Recapitulating endocrine cell clustering in culture promotes maturation of human stem-cell-derived β cells

Abstract

Despite advances in the differentiation of insulin-producing cells from human embryonic stem cells, the generation of mature functional β cells in vitro has remained elusive. To accomplish this goal, we have developed cell culture conditions to closely mimic events occurring during pancreatic islet organogenesis and β cell maturation. In particular, we have focused on recapitulating endocrine cell clustering by isolating and reaggregating immature β-like cells to form islet-sized enriched β-clusters (eBCs). eBCs display physiological properties analogous to primary human β cells, including robust dynamic insulin secretion, increased calcium signalling in response to secretagogues, and improved mitochondrial energization. Notably, endocrine cell clustering induces metabolic maturation by driving mitochondrial oxidative respiration, a process central to stimulus-secretion coupling in mature β cells. eBCs display glucose-stimulated insulin secretion as early as three days after transplantation in mice. In summary, replicating aspects of endocrine cell clustering permits the generation of stem-cell-derived β cells that resemble their endogenous counterparts.

Fig. 1 |. Generation of islet-like eBCs.

a, A schematic representation of the directed differentiation of hESCs to pancreatic β cells from d0 to d26–28. The process recapitulates embryonic development and includes a strategy of reaggregation of immature β-like cells to mimic β cell coalescence during islet formation. Immature INS^GFP+ β-like cells are isolated by FACS from d19–20 spheres, and subsequently reaggregated into 100 μm-sized eBCs in Aggrewells. Culture conditions are shown beneath each arrow. Bright-field and fluorescence images of eBCs in Aggrewells are shown; scale bars, 200 μm. ES, embryonic stem cells (dark grey); GT, gut tube (pink); PFG, pancreatic foregut (orange); Panc. prog., pancreatic progenitors (blue); INS-GFP⁺, GFP-positive insulin-producing cells (green); Non-panc., non-pancreatic cells (purple). b, Immunofluorescent staining of d26–27 eBCs for C-peptide (C-PEP), glucagon (GCG), somatostatin (SST) and nuclei (DAPI), indicating that the majority of cells are C-peptide⁺ and monohormonal. White arrows indicate C-peptide⁺/glucagon⁺ or C-peptide⁺/ somatostatin⁺ cells. Scale bar, 100 μm. Images are representative of four independent experiments. c, Quantitative analysis of d20 clusters and d26–27 eBCs by flow cytometry showing enrichment and confirming that the majority of the cells are C-peptide⁺/NKX6.1⁺/glucagon⁻ in eBCs. Flow experiments were repeated 10 times independently with similar results. d, Flow cytometric quantification of co-expression of C-peptide and various key β cell markers such as NKX6.1, PDX1, CHGA, NEUROD1, PAX6, ISL1 and NKX2.2, and C-peptide⁺/glucagon⁻ cells before (d19–20, black circles: C-pep⁺/NKX6.1⁺, n = 10; C-pep⁺/PDX1⁺, n = 9; C-pep⁺/ CHGA⁺, C-pep⁺/NEUROD1⁺, n = 4; C-pep⁺/PAX6⁺, n = 6; C-pep⁺/ISL1⁺, C-pep⁺/NKX2.2⁺, n = 3; C-pep⁺/GCG⁺, C-pep⁻/PDX1⁺, C-pep⁻/NKX6.1⁺, PDX1⁺/ NKX6.1⁺/C-pep⁻, n = 7; C-pep⁺/GCG⁻, n = 8 biological samples) and after sorting and reaggregation (eBCs, green squares: C-pep⁺/NKX6.1⁺, C-pep⁺/GCG⁻, n = 10; C-pep⁺/PDX1⁺, C-pep⁺/NKX2.2⁺, n = 4; C-pep⁺/CHGA⁺, C-pep⁺/NEUROD1⁺, n = 6; C-pep⁺/PAX6⁺, C-pep⁺/ISL1⁺, n = 3; C-pep⁺/GCG⁺, n = 8; C-pep⁻/ PDX1⁺, C-pep⁻/NKX6.1⁺, PDX1⁺/NKX6.1⁺/C-pep⁻, n = 7 biological samples). Data are presented as mean± s.e.m. See Supplementary Table 6 for source data. **P<0.01, ***P< 0.001 determined using the Holm−Sidak method for multiple t two-sided tests, with α = 0.05. See the Methods section for exact P values.

Fig. 2 |. eBCs exhibit functional characteristics similar to human islets in vitro.

a, Dynamic secretion of C-peptide in response to stimulation with 20 mM glucose, 10 nM exendin-4 (Ex-4) and 30 mM KCl in an in vitro perifusion assay with a starting basal glucose concentration of 2.8 mM. n = 3 independent samples. Data are presented as mean ± s.e.m. b, Cytosolic calcium signalling in response to alternating high (20 mM) and low (2.8 mM) glucose followed by KCl (30 mM) stimulation as measured by Fura-2/AM fluorescence emission intensity. Plots are population measurements from individual whole clusters (not pre-selected single cells). c, Calcium signalling and insulin secretion response of eBCs to tolbutamide, a sulfonylurea drug that blocks ATP-sensitive K⁺ channels. Calcium signalling analyses were performed with 5 independent samples of d20 clusters, 6 independent samples of eBCs and 6 independent islet preparations (Supplementary Fig. 3e,f). See Supplementary Table 6 for source data.

Fig. 3 |. Enrichment of β cell function/maturation related genes and repression of disallowed genes in eBCs.

a, A schematic illustration depicting isolation of INS^GFP-high+ cells from d20 clusters and eBCs for RNA sequencing (RNA-seq). b, Selected pathways enriched in eBC INS^GFP-high+ cells (green) and d20 INS^GFP-high+ cells (black) by GSEA performed with ‘Hallmark’ (left) and ‘KEGG’ (right) gene sets from the Molecular Signatures Database (MSigDB). NES, normalized enrichment score. A full list of pathways is shown in Supplementary Table 5. c, Enrichment plots with genes in the leading edge of each pathway indicated on the right side of the respective trace. Note that only INS^GFP-high+ cells were analysed for both the d20 immature and eBC populations. GSEA uses Kolmogorov–Smirnov statistics to calculate P values. d, Transmission electron micrograph of β cells present in d20 immature clusters and β cells found in eBCs. Higher magnification images (panels on the right of each main image) show representative types of individual granules present in the samples. Arrowheads, mature granules; arrows, immature granules. Scale bar, 1 μm. Images are representative from two independent experiments. e, Quantification of immature and mature granules using the metric described in d and expressed as immature granules as a percentage of total granules. n = 8 sections were analysed in each sample. Data are expressed as mean ± s.e.m. P < 0.0001 determined by two-tailed unpaired t-test with Welch’s correction. f, Percentage of pro-insulin to C-peptide in d20 immature clusters and eBCs. Immature d20 clusters, n = 10 independent samples; eBCs, n = 9 independent samples. Data are expressed as mean ± s.e.m. P = 0.0003 determined by two-tailed Welch’s t test. g, Bisulfite sequencing analysis for the disallowed genes HK1 and LDHA at the indicated loci in eBCs. Each parallel line is an independent clone. Filled circles represent fully methylated and open circles represent hypomethylated CpGs. These regions are mostly methylated at both loci in eBCs. h, Gene expression of HK1 and LDHA in eBCs relative to human islets. eBCs, n = 4 independent samples; human islets, n = 6 independent samples. Data are expressed as mean ± s.e.m. P values determined by two-tailed Welch’s t-test. ns, not significant. See Supplementary Table 6 for source data.

Fig. 4 |. β cells in a highly enriched endocrine niche of eBCs are distinct from immature β cells in a progenitor-rich niche.

a, A schematic illustration depicting the isolation of INS^GFP-high+ cells via FACS from d20 immature clusters, d26–27 NECs and d26–27 eBCs for RNA-seq. Note that only INS^GFP-high+ cells were used for transcriptome analysis. b, Heat map of differentially expressed genes between the three types of β cells. Hierarchical clustering indicates that d20 immature β-like cells (black) and NEC β cells (brown) are more closely related than β cells from eBCs (green) (n = 2 independent samples). c, GSEA of the three types of β cells with gene sets from MSigDB biological process ontology (GO) reveals various sub-processes of oxidative phosphorylation (OXPHOS) that are highly enriched in β cells isolated from eBCs compared to β-like cells of d20 clusters (left) or NECs (right). GSEA uses Kolmogorov–Smirnov statistics to calculate P values. d, Heat map of GSEA leading edge genes in electron transport chain ontogeny. Sample FPKM values are normalized across every gene (n = 2 independent samples). e, Scatter plot of log₁₀ FPKM values for all coding transcripts present in β cells of eBCs and β cells isolated from adult human islets. Pearson correlation coefficient (r) = 0.86, P < 2.2 × 10⁻¹⁶. f, Venn diagram of differentially expressed genes between human islet β cells and d20 β-like cells, NEC β cells and eBC β cells. eBC β cells are more closely related to human islet β cells by 967 genes (compared to NEC β cells) or 843 genes (compared to d20 β-like cells). DEG, differentially expressed genes.

Fig. 5 |. eBCs possess functionally mature mitochondria.

a–c, Mitochondrial respiration assessed by a Cell Mito Stress Test using a Seahorse XFe24 Bioanalyzer. a, Oxygen consumption rate (OCR) was first measured under basal conditions (2.8mM glucose) followed by sequential addition of either 2.8mM (blue line) or 20mM glucose (red line), 5μM oligomycin (oligo), 1 μm FCCP and 5 μm rotenone + antimycinA (rot/anti-M). eBCs show similar OCR profiles as human islets, whereas d20 immature clusters do not increase their OCR following glucose stimulation. d20 immature clusters: 2.8mM glucose, n = 5 independent samples; 20mM glucose, n = 7 independent samples. eBCs: 2.8mM glucose, n = 5 independent samples; 20mM glucose, n = 6 independent samples. Human islets: 2.8mM glucose, n = 6 independent samples; 20mM glucose n = 7 independent samples. Data are expressed as mean±s.e.m. b, Basal OCR of d20 immature clusters (black), eBCs (green) and human islets (red). Human islets, n = 7 independent samples; eBCs, n = 5 independent samples; d20 immature clusters, n = 4 independent samples. Data are expressed as mean± s.e.m. ****P = 0.0001. P values determined by one-way ANOVA with Dunnett test for multiple comparisons versus eBCs. ns, not significant. c, Dynamic mitochondrial energization as monitored by quenching of rhodamine-123 fluorescence. The rate of fluorescence decay/quenching of rhodamine-123 fluorescence is directly proportional to mitochondrial membrane potential. Representative experiment is shown. Mitochondrial membrane potential analyses were repeated with three distinct hESC β cell differentiations independently with similar results. Of note, the profile of mitochondrial potential changes on glucose stimulation is distinct between rodent and human β celIs. d, Single-cell end-point analyses of mitochondrial potential in the indicated populations measured by flow cytometry after incubation with 2.8 mM glucose or 20 mM glucose and stained with the mitochondrial membrane potential indicator dye, Mito Tracker Red CMXRos. Upper panel shows Mito Tracker staining in C-peptide⁺ cells and lower panel shows Mito Tracker staining in C-peptide⁺/NKX6.1+double-positive cells. The experiment was repeated with three independent differentiations. Ratio of median fluorescence intensity at 20 mM and 2.8 mM glucose is shown (mean± s.e.m.). See Supplementary Table 6 for source data.

Fig. 6 |. Endocrine cell clustering improves mitochondrial morphology and increases mitochondrial numbers.

a, GSEA traces demonstrating enrichment of the inner mitochondrial membrane complex signature in the cellular component ontology gene sets in INS^{GFP-high +} cells of eBC compared to those in d20 clusters (left) and NECs (right). GSEA uses Kolmogorov-Smirnov statistics to calculate P values. b, Representative transmission electron micrographs of mitochondria in d20 immature β-like cells and eBC β cells. The mitochondrial cristae are denser and more tightly folded in eBCs. Scale bar, 250 nm. Images are representative from two biologically independent samples. c, Inner mitochondrial membrane length was calculated using ImageJ software and represented as length in pixels in eBC β cells and d20 immature β-like cells. The image on the right shows an example mitochondrion in which inner mitochondrial membrane length was marked with the ImageJ tool for quantification (yellow lines). Scale bar, 250 nm. eBC β ce11s, n = 44 mitochondria; immature β cells, n = 34 mitochondria. Data are expressed as mean ± s.e.m. **P< 0.01 determined by two-tailed unpaired t-test with Welch’s correction. d, Flow cytometry with MitoID was used to assess mitochondrial mass in the indicated populations. Left: flow histogram of whole populations stained with MitoID. Right: flow histogram of C-pep⁺/NKX6.1⁺ double-positive cells within each population stained with MitoID. The experiment was repeated with three independent differentiations. e, Quantification of staining with MitoID. eBCs, n = 4 independent biological samples; NECs, n = 3 independent biological samples; human islets, n = 4 independent biological samples. Data are presented as mean fluorescence intensity ± s.e.m. *P< 0.05 determined by one-way ANOVA with Dunnett test for multiple comparisons versus eBCs. ns, not significant. f, Mitochondrial DNA (mtDNA) content assessed by the ratio of mtDNA/nuclear DNA by qPCR for the mitochondrial 16S rRNA gene and the nuclear β2 microglobulin gene. n = 3 independent samples. Data are presented as mean ± s.e.m. *P< 0.05 determined by two-tailed unpaired t-test. See Supplementary Table 6 for source data. g, Levels of ERRγ determined as FPKM in RNA-seq data from INS^{GFP-high +} from eBCs, NECs and d20 immature clusters. Data shown as fold change between eBC β/d20 β-like and eBC p/NEC β. P and FDR-corrected Q values are indicated. Cuffdiff uses a theoretic metric derived from the Jensen–Shannon divergence to calculate P values.

Fig. 7 |. eBCs are functional in vivo as early as 3 days after transplant, retain function long term and do not form tumours.

a,b, 700 eBCs (~0.7 × 10⁶ cells) were transplanted under the kidney capsule of non-diabetic male NSG mice. a, Grafts removed 3 days post-transplant (top) or 48 days post-transplant (bottom) were stained for C-peptide (C-PEP), glucagon (GCG) and somatostatin (SST) (left), or H&E (right). White arrows indicate GCG⁺/C-PEP⁺ double-positive cells. White arrowheads mark monohormonal GCG⁺ or SST⁺ cells. Black arrowheads in the H&E images indicate blood cells. Scale bars, 100 μm. Images are representative of five independent experiments. b, In vivo glucose challenge test at indicated days after transplant. Human C-peptide levels in the serum were measured after an overnight fast (grey) and again 30 min after an intraperitoneal glucose injection (black). The numbers on the x axes indicate individual animals. ND, not detected. c, The intraperitoneal glucose tolerance test (IPGTT) was performed 30 days after transplantation with eBCs (green, n = 5 animals), NECs (brown, n = 4 animals), d20 clusters (black, n = 5 animals), human islets (red, n = 3 animals) and control non-transplanted NSG mice (no cells, dotted line, n = 4 animals). Statistics determined two-way repeated measures ANOVA with Sidak’s multiple comparison tests. *P< 0.05, ***P< 0.001, NEC versus eBCs; ^†P< 0.001, no cells versus eBCs; ^&P< 0.05, d20 versus eBCs. Area under the curve was determined for each group. *P< 0.05, **P< 0.01, ****P< 0.0001, ns = not significant determined by one-way ANOVA with Dunnett test for multiple comparisons versus eBCs. Data presented as mean ± s.e.m. d, Random fed glucose measurements taken from STZ-treated control NSG mice (no cells, black) or NSG mice transplanted with 6,000 eBCs (green). n = 3 animals. One of the diabetic control non-transplanted mice expired on day 24. Data presented as mean ± s.e.m. See Supplementary Table 6 for source data. e, 4,000 eBCs (~4×10⁶ cells) were transplanted under the kidney capsule of non-diabetic male NSG mice. Human C-peptide levels following fasting (grey) and 60 min after an intraperitoneal glucose bolus (black) were measured 10 days and 8 months after transplant. f, Images of kidneys from mice transplanted with 4×10⁶ cells each of d20 immature clusters 3 months post-transplant (left) or eBCs 8 months post-transplant (right). g, Immunofluorescent images of eBC grafts 8 months after transplant stained with C-peptide/glucagon/somatostatin (left) and C-peptide/PDX1/NKX6.1 (right). Images are representative of five independent experiments.