Efficient control of enterochromaffin versus islet differentiation from human pluripotent stem cell-derived pancreatic progenitors

Abstract

Knowledge of the molecular cues guiding pancreatic development is critical to developing beta cell replacement therapies for the treatment of diabetes. We compare different methods of pancreatic endocrine differentiation from human pluripotent stem cells (hPSCs) and establish sequences of patterning that can selectively increase the frequencies of islet-like or off-target enterochromaffin (EC)-like cells, thereby significantly increasing islet-like cellular yield and glucose-stimulated insulin secretion. Using a model of disrupted murine islet development that gives rise to pancreatic EC-like cells, we identify persistent Neurogenin 3 (NGN3) expression as a conserved feature associated with human and murine pancreatic EC-like cell differentiation. Finally, by comparing the phenotypes obtained through different patterning strategies, we observe that endocrine subtypes can vary significantly in their expression of canonical lineage markers. In addition to expanding our understanding of pancreatic endocrine lineage allocation and identity, these findings establish a logical differentiation framework to guide the generation of designer hPSC-islets for research and therapeutic applications.

How different islet lineages emerge within pancreatic islets remains largely unknown. Here, using hPSC and mouse models, the authors identify mechanisms that influence islet versus enterochromaffin lineage specification and islet functionality.

Introduction

The ability to differentiate human pluripotent stem cells (hPSCs) into pancreatic endocrine cells in vitro offers unprecedented access to islet-like cells (hPSC-islets), both for research into their function and pathology, and for the therapy of type 1 diabetes (T1D). Although many pathways that promote the differentiation of hPSC-derived pancreatic progenitors (hPSC-PPs) into pancreatic endocrine cells have been identified^1–7, little is known about the molecular cues that direct hPSC-PPs towards specific islet lineages. As a result, hPSC-PPs can be differentiated into cells with nearly ubiquitous expression of endocrine markers such as synaptophysin¹, NKX2-2^1,8 and chromogranin A (CHGA)^1,8,9, but the allocation of these endocrine cells towards each hormonal lineage, including insulin-producing beta-like cells, remains uncontrolled and difficult to modulate. This is a challenge for therapeutic and research applications of hPSC-islets, which may also require a regulated composition of islet endocrine subtypes to ensure a consistent safety and functional profile after transplantation.

One of the endocrine fates spontaneously acquired during hPSC-islet differentiation is characterized by expression of the vesicular serotonin (5-hydroxytryptamine) transporter SLC18A1¹⁰. This gene is not known to be expressed in human pancreatic islets, but rather is restricted in derivatives of the definitive endoderm to enterochromaffin (EC) cells of the stomach and intestine¹¹. Although the impact of these hPSC-derived EC-like cells on hPSC-islet function remains unclear, these cells are generally considered contaminants and are thus undesired. Conversely, given the rarity of EC cells in intestinal organoids generated from human pluripotent^12–14 and intestinal^15–17 stem cells, the ability to efficiently generate EC-like cells from pancreatic models could offer the possibility for disease modeling and therapy. A detailed understanding of the developmental mechanisms underlying the specification of EC-like cells from hPSC-PPs thus represents an unmet need in the field of hPSC-islet biology.

Here, we describe our efforts to understand mechanisms regulating enterochromaffin versus islet lineage allocation. After observing that enterochromaffin cells can emerge from pancreatic-committed progenitors in vitro and in vivo, we identify the MAPK/ERK and BMP pathways as candidate modulators of endocrine lineage specification. Building on these findings, we then find that the method of hPSC-PP differentiation synergizes with the factors later applied to induce endocrine differentiation to determine the final endocrine composition. These findings provide a valuable framework for directing hPSCs towards specific endocrine subtypes, and provide a platform for exploring enterochromaffin cell biology.

Results

Latrunculin A improves specification of primitive streak and enhances endoderm and mesoderm differentiation

An important limitation in our protocol for the directed differentiation of hPSCs towards the pancreatic lineage (Fig. 1a) is that the differentiation of hPSCs into day 1 Brachyury (BRA)⁺ primitive streak-like cells (PS), the first step in the progression of hPSCs into definitive endoderm (DE), is highly influenced by the confluency of plated cells (Fig. 1b–d)^18,19. Recently, the organization of the actin cytoskeleton around tight junctions in hPSCs grown to confluency was found to limit primitive streak differentiation²⁰. In order to reduce variability related to density, we thus cultured hPSCs (day 0) at low and high densities (Fig. 1b), with or without the actin destabilizer Latrunculin A (LatA) on the first 24 h of differentiation (day 0 to 1). As expected, high-density cells exhibited an impairment in their ability to generate BRA⁺ cells (Fig. 1c, d). This effect was completely abrogated with LatA treatment at day 0, yielding nearly pure populations of BRA⁺ primitive streak cells 24 h later (Fig. 1c, d). At day 3 of differentiation, LatA-treated high-density hPSC cultures generated similar frequencies of CXCR4⁺KIT⁺ cells (Supplementary Fig. 1a), but significantly higher frequencies of FOXA2⁺SOX17⁺ DE-like cells (Fig. 1e, f). At day 13, LatA-treated cells had a greater frequency of PDX1⁺NKX6-1⁺ cells (Fig. 1g, h) and a significant reduction in the PDX1⁻NKX6-1⁻ population (Supplementary Fig. 1b). By day 23 of differentiation, LatA-treated cells exhibited enhanced differentiation into CHGA⁺ endocrine cells (Supplementary Fig. 1c), associated with a greater frequency of NKX6-1⁺CPEP⁺ beta-like cells compared to the control protocol (Fig. 1i, j). Thus, the effects of improved PS differentiation with LatA resulted in enhanced hPSC-PP and hPSC-islet differentiation, which was also confirmed in two independent hPSC lines (Supplementary Fig. 1d–i).

Fig. 1. Latrunculin A improves primitive streak differentiation. Also see Supplementary Fig. 1.

a Schematic describing the sequence of factors applied during hPSC-islet differentiation, with or without LatA. Created in BioRender. Misra, P. (2026) https://BioRender.com/gqq1211. b Light microscopy image of hPSCs cultured at low and high density prior to differentiation. Scale bar length = 100 µm. Representative images from n = 6 differentiations. c Representative flow cytometry plots for BRACHYURY (BRA) and SOX2 expression at 24 h after the initiation of differentiation. d Quantification of BRA⁺ cells at 24 h after the initiation of differentiation (n = 5 independent differentiations). One-way ANOVA followed by Holm-Sidak correction. Adjusted p-values for all comparisons performed are shown in the figure. e–j Representative flow cytometry plots and quantification of FOXA2 and SOX17 at day 3 (e, f, n = 7 independent differentiations), NKX6-1 and PDX1 at day 13 (g, h, n = 7 independent differentiations), and NKX6-1 and CPEP at day 23 (i, j, n = 6 independent differentiations) of differentiation with or without LatA. For panels (f, h, and j), distributions were compared by a one-tailed paired t test. k Schematic describing the sequence of factors applied during hPSC-lung progenitor differentiation, with or without LatA. Created in BioRender. Misra, P. (2026) https://BioRender.com/8tjs70k. l, m Representative flow cytometry plots (l) and quantification (m) for NKX2-1 at day 13 (n = 3 independent differentiations). One-tailed paired t test. n Immunofluorescence of control and LatA-treated cells differentiated to NKX2-1⁺ lung progenitors. The dashed white box in the merged panels is magnified in the DAPI and NKX2-1 panels. Representative images from n = 3 differentiations. For panels (d, f, h, j, and m), all data are presented as mean values +/− SEM. Legend: ATRA, all-trans retinoic acid; C-Peptide, (CPEP); CTRL, control; DBZ, Dibenzazipine; DM, Dorsomorphin; LatA, Latrunculin A; LDN, LDN193189; PD03, PD0325901; RPSX, Repsox; T3, 3,3′,5-Triiodo-L-thyronine.

To determine whether LatA bestowed an enhanced competency to form multiple lineages or specifically promoted pancreatic commitment, DE generated from control and LatA-treated hPSCs were differentiated to NKX2-1⁺ lung progenitors according to a previously published protocol (Fig. 1k)²¹, and higher frequencies of NKX2-1⁺ lung progenitors were observed with LatA (Fig. 1l–n). Similarly, using a previously published mesoderm protocol²², we observed that inclusion of LatA at day 0 significantly enhanced the frequency of PDGFRA⁺ cells by day 3 (Supplementary Fig. 1j, k). Taken together, our findings demonstrate a role for LatA in broadly supporting early differentiation of hPSCs into primitive streak and downstream mesodermal and endodermal lineages. Given these results, all subsequent experiments were performed using LatA on day 0 of differentiation.

Activin A, all-trans retinoic acid, and WIKI4 synergistically pattern definitive endoderm towards NKX6-1⁺ pancreatic progenitors

We next focused on improving the patterning of newly formed endoderm (days 3-6, Fig. 2a), as morphogenic stimuli encountered at this stage are known to be particularly influential in downstream pancreatic commitment^2,3,23–26. Based on previous reports of morphogens that influence pancreatic specification^3,25,27,28, we treated day 3 DE-like cells for 3 days with different combinations of the tankyrase/canonical Wnt inhibitor WIKI4, all-trans retinoic acid (ATRA), and Activin A in our base media containing FGF10 and Dorsomorphin (Fig. 2a). By day 13 of differentiation, the combination of all three factors (ARW) gave rise to the highest percentage of PDX1⁺NKX6-1⁺ cells with a mean of 92% compared to 64% in the protocol without any of these added compounds (Fig. 2b, c and Supplementary Fig. 2a). This combination also gave rise to the lowest frequencies of CDX2⁺, NEUROG3⁺, and CHGA⁺ cells at day 13 (Fig. 2c and Supplementary Fig. 2b–d).

Fig. 2. High NKX6-1 expression does not prevent EC-like cell specification. Also see Supplementary Fig. 2.

a Schematic describing the design of panels (b–e), with or without Activin A, WIKI4, and ATRA, in addition to all factors present in Fig. 1a. Created in BioRender. Misra, P. (2026) https://BioRender.com/cyuyccb. b Representative flow cytometry plots for NKX6-1 and PDX1 expression at day 13 of differentiation across all conditions tested. c Frequencies of PDX1⁺NKX6-1⁺, CDX2⁺, CHGA⁺, and NGN3⁺ populations at day 13 across all conditions tested as assessed by flow cytometry (n = 4 independent differentiations). d Representative flow cytometry plots for NKX6-1 and CPEP expression at day 23 of differentiation across all conditions tested. e Frequencies of NKX6-1⁺CPEP⁺ (n = 5 independent differentiations), CHGA⁺ (n = 5 independent differentiations), SLC18A1⁺ (n = 4 independent differentiations) and GCG⁺ (n = 4 independent differentiations) populations at day 23 as assessed by flow cytometry. For panels (c, e), all data are presented as mean values +/- SEM. f Schematic describing breeding strategy used to generate Isl1-CKO (Neurod1-Cre; Isl1^fl/fl) mice. Created in BioRender. Misra, P. (2026) https://BioRender.com/7ma8k1c. g, h Representative immunofluorescence images of Control and Isl1-CKO mouse islets from n = 5 independent mice for each genotype. Scale bars represent 50 μm. i, j Quantification of SLC18A1⁺ (i, n = 45 areas evaluated in Control and n = 36 in Isl1-CKO) and 5HT⁺ (serotonin⁺) (j, n = 36 areas evaluated in Control and n = 42 in Isl1-CKO) area in Control vs. Isl1-CKO mice at P9 from n = 5 independent mice each. Unpaired one-tailed Student’s t test with Welch’s correction. Legend: 5HT, serotonin; A, Activin A; ATRA or R, all-trans retinoic acid; DBZ, Dibenzazipine; DM, Dorsomorphin; LDN, LDN193189; PD03, PD0325901; P9, postnatal day 9; RPSX, Repsox; W, WIKI4.

We next studied how these protocol modifications affected subsequent endocrine commitment using a 2D/planar culture format that included tankyrase inhibition in the endocrine induction media (Fig. 2a) to improve endocrine and beta cell differentiation⁴. However, despite the large differences in NKX6-1 expression at day 13, we observed only minor differences in hPSC-islet composition at day 23 (Fig. 2d, e and Supplementary Fig. 2e–g). Strikingly, all protocols generated a high frequency of so-called off-target SLC18A1⁺ EC-like cells (Fig. 2e and Supplementary Fig. 2h). These findings suggest that improving pancreatic commitment at the hPSC-PP stage, as assessed by NKX6-1 expression at day 13, does not prevent specification of EC-like cells.

Given these findings, we next sought to determine whether pancreatic progenitors that arise during development in vivo might have a latent potential to generate EC cells. We and others have previously found that the absence of Isl1 can promote the emergence of apparent hormone-negative cells within the islets^29,30, some of which express NKX6-1³¹. Interestingly, ISL1 has an established role as a suppressor of EC specification within the intestine³². To determine whether the hormone-negative cells in islets of Isl1 knockout mice might in fact be EC cells, we examined the phenotype of islets in Rosa26-tdTomato reporter Ai14 mice with a conditional Isl1 knockout (Isl1-CKO) following Neurod1 expression (Neurod1-Cre; Isl1^fl/fl, Fig. 2f)³¹. In this model, Isl1 knockout only occurs after Neurod1 is expressed, and its loss should therefore not compromise progenitor specification and commitment to pancreatic lineages.

First, we assessed Slc18a1 expression in our previously-published³¹ bulk RNA-sequencing data of FACS-sorted tdTomato⁺ endocrine cells from Control (Neurod1-Cre; Isl1^fl/+) and Isl1-CKO (Neurod1-Cre; Isl1^fl/fl) mice. Although transcript levels were low and comparable between Control and Isl1-CKO mutants at embryonic day 14.5 (E14.5, Supplementary Fig. 2i), we observed considerable Slc18a1 upregulation in Isl1-CKO mice at postnatal day 9 (P9), along with two additional EC markers, Tph1 and Tac1 (Supplementary Fig. 2j–l). Using immunofluorescence, we found widespread expression of SLC18A1 and serotonin in tdTomato⁺ islet cells of Isl1-CKO mice, compared to minimal expression in Control islets (Fig. 2g–j). Considering that the genetic perturbations in Isl1-CKO pancreatic cells occur only after endocrine commitment has begun and should therefore not perturb progenitor specification, these findings support the hypothesis that primary fetal pancreatic progenitors can give rise to EC cells.

PD03 treatment during endocrine specification promotes EC-like cell differentiation

Having established that primary pancreatic progenitors harbor the potential for EC differentiation, we next sought to identify whether factors in our endocrine induction media (days 13–16) might modulate the generation of SLC18A1⁺ cells. We first evaluated the MEK1/2 inhibitor PD03 and the tankyrase inhibitor WIKI4, which were already present in our protocol (Fig. 1a), along with LatA, which had been recently described to enhance endocrine differentiation of pancreatic progenitors cultured in planar format (Supplementary Fig. 3a)⁵. We omitted SANT-1 and the BMP inhibitor LDN, which had unclear roles at this stage, in order to improve interpretability of the signaling logic, and additionally included FGF10 to improve cell yield³³. With these conditions, we observed a general trend towards greater CHGA expression when PD03, LatA, and WIKI4 compounds were all included (Supplementary Fig. 3b), mirrored by an increase in the frequency of NKX6-1⁺CPEP⁺ co-expressing cells (Supplementary Fig. 3c). Strikingly, we observed a large increase in the frequency of SLC18A1⁺ cells when LatA and PD03 were used in combination (Supplementary Fig. 3d). In contrast, the frequencies of ISL1⁺ or ARX⁺ (putative early alpha-like cells) populations remained relatively unchanged (Supplementary Fig. 3e, f). Interestingly, PD03 was also previously found to increase EC specification in intestinal organoids³⁴. To confirm this effect in our system, we tested different durations of PD03 treatment in a base protocol containing WIKI4 and LatA, and observed a large increase in SLC18A1 expression when PD03 was included throughout the endocrine commitment stages (days 13–22, Supplementary Fig. 3g, h). Thus, PD03 appeared to bias the differentiation of endocrine progenitors towards the EC lineage.

Recently, several groups described close transcriptional similarity between beta-like and EC-like cells generated from hPSC-PPs^35,36. One of these studies observed that both populations exist along a spectrum of phenotypes with overlapping expression of EC and canonical beta cell markers, including NKX6-1 and CPEP³⁵. Examining the range of phenotypes generated in our differentiations with the various combinations of PD03, WIKI4, and LatA, we observed that > 80% of all SLC18A1⁺ EC-like cells expressed NKX6-1, and that this expression pattern remained stable between days 23 and 29 (Supplementary Fig. 3i). Meanwhile, over half of EC-like cells expressed CPEP, though this expression decreased over time (Supplementary Fig. 3j). Similarly, nearly half of SLC18A1⁺ cells co-expressed both NKX6-1 and CPEP at day 23, and though this frequency had decreased by day 29, it remained considerable (Supplementary Fig. 3k). Given these observations, we speculate that triple-positive (SLC18A1⁺NKX6-1⁺CPEP⁺) cells represent an intermediate population that may resolve to SLC18A1⁺NKX6-1⁺CPEP⁻ EC-like cells at later time points. We thus modified our marker-based definition of beta-like cells to include the absence of SLC18A1 to identify cells with the highest confidence of having committed to the pancreatic beta cell lineage (SLC18A1⁻NKX6-1⁺CPEP⁺).

FGF2, BTC, and BMP4 suppress SLC18A1 expression and promote differentiation of islet lineages

After finding greater EC-like differentiation upon treatment with PD03, which has been shown to inhibit MEK1/2 activity³⁷, we wondered whether stimulating this pathway could instead reduce the emergence of these cells. Two candidate MEK/ERK-activating ligands from different molecular families were thus tested for their ability to suppress EC lineage differentiation: FGF2³⁸ and Betacellulin (BTC)^9,39–41 (Fig. 3a). We treated day 13 hPSC-PPs with each ligand for short (days 13–16) or prolonged (days 13-23) durations in the presence of both LatA and WIKI4 to maximize endocrine commitment. Across these conditions, we observed no significant impact of FGF2 and BTC on endocrine commitment at day 29, with nearly all cells in all treatment conditions expressing the pan-islet marker NKX2-2⁴² (Supplementary Fig. 3l). In contrast, all conditions reduced the proportion of SLC18A1⁺ cells (Fig. 3b, c) and variably increased the frequency of ISL1⁺ cells (Fig. 3b, d). Neither ligand affected the frequency of SLC18A1⁻NKX6-1⁺CPEP⁺ beta-like cells (Supplementary Fig. 3m), but they instead generally increased the proportion of cells expressing the alpha-lineage homeobox ARX^+43,44 (Supplementary Fig. 3n). We also observed that FGF2 treatment was specifically associated with an increase in both CPEP⁻GCG⁻SST⁺ (Supplementary Fig. 3o) and SST⁺HHEX⁺ (Fig. 3e, f) cells, suggesting enhanced delta lineage specification⁴⁵.

Fig. 3. FGF2, BTC, and BMP4 suppress SLC18A1 expression and promote differentiation of islet endocrine lineages. Also see Supplementary Fig. 3.

a Experimental design for panels b-f, testing the effects of FGF2 and BTC. Created in BioRender. Misra, P. (2026) https://BioRender.com/09lruek. b Representative flow cytometry plots for ISL1 and SLC18A1 in hPSC-islets exposed to FGF2 or BTC. c–e Quantification of hPSC-islet expression of SLC18A1 (c, n = 5), ISL1 (d, n = 5), and co-expression of HHEX and SST (e, n = 5), as determined by flow cytometry. f Representative flow cytometry plots for HHEX and SST across all conditions tested. g Experimental design for panels (g–k), testing the effects of BMP4 and BMP signaling antagonists (LDN and NOGGIN). Created in BioRender. Misra, P. (2026) https://BioRender.com/fga3ox3. h Representative flow cytometry plots for ISL1 and SLC18A1 expression in hPSC-islets exposed to BMP4 at various timepoints throughout endocrine differentiation. i, j Quantification of hPSC-islet frequencies of SLC18A1 (i, n = 5), and ARX (j, n = 5), as determined by flow cytometry. k Representative flow cytometry plots for ARX expression in hPSC-islets exposed to BMP4 at various timepoints throughout endocrine differentiation. l Experimental schematic for panels M-N, testing the additive effects of PD03 and LDN on EC-like cell differentiation. Created in BioRender. Misra, P. (2026) https://BioRender.com/9kwyzvz. m Representative flow cytometry plots for SLC18A1 expression vs. FSC of d29 cultures differentiated with PD03 and/or LDN. n Quantification of SLC18A1 frequency in day 29 hPSC-islets across conditions tested, as assessed by flow cytometry (n = 5). For panels (c–e, i, j, and n), all data are presented as mean values +/− SEM. Comparisons in all panels were done using paired one-way ANOVA followed by Holm-Sidak’s post-hoc test, comparing each condition against the control. Only significant comparisons are shown. Adjusted p-values for conditions not shown (compared to base): d) both comparisons not shown, p = 0.13. e F2 d13-16, p = 0.41; all other comparisons with base not shown, p = 0.44. i NOG d13-16 and NOG d16-23, p = 0.57; all other comparisons with base not shown, p = 0.32. j BMP4 d16-23, 0.08; LDN d13-16, 0.21; all other comparisons with base not shown, p = 0.93. Additional details for experimental schematics in panels (a, g, l): From day 0 to day 13, factors present in Fig. 1a, including LatA at day 0 were included. From days 13–16, T3, Y-27632, Heparin, FGF10, ATRA, Repsox and DBZ were included in addition to the specific factors shown. From day 16–23, T3, Heparin, FGF10, ATRA and Repsox were included in addition to the specific factors shown. Legend: BTC, betacellulin; CTRL, control; d, day; DBZ, dibenzazepine; EC, enterochromaffin; F2, FGF2; ISX9, isoxazole-9; LatA, Latrunculin A; LDN, LDN193189; PD03, PD0325901; T3, 3,3′,5-Triiodo-L-thyronine.

We next evaluated the effect of BMP pathway modulation on EC commitment, since the initial EC-biased differentiation protocol included the small molecule BMP receptor inhibitor LDN (Fig. 3g). Compared to the BMP antagonist NOGGIN (NOG) and the canonical BMP ligand BMP4, we observed no differences in NKX2-2 expression (Supplementary Fig. 3p). In contrast, BMP4 stimulation significantly decreased SLC18A1 expression (Fig. 3h, i) and increased ISL1 expression (Fig. 3h and Supplementary Fig. 3q). The shift from SLC18A1 expression to ISL1 expression with BMP4 treatments was associated with significant increases in CPEP⁻GCG⁻SST⁺ cells, though this effect was more modest than that seen with FGF2 (Supplementary Fig. 3r). Meanwhile, BMP4 treatment led to more notable increases in the alpha-lineage marker ARX (Fig. 3j, k) along with decreases in beta-like cells (Supplementary Fig. 3s), specifically when applied between days 13–16. We observed no significant effect with NOG treatment on any of the lineages characterized, while the more potent but potentially less-specific LDN⁴⁶ led to small increases in SLC18A1 expression (Fig. 3i).

Our experiments had thus far identified a surprising degree of endocrine plasticity among hPSC-PPs, particularly with regard to the EC lineage. To obtain greater control over endocrine specification, we tested the small molecule Isoxazole-9 (ISX9), which could enhance both Neurog3 along with the endocrine lineage determinant Pax4 in murine intestinal organoids⁴⁷, and has also been used in two hPSC-islet differentiation protocols that induced high frequencies of NKX6-1⁺CPEP⁺ cells^6,48. Compared with LatA, which had also been shown to increase NEUROG3 expression⁵, ISX9 yielded higher frequencies of SLC18A1⁻NKX6-1⁺CPEP⁺ beta-like cells and CHGA⁺ endocrine cells, a trend towards more SLC18A1⁺ EC-like cells, and fewer ARX⁺ alpha-lineage cells (Supplementary Fig. 3t–w). Given these effects on SLC18A1 expression, we next sought to consolidate the use of ISX9 with MEK and BMP inhibition to generate hPSC-islets highly enriched with EC-like cells. Using a base protocol containing ISX9, we tested PD03 and LDN, alone or in combination (Fig. 3l). By day 26, dramatic upregulation of SLC18A1 expression was observed in cells treated with all three compounds, reaching as high as 80% in some experiments (Fig. 3m, n). This was accompanied by a decrease in ISL1⁺ (Supplementary Fig. 3x) and SLC18A1⁻NKX6-1⁺CPEP⁺ cells (Supplementary Fig. 3y). Meanwhile, relatively minor differences were observed in the frequency of ARX⁺ cells between these conditions (Supplementary Fig. 3z). Overall, these findings demonstrate how rational combinations of small molecules and pathway modulators can be used to direct hPSC-PPs towards either islet or EC lineages.

Endocrine composition of hPSC-islets is influenced by the interaction between early (pre-progenitor) and late (post-progenitor) patterning

Equipped with a better understanding of factors that guide hPSC-PPs towards specific endocrine lineages, we revisited the impact of early endodermal patterning on endocrine lineage allocation. We first directed DE-like cells towards PDX1⁺ progenitors with either control, suppressed (using high-dose ATRA at 2 µM⁴⁹ and the TGFβ inhibitor SB-431542²⁴ from days 8-13), or enhanced (using ATRA, Activin A and WIKI4 from days 3-6 as in Fig. 2a) levels of NKX6-1⁺ (Fig. 4a). As expected, significant differences in NKX6-1 and SOX9 expression were observed at day 13, although most cells were SOX9⁺ in all protocols (Fig. 4b–d). Meanwhile, the suppressed protocol contained the highest number of precocious NKX2-2⁺ endocrine cells at this stage (median 26%), compared with rare frequencies (median 3%) in the NKX6-1-enhanced protocol (Fig. 4b, e). Next, we leveraged our earlier findings to design protocols to pattern each of these progenitors towards the alpha (LatA with BMP4), beta (ISX9 with FGF2/BTC and LDN), and EC (ISX9 with LDN and PD03) lineages (Fig. 4a, f, g).

Fig. 4. Endocrine composition of hPSC-islets is influenced by the interaction between early (pre-progenitor) and late (post-progenitor) patterning. Also see Supplementary Fig. 4.

a Experimental design for pre-progenitor (left) and post-progenitor patterning (right) patterning. From day 0 to day 13, factors present in Fig. 1a including LatA at day 0 were included. From days 13–16, T3, Y-27632, Heparin, FGF10, ATRA, Repsox, WIKI4 and DBZ were included in addition to the specific factors shown. From day 16–23, T3, Heparin, FGF10, ATRA and Repsox, WIKI4 and DBZ were included in addition to the specific factors shown. Created in BioRender. Misra, P. (2026) https://BioRender.com/mn3k3lc. b Representative flow cytometry plots for PDX1 vs. NKX6-1 (top), and SOX9 vs. NKX2-2 expression (bottom) at day 13 of differentiation. c–e Quantification of frequencies of PDX1⁺NKX6-1⁺ (c), PDX1⁺SOX9⁺ (d), and NKX2-2⁺ (e) cells at day 13 of differentiation (n = 5). All comparisons were done using paired one-way ANOVA followed by Holm-Sidak’s post-hoc test, comparing each combination of conditions. f Representative flow cytometry plots for endocrine clusters derived from NKX6-1-suppressed (top row), control (middle row) and NKX6-1-enhanced (bottom row) progenitors subjected to either alpha (left column), beta (middle column), or EC (right column) patterning. The phenotype of associated progenitors is shown in the left-most column. g Stacked bar plot demonstrating the composition of alpha-like (ARX⁺GCG⁺), beta-like (SLC18A1⁻NKX6-1⁺CPEP⁺), delta-like (HHEX⁺SST⁺), gamma-like (ARX⁺PPY⁺), other ARX⁺ (ARX⁺GCG⁻PPY⁻), and EC-like (SLC18A1⁺) cells resulting from each progenitor-patterning combination tested. The value shown for each population is the median frequency observed, as quantified in Supplementary Fig. 4a–h. h Quantification of total cell yield per well at day 13 (progenitor stage) vs day 26 (endocrine cluster stage) for each progenitor-patterning combination tested (n = 4 independent differentiations for control-alpha; n = 5 independent differentiations for all other alpha combinations, all EC-combinations, and suppressed-beta; n = 6 independent differentiations for all other beta conditions and progenitors). One-way ANOVA followed by Holm-Sidak’s post-hoc test. i Static glucose-stimulated insulin secretion test comparing insulin secretion index (secretion at 16.7 mM glucose over 2.8 mM glucose) of EC-enriched (NKX6-1-suppressed progenitors followed by EC patterning) and beta-enriched (NKX6-1-suppressed progenitors followed by EC patterning) hPSC-islets. Ratio paired one-tailed Student’s t test for mean values across each of n = 9 independent sets of hPSC-islet clusters. For panels (c–e, h, and i), all data are presented as mean values +/− SEM. Legend: ACT, Activin A; ACTi, Activin receptor inhibitor (SB431542); ATRA, all-trans retinoic acid; CTRL, control; EC, enterochromaffin; ISX9, Isoxazole 9; LatA, Latrunculin A; LDN, LDN193189; PD03, PD0325901.

As expected, we observed that alpha patterning generated higher levels of ARX compared to beta and EC patterning (Supplementary Fig. 4a). Among progenitors exposed to alpha patterning, NKX6-1-suppressed progenitors generated the highest frequency of ARX⁺GCG⁺ alpha-like cells (Fig. 4f, g and Supplementary Fig. 4b), and higher frequencies of ARX⁺CPEP⁺ and ARX-dependent gamma-like (ARX⁺PPY⁺) cells (Supplementary Fig. 4c–e). We also studied the effects of alpha patterning in the HES3-INS^GFP/+GCG^mCherry/+ hESC line, which directly reports on transcriptional activity from the INS and GCG loci⁵⁰. As with the H1 hESC line, alpha patterning produced hPSC-islets with striking enrichment of GCG^mCherry+ cells (Supplementary Fig. 4f), supporting the generalizability of these findings.

Next, the greatest degree of beta-like differentiation was achieved with the NKX6-1-enhanced progenitors exposed to beta patterning conditions (Fig. 4f, g and Supplementary Fig. 4g). Surprisingly, NKX6-1-suppressed progenitors exposed to beta patterning demonstrated instead an enhanced capacity for differentiation into delta-like (SST⁺HHEX⁺) cells (Fig. 4f, g and Supplementary Fig. 4h). Meanwhile, EC patterning successfully generated SLC18A1⁺ cells across all progenitor conditions (Fig. 4f, g and Supplementary Fig. 4i), associated with a commensurate decrease in ISL1 expression (Supplementary Fig. 4j). Notably, NKX6-1-suppressed progenitors generated the highest frequency of SLC18A1⁺ cells (Fig. 4f, g and Supplementary Fig. 4i) and the lowest frequency of ISL1⁺ cells (Supplementary Fig. 4j), while the opposite trend was observed for NKX6-1-enhanced progenitors. We validated these trends using immunofluorescence, where we additionally observed that SLC18A1⁺ cells preferentially localized on the periphery of hPSC-islet clusters in all conditions (Supplementary Fig. 5a).

Interestingly, although all three progenitors appeared as flat sheets of cells on day 13, both beta and EC patterning led to spontaneous self-aggregation of nearly all cells into well-defined spheroid clusters by day 23, reminiscent of the appearance of isolated pancreatic islets (Supplementary Fig. 5b). We validated these results in a different induced pluripotent stem cell line, which reproduced the findings described above in the H1 hESC line for each progenitor-patterning combination (Supplementary Fig. 5c, d).

When monitoring cell counts, we observed that the NKX6-1-suppressed protocol generated a high number of progenitors on day 13 (Fig. 4h), leading to the generation of high numbers of EC-like cells with both beta and EC patterning (Supplementary Fig. 5e). Importantly, only EC patterning of these progenitors demonstrated high EC-like cell purity in addition to yield (Fig. 4f, g and Supplementary Fig. 4i), which on average was 850,000 EC-like cells out of an average of 1,020,000 total cells per 3.5 cm² well (one well of a 12-well plate). Meanwhile, although the yield of NKX6-1-enhanced progenitors was lower compared to the other protocols, beta patterning enabled the generation of islet-like cells enriched in beta cells without sacrificing cell yield (Fig. 4h). We quantified the absolute number of beta-like cells from this combination to be on average 700,000 cells per 3.5 cm² well, out of an average of 1,600,000 cells per well. When compared to a recent protocol designed to support cell survival during endocrine commitment⁵¹, this 2D differentiation method for beta-like cells generated higher frequencies and yield of beta-like and a lower frequency of EC-like cells (Supplementary Fig. 5f–h). Importantly, the insulin-secreting functionality of the beta-enriched protocol was significantly higher than the EC-enriched protocol, supporting the importance of reducing the EC frequency within hPSC-islets (Fig. 4i).

Pancreatic EC-like cell differentiation is characterized by prolonged Neurogenin 3 expression in vitro and in vivo

Although beta-like and EC-like cells required distinct conditions for optimal differentiation, they shared several features, including: a) co-expression of NKX6-1 and INS, b) enhanced differentiation with use of ISX9, and c) a speculated shared requirement for PAX4. To better understand the differences between hPSC-PP differentiation into beta-like and EC-like cells, we subjected NKX6-1-enhanced progenitors to EC-like or beta-like patterning, and studied the kinetics of endocrine-related proteins starting from the hPSC-PP stage (Fig. 5a–c and Supplementary Fig. 6a). With both programs, a rapid decrease in the pancreatic progenitor marker SOX9 (Fig. 5d) was associated with increases in NKX2-2 (Fig. 5e) and CHGA (Fig. 5f). We observed low expression of ARX in both conditions, perhaps due to the suppressive effects of ISX9 and LDN (Supplementary Fig. 6b), while NKX6-1 decreased initially in both conditions until day 15, after which it increased again (Supplementary Fig. 6c). Overall, these analyses suggest that endocrinogenesis occurs largely with similar kinetics in both EC-like and beta-like differentiation.

Fig. 5. Pancreatic EC-like cell differentiation is characterized by prolonged NEUROG3 expression in vitro and in vivo. Also see Supplementary Fig. 5.

a Representative flow cytometric plots characterizing SLC18A1 and ISL1 expression of NKX6-1-enhanced hPSC-PPs undergoing either beta cell or EC patterning, as in Fig. 4a. b–g Quantification of the frequencies of SLC18A1⁺ (b), SLC18A1⁻NKX6-1⁺CPEP⁺ (c), SOX9⁺ (d), NKX2-2⁺ (e), CHGA⁺ (f), and NGN3⁺ (g) cells following beta cell or EC differentiation (n = 3 each). For panels (b–g), all data are presented as median values +/− SD. h Representative flow plots for NGN3 and NKX2-2 expression in hPSC-PPs undergoing either beta cell or EC patterning. i Experimental design for single-cell RNA sequencing of Neurod1 (tdTomato)⁺ endocrine cells in Control and Isl1-CKO Ai14 mouse islets. Created in BioRender. Misra, P. (2026) https://BioRender.com/21jfb5p. j UMAP of P9 endocrine single cells from Control and Isl1-CKO mice. k Split UMAP demonstrating clusters of endocrine cells from Control and Isl1-CKO mice. l Comparison of frequency of Slc18a1-expressing cells (count > 0) in each cluster of endocrine cells, stratified by mouse genotype. m Dot plot highlighting expression patterns of endocrine lineage genes across each cluster. Legend: EC, enterochromaffin; FACS, fluorescence-activated cell sorting; PPs, pancreatic progenitors.

NGN3 (encoded by NEUROG3) is a transcription factor that is transiently required for both pancreatic and intestinal endocrine differentiation^33,52,53. We observed that NGN3 expression was minimal at day 13 in both conditions but rapidly increased and peaked within 48 hours (Fig. 5g, h). Notably, while beta patterning was characterized by a rapid decline in NGN3 to <10% by day 20, progenitors undergoing EC patterning exhibited a slower decline, retaining levels over 20% on day 20 (Fig. 5g, h). To determine whether perturbed Neurogenin 3 kinetics might be a generalized feature of pancreatic EC differentiation, we examined its expression in Isl1-CKO mice. Consistent with EC-like differentiation in hPSC-islets, we observed upregulation of Neurog3 transcript in bulk RNA-seq of Isl1-CKO mice at P9 (Supplementary Fig. 6d), compared to similar expression in the two groups at day E14.5 (Supplementary Fig. 6e).

To gain further molecular insight into the specification of pancreatic EC-like cells, we analyzed single-cell RNA sequencing data of tdTomato (Neurod1)⁺ islet cells isolated from P9 Isl1-CKO and Control mice⁵⁴ (Fig. 5i, j). In mutants, we observed the distinct absence of Sst⁺ clusters along with a unique cluster of EC-like cells not found in controls with very high expression of Slc18a1 (Fig. 5k, l and Supplementary Fig. 6f–h). These EC-like cells expressed high levels of Fev (Fig. 5m), which is also a notable characteristic of primary EC cells⁵⁵. In addition, EC-like cells also expressed Ins1, Ins2, and Nkx6-1 (Fig. 5m), similar to EC-like cells in hPSC-islets (Supplementary Fig. 3i, j)^10,35,36. With regards to Neurog3, we again observed increased expression, both in terms of the number of Neurog3⁺ cells and also in terms of their level of expression (Supplementary Fig. 6i, j). Its expression was primarily within progenitors and delta cells in Control mice, while mutants exhibited Neurog3 expression within progenitors and in EC-like cells (Fig. 5m). Taken together, our studies in murine and hPSC-derived pancreatic progenitors suggest that EC-like cells that emerge within pancreatic tissues share a molecular phenotype characterized by expression of Nkx6-1 and Ins-related transcripts, as well as prolonged Neurog3 kinetics.

Endocrine patterning of hPSC-PPs influences both the frequency and phenotype of derived cell types

Finally, we sought to determine how endocrine patterning of hPSC-PPs affects the phenotype of each endocrine lineage. To test this, we differentiated NKX6-1-enhanced hPSC-PPs towards either the islet or the EC-like lineage. We omitted LDN in the islet protocol to limit the emergence of SLC18A1⁺ cells, thereby maximizing the contrast between the two conditions. Using fixed single-cell profiling (Fig. 6a), we clearly identified the 5 major islet cell types (alpha-like, beta-like, delta-like, epsilon-like and gamma-like) at the expected frequencies (Fig. 6b, c and Supplementary Fig. 7a). In addition, two distinct clusters of EC-like cells were obtained, one of which expressed NEUROG3 (Fig. 6b, c). Several clusters expressed the gastric endocrine transcription factor NKX6-3⁵⁶, including a cluster that co-expressed markers characteristic of gastric endocrine G cells, including GAST, SLC6A4, and ACVR1C⁵⁷, which we tentatively labeled as G-like. The remaining clusters included CXCR4⁺ subsets of alpha- and beta-like cells, a previously reported¹⁰ group of PHOX2A⁺ endocrine cells, and several pancreatic ductal-like clusters variably expressing KRT19, SOX9, and HES1 (Fig. 6b, c).

Fig. 6. Endocrine patterning of hPSC-PPs influences both the frequency and phenotype of derived cell types. Also see Supplementary Fig. 6 and Supplementary Table 1.

a Schematic depicting experimental design for single cell RNA profiling of EC-enriched vs islet-enriched hPSC-islets. Created in BioRender. Misra, P. (2026) https://BioRender.com/aguoykm. b UMAP of integrated samples. c Dot plot highlighting expression patterns of marker genes across each cluster. d UpSet plot depicting the EC marker genes from primary human intestinal EC, mouse Isl1-CKO EC-like, and hPSC-PP derived EC-like cells (hPSC EC-like). Set size refers to the number of genes in each gene set, and intersection size demonstrates the number of genes that overlap only between the gene sets linked in the plot below the bars. e Correlation plot showing the degree of transcriptional correlation between each combination of clusters. f Expression scatter plot highlighting major transcriptional differences in EC identity genes between the more mature EC-like cluster and NEUROG3⁺ EC-like cluster. g–j Volcano plots showing genes that are differentially expressed by the Wilcoxon Rank Sum test between EC-patterning and islet-patterning within the alpha-like (g), beta-like (h), EC-like (i), and NEUROG3⁺ EC-like (j) clusters. Legend: EC, enterochromaffin; ISX9, Isoxazole 9; LDN, LDN193189; PD0325, PD0325901.

We and others^10,36,58 have observed that EC-like cells in hPSC-islets express NKX6-1, a finding that appears to be conserved in the pancreatic EC-like cells of Isl1-CKO mice (Fig. 5m), and that appears to be distinct from intestinal EC cells. To better understand how the pancreatic EC phenotype compares to intestinal EC cells, we compared the transcriptional profiles of hPSC-derived EC-like cells, Isl1-CKO mice, and primary human intestinal ECs. To do this, we first derived a consensus transcriptional signature of intestinal EC cells based on 3 publicly available datasets of single-cell RNA sequencing performed on human intestinal cells, including two that primarily examined fetal intestine^59–61. This included 63 genes identified as EC markers in at least two datasets, with 16 genes common to all 3 datasets (Supplementary Fig. 7b and Supplementary Table 1). This fully conserved set was then compared with markers of murine (Isl1-CKO) and human (hPSC-islet) pancreatic EC-like cells, revealing that 13 out of 16 intestinal EC genes were also markers of hPSC-derived EC-like cells (Fig. 6d), suggesting that EC-like cells in hPSC-islets largely share the conserved identity of intestinal EC-like cells. Notably, NKX6-1 was one of the genes exclusively shared by EC-like cells in Isl1-CKO and hPSC-islets.

We next explored the phenotype of NEUROG3-positive EC-like cells. We observed that the NEUROG3-positive EC-like cells had expression patterns that correlated most closely with EC-like cells, as expected (Fig. 6e), although NEUROG3-positive EC-like cells exhibited higher levels of genes associated with the endocrine progenitor state, including developmental homeoboxes PROX1 and RUNX1T1⁵⁵ as well as the NEUROG3-associated orphan receptor GPR160⁶² (Fig. 6f). Meanwhile, NEUROG3-negative EC-like cells exhibited higher expression levels of genes associated with EC-like maturation, including MME, and ITPR3⁵⁸, as well as two receptors which have recently been shown to be critical for functional integration of EC cells with other enteroendocrine cells: NPY1R¹⁵ and GLP1R⁶³ (Fig. 6f). Moreover, NEUROG3-negative EC-like cells generally expressed higher levels of genes in the conserved EC gene set (Supplementary Fig. 7c).

Finally, we examined the effects of endocrine patterning on gene expression within each cluster (Fig. 6g–j). For both alpha-like and beta-like clusters, we observed a pattern of increased islet identity with islet patterning, with beta-like cells upregulating PAX4 and alpha-like cells increasing expression of ARX-lineage markers, including DPP4, PPY, and PYY. In contrast, EC-like cells exhibited ectopic expression of INS with islet patterning. Meanwhile, EC patterning appeared to induce atypical markers in islet lineages, including GIP in beta-like cells and GAST in alpha-like cells, while increasing SLC18A1 and SLC18A2 in EC-like cells. We observed a similar trend in NEUROG3⁺ EC-like cells as well, with increased SLC18A1 and NKX6-1 with EC patterning (Fig. 6j), while islet patterning induced the ductal/progenitor markers SOX9 and CA2. Taken together, these findings raise the possibility that endocrine patterning influences both the frequency and the phenotype of cell types generated in vitro.

Discussion

The observation that EC-like cells can emerge in hPSC-islets highlights an important gap in our understanding of endocrine differentiation, limiting the optimal use of this technology for research or therapy. In this study, we demonstrate that the EC lineage can be readily obtained from well-committed pancreatic progenitors in vitro and in vivo, and identify specific strategies to enrich or deplete them in hPSC-islets. We demonstrate that soluble factors can guide differentiation of hPSC-PPs towards different endocrine programs, but also that hPSC-PPs display strong inherent biases towards specific endocrine lineages, and these two elements jointly determine the final hPSC-islet composition. This work suggests a logical framework to guide the generation of designer endocrine clusters with enrichment of desired cell types, including the EC lineage.

Through decades of research on pancreatic differentiation in a variety of systems, few reports have elucidated mechanisms of endocrine lineage allocation. An early study controlling the timing of endocrine commitment in the murine fetal pancreas found that the emergence of each islet cell type occurred in temporally defined windows, suggesting that progenitors pass through phases of competence for each endocrine lineage⁶⁴. Supporting the notion of inherent biases, another study observed that uncommitted pancreatic progenitors exhibit epigenetic heterogeneity with regard to transcription factors that control the differentiation of specific endocrine lineages⁶⁵. Although much remains unknown regarding how these biases arise, the transcription factor NKX6-1 has been strongly associated with influencing lineage decisions at multiple stages in the developing pancreatic anlage^66–71. Our study supports these reports by demonstrating that NKX6-1 marks hPSC-PPs with a higher propensity to form pancreatic islet-like cells enriched with beta-like cells, while NKX6-1-deficient progenitors preferentially generate the alpha, delta, and EC lineages.

Adding further complexity to this system, there have been reports that factors present after progenitor specification may also influence endocrine lineage allocation. One study using culture of E12.5 fetal mouse explants, largely composed of multipotent pancreatic progenitors, found that ErbB ligands can modulate the pattern of hormone expression following spontaneous endocrine differentiation ex vivo⁷². More recently, matrix-derived factors⁷³ and FGF ligands⁷⁴ were identified as being capable of modulating hormone expression in differentiating hPSC-PPs. Our study adds to this literature by identifying multiple soluble factors capable of modulating endocrine lineage allocation of hPSC-PPs. Unique to our study, we simultaneously characterized the single-cell expression of lineage-specific transcription factors in addition to hormone peptides, allowing us to track endocrine differentiation with greater precision through a broad range of conditions. This methodology was instrumental in allowing us to identify methods to direct hPSC-PPs selectively towards the alpha (ARX⁺GCG⁺), beta (SLC18A1⁻NKX6-1⁺CPEP⁺), delta (SST⁺HHEX⁺), and EC (SLC18A1⁺) lineages. Importantly, we note that co-expression of NKX6-1 and CPEP may be insufficient to identify beta-like cells, as we observed their expression within SLC18A1⁺ and ISL1⁻ cells.

One surprising finding was the preferential emergence of delta-like (HHEX⁺SST⁺) cells from NKX6-1-suppressed progenitors subjected to beta patterning conditions. We note that, within the ISL1-dependent endocrine lineages, the beta and delta lineages share a requirement for PAX4⁷⁵ while alpha cells require the transcription factor ARX⁴⁴, and that these transcription factors are mutually inhibitory⁴³. We thus speculate that the beta/delta (PAX4⁺) lineages develop under control of a shared program that is at least partially inducible by extracellular signals, in a manner cell-autonomously modulated by NKX6-1 expression at the hPSC-PPs stage. Considering that NKX-6-1-suppressed progenitors had an increased propensity to form EC-like cells in addition to alpha and delta cells, these findings suggest a gradual restriction of plasticity towards the beta lineage as PDX1⁺ progenitors mature into NKX6-1⁺ hPSC-PPs, which is summarized in Supplementary Fig. 7d.

An important focus of this work was the SLC18A1⁺ EC-like lineage. Although some groups consider these cells to be aberrant in hPSC-islets⁷⁶, EC cells play key roles within the gastrointestinal tract in regulating gut motility and interfacing with the nervous system, and they may also mediate responses to the gut microbiome. Using a murine model harboring a conditional knockout of Isl1 following Neurod1 expression³¹, we observed the emergence of the EC lineage within the murine pancreatic anlage. This raises the possibility that fetal pancreatic progenitors may harbor a latent potential for EC differentiation that is normally suppressed. We demonstrate that EC-like cells can efficiently be derived from hPSCs through a pancreatic progenitor intermediate, with the highest yield resulting from PDX1⁺NKX6-1⁻ progenitors. Importantly, although EC-lineage cells obtained through pancreatic progenitors express many genes characteristic of intestinal EC cells, including FEV, SLC18A1, and TPH1, they also express exhibit a unique phenotype, most notably characterized by expression of NKX6-1, which remains of unclear significance.

Our findings also provide insight into the relationship between beta-like and EC-like cells obtained from hPSC-PPs. Although EC cells are believed to share the beta/delta lineage requirement for PAX4⁷⁷, and in our protocols were found to express a variety of beta cell markers, including NKX6-1 and CPEP, we observed EC-like cells and beta-like cells preferentially emerged from different progenitors and using different endocrine patterning steps. Thus, while these findings do not rule out the possibility that the EC-like cells might be capable of further differentiation into beta-like cells as previously proposed³⁶, EC-like and beta-like cells do appear to be developmentally distinct lineages. Moreover, by comparing our EC-enriched hPSC-islets with the islets of Isl1-CKO mice, we identify prolonged Neurogenin 3 expression as a conserved feature of pancreatic EC-like cells in mice and humans. Future studies will be needed to clarify the role of Neurogenin 3 expression patterns on endocrine lineage allocation in the pancreas.

Finally, it is worth noting that our dissection of the molecular processes regulating endocrine lineage diversification in hPSC-PPs required strong early endodermal commitment to prevent confounding contributions from off-target lineages. Previous studies have established that high hPSC seeding density at day 0 of differentiation can reduce the efficiency of DE differentiation^18,78,79. Based on findings from one study that identified a connection between high-density cultures, actin stress-fiber formation, and impaired differentiation into BRA⁺ primitive streak²⁰, we speculated that cytoskeletal disruption might negate some of this variability and provide an effective low-density signaling environment irrespective of actual cell density. Indeed, treating high-density hPSCs with the actin destabilizer LatA restored efficient progression to BRA⁺ primitive streak, thereby improving subsequent directed differentiation into both KDR⁺ PDGFRA⁺ mesoderm and FOXA2⁺SOX17⁺ DE. Notably, the effect of LatA on improved DE differentiation at high starting cell density resulted in improved pancreatic differentiation across multiple cell lines, and also improved differentiation towards NKX2-1⁺ lung progenitors. These findings support the use of LatA treatment to reduce variability in pancreatic differentiation related to initial seeding density.

Several notable factors may limit the generalizability of these findings. First, we tested differentiation conditions predominantly in the context of a planar culture system, while many protocols, including those employed in clinical trials, rely on 3-dimensional suspension culture to generate hPSC-islets^40,80 and their intermediates⁸¹. Given the well-established effects of cytoskeletal signaling⁵, mechanotransduction⁸², and cell culture platforms¹ on endocrine commitment, the use of planar culture systems could represent an important factor that modulates endocrine lineage allocation in response to other extracellular signals. Second, we clearly identify that hPSC-PPs generated through different methods may respond differently when exposed to factors in later endocrine induction media, suggesting that our framework may yield different results when applied to protocols that differ from ours. Finally, our study focused primarily on how pancreatic progenitors commit to various pancreatic endocrine lineages and did not evaluate the functional impacts of all the conditions tested. Given the therapeutic importance of this technology, studying the functional impacts of protocol modifications in early differentiation stages will be an important future direction of this work.

In conclusion, we demonstrate that the EC lineage can arise from pancreatic-committed progenitors in vitro and in vivo, and present strategies to enrich or deplete this cell type within hPSC-islets. This study addresses an important gap in our understanding of how hPSC-PPs commit to specific endocrine lineages and provides a framework for the rational design and optimization of hPSC-islet differentiation protocols for research or therapeutic purposes.

Methods

Human pluripotent stem cell maintenance

The human embryonic stem cell (hESC) line H1 (male) was obtained from WiCell Research Institute (Madison, WI) and the HES3-INS^GFP/+GCG^mCherry/+ hESC line (female) was obtained as a kind gift from Dr. Edward Stanley (Murdoch Children’s Research Institute). The human induced pluripotent stem cell line LiPSC-GR1.1⁸³ (male) was obtained from the National Institute of Neurological Disorders and Stroke Human Cell and Data Repository (NHCDR). The functionally wild-type (RFXANK^+/−) Elf1 hPSC line (female) was obtained as a gift from Dr. David Russell (University of Washington, Seattle). All experiments using hPSCs were approved by the Stem Cell Oversight Committee and the University Health Network Research Ethics Board.

Cells were cryopreserved in a media composed of 50% IMDM (Life Technologies #12440) or StemMACS iPS-brew XF, 40% Fetal Bovine Serum, and 10% DMSO. Cryopreserved hPSCs were thawed in DMEM:F12 at a ratio of 1 mL cells: 9 mL media, centrifuged at 233 G for 5 minutes, and resuspended in StemMACS iPS-brew XF with 10 µM ROCK1/2 inhibitor (Y-27632). Cells were maintained in monolayer culture in StemMACS iPS-brew XF media with daily media changes. Upon reaching full confluency, cells were passaged by single cell dissociation using TrypLE (Life Technologies #12605) with 1% Bovine Pancreas DNase I (stock concentration 1 mg/mL, Millipore Sigma #260913), centrifuged for 5 min at 233 G, and resuspended in StemMACS iPS-brew XF with 10 µM ROCK1/2 inhibitor. Cells were split at a cell line-specific ratio to ensure confluency within 2-3 days, upon which differentiation was initiated.

Human pluripotent stem cell differentiation

On day 0, confluent cultures of hPSCs in monolayer format were washed with Calcium- and Magnesium-free dPBS (Corning #20-031-CV) and treated with components of stage 1 media. The media used for each stage are specified in Supplementary Table 2, with alterations specified in the text where relevant. For reference, the media used for the differentiation according to Balboa et al.⁵¹ is also provided in Supplementary Table 2. All media were supplemented with 1% vol/vol Penicillin-Streptomycin (ThermoFisher Scientific #15070063).

On day 13, some experiments were aggregated by single cell dissociation using TrypLE with 1% DNase I, centrifuged for 5 minutes at 233 G, and resuspended in base medium at 2 million cells/mL supplemented with 1% DNase I.

For lung differentiation, cells were differentiated as previously published²¹. For mesodermal differentiation, hPSCs were differentiated using a protocol modified from Shi et al.²². Briefly, hPSCs were passaged at 1-1.2 × 10⁶ cells per 12-well plate. Differentiations were started 24 hours later in a base media containing MCDB131, 0.25% bovine serum albumin, 1% L-glutamine, and 0.5% insulin-transferrin-selenium (ITS) supplement. On day 1, base media was supplemented with Activin A (50 ng/mL), BMP4 (25 ng/mL), FGF2 (25 ng/mL), CHIR99021 (5 μM), with or without Latrunculin A and/or Y-27632. On days 1-2, all-trans retinoic acid (100 nM), SB431542 (10 μM), LDN193189 (100 nM) and FGF2 (25 ng/mL) were added to base media. Cells were assessed for PDGFRA and KDR by flow cytometry on day 3 of differentiation.

In vitro glucose-stimulated insulin secretion assay

Under direct visualization using a low-power stereomicroscope, 20–30 hPSC-islets were individually selected and transferred to one well of a V bottom 96-well plate. 3 sets of hPSC-islets were analyzed per condition. All subsequent steps were performed using a base KRB media containing 0.1% fatty acid-free bovine serum albumin (BSA) buffered to a pH of 7.4. Cells were first rested in a low glucose media (2.8 mM) for 30 minutes to 1 hour. Cells were then spun down at 233 g for 1 minute, the supernatant was carefully removed, and fresh low glucose (2.8 mM) media was added to each well. After 1 hour, cells were again spun down at 233 g for 1 minute, supernatant was carefully collected, and fresh high-glucose (16.7 mM) media was added to each well. After 1 hour, cells were spun down at 233 g for 1 minute, the supernatant was carefully collected, and cells were rested in fresh low-glucose (2.8 mM) media for 30 minutes. Cells were then exposed to KCl media containing potassium chloride (30 mM) and high glucose (16.7 mM) for 1 hour, after which the supernatant was collected. For the low glucose, high glucose, and KCl steps, human insulin content was quantified using the Ultra-Sensitive low-range HTRF kit (Cisbio #62IN2PEG). 3 technical replicates were taken for each of 3 biological replicates, and experiments were analyzed by ratio-paired one-tailed t test.

Mouse model

Animal experiments were conducted according to protocols approved by the Animal Care and Use Committee of the Institute of Molecular Genetics, Czech Academy of Sciences (protocol # 878/2022). All experiments were performed with littermates (males and females). Sex was not determined for P9 mice used in this study, as tissues were collected prior to sex identification and samples were pooled for downstream analyses. Consequently, data could not be reported disaggregated by sex. We used the mouse model with a conditional deletion of Isl1, Isl1-CKO, with a Neurod1^Cre/+;Isl1^loxP/loxP genotype, described previously³¹. Using the tdTomato reporter line (Ai14, B6.Cg-Gt(ROSA)26Sor^{tm14(CAG−tdTomato)Hze}, Stock No: 7914 Jackson Laboratory) for our analyses, we compared Control and Isl1-CKO mice³¹. Lines were maintained on a C57BL/6 background. Genotyping was performed by PCR on tail DNA (primers are indicated in Supplementary File S4). Mice were kept under standard experimental conditions with a constant temperature (23–24 °C) and fed on soy-free feed (LASvendi, Germany).

Flow cytometry

Sample preparation

hPSC-derived cultures were incubated with TrypLE Express with 1% DNase I at 1 mg/mL for 3–10 minutes at 37 °C (7–10 minutes for hPSC-islets), then gently dissociated with a P1000 pipette. An equal volume of staining buffer containing 10% heat-inactivated fetal bovine serum (FBS) and 90% PBS without calcium and magnesium, supplemented with DNase (100 µL/mL), was added to stop the dissociation reaction. Antibody staining was performed in U-bottom 96-well plates.

Intracellular staining

After sample preparation, cells were incubated with Zombie Violet Fixable Viability dye (diluted at 1 µL/mL in PBS without Calcium and Magnesium) for 20 minutes at room temperature. Cells were centrifuged and then resuspended with Cytofix/Cytoperm (BD #555722) solution for 20 minutes at room temperature. Following removal of Cytofix/Cytoperm, cells were resuspended in primary antibody solution (Supplementary File S3) for at least 1 hour at room temperature (most antibodies) or overnight at 4 °C (any solution containing anti-human NKX6-1 primary antibody). Cells were washed once, resuspended in secondary antibody solution (Supplementary File S3), and left to incubate for 20 minutes at room temperature. Cells were then washed once and resuspended in staining buffer for analysis. All washes and antibody dilutions were performed with Perm/Wash solution (10% BD Perm/Wash 10X stock solution, 90% PBS without Calcium and Magnesium). Due to cross-reactivity of donkey anti-mouse secondary antibodies against rat primary antibodies, solutions containing rat primary antibodies were always added after incubations with mouse primary antibodies and anti-mouse secondary antibodies were completed.

Analysis

Cells were analyzed on a BD LSRFortessa flow cytometer. Experimental data were analyzed using FlowJo Software (BD). For all experiments, gating was first performed as follows, exemplified in Supplementary Fig. 8: First, live cells were selected (Zombie Violet-negative), then doublets were excluded, and finally gating was performed on the fluorescent channels of interest. Fluorescence minus one gates were used to help identify positive populations. Fluorophores detected are displayed in the axes for each plot shown in the figures, with the following abbreviations: A647, Alexa 647; A488, Alexa 488; A594, Alexa 594.

Immunofluorescence

Sample preparation

Monolayer cells undergoing lung differentiation were washed with dPBS without calcium and magnesium, and fixed with paraformaldehyde (PFA, Electron Microscopy Sciences #15710-S, diluted down to 4% using dPBS without calcium and magnesium) for 20 min at room temperature. Cells were again washed with dPBS twice, and subsequently permeabilized using 0.5% Triton-X (Millipore Sigma #T9284) diluted in dPBS without magnesium and calcium.

Clusters generated using the H1 cell line were harvested on day 26 of differentiation and fixed in 4% PFA for 20 min at room temperature. They were then embedded in agarose and paraffin, and 4 μm sections were cut by the Toronto General Hospital Pathology Research Program Laboratory. Sections were de-paraffinized using xylene and rehydrated in serial dilutions of absolute alcohol. Antigen retrieval was performed with 0.01 M citrate buffer, pH 6.0.

Antibody labeling

Samples were first incubated with blocking solution containing 10% donkey serum and 2% BSA at room temperature for 30–45 minutes (lung monolayer) or permeabilized with 0.5% TritonX-100 before incubating with blocking solution containing 3% donkey serum and 0.1% BSA at room temperature for 1 hour (hPSC-islet clusters). Cells were washed with dPBS without calcium and magnesium once, and then incubated with primary antibodies in a buffer containing either 0.05% Triton-X 100 and 2% BSA (lung monolayer) or 0.1% TWEEN-20 and 0.1% BSA (hPSC-islets), overnight at 4 °C. Samples were then washed with dPBS without calcium and magnesium twice and incubated with secondary antibodies in a buffer containing either 0.05% Triton-X 100 and 2% BSA (lung monolayer) or 0.1% TWEEN-20 and 0.1% BSA (hPSC-islets) at room temperature for 45–60 minutes. Cells were washed with dPBS without calcium or magnesium once, and then incubated with DAPI (1:100,00 in dPBS without calcium and magnesium) for 15–20 minutes. Cells were imaged on an EVOS microscope (ThermoFisher Scientific).

Cell clusters generated from the HES3-INS^GFP/+GCG^mCherry/+ line were mounted on an 8-well chamber slide (ibidi) and imaged with the Leica TCS SP8 confocal microscope at the Advanced Optical Microscopy Facility (AOMF), University Health Network, Toronto. Brightness, contrast, and color adjustments were made using Leica Application Suite X (LAS X). The Leica Application Suite X (LAS X) Office software and FIJI (a distribution of ImageJ) were used to adjust the brightness, contrast, and color of images.

Mouse islet antibody labeling

For vibratome sections, dissected tissues were fixed in 4% PFA, embedded in 4% agarose gel, and sectioned at 80 µm on a Leica VT1000S vibratome. The nuclei were, in specific cases, counterstained with Hoechst 33342. Image acquisition was completed using the Zeiss LSM 880 NLO scanning confocal microscope with the ZEN lite program. The expression of serotonin and SLC18A1 was quantified in all viewing areas of a vibratome section from the central part of the pancreas with the largest pancreatic footprint in Isl1-CKO and Control mice (n = 5 pancreases per genotype) using the thresholding tool in ImageJ. The results were expressed as the percentage of the serotonin⁺ or SLC18A1⁺ area relative to the tdTomato⁺ area.

Single-cell RNA profiling of hPSC-islets

Sample preparation

hPSC-islets were dissociated using TrypLE (Life Technologies #12605) with 1% DNAse I for 10 minutes. Cells then underwent viability enrichment through magnetic sorting using the Miltenyi Biotec Dead Cell Removal Kit (#130-090-101). Cells were fixed for 24 hours according to the 10X Genomics protocol for cell fixation (protocol #CG000478). Cryopreserved, fixed cells were thawed and underwent fluorescence-activated cell sorting (FACS) to enrich for singlets. At least 75,000 cells per sample were submitted for the probe-based Chromium Single-Cell Gene Expression Flex RNA profiling assay.

Analysis

All analyses were performed using the R programming language (v4.3.0). First, ambient RNA was excluded using the SoupX algorithm⁸⁴ using a fixed threshold of 15%. Next, integer count matrices pre-filtered to remove empty and low-quality barcodes were furthered pre-processed and integrated using the standard Seurat (v4.3.0) workflow⁸⁵. Barcodes expressing fewer than 5000 different transcripts were excluded. Because mitochondrial reads are underrepresented in the Chromium Single Cell Gene Expression Flex RNA profiling assay, stressed and dying cells were identified and excluded if heat shock protein-related transcripts exceeded 0.5% of any individual barcode’s library. Seurat’s SCTransform() algorithm was used for normalization. Prior to downstream analysis, doublets were inferred and excluded using a combination of DoubletFinder⁸⁶ and SCDS⁸⁷. The grouping of cells into meaningful clusters was done through integration of datasets from each condition, followed by iterative clustering, sub-clustering, and cluster merging using native Seurat functions and marker gene identification. Additional data processing was performed using the dplyr (v1.1.2)⁸⁸, purrr (v1.0.1)⁸⁹ and stringr (v1.5.0)⁹⁰ packages. Data visualization was performed using the ggplot2 (v3.4.2)⁹¹, scCustomize (v1.1.1)⁹² and SCpubr (v1.1.2)⁹³ packages. VennDiagram (v1.7.3)⁹⁴ was used to calculate overlaps in gene sets, while ggvenn (v0.1.10)⁹⁵ was used to generate Venn Diagrams.

Single-cell RNA sequencing of Isl1-CKO islet cells

Sample preparation

The Isl1-CKO and Control pancreases from 9-day-old (P9) pups (n = 6 mice per genotype) were used for the preparation of endocrine cells, as published elsewhere⁵⁴. Pancreases were perfused by Collagenase I (1 mg/ml; Sigma-Aldrich) in HBS (Merck) supplemented with Actinomycin D (25 µg/ml; Merck) to stop transcription. The expanded pancreases were digested with 0.8 mM Collagenase I at 37 °C for ~ 12 minutes. Collagenase-released endocrine cells were centrifuged (140 rcf, 4 °C, 4 minutes), washed with 2 ml of HBS with Actinomycin D (2.5 µg/ml; Merck), and then with Dulbecco’s PBS. Pancreatic endocrine cells were then treated by 0.05% trypsin/0.53 mM EDTA solution⁹⁶ supplemented with Actinomycin D (25 µg/ml) at 37 °C for 5 min. Trypsinization was stopped by adding FACS buffer [PBS, 10 mM EGTA supplemented with 10% FBS(Merck)]. Cells were gently dispersed mechanically using a P1000 pipette, spun down (800 rcf, 4 °C, 5 minutes), and resuspended in FACS buffer (PBS, 10 mM EGTA, supplemented with 2% FBS)⁹⁶. Cell suspensions were filtered through 40 µm nylon mesh, and immediately tdTomato⁺ cells were sorted using a flow cytometer (BD FACSAria™ Fusion), through a 100 µm nozzle in 20 psi, operated with BD FACSDiva™ Software.

Single-cell suspension (1000 cells/µl) was used for library preparation: Chromium Next GEM Single-Cell 3’ Reagent Kits v3.1 (10 × Genomics, Pleasanton, CA) was used to prepare the sequencing libraries, and the protocol was performed according to the manufacturer’s instructions. The concentration and quality of the libraries were measured using Qubit dsDNA HS Assay Kit (Invitrogen) and Fragment Analyzer HS NGS Fragment Kit (#DNF-474, Agilent). The libraries were pooled and sequenced in paired-end mode using Illumina NovaSeq 6000 SP Reagent Kit, Read 1 containing a barcode and a UMI, and Read 2 covering the sequence of interest. The reads were processed as described previously⁹⁷. Additional details on bioinformatics analyses can be found in Supplementary Methods.

Analysis

The single-cell sequencing bioinformatics pipeline was lacking any of the expected sequences in read 1 (known cell barcode, adapter sequence (W1), or beginning of the poly-T tail). The sequencing data were aligned to the reference mouse genome GRCm38 and annotated (GENCODE version M8 annotation) by STARsolo (STAR version 2.7.3a)⁹⁸. EmptyDrops function (DropletUtils R package (version 4.1.1))⁹⁹ with a threshold of 100 UMIs and FDR ≤ 0.001 was applied to preserve only cell-containing droplets. The final number of sequenced cells was 1025 cells for Isl1-CKO and 1316 cells for the Control. The data were further processed using the Seurat R package. Data were SCTransformed and integrated (excluding mitochondrial and ribosomal genes, prefixed by mt- or Rps/Rpl, respectively). Uniform Manifold Approximation and Projection (UMAP) was used to visualize 10 principal components (PC), which were subsequently clustered (FindNeighbors and FindClusters functions, UMAP resolution 0.2 and 0.3 for Control and Isl1-CKO, respectively). Clusters were annotated based on the expression of known marker genes of the expected cell populations, and their correspondence to the markers found by the FindAllMarkers function (at least 80% cells in the cluster expressing the markers). DoubletFinder R package (v2.0.3) was used for the identification of droplets potentially containing more than one cell. Clusters expressing ambiguous markers and containing a higher number of doublets were filtered out of the data set.

Reanalysis of previously published data

Bulk RNA sequencing of Control and tdTomato⁺ cells from mice was obtained as previously described³¹. Raw counts were processed using the DESeq2 pipeline (v1.34.0) to obtain normalized counts¹⁰⁰ and visualized using ggplot2⁹¹.

Single cell RNA sequencing data of primary human enteroendocrine cells (EECs) were obtained from https://www.gutcellatlas.org⁵⁹ (subsetting the EECs from the raw count matrix using provided annotations), from https://zenodo.org/records/5457926⁶⁰ (using the provided EEC subset of the background-corrected count matrix) and from GSE156796⁶¹ (using the provided annotation to extract the raw count matrix for intestinal cells, followed by filtering for minimum of 200 expressed features, normalization using SCTransform(), clustering, and subsetting of the single cluster expressing endocrine markers such as CHGA, CHGB, NKX2-2, NEUROD1). Each EEC dataset was subsequently processed using the same pipeline: normalization using SCTransform() from the Seurat package (v4.3.0), clustering over the first 30 components of principal component analysis, and identification of the EC cluster using the canonical EC markers TAC1, TPH1, SLC18A1, and LMX1A. Marker genes of EC cells (compared only to other EECs) were identified using the FindMarkers() function from Seurat to identify genes upregulated (avg_log2FC > 0) with an adjusted p-value of less than 0.05.

Statistical analysis and illustrations

GraphPad Prism (v 9.1.4) was used for statistical analysis and data visualization. Statistical tests employed along with sample sizes are indicated in the figure captions. For all statistical tests, significance is labeled as follows: ns, non-significant; *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001. Unless indicated otherwise, all error bars represent the standard error of the mean. All data points represent independent biological replicates. Experiments were repeated at least three times to obtain biological replicates, taken from different samples. Experimental schematics were designed using BioRender (www.biorender.com).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization, P.S.M., E.C.M., G.P., and M.C.N; Methodology: P.S.M., E.C.M., R. S., V.F., R.B., G.P., and M.C.N.; Formal analysis, P.S.M. and V.F.; Investigation, P.S.M., E.C.M., H.H., A.C., J.L., F.S., A.O., R.S., Y.S., V.F., and R.B.; Data Curation – P.S.M. and V.F.; Writing – Original Draft, P.S.M. and M.C.N.; Writing - Review & Editing, all authors; Visualization, P.S.M. and V.F.; Funding acquisition, G.P. and M.C.N.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The single cell transcriptomics data generated in this study have been deposited in the NIH Gene Expression Omnibus (GEO) repository under accession codes GSE269266 for single cell RNA profiling of hPSC-derived endocrine cells and GSE268657 for sorted endocrine cells from Control and Isl1-CKO mice. Source data are provided in this paper.

Competing interests

P.S.M., E.C.M., and M.C.N. are co-inventors on 2 patent applications related to this work. All other authors have no competing interests to declare.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70666-y.