Functional, metabolic and transcriptional maturation of human pancreatic islets derived from stem cells

SC-islets present organotypic cytoarchitecture and function

We devised an optimized differentiation protocol by combining previous advances in the generation of SC-islets^8,9,13,14 (Fig. 1a). Noteworthy differences to the most widespread protocols^8,9 include: (1) differentiation of hPSCs in adherent conditions until the pancreatic progenitor stage (S4); (2) optimized S4 step including nicotinamide, epidermal growth factor, Activin A and a ROCK inhibitor^13,14; (3) a microwell aggregation step that results in ≈ 80% PDX1⁺NKX6-1⁺ pancreatic progenitor population in uniformly sized clusters (Supplementary Fig. 1a); and (4) improved final maturation stage (S7), carried out in suspension culture. This S7 maturation step omits ALK5 inhibitor³ and contains an antiproliferative¹⁵ aurora kinase inhibitor ZM447439 (ZM) (adapted from Patent WO2017222879A1), in addition to previously described components triiodothyronine (T3) and N-acetyl cysteine (NAC)⁹.

To determine the impact of the S7 maturation step on SC-islet function, we systematically characterized hPSC-derived SC-islet maturation from the beginning of S7 (S7w0) to the end of its sixth week (S7w6), using a series of morphometric, functional, metabolomic and transcriptomic analyses. At S7w0, SC-islets contained ≈40% insulin-positive monohormonal cells (INS⁺)—a proportion that remained relatively stable until S7w6 (Fig. 1b,c and Supplementary Fig. 1b,c). At S7w0 and S7w1, SC-islets contained ≈15–20% cells coexpressing insulin and glucagon (INS⁺GCG⁺)—a proportion that was reduced to <5% by S7w3. Concomitantly, the number of single-positive GCG⁺ cells increased from ≈5% to ≈40–50% (Fig. 1c and Supplementary Fig. 1b), consistent with previous studies demonstrating polyhormonal to alpha cell differentiation in vitro and in vivo^{7,10,11,13,16,17,18,19,20}. Somatostatin (SST) positive cells were present at S7w0 at around 4% and the proportion remained unchanged until S7w6 (Fig. 1c).

As functional maturation of beta cells is linked to reduced proliferation^21,22, we examined markers of cell proliferation during the S7 and observed an 80% reduction (from 2.1% to 0.46%) in Ki-67⁺ INS⁺ cells (Fig. 1b,d). Critically, reduced proliferation was dependent on the use of ZM, NAC and T3 in S7 medium (Fig. 1e), while the proportions of INS⁺ and GCG⁺ cell populations were not affected (Fig. 1f). Stem-cell-derived enterochromaffin-like (SC-EC) cells have been reported to arise as an undesired byproduct of SC-islet differentiation⁷. We detected ≈13% SC-EC (SLC18A1⁺) cells at S7w0—a proportion that decreased steadily to ≈6.5% by S7w6 (Fig. 1g). Notably, this decrease was dependent on the presence of ZM in S7 medium (Fig. 1h).

SC-islet cytoarchitecture changed profoundly during S7 maturation. A high proportion of INS⁺ cells localized to the SC-islet periphery at S7w0–w1, but, by S7w3, SC-islets were polarized, with GCG⁺ and INS⁺ cells clustered separately (Fig. 1b). However, by S7w6, the cytoarchitecture varied from core-mantle organization (Supplementary Fig. 1d) to intermingled clusters of GCG⁺ and INS⁺ cells (Fig. 1b). Quantitatively, this reorganization resulted in an increased number of cell–cell contacts between GCG⁺ and INS⁺ cells from S7w0 to S7w6 (Supplementary Fig. 1e). Similar cytoarchitectural rearrangements have also been described during human fetal pancreatic islet development^23,24,25.

While beta cell numbers remained unchanged during the first 3 weeks of S7, the insulin content of SC-islets increased fourfold (Supplementary Fig. 1f). Concurrently, SC-islet beta (SC-beta) cells progressively acquired dense core insulin granules with ultrastructural morphology resembling those of primary beta cells (Fig. 1i).

Adult primary islets are characterized by a tightly controlled, biphasic insulin secretion response to increases in glucose²⁶. This is controlled through a metabolic response to glucose through K_ATP-channel activity (the triggering pathway), and modulated through neurohormonal and metabolic amplifying pathways²⁷. At S7w0, high glucose concentrations alone (16.7 mM) failed to trigger insulin secretion. However, treatment with the GLP-1 analog exendin-4 and membrane depolarization with high K⁺ both triggered pronounced secretory responses. From S7w2 onwards, SC-islets displayed biphasic glucose-stimulated insulin secretion (GSIS) responses similar to primary islets, with gradual increases in the magnitude of the response until S7w6 (Fig. 1h). Of note, the primary islets in this study had secretory responses representing the lower end of responses recorded in previous studies²⁸ and in publicly available databases²⁹. The SC-islets sustained their second phase response for >70 min (Supplementary Fig. 1g). The acquisition of SC-islet function was replicated in two additional human iPSC-lines (Supplementary Fig. 1h) demonstrating the robustness of the maturation protocol. Omission of either ZM or NAC from S7 medium attenuated GSIS responses, while omitting all additives nearly abolished it (Fig. 1k)—an effect mostly explained by higher insulin release in low glucose (Supplementary Fig. 1i).

Immature fetal and infantile primary beta cells are unable to suppress their insulin secretion in low glucose³⁰. Beta cell functional maturity is thus reflected by the glucose concentration threshold that triggers insulin secretion²⁸. Immature S7w0 SC-islets released higher levels of insulin in low glucose (Supplementary Fig. 1j). This basal release could be reduced with the K_ATP-channel opener diazoxide (Supplementary Fig. 1k), suggesting inappropriate K_ATP-channel closing in basal conditions at S7w0. We assessed insulin secretion thresholds also by gradually increasing glucose concentration in perifusion assays. SC-islets at S7w0 showed no glucose-induced insulin release, whereas at S7w2 they responded at unphysiologically low glucose concentrations of ≈3 mM. However, at S7w3 and S7w6 they reached the adult threshold of ≈5 mM glucose (Fig. 1l). This shift was also reflected in the glucose concentration eliciting the half-maximal secretory response, (5.6, 6.7 and 8.1 mM, at S7w2, S7w3 and S7w6, respectively) (Supplementary Fig. 1l).

These results demonstrate the generation of SC-islets in vitro, with biphasic glucose-dependent insulin secretion similar to that of adult islets. Functional maturation correlated with changes in SC-islet architecture and cell composition, but not with an increase in beta cell mass.

Functional insulin secretion machinery in SC-beta cells

To better understand the mechanisms of SC-islet glucose sensitivity, we dissected the stimulus-secretion coupling machinery of SC-islet beta cells with measurements of ion channel conductance, cytoplasmic Ca²⁺ and cAMP concentrations, as well as exocytosis. Patch-clamp recordings showed that S7w3 SC-beta cells fired action potentials (Fig. 2a), with 11 of 16 cells active in 3 mM glucose. In S7w6 cells, 1 of 17 cells fired action potentials in 3 mM glucose, which increased to 4 active cells in 16 mM glucose. SC-beta cells had Ca²⁺– and Na⁺-currents with voltage dependences similar to those in primary human beta cells (Fig. 2b,c). Ca²⁺-current amplitude was similar in both cell types, while Na⁺-currents were about twofold larger in SC-beta cells (Fig. 2c). K_ATP-channel dependent K⁺-conductance of S7w3 SC-beta cells was quantified using symmetric voltage-steps (Fig. 2d) or ramps (Fig. 2e). In 3 mM glucose, the membrane conductance was, on average, 53 ± 4 pS/pF (n = 50 cells) and increased in the presence of diazoxide in 49/50 cells to 273 ± 30 pS/pF (n = 50 cells). Both values are similar to those previously reported for human beta cells³¹.

The cytoplasmic Ca²⁺ concentration ([Ca²⁺]_i) was recorded from individual cells in SC- and primary islets loaded with a fluorescent Ca²⁺ indicator. In S7w3 SC-islets, a subset of cells showed [Ca²⁺]_i oscillations in low glucose with little change in response to high glucose (Fig. 2f). Other cells showed low and stable [Ca²⁺]_i at low glucose, with increased, and often oscillatory, [Ca²⁺]_i in high glucose (Fig. 2f). Primary islet cells also behaved heterogeneously but a higher proportion responded to elevated glucose (Fig. 2f,g). K_ATP-channel opening with diazoxide reduced, and closure with tolbutamide increased, [Ca²⁺]_i in both SC-islets and primary islets with more pronounced responses in the latter (Fig. 2f,g). Depolarization with 30 mM K⁺ increased [Ca²⁺]_i in all cells with similar magnitude in SC- and primary islets (Fig. 2f,g). The overall [Ca²⁺]_i responses or fraction of glucose-responsive cells in SC-islets did not change consistently during prolonged S7 maturation (range 42–74%; Supplementary Fig. 2a,b). However, among the glucose-responsive cells, basal [Ca²⁺]_i decreased from 47 ± 0.4 to 24 ± 0.2% of the K⁺-stimulated level and the increase induced by glucose stimulation improved from 5.2 ± 0.1% at S7w0 (n = 1,091 responsive cells) to 20.8 ± 0.4% at S7w7 (n = 1,659 responsive cells), consistent with the observed reduction of basal secretion and improved stimulation index. Since these unbiased analyses of indicator-loaded cells inevitably include a fraction of nonbeta cells, experiments were also performed with S7w7 SC-islets expressing the genetically encoded Ca²⁺ reporter R-GECO1 under insulin promoter control. Recordings from SC-beta cells identified by R-GECO1 expression confirmed the response heterogeneity while also highlighting that at S7w7, 79% of the beta cells showed glucose-induced [Ca²⁺]_i increases dependent on K_ATP-channel closure (n = 130 cells; Fig. 2h and Supplementary Fig. 2c).

In the presence of low glucose, tolbutamide increased [Ca²⁺]_i in both primary and SC-islets, but again to a higher degree in primary islet cells (Supplementary Fig. 2d,e). High glucose in the continued presence of tolbutamide caused a slight [Ca²⁺]_i increase in SC-islets and a decrease in primary islet cells. Despite this lowered [Ca²⁺]_i, glucose amplified secretion under these conditions (Fig. 3g). Treatment with exendin-4 did not alter [Ca²⁺]_i in primary islets and had only a weak tendency to increase [Ca²⁺]_i in SC-islets (Supplementary Fig. 2d,e).

The concentration of submembrane cAMP ([cAMP]_m)—a modulator of insulin exocytosis—was recorded in single cells in intact SC- and primary islets. High glucose induced a small, and exendin-4 a much more pronounced, increase in [cAMP]_m in both preparations (Fig. 2i–k) and in SC-beta cells identified by insulin-promoter-driven expression of R-GECO1 (Supplementary Fig. 2f), indicating that cAMP signaling in SC-islets closely resembles that in primary human islets.

Single-cell exocytosis was measured as membrane capacitance changes using patch clamp. A train of depolarizations (14 × 200 ms) resulted in identical capacitance increases (ΔC) of 0.087 ± 0.012 fF/pF (n = 80) in S7w3 beta cells and 0.084 ± 0.013 fF/pF (n = 39) in primary beta cells (Fig. 2l–m). Cell size, as assessed by the initial cell capacitance (Cm0), was slightly larger for SC-cells than for primary beta cells (by 27 %, P = 0.0001, unpaired t-test) (Supplementary Fig. 2f).

Docking and exocytosis of insulin granules at the plasma membrane were studied by total internal reflection (TIRF) microscopy (Fig. 2n). Depolarization of S7w0 cells with elevated K⁺ (in the presence of diazoxide to prevent spontaneous depolarization) released 0.063 ± 0.008 granules (gr) μm⁻² (n = 68 cells; Fig. 2o). Exocytosis proceeded initially with a burst (4 × 10⁻³ gr μm⁻² s⁻¹, <10 s) and later decreased (to <0.7 × 10⁻³ gr μm⁻² s⁻¹). We consistently observed more than a doubling of K⁺-stimulated exocytosis following exendin-4 treatment (0.14 ± 0.01 gr μm⁻², n = 71 cells; Fig. 2o,p). All exocytosis values are similar to those reported in primary beta cells³².

Notably, in S7w0, spontaneous exocytosis (no diazoxide) was similar in 3 mM and 10 mM glucose (0.045 ± 0.004 gr μm⁻², n = 13 versus 0.041 ± 0.002 gr μm⁻², n = 12) (Fig. 2q). In contrast, at S7w6, basal exocytosis was lower (0.025 ± 0.002 gr μm⁻², n = 40) and doubled when glucose was raised to 10 mM (0.051 ± 0.003 gr μm⁻², n = 41; Fig. 2q).

The density of docked granules was ~ 0.6 gr μm⁻² in S7w3 cells (Supplementary Fig. 2g–k), which is identical to values reported for primary beta cells³². Treatment with exendin-4 slightly increased docked granules when exocytosis was prevented with diazoxide (Supplementary Fig. 2g).

In summary, these analyses showed that SC-beta cells are equipped with the necessary ion channels, exocytosis components and intracellular signaling machinery required for fine-tuned regulation of insulin secretion.

SC-islets exhibit immature mitochondrial glucose coupling

As SC-islets display functionally mature exocytotic machinery, we next sought to uncover the extent of metabolic coupling to insulin release. Glucose-induced mitochondrial respiration is another characteristic feature of functional adult islets, which correlates with GSIS^2,33,34,35. We assayed oxygen consumption rate (OCR) during glucose stimulation (Fig. 3a) and observed that glucose increased mitochondrial respiration in primary islets but not in SC-islets, despite similar insulin secretion dynamics (Fig. 1j). This lack of respiratory response to glucose was not explained by aberrantly low or high basal respiration rates in SC-islets (Fig. 3b). In contrast, SC-islets responded with increased respiration rates and insulin secretion to high concentrations of pyruvate, while primary islets remained unresponsive (Fig. 3c,d). This is indicative of a retention of immature metabolic characteristics in SC-islets as genes responsible for pyruvate sensitivity are ‘disallowed’ in adult islets³⁶. Direct stimulation of mitochondrial metabolism using glutamine and leucine triggered similar insulin release in both primary islets and SC-islets, while the increase in respiration rates was slightly higher for primary islets (Fig. 3e,f).

Since oxidative glucose metabolism is considered essential for the activation of the triggering pathway of insulin secretion, we next sought to clarify if a compensatory metabolic amplifying pathway may help explain SC-islet function despite the low oxidative metabolic response. We therefore exposed SC-islets and adult islets to high glucose under tolbutamide stimulation to determine the degree of insulin secretion occurring independently from K_ATP-channel closure. S7w3 and w6 SC-islets demonstrated a stronger initial insulin secretion response to tolbutamide than adult islets. Subsequent glucose-dependent metabolic amplification was detected in S7w3 SC-islets, but it was transient and lower compared with the initial K_ATP-channel dependent secretion (Fig. 3g). Conversely, adult islets displayed a sustained K_ATP-channel independent glucose-responsive amplification more similar in magnitude to that of K_ATP-channel dependent secretion, as has been reported in previous studies²⁸.

Taken together, glucose processing seems aberrant or immature in SC-islets, resulting in undetectable mitochondrial respiratory responses. However, mitochondrial activity seems intact since respiration increases and dynamic insulin release can be elicited with other direct mitochondrial substrates, suggesting this discrepancy is not due simply to a low proportion of beta cells in SC-islets. Glucose-dependent insulin secretion independent of the K_ATP-channel is weakly present in SC-islets, suggesting metabolic amplification is a minor factor in explaining the discrepancy between robust insulin secretion and weak glucose-responsive respiration.

SC-islets demonstrate an immature glucose metabolism

To further probe the discrepancy between SC-islet functionality and low respiratory coupling we investigated how glucose metabolism differed between primary islets and SC-islets. We performed metabolite tracing analyses using uniformly labeled [U-¹³C6]-glucose comparing S7w0, w3 and w6 SC-islets together with primary adult islets, under low (3 mM) and high labeled glucose (17 mM) conditions (Fig. 4a).

Beta cell glucose-sensing is mediated in part by the hexokinase step of glycolysis³⁴. Over the course of SC-islet maturation, we detected a reduction in the ratio of labeled glucose-6-phosphate (G6P) to labeled glucose under both low and high glucose conditions (Fig. 4b), suggesting a tighter control of glucose uptake and phosphorylation. This pattern is also evident from the relative abundances of both labeled G6P and residual labeled glucose under low glucose conditions (Supplementary Fig. 3a). Only primary islets displayed a trend for glucose-concentration-dependent increases in the G6P/glucose ratio, which may indicate a more complete degree of regulation of the hexokinase step. We also observed reduced labeled levels of 3-phosphoglycerate (3-PG) and phosphoenolpyruvate (PEP) in SC-islets compared with primary islets, despite similar levels of labeled dihydroxyacetone phosphate (DHAP) (Fig. 4c). These results are consistent with a proposed glycolytic ‘bottleneck’ due to reduced GAPDH activity¹². A significant decrease in the production of labeled lactate was a strong characteristic of SC-islet maturation during S7 (Fig. 4d) and is in agreement with studies that link lactate overproduction to reduced GSIS^37,38. The diversion of 3-PG into de novo serine and glycine biosynthesis, which is low in primary islets but significantly higher in less mature SC-islets, is another possible avenue of aberrant glucose metabolism (Fig. 4e).

SC-islets showed increased labeled glucose incorporation into the core tricarboxylic acid (TCA) cycle metabolites citrate, alpha-ketoglutarate (αKG), fumarate and malate upon stimulation with high glucose, but the response was clearly lower than in primary islets (Fig. 4g). Ratiometric analyses of labeled lactate:pyruvate and labeled cis-aconitate:pyruvate further demonstrated that metabolic trafficking of pyruvate is biased towards lactate production in SC-islets, and citrate/isocitrate formation in primary islets (Fig. 4f). We inferred flux through the TCA cycle by tracing the degree of ¹³C-glucose-derived carbon incorporation into each TCA metabolite. Primary islets showed enhanced oxidative TCA cycling for citrate, αKG, fumarate and malate compared with SC-islets (Supplementary Fig. 3b–e), as well as enhanced flux through the anaplerotic pyruvate carboxylase reaction, which resulted in a high proportion of M+3 malate and fumarate isotopologues (Supplementary Fig. 3b–e). A ratiometric analysis of M+2/M+3 malate isotopologues indicated an increase in the bias towards the anaplerotic pyruvate carboxylase reaction during SC-islet maturation, towards the level seen in primary adult islets (Supplementary Fig. 3f).

Aspartate and glutamate are components of the malate–aspartate redox shuttle, a key constituent of beta cell metabolism that supports glucose-stimulated insulin secretion³⁹. Primary islets used a significantly higher proportion of glucose-derived carbons to generate these amino acids than SC-islets (Fig. 4h), as well as a significantly higher amount of labeled glutathione (GSH) (Fig. 4i). In contrast, the synthesis of labeled glycine (another GSH component) was barely detectable in primary islets (Fig. 4e). Of note, primary islets maintained higher total levels of reduced and oxidized forms of glutathione and the electron carriers NAD and NADP (Supplementary Fig. 3h). This is also reflected in the glucose concentration-dependent shifts in NAD⁺/NADH ratio, which were significantly more responsive in primary islets than in SC-islets (Supplementary Fig. 3g).

We next determined the extent of glucose-dependent changes in ATP/ADP ratio, an important determinant of the K_ATP-channel-dependent triggering pathway. SC-islets displayed a low, but nonsignificant, increase in the ATP/ADP ratio following glucose stimulation, as determined by metabolomic data and enzymatic assay (Supplementary Fig. 3i-j). In contrast, primary islets displayed a significant degree of glucose coupling to ATP/ADP ratio shifts.

Other metabolic pathways have been proposed to work in concert with the canonical oxidative phosphorylation-coupled insulin secretion model. The PEP cycle is one such model that has been suggested to function through anaplerotic regeneration of PEP from mitochondrial oxaloacetate⁴⁰. A hallmark of such cycling is the presence of M+2 labeled PEP and pyruvate, as oxidative generation of M+2 oxaloacetate would also be used in PEP regeneration. However, we were unable to detect M+2 PEP in SC-islet or primary islet samples (Supplementary Fig. 3k), and only a low percentage of M+2 pyruvate (<5%), which is in agreement with another recent study¹².

Reductive carboxylation of αKG to isocitrate and citrate via the IDH2 enzyme to fuel cytosolic redox reactions has been proposed recently as another mechanism of modulating insulin release in beta cells^41,42,43. Using ¹³C5-glutamine labeling, we observed that such reactions do occur in SC-islets, demonstrated by the high degree of M+5 cis-aconitate enrichment, an isotopologue that could only be generated by such a reductive carboxylation reaction (Supplementary Fig. 3l). By tracking the isotopologue profile of M+3 malate, we could infer the export of citrate (or isocitrate) from the mitochondria as a component of the pyruvate–citrate, pyruvate–malate and/or glycerolipid/FFA cycle (Supplementary Fig. 3l). We detected the generation of labeled pyruvate following labeled glutamine treatment, demonstrating some degree of pyruvate regeneration from TCA metabolites (Supplementary Fig. 3l).

Thus, primary human islets and SC-islets differ not only in their core TCA cycle turnover and respiration rates under glucose stimulation, but also in the production of TCA-derived metabolites and redox pathway components. Despite the differences in both glycolytic and mitochondrial glucose metabolism (Fig. 4j), SC-islets do display dynamic glucose-sensitive insulin secretion responses.

SC-islets control the glycemia of mice in vivo

To investigate the in vivo functional potential of immature (S7w0) and more mature (S7w3) SC-islets, we implanted them under the kidney capsule of nondiabetic mice^{2,9,10,44,45,46} (Supplementary Fig. 4a). Circulating human C-peptide was detectable at 1 month postengraftment in all engrafted mice. However, mice engrafted with S7w3 SC-islets demonstrated twofold higher human C-peptide levels at 2 and 3 months than S7w0 engrafted mice (Supplementary Fig. 4b). Correspondingly, blood glucose levels at 3 months were lower in S7w3 SC-islet engrafted animals (Supplementary Fig. 4c) and reached the human glycemic set point (4.5 mM) by 3 months postengraftment, as reported in primary islet engraftment studies⁴⁷. Glucose tolerance tests showed regulated insulin secretion in response to glucose injection in mice carrying both types of grafts (Supplementary Fig. 4d), but the glucose clearance was more rapid in S7w3 engrafted mice (Supplementary Fig. 4e,f). Next, we tested whether the S7w3 SC-islet grafts could sustain normoglycemia after streptozotocin (STZ)-induced loss of endogenous mouse beta cells. Glucose tolerance tests before and after STZ treatment (after 4 months of engraftment, with assays at 5 months postengraftment) showed that both control and STZ-treated animals presented robust glucose-regulated C-peptide secretion (Supplementary Fig. 4g). Despite C-peptide levels being lower in the STZ group, the glucose levels were similarly controlled in both groups (Supplementary Fig. 4h). The proportions of INS⁺ and GCG⁺ cells in the graft were not affected by the STZ treatment (Supplementary Fig. 4i,j). After removal of the engrafted kidney, the blood glucose levels increased sharply (Supplementary Fig. 4k), demonstrating that the engrafted SC-islets were actively controlling the glycemia of the diabetic mice. Extended in vitro culture in S7 conditions thus confers SC-islets a degree of maturation that results in improved functionality upon engraftment in vivo.

SC-islets transcriptionally mature in vitro and in vivo

To investigate the transcriptional changes associated with in vitro and in vivo SC-islet maturation, we performed single-cell RNA (scRNA) sequencing on SC-islets during in vitro differentiation (S5, S7w0, S7w3 and S7w6), as well as SC-islet grafts retrieved at 1, 3 and 6 months postengraftment (Fig. 5a). We obtained a dataset comprising 38,978 cells, which we integrated with previously published datasets from S5 hPSC-derived cells (4,458 cells) and human adult islets (19,435 cells)^48,49. The full integrated dataset had a total of 62,871 cells, including 46,261 endocrine cells that were selected for further study (Supplementary Fig. 5a–d and Supplementary Table 1; Methods).

The endocrine cell dataset was clustered according to time of origin and cell identity (Fig. 5b,c, Supplementary Fig. 5e-g and Supplementary Table 1). These populations reconstructed a differentiation continuum, from multipotent pancreatic progenitors, through intermediate differentiating stages and finally into beta (44%), alpha (33%), delta and gamma cells (Fig. 5c–d, Supplementary Tables 1 and 2 and Supplementary Fig. 5f–i). We also detected a cell population expressing FEV (a beta cell developmental transcription factor) that could represent SC-EC (Supplementary Fig. 5i). To cross-reference this and other cell types, we integrated our dataset with those described by Veres et al.⁷. Most of the cell types identified in our dataset clustered with the equivalent cell populations from the Veres et al. dataset (Supplementary Fig. 7). The cells identified as SC-EC by Veres et al. also clustered together with our FEV -expressing SC-EC cells, suggesting that they are the same cell type. However, the proportion of SC-EC cells identified in our dataset declined to undetectable levels during SC-islet maturation (Supplementary Fig. 7a).

We then focused on SC-beta cell subpopulations to determine the transcriptional changes associated with functional maturation (Fig. 5d–e and Supplementary Fig. 5j). Principal component and correlation analyses indicated that S7w3 and S7w6 SC-beta cells were transcriptionally more similar to grafted and adult beta cells than to S7w0 SC-beta cells (Fig. 5g and Supplementary Fig. 5k). We next examined the average expression of known mature beta cell marker genes across each individual sample^{7,10,50,51,52,53} (Fig. 5f). INS, G6PC2 and SIX2 gene expression increased early in in vitro culture, whereas other mature beta cell markers such as HOPX, UCN3, IAPP, CPE and FXYD2 were upregulated only upon engraftment. CHGB and MAFA expression was sharply upregulated at 6 months postengraftment, suggesting the need of extended in vivo maturation for the upregulation of these genes. Interestingly RBP4 and SIX3 were detected primarily only in adult beta cells (Fig. 5f).

Beta cell differentiation is not a synchronous and homogeneous process, as evidenced by the coclustering of beta cells from different stages of maturation (Fig. 5h). To reduce the interference introduced by heterogeneous populations, SC-beta cells were unbiasedly clustered by transcriptional similarity, rather than by time of origin, into ‘SC-early’, ‘SC-late’ and ‘Adult beta’ categories (Fig. 5i). As expected, SC-beta cell proportions in the SC-late and Adult beta categories increased with the progression of time in vitro and in vivo (Fig. 5j). Cells in the Adult beta category presented higher expression of genes related to insulin secretion (PCSK1, CPE, CHGB, ABCC8, FXYD2), beta cell maturation (MAFA, GDF15) and oxidative phosphorylation (OXPHOS) (Supplementary Table 3).

RNA velocity estimation (Fig. 5k) and pseudotemporal ordering (Fig. 5l) of beta cell subpopulations enabled us to infer a differentiation trajectory to investigate the genes differentially regulated upon beta cell maturation (Supplementary Fig. 5l–m and Supplementary Table 4). The expression of genes associated with pancreatic beta cell maturation and insulin secretion increased with pseudotime, together with ribosomal and HLA genes (Fig. 5m). Glutathione metabolism genes were also upregulated with pseudotime (Fig. 5m and Supplementary Fig. 5n), consistent with our metabolomic findings (Fig. 4i). Conversely, the expression of genes related to mTORC1 and MAPK signaling, cholesterol homeostasis, mitosis and MYC targets decreased with pseudotime (Fig. 5n). Genes associated with axon guidance (ROBO1, ROBO2) and adherens junctions were downregulated with pseudotime, indicating changes in cell migration, adhesion and cytoskeletal properties consistent with the observed cytoarchitectural changes upon maturation (Fig. 5n, Supplementary Fig. 5l–m and Supplementary Table 4). We calculated average expression levels for the genes in these processes to understand their dynamics across our dataset, showing that these mature beta program scores increased with time in vitro and in vivo, while mTORC1 and mitosis programs decreased (Fig. 6a and Supplementary Fig. 5o).

During SC-islet maturation following engraftment, SC-beta cells further upregulated genes related to beta cell maturation (G6PC2, UCN3, MAFA), insulin secretion (CHGB, GABRA2), cAMP signaling, OXPHOS and fatty acid metabolism. Whereas genes associated with mTORC1 signaling, cholesterol homeostasis and beta cell developmental transcription factors (FEV, ISL1) were downregulated (Fig. 6b–d and Supplementary Table 5). We performed an integrative analysis to understand the transcriptional differences between primary adult beta cells and the endpoints of SC-beta maturation in vitro (S7w6) and in vivo (M6) (Supplementary Fig. 5l–m and Supplementary Table 4). Glycolysis, OXPHOS and mTORC1 signaling-related genes were upregulated in more mature cells, while genes related to MAPK-, WNT- and estrogen-signaling pathways were downregulated (Supplementary Table 6 and Supplementary Fig. 6a–e). The expression of voltage-gated Na⁺ channel subunits was also downregulated upon extended maturation (Supplementary Fig. 6f), consistent with our electrophysiology findings (Fig. 2c).

Transcriptional markers of in vitro SC-beta maturation

We then investigated the transcriptional changes specifically occurring during in vitro maturation (S7w0 to S7w6) to better understand the acquisition of in vitro function. We found that beta cell maturation markers (IAPP, G6PC2, GLIS3), together with exocytosis-related genes (KCNJ11, RAB27A, VAMP8), unfolded protein response pathway genes (TRIB3, DDIT3, HSPA5)⁵⁴ and immediate-early transcription factors (FOS, JUN) were all significantly upregulated (Fig. 6e–f and Supplementary Table 7). Transcription factors involved in beta cell differentiation (SIX2, HOPX, ZBTB20)^51,55,56 and cell cycle inhibition (CEBP transcription factor family⁵⁷, and SCRT1^58,59), were also upregulated in S7w6 SC-beta cells, together with ligands (WNT4, TFF3)^4,60 and receptors associated with beta cell function (GCGR, GABRA2, FFAR1)⁶¹ (Fig. 6f). In line with the functional results, these transcriptional changes overall suggest that S7w6 beta cells present improved insulin production and exocytosis (Fig. 1j and Fig. 2q), which are associated with increased endoplasmic reticulum-stress levels and reduced proliferation (Fig. 1d)—all important hallmarks of mature beta cells^22,54.

Glycolysis- and TCA cycle-related genes were upregulated during in vitro maturation, concomitantly with the increased expression of genes involved in noncanonical coupling processes that may act to trigger insulin exocytosis (PC, PCK2, SLC25A1, IDH1) (Fig. 6g). Contrastingly, the expression of disallowed genes such as HK1, monocarboxylate transporter SLC16A1 (MCT1) and lactate dehydrogenase isoform A (LDHA) were reduced during maturation in vitro (Fig. 6h–i). These results are consistent with our functional and metabolomics findings, strengthening the notion of tighter control of glycolytic flux and reduced trafficking of glucose into lactate upon SC-beta cell maturation (Fig. 1l and Fig. 4b–d).

Given the heterogeneity observed in Ca²⁺ signaling, we investigated to what extent SC-islet functionality could be driven by a subpopulation of SC-beta cells. Mature beta cell marker expression was indeed heterogeneous across SC-beta cells (Supplementary Fig. 6g). We therefore calculated a gene expression score to classify them into high and low ‘mature beta signature’ (Supplementary Table 8). Consistent with our previous analyses, genes associated with insulin secretion and OXPHOS were upregulated in the mature beta signature high cells, suggesting that this subpopulation could represent beta cells better suited for improved functionality (Fig. 6j).

We have made our single-cell datasets available via an interactive single-cell portal to facilitate the access and exploration of this resource (https://singlecell.broadinstitute.org/single_cell/study/SCP1526/).