells (red distributions) during several weeks of stage 6, shown as violin plots of SC-β or SC-α cells from that particular time point. The connected lines connect the medians of each population at each time point. b, Identification of genes enriched in cadaveric-islet α cells and islet β cells, from previously published data. c, Heat map of expression level of genes from b, shown for islet α, SC-α, SC-β and islet β cells. d, Genes enriched in islet β cells are upregulated in SC-β cells, and genes enriched in α cells are upregulated in SC-α cells. The displayed P value is computed using a two-sided Wilcoxon rank-sum test. In the box plot, boxes extend from first to third quartiles, whiskers extend from 5th to 95th percentiles, central line indicates median and box notching indicates 95th percentile confidence interval for median.

Extended Data Fig. 5 |. SC-EC cells secrete serotonin and exist in other protocols.

a, Schematic of serotonin synthesis from tryptophan. Enterochromaffin cells use TPH1, whereas serotoninergic neurons use TPH2 for the first and rate-limiting synthesis step. b, Serotonin release during sequential challenges of low and high glucose, followed by KCl depolarization. Top, clusters from three independent SC-β-cell protocol differentiations. Bottom, human cadaveric islets from two donors. Symbols show values of individual replicates for each sample (different clusters from the same sample are split and measured separately). P values computed using two-sided Wilcoxon rank-sum test. NS, non-significant with P > 0.05. c, d, Expression of enterochromaffin marker genes (shown in blue) is detectable in bulk RNA-sequencing data (from a previous publication ), and enriched via sorting of NKX6.1–GFP ⁺ cells, shown as fold change, mean expression and differential expression q-values. Positive fold change indicates higher expression in NKX6.1–GFP ⁺ cells. Enrichment of SC-EC-cell markers is comparable to β-cell markers (shown in purple) and opposite of α-cell markers (shown in red). All values shown are directly reproduced from results that were previously computed and deposited. e, Flow cytometry shows that SLC18A1 is co-expressed with NKX6.1⁺ in SC-EC cells of v8 SC-β-cell protocol differentiations. This example is representative across more than 100 independent differentiations. f, Comparison of gene expression between wild-type mouse islets and mouse islets 25 weeks after β-cell-specific Polycomb Repressive Complete 2 ablation via Eed knockout. Purple genes are examples of downregulated β-cell-identity genes, blue genes represent serotonin–enterochromaffin signature. q-values are FDR-corrected (α = 0.05) P values from Limma differential expression analysis.

Extended Data Fig. 6 |. Characterization of non-endocrine cells from stage-6 time course.

a, b, t-SNE projection of non-endocrine cells from stage-6 time course, shaded by collection day (a) or by genes relevant to cell identity (b). Expression is normalized relative to maximum value, and smoothed over neighbouring cells. c, t-SNE projection shaded by assigned cluster and bar charts of cellular fraction in each cluster by week of differentiation. d, Gene expression of population-specific markers for each subpopulation of non-endocrine cells. Shading displays mean expression (z-normalized TPM) and diameter denotes fractional expression.

Extended Data Fig. 7 |. Re-aggregation is a scalable, function-preserving method to enrich for endocrine cells.

a, Schematic of re-aggregation procedure to remove non-endocrine cells. Cells are enzymatically dissociated and re-aggregated during continued suspension culture. Non-endocrine cells fail to adhere and are removed by filtration. b, Schematic of CD49a enrichment procedure to produce SC-β cell enriched clusters. Dissociated cells are stained with anti-CD49a PE-conjugated antibody, incubated with anti-PE magnetic microbeads and magnetically separated. The enriched cells are re-aggregated in six-well plates on a rocker. c, t-SNE projection of cells sequenced from native and re-aggregated clusters from a single differentiation show a strong depletion of the non-endocrine population. Cells in both panels were differentiated using protocol v8. d, Immunofluorescence staining for C-peptide, GCG and SLC18A1 shows distinct neighbourhoods in re-aggregated clusters (protocol v8). Images shown are maximum intensity projections from z-stacks. Each panel shows separate representative clusters stained for all markers. Scale bars, 100 μm. e, f, Representative flow cytometry analysis of endocrine cell abundance (from protocol v8), before and after re-aggregation. Endocrine cells express CHGA. g, Summary of population composition (as assayed by flow cytometry) in 60 re-aggregated and 41 native independent differentiations, carried out using protocol v8. Re-aggregations were carried out in spinner flasks. P value computed using two-sided Wilcoxon rank-sum test. In g, h box plots, boxes extend from first to third quartiles, whiskers extend from 5th to 95th percentiles, central line indicates median and box notching indicates 95th percentile confidence interval for median. h, Stimulation index (insulin released at 20 mM glucose versus insulin released at 2 mM) of 52 independent protocol v8 differentiations, with paired native versus re-aggregated comparisons. P value computed using two-sided Wilcoxon signed-rank test. i, Complete data for static GSIS assays, performed as in Extended Data Fig. 2, corresponding to stimulation indices shown in Fig. 4d. Circles are individual technical triplicates and bars show mean of those triplicates. j, Dynamic perifusion assay of glucose-responsive insulin secretion of human islets, native SC-β-cell clusters (stage 6, day 22, v8) and matched SC-β islets produced via magnetic sorting for CD49a. Each point is the mean of three technical replicates, and the vertical bar indicates the s.e. across those triplicates. k, Area under the curve comparing the first low-glucose stimulation and the high-glucose stimulation, normalized to equal effective time in each treatment.

Extended Data Fig. 8 |. Stage-5 time-course markers and progenitor population heterogeneity.

a, b, t-SNE projection of stage-5 time-course data shaded by collection day (a) and by population-marker genes (b). Expression is normalized relative to maximum value, and smoothed over neighbouring cells. c, Pseudotime analysis of day 0 (top) and day 1 (bottom) progenitor cells. Shading on each t-SNE shows assigned pseudotime value of each cell. d, Pseudotime ordering of progenitor cells from stage 5 day 0 (top row) and day 1 (bottom row), showing population heterogeneity among early progenitors. Individual cells are shown as dots, shaded as in c. Gene expression predicted from pseudotime regression shown as overlaid line. e, Summary of stage-5 (day 0) heterogeneity captured by pseudotime analysis. Fold change between start and end of pseudotime ordering. q-value from likelihood ratio test of model with and without pseudotime. f, Heat map of receptors, ligands and signalling effectors that are dynamically expressed across stage-5 populations. Shading displays mean expression (z-normalized TPM) and diameter denotes fractional expression.

Fig. 1 |. Single-cell RNA sequencing of in vitro β-cell differentiation.

a, Summary of cell populations identified by flow cytometry at the end of stages 3–6 of the SC-β-cell protocol described in ref.. b, Use of inDrops to sample cells from several time points of the same differentiation. c, Expression profiles of developmentally relevant genes and markers across cell types identified in SC-β-cell differentiation. The shading displays mean expression as z-normalized transcripts per million mapped reads (z-norm. TPM), and diameter denotes fractional expression. d–g, t-SNE projections of cells sampled from the end of stages 3 through to stage 6 of protocol x1 (see Extended Data Table 1 for definition of protocols). Cells are coloured according to their assigned cluster. Horizontal bars indicate cell-type proportions.

Fig. 2 |. SC-β cells maintain identity and gain maturation-marker expression during extended culture in stage 6.

a, Experimental design for studying functional and transcriptional changes during stage 6 of protocol v8 (see Extended Data Table 1). b, Glucose-stimulated insulin secretion showing consecutive low-glucose (2.8 mM) and high-glucose (20 mM) challenges for three independent differentiations (diff.) over a period of five weeks. c, Stimulation indices (insulin released at 20 mM glucose versus insulin released at 2 mM glucose) for data in b. d, t-SNE projection of 38,494 cells from 6 time points that span 5 weeks of stage 6. Cells are coloured according to their assigned type. Vertical bars show population ratios in each week. e, Expression of endocrine-marker genes. f, Correlation of expression profiles for each major cell type, broken down by week. Cell type colours match those in d. g, Top, pseudotime order for SC-β cells shown on the t-SNE plot. Bottom, distribution of SC-β-cell pseudotime order, stratified by sampling week. h, Identification of dynamic genes along SC-β-cell pseudotime. Fold change compares start and end of pseudotime trajectory. q-values are false-discovery-rate (FDR)-adjusted (α = 0.001) P values from a likelihood ratio test comparing full and reduced models (Methods). i, Expression of selected genes shown along SC-β-cell pseudotime. Each dot represents expression of a cell, sorted and shaded as in g. Line shows result of pseudotime regression.

Fig. 3 |. Characterization of SC-EC cells.

a, Comparison of SC-β- and SC-EC-cell gene-expression profiles. Blue genes are enterochromaffin markers or are required for serotonin synthesis. b, Expression levels for SC-EC-cell-enriched genes across in vitro populations (top) and human pancreatic endocrine cells (bottom). Shading displays mean expression (TPM, log-scaled) and diameter denotes fractional expression. c, d, Immunofluorescence staining for SC-EC-cell markers showing co-localization with serotonin, using the v8 protocol. Scale bars, 100 μm. e, Immunofluorescence staining of graft tissue recovered eight weeks after transplantation of (protocol v4, see Extended Data Table 1) stem-cell islet clusters shows SC-EC persist after transplantation.

Fig. 4 |. Purification of SC-β cells with CD49a.

a, Expression of CD49a in stage-6 time-course data. b, Immunofluorescence for SC-β-cell (top) and endocrine-cell (bottom) markers of native, unsorted re-aggregated and CD49a⁺ sorted re-aggregated clusters. Scale bars, 100 μm. c, Flow cytometry quantification of SC-β cells (C-peptide⁺NKX6.1⁺) and SC-EC cells (SLC18A1⁺) fractions in three matched conditions for five biologically independent differentiations using protocol v8. d, Stimulation index for the differentiations shown in c. In c, d, symbol shows mean, and error bars (where shown) correspond to s.e. across three independently re-aggregated (re-agg) biological replicates. P values are from (two-sided) dependent t-test.

Fig. 5 |. A high-resolution map of in vitro endocrine induction.

a–c, t-SNE projection of 51,274 cells, shaded according to sampling time within stage 5 (a), to expression of NEUROG3 (b) and to their assigned cell types (c). Arrows in c indicate key lineage bifurcations. Endo. ind., endocrine induction. d, Fraction of cells from each cluster in c for each day of both independent differentiations. e, t-SNE shading of branch assignment and pseudotime value of each cell on the path from NKX6.1⁺ progenitors to SC-β cells and SC-EC cells. f, Expression of selected marker genes along pseudotime ordering from e. Dots show expression in single cells, sorted and shaded according to pseudotime order. Lines show regression on pseudotime for each branch (blue, SC-EC cell; purple, SC-β cell). g, Genes with significant branch-specific expression pattern. q-values are FDR-adjusted (α = 0.001) P values from likelihood ratio test comparing branched and non-branched models (Methods). h, Mean expression values of transcription factors for clusters presented in c, d. Shading displays mean expression (z-normalized TPM) and diameter denotes fractional expression. i, Proposed developmental model for the key cell types produced by SC-β-cell protocol.