Modeling in vivo induction of gastric insulin-secreting cells using transplanted human stomach organoids

Summary

Insulin-dependent diabetes could be treated by supplying patients with primary pancreatic islets or other types of insulin-secreting cells. Functional insulin-secreting cells can be induced in situ from the murine stomach using defined genetic factors, offering a promising method to directly produce autologous insulin-secreting cells. Here, we modeled whether such gastric insulin-secreting (GINS) cells could be generated in vivo from human stomach tissues. We produced human gastric organoids (hGOs) from human embryonic stem cells engineered with inducible expression of reprogramming factors. The hGOs were stably transplanted for 6 months and showed robust cytodifferentiation resembling the human stomach in structure and cellular composition. Upon hGO maturation in vivo, we activated the reprogramming factors and observed the formation of insulin⁺ cells, which secreted insulin into the circulation and ameliorated experimental diabetes. Our modeling indicates that GINS cells can be induced from human stomach tissues in vivo, warranting further therapeutic development for this technology.

Highlights

• •Engrafted human gastric organoids (hGOs) model human stomach in vivo
• •NPM factor induction converts hGOs into insulin⁺ β-like cells in vitro and in vivo
• •In situ gastric reprogramming offers potential for autologous diabetes cell therapy

In this article, Lu and colleagues demonstrate that human gastric organoids grafted in mice can be induced to generate insulin-secreting cells by NPM factors and ameliorate experimental diabetes, providing proof of principle for in situ human gastric reprogramming as a potential autologous therapy for diabetes.

Introduction

Generating functional insulin-secreting cells is a major goal in developing cell therapies to treat diabetes. Human embryonic stem cells (hESCs) and induced pluripotent stem cells (hiPSCs) can be differentiated in vitro into organoids resembling human pancreatic islets (sc-islets) (Balboa et al., 2022; Hogrebe et al., 2021; Pagliuca et al., 2014; Rezania et al., 2014). Transplantation of the sc-islets into animal models led to functional maturation and regulated insulin secretion (Augsornworawat et al., 2020; Maxwell et al., 2022). Sc-islets have now entered clinical trials to treat type 1 diabetes (T1D), a disease in which β cells are selectively destroyed by autoimmunity (Hogrebe et al., 2023; Ramzy et al., 2021; Reichman et al., 2025; Shapiro et al., 2021; Wang et al., 2024). However, the strong systemic immunosuppression necessary to control both allo- and auto-immunity poses severe health challenges such as increased infection and cancer risks. Thus, the vast majority of patients with T1D are excluded from allogenic sc-islet therapy (Blau et al., 2015; Ryan et al., 2004). Emerging technologies aimed at delivering local immune protection, such as encapsulation and immune-evasive sc-islet generation, are being developed, but their efficacy and safety have yet to be fully evaluated (Desai and Shea, 2017).

Advancements in the reliable production of autologous insulin-secreting cells have the potential to revolutionize cell therapies for diabetes by removing the necessity for allogenic immunosuppression, although autoimmune protection may still be required. One promising approach to produce autologous insulin-secreting cells is by in vivo induction from the gut mucosa (Ariyachet et al., 2016; Chen et al., 2014). Deletion of Foxo1 from murine and hESC-derived intestinal endocrine progenitors induced insulin⁺ cells (Bouchi et al., 2014; Du et al., 2022; Talchai et al., 2012). Expression of NPM (NEUROG3, PDX1, and MAFA) reprogramming factors can also induce functional insulin-secreting cells from the murine gut, with the stomach yielding the highest density of insulin⁺ cells (Ariyachet et al., 2016; Chen et al., 2014). The induced murine gastric insulin-secreting (GINS) cells resemble pancreatic β cells in both molecular properties and functionality by ameliorating experimental diabetes. In a more recent study, we showed that human gastric progenitor cells cultured from biopsy samples can be expanded and differentiated into functional and transplantable GINS organoids that effectively improve diabetes outcomes (Huang and Zhou, 2023; Huang et al., 2023). Despite these advances, it remains unclear whether GINS cells can be induced from human gastric mucosa residing in a complex in vivo milieu. For instance, gastric epithelial cells are heterogeneous and surrounded by mesenchymal cells that secrete many signaling factors. To evaluate in vivo GINS cell induction, we turned to human gastric organoids (hGOs) derived from hESCs. The hGOs are composed of both epithelial and mesenchymal components and resemble “mini-stomachs” in structure and cellular composition. In this study, we generated engineered hGOs and transplanted them into mouse hosts for in vivo maturation, upon which NPM factors were activated. We observed the formation of abundant GINS cells in the transplanted hGOs, which secreted human insulin into the murine hosts and ameliorated experimental diabetes. Our data demonstrate that it is possible for human stomach mucosa to produce functional GINS cells in vivo upon genetic factor induction. Continued development of this method may yield a useful technology to generate autologous insulin-secreting cells for replacement therapies in diabetes.

Results

NPM-hGO generation and validation

Although gene expression from the AAVS1 locus in differentiated cells is often heterogeneous, this site poses minimal risk of disrupting essential genes and has been widely used in gene-editing studies (Bhagwan et al., 2020; Hamilton et al., 2004; Klatt et al., 2020; Yada et al., 2017; Zhu et al., 2014a, 2014b). To produce hGOs with inducible expression of NPM factors (thereafter referred to as NPM-hGOs), we first engineered hESC lines using a published protocol to target two constructs into the human AAVS1 safe-harbor locus (also known as PPP1R12C) (Figure 1A) (Zhu et al., 2014a, 2014b). One construct carries the NPM factors and a Cherry fluorescent protein in a polycistronic construct behind a TetO-inducible promoter. The second construct carries the CAG promoter driving rtTA expression (Figure 1A). We electroporated the two donor plasmids into the HUES8 hESC line. Double selection with puromycin and neomycin yielded 4–8 clones per electroporation. Upon doxycycline (Dox) treatment, immunohistochemistry showed co-expression of NPM proteins and Cherry in all clones analyzed (Figures 1B and 1C).

Figure 1.

Generation of hGOs from hESCs with inducible expression of NPM factors

(A) Schematic diagram of experimental strategy to target the human AAVS1 locus in hESCs with two gene-trap donor plasmids to enable inducible expression of NPM factors. SA, splice acceptor; 2A, viral peptide that mediates polycistronic gene expression; Puro, Neo, puromycin and neomycin-resistant genes; NPMCherry, polycistronic expression cassette that carries NPM factors and cherry; CAG, constitutive promoter. Black boxes: exons of the AAVS1 (PPP1R12C) locus.

(B) Diagram of the inducible expression system showing activation of NPM and Cherry expression in the presence of doxycycline (Dox).

(C) Immunohistochemistry of one NPM-hES clones showed co-expression of NPM factors with Cherry after Dox induction whereas no NPM or Cherry was observed without Dox.

(D) Schematic diagram of the key steps and factors used to differentiate hESCs into hGOs.

(E) Examples of hGOs grown in Matrigel at an early and a late differentiation step. Scale bars: 100 μm.

(F) Immunohistochemistry of hGOs grown in vitro showed expression of the antral gastric marker PDX1 and the progenitor marker SOX9 in E-cadherin⁺ (ECAD) epithelial cells. Scale bars: 100 μm.

(G and H) Many MUC5AC⁺ mucus cells and a small number of hormone-positive enteroendocrine cells were present in the cultured hGOs. The endocrine cells were detected by co-staining with a mixture of antral gastric hormone antibodies (GAST, SST, and GHRL). Scale bars: 100 μm.

(I) qPCR analysis showed significant activation of antral gastric markers TFF2, MUC5AC, and GAST. One asterisk: p < 0.05. n.s: not significant. n = 4 independent samples. Data are mean ± SD; two-tailed unpaired t test.

See also Figure S1.

We selected two NPM-hESC clones and differentiated them into human antral hGOs using an established protocol (Figures 1D and S1A). The differentiation procedure consists of four stages lasting 30 days in total. In the first stage, treatment of hESCs in a monolayer with Activin A led to the formation of definitive endoderm (DE) (88.17% ± 6.5% SOX17⁺FOXA2⁺, Figure S1B). Subsequently, a cocktail of signaling factors including the Wnt pathway activator CHIR99021, the bone morphogenetic protein (BMP) inhibitor Noggin, FGF4, and retinoic acid (RA) patterned the DE into posterior foregut (stage 2, 99.3%3 ± 0.82% SOX2⁺, Figure S1C). At the foregut stage, small epithelial spheres emerged from the monolayer cultures (Figures 1E and S1A). After embedding the foregut spheres into 3D Matrigel domes, treatment with RA, Noggin, and epidermal growth factor (EGF) induced antral stomach fate (stage 3, 84.5% ± 3.87% GATA4⁺, 80.25% ± 5.9% SOX2⁺, and 97.75% ± 2.22% PDX1⁺, Figure S1D). The stomach spheres continued to grow in Matrigel so that by day 30 of differentiation, hGOs of 2–5 mm in diameter formed (Figures 1E and S1A).

The antral stomach is mainly populated by mucus-secreting cells and hormone-secreting enteroendocrine cells, whereas most acid-secreting parietal cells are present in the corpus stomach. PDX1, a protein with high levels of expression in pancreatic β cells, is also expressed at low levels in the antrum and was detectable in the CDH1⁺ (also known as E-cadherin or ECAD) epithelial layer of the hGO spheres (Figure 1F). Many epithelial cells also expressed SOX9, a stomach stem/progenitor cell marker (Figure 1F). Immunohistochemistry revealed an abundance of MUC5AC⁺ mucus-secreting cells and a small number of hormone-positive enteroendocrine cells in the cultured hGOs (Figures 1G and 1H). qPCR confirmed upregulation of the stomach-specific genes TFF2, MUC5AC, and gastrin (GAST) (fold increase over control: [1.67 ± 0.14] × 10⁶, [2.13 ± 0.54] × 10², 15.6 ± 1.9, respectively) (Figure 1I). These data indicate the successful generation of hGOs that resemble the human antral stomach. We did not observe significant differences in hGO formation from the two NPM-hESC lines except for GAST mRNA expression (Figure S1F).

Activation of NPM factors in hGOs in vitro led to formation of insulin⁺ cells

We induced the expression of NPM factors in hGOs for 7 days (day 27 to day 34 of hGO culture) with Dox and evaluated the induction of insulin and other pancreatic β cell markers. Dox treatment resulted in Cherry expression in the hGOs and scattered c-peptide⁺ (cPEP⁺) cells within the ECAD⁺ epithelial layer (Figure 2A). All cPEP⁺ cells co-expressed Cherry (Figure 2B), whereas no insulin (INS) was detectable in hGOs without Dox induction (Figure 2A). Among Cherry⁺ cells, 36% were cPEP⁺, and all of these were epithelial (EPCAM⁺; Figure 2C). RNA sequencing (RNA-seq) analysis confirmed strong induction of INS gene and significant upregulation of key human β cell markers including NEUROD1, RFX6, ABCC8, PCSK1 (prohormone convertase), GCK (glucokinase), MAFB, PAX6, NKX2-2, MNX1, and other genes involved in β cell development and insulin secretion after Dox induction (Figures 2D and 2E).

Figure 2.

Induction of insulin⁺ cells from NPM-hGOs in culture

(A) Scattered c-peptide⁺ (cPEP⁺) cells appeared in ECAD⁺ epithelia of NPM-hGOs after Dox treatment.

(B) cPEP⁺ cells expressed Cherry, indicating their origin from NPM-expressing cells. Scale bars: 100 μm.

(C) Flow cytometry shows that 36% of the Cherry⁺ ECAD⁺ cells in the NPM-hGOs treated with Dox express cPEP.

(D) RNA-seq was performed on FACS-purified mCherry⁺ cells from NPM-hGOs treated with Dox, compared to untreated (−Dox) controls. The analysis revealed significant upregulation of insulin and a panel of key β cell genes following Dox induction. One asterisk: adjusted p < 0.05; two asterisks: adjusted p < 0.01; three asterisks: adjusted p < 0.001. n = 4 independent samples. Adjusted p value by Wald test using DEseq2. Data are mean ± SD.

(E) Heatmap of representative β cell marker expression in NPM-hGO treated with Dox.

(F) t-SNE visualization of hGOs treated with Dox for 10 days. GP, gastric progenitors.

(G) Relative expression of cell-type-specific markers. The shading displays scaled average gene expression, and diameter denotes fractional expression.

(H) Heatmap showing gene expression clusters along the pseudotime trajectory from GP to β-like cells. Select significant GO terms enriched in each gene cluster are shown. p value calculated by hypergeometric distribution followed by Benjamini-Hochberg adjustment. Key transcription factors are highlighted.

(I) Static glucose-stimulated insulin secretion assay showed enhanced human insulin secretion at high glucose (20 mM) vs. low glucose (2 mM) from NPM-hGOs derived from two independent lines of hGOs. Data shown as human insulin (ng) per sample per hour. 10 hGOs or 10 medium-sized human islets per sample. Student’s t test.

To gain insights into the cell identities of hGOs overexpressing NPM factors, we employed single-cell RNA sequencing (scRNA-seq) to analyze the transcriptomes of 8,363 mCherry⁺ EPCAM⁺ epithelial cells from hGOs treated with Dox for 10 days. The primary populations identified included β-like cells (51.0%) and gastric progenitors (39.8%) (Figure 2F). Additionally, a minor population of δ-like cells (5.0%) and ε-like cells (4.2%) was observed, while only a few glucagon⁺ (GCG⁺) cells were induced (Figure 2F). Both β- and δ-like endocrine cells expressed key markers of their islet counterparts, including INS, PDX1, SCG2, and GCK in β cells and SST and HHEX in δ cells (Figure 2G). To assess the conversion of gastric cells to β-like cells, we constructed a developmental trajectory utilizing pseudo-time ordering to identify genes that exhibit differential expression as cells progress along this trajectory (Figure 2H). Gene Ontology (GO) term analysis revealed a rapid downregulation of gastric progenitor pathways, including those related to digestion and WNT signaling, while pathways associated with histone modification and stem cell differentiation were activated when cells exited the gastric progenitor stage. Functional pathways of β cells, such as protein secretion and oxidative phosphorylation, emerged later in the developmental process (Figure 2H; Table S2).

Next, we evaluated glucose responsiveness of the induced insulin⁺ cells differentiated from two separate NPM-hGO clones. hGOs treated with Dox for 10 days in vitro were partially dissociated to expose the inner epithelial layer for the glucose-stimulated insulin secretion (GSIS) assay, which detected robust human insulin secretion of hGOs to high-glucose challenges (Figure 2I). Thus, engineered NPM-hGOs can be induced to produce functional GINS cells in vitro.

NPM-hGOs can be maintained in vivo for months and show gastric cytodifferentiation

To evaluate their stability and cytodifferentiation, we transplanted hGOs into highly immunocompromised NSG mice inside the epididymal fat pad. The transplanted hGOs from both sites can be recovered for up to 6 months, the longest duration tested (Figure 3A). The hGOs did not show substantial growth in either transplantation site and remained relatively small in size (2–5 mm in diameter), consistent with other studies (Eicher et al., 2022; McCracken et al., 2014). All transplanted hGOs had a thick mesenchyme layer surrounding the epithelium (Figure 3A). The epithelia were often simple single-layered structures, although gland-like epithelial folds were present in many samples (Figure S2). We observed extensive cytodifferentiation in the transplanted hGOs, with a preponderance of MUC5AC⁺ cells (Figure 3B). Many hormone-positive endocrine cells were detected, including the major antral endocrine cell types that express GAST, ghrelin (GHRL), and somatostatin (SST) (Figures 3D–3F). These endocrine cells co-expressed the antral epithelial marker, PDX1 (Figure 3C), and we also detected SOX9⁺ progenitors (Figure 3G).

Figure 3.

hGOs can be transplanted and maintained long-term in vivo

(A) Immunohistochemistry of a representative hGO harvested 6 months after transplantation showed many hormone-positive enteroendocrine cells (detected using a mixture of antibodies against the major antral gastric hormones GAST, GHRL, and SST). The antral marker PDX1 marks epithelial cells. Scale bars: 100 μm.

(B–G) Transplanted hGOs showed large numbers of MUC5AC⁺ mucus cells (B), enteroendocrine cells that express PDX1 (C), SST (D), GAST (E), and GHRL (F), and SOX9⁺ stem/progenitor cells (G), indicating robust cytodifferentiation and persistence of the stem/progenitor compartment.

See also Figure S2.

NPM-hGOs produce functional GINS cells in vivo upon NPM activation

Two weeks after transplanting NPM-hGOs into the fat pad, we administered Dox for 2 weeks in the drinking water to induce GINS cells (Figure 4A). Three weeks after the end of Dox treatment, immunohistochemistry revealed an abundance of GINS cells in the hGO epithelia (Figures 4B and 4C). Most GINS cells expressed key β cell markers such as PCSK1, PAX6, NKX2-2, and MAFB (Figures 4D–4G). The activation of MAFB in the induced GINS cells is notable as it is expressed in native human but not murine β cells. Induction of GINS cells in the hGOs was heterogeneous. Sixty percent of them (16 out of 26) had significant numbers of GINS cells whereas the others had relatively few. In contrast, none of the control hGOs examined had any insulin⁺ cells. Expression of SOX9 and MKI67, a proliferation marker, persisted in the crypt-like compartments of the hGOs (Figures 4H and 4I), indicating continued presence of gastric stem/progenitor cells. Strong induction of MAFB and NKX2-2 (54.5% ± 5.9% and 59.7% ± 1.0% of insulin⁺ cells, respectively) was mostly observed in hGOs with high density of GINS cells, whereas PAX6 and PCSK1 were more uniformly induced (74.5% ± 2.8% and 58.3% ± 7.3% of insulin⁺ cells, respectively) in all hGOs (Figure 4J). Most grafted mCherry⁺ cells expressed insulin (95.3% ± 2.1%) (Figures 4J and 4K), indicating maturation after engraftment. The majority of induced GINS cells were mono-hormonal, although a subset co-expressed antral gastric hormones (Figures 4L–4P).

Figure 4.

NPM-hGOs can generate GINS cells in vivo

(A) Schematic diagram of inducing GINS cells in transplanted hGOs in comparison with untreated control (−Dox).

(B and C) Immunohistochemistry of a representative NPM-hGO showed induction of many insulin⁺ cells 3 weeks after Dox treatment. Scale bars: 100 μm.

(D–G) Many induced insulin⁺ cells in hGOs co-expressed the β cell markers PCSK1, PAX6, NKX2-2, and MAFB. Scale bars: 100 μm.

(H and I) Expression of SOX9 and MKI67 persisted in crypt-like compartments in hGOs after Dox treatment, indicating preservation of stem/progenitor cells.

(J) Quantification of the immunohistochemistry data shown in (D)–(G) and (K). Over 500 insulin⁺ cells counted from at least 10 independent sections. Data are mean ± SD.

(K–O) Immunohistochemistry showed that the vast majority of insulin⁺ cells co-localized with Cherry (K). A small subset of the insulin⁺ cells co-expressed antral hormones GHRL (L), GAST (M), SST (N), and very rarely, GCG (glucagon) (O).

(P) Quantification of the immunohistochemistry data shown in (K)–(O). Over 500 insulin⁺ cells counted from at least 10 independent sections. Data are mean ± SD.

See also Figure S3.

We next evaluated whether induced GINS cells from the hGOs can secrete human insulin into the circulation and ameliorate experimental diabetes in the mouse (Figure 5A). The SCID-Beige mice used for this experiment had a high mortality rate when administered with a single high dose of streptozotocin (STZ, 160–180 mg/kg body weight), so the mice were treated with three low-dose STZ (50 mg/kg body weight) over 5 days. The mice developed hyperglycemia (in the range of 350–450 mg/dL blood glucose), followed by a gradual and partial recovery over 3–4 weeks, before stabilizing at 200–300 mg/kg blood glucose (Figure 5B, black line). Mice transplanted with NPM-hGOs were allowed to recover for 2 weeks after surgery, followed by 3 rounds of low-dose STZ, which resulted in hyperglycemia (Figure 5B). The experimental group was then administered Dox in their drinking water for 5 weeks to induce the formation of GINS cells in the hGOs. Compared with the control group, which received no Dox (Figure 5B, black line), there was a rapid amelioration of blood glucose in the Dox-treated animals, which was maintained until the end of the experiment at approximately 6 weeks (Figure 5B, red line). Due to the spontaneous decline of hyperglycemia in the control group, statistically significant difference of blood glucose levels between the +Dox and the −Dox groups can only be established for the first 2 weeks. Consistent with the blood glucose monitoring data, glucose tolerance test showed significant improvement in the Dox-treated group (Figure 5C). Human insulin was detected in the sera of Dox-treated animals but not in the controls (Figures 5D and S3). Most of the grafted tissue treated with Dox secreted insulin in response to glucose (Figure 5D). Immunohistochemistry showed that induced GINS cells in the hGOs were closely associated with PECAM1⁺ vascular cells (Figure 5E), consistent with a previous report that induced β-like cells can recruit and remodel the vasculature (Huang et al., 2023). These data collectively indicate that GINS cells can be induced in vivo from transplanted hGOs and secrete insulin into the circulation.

Figure 5.

Transplanted NPM-hGOs secrete insulin and ameliorate experimental diabetes

(A) Schematic diagram of the experimental strategy. NPM-hGOs were transplanted into immune-compromised SCID-beige mice. After 2 weeks, the animals were treated with three doses of streptozotocin (STZ, 50 mg/kg body weight) to induce hyperglycemia. One cohort of the animals were administered Dox in drinking water for 2 weeks whereas the control cohort was not treated. The animals were monitored for 5 more weeks before transplant removal and analysis.

(B) Blood glucose monitoring showed that the control animals (without Dox, black line) developed hyperglycemia after STZ, which was followed by partial recovery. Animals transplanted with hGOs and received Dox showed rapid reversal of hyperglycemia (with Dox, red line), which was maintained until the end of the experiment. One asterisk: p < 0.05; two asterisks: p < 0.01; three asterisks: p < 0.001. n = 6 mice per group. Data are mean ± SEM; two-way ANOVA with Dunnett’s test.

(C) Intraperitoneal glucose tolerance test 1 week after Dox treatment showed improved glucose tolerance in animals treated with Dox (+Dox, red line) compared with controls (Dox, blue line). One asterisk: p < 0.05; two-way repeated measures ANOVA with Dunnett’s test. Data are mean ± SEM.

(D) Circulating human insulin was detected in the sera of Dox-treated mice after 6 h fasting but not in control mice. Student’s t test (left). Human insulin from individual mouse was measured after overnight fasting (0 min) and 30 min after glucose challenge. Paired t test (right). n = 6 mice. Data are mean ± SEM.

(E) Immunohistochemistry showed close juxtaposition of insulin⁺ cells and PECAM1⁺ vascular cells in the transplanted hGOs. Scale bars: 100 μm.

Discussion

In this study, we generated human antral stomach organoids from hESCs. Consistent with prior reports, the hGOs can be transplanted, remain stable in vivo for several months, and show extensive cytodifferentiation (Broda et al., 2019a, 2019b; Eicher et al., 2022). Activation of the NPM factors in the transplanted hGOs led to the formation of GINS cells, circulating insulin, and amelioration of hyperglycemia. These data suggest the feasibility of inducing GINS cells in situ in the human stomach. In situ GINS induction may be further developed into a therapeutic approach to generate autologous insulin-secreting cells to treat diabetes. We envision that in future clinical studies, the NPM factors will be delivered into the human stomach mucosa by focal injections of mRNAs or adeno-associated viral vectors. We currently cannot evaluate these delivery methods due to the inability to micro-inject the small hGOs embedded in the fat pad.

While this study is focused on antral hGOs, the ability to induce the human corpus stomach to produce GINS cells in vivo remains unknown. Nonetheless, cultured human corpus gastric progenitor cells can be differentiated into functional GINS organoids (Huang et al., 2023), so in vivo corpus GINS cell induction is a possibility.

A technical limitation of this study is that long-term hyperglycemia was not maintained at a steady level in the transplanted mice, which showed partial recovery. Nevertheless, statistically significant amelioration of hyperglycemia was achieved in the mice transplanted with NPM-hGOs during the time frame of consistent hyperglycemia in the control mice (day 10–24). Together with the presence of circulating human insulin in these mice, our data showed that the induced GINS cells are functional. Another limitation of this study is the use of only a single hESC line (HUES8), which leaves the translational potential for autologous in situ therapy uncertain. To better evaluate the applicability of this approach, future studies using induced pluripotent stem cells derived from multiple donors will be essential.

It is notable that the induced GINS cells were scattered in the epithelium and did not aggregate into islet structures. In our study of reprogramming pancreatic acinar cells into β-like cells in vivo, we observed that the induced β-like cells, although initially diffused among the parenchyma, aggregated into islet-like structures after 2 months (Li et al., 2014a, 2014b). It is possible that the GINS cells in hGOs may also aggregate into islet-like structures if given a longer time. Future studies will be needed to evaluate this possibility.

Although islet transplantation can effectively restore glycemic control, its broader application is limited by donor scarcity and the need for systemic immunosuppression. hESC-derived islets address the issue of limited donor availability (Reichman et al., 2025) but still require immunosuppression, increasing the risk of infection and malignancy. Autologous cell therapies using hiPSC- or human gastric stem cell-derived islets hold great promise; however, they are hindered by lengthy quality control processes and variable differentiation efficiency depending on the cell source. In contrast, in situ reprogramming offers a potentially lower cost alternative to ex vivo approaches and circumvents the need for systemic immunosuppression, although some form of autoimmune protection may still be required. Our study provides proof of principle that human stomach tissue can be reprogrammed in vivo to produce functional insulin-secreting cells, laying the foundation for future development of an autologous, in situ therapeutic strategy for T1D.

Methods

Mouse strains

Immunodeficient non-obese diabetic/SCID/IL2rg^null (NSG) and SCID-Beige mouse strains were obtained from The Jackson Laboratory and Beijing Vital River Laboratory Animal Technology Co., Ltd. All animal experiments are approved by the Harvard Institutional Animal Care and Use Committee and the Institutional Animal Care and Use Committee of Peking University.

Generation of genetically engineered human pluripotent stem cells

hESC lines HUES8 and HUES9 were obtained from Harvard Stem Cell Institute. HUES8 and HUES9 were maintained as colonies and cultured in feeder-free conditions on Geltrex (Life Technologies, A1413202) in mTesR1 media (STEMCELL Technologies) with mTesR1 supplement. Cells were passaged every 4 days with TrypLE Express Enzyme (Gibco). The genetic engineering of hESC lines was carried out as previously described (Zhu et al., 2014a, 2014b). In brief, the AAVS1-TetO-MPNcherry plasmid was generated by gateway cloning LR reaction of the pENTR-NPMcherry plasmid with the AAVS1-tre-tight-DEST vector. AAVS1-TetO-NMcherry and AAVS1-Neo-rtTA plasmids were electroporated into HUES8 and HUES9. The electroporated cells underwent two rounds of antibiotic selection (500 μg/mL geneticin for 4 days and 0.5 μg/mL puromycin for 3 days).

Differentiation of hESCs into hGOs and GINS induction

Generation of hGOs was carried out as previously described (Broda et al., 2019a, 2019b). In brief, hESCs were plated as single cells in a Geltrex-coated 24-well dish at a density of 210,000 cells per well in mTesR1 with ROCK inhibitor Y-27632 (5 μM; Stemgent). Starting on the next day (day 2), cells were exposed to Activin A (100 ng/mL; PeproTech) for 3 days in RPMI 1640 media (Invitrogen) containing increasing concentrations of 0%, 0.2%, and 2.0% define fetal bovine serum (dFBS; Invitrogen) for DE induction. BMP4 (50 ng/mL; R&D Systems) was added on the first day of DE induction. Following DE induction, cells were cultured in FG media: RPMI 1640 media containing 2.0% dFBS, CHIR99021 (2 μM; Tocris), FGF4 (500 ng/mL; R&D Systems), and Noggin (200 ng/mL; R&D Systems). The media was changed every day. After 3 days, foregut spheroids were seen in the wells. On the third day of CHIR/FGF/Noggin treatment, retinoic acid (RA, 2 μM; Sigma-Aldrich) was added. The spheroids were collected, re-suspended in 35 μL Matrigel (Corning), and plated in a three-dimensional droplet. Spheroids were overlaid with ST1 media: Advanced DMEM/F12 with N2 (Invitrogen), B27 (Invitrogen), L-glutamine, penicillin/streptomycin, and EGF (100 ng/mL; R&D Systems). For the first 3 days, RA and Noggin were added to the ST1 media. Media was replaced every 3–4 days. At day 20, organoids were collected and re-plated in fresh Matrigel at dilution of 1:15. At day 27, Dox (1 μg mL⁻¹) was added to the medium to induce the expression of NPM factors.

Antibodies and immunohistochemistry

Tissues were processed as described. Briefly, hGOs or grafted tissues were fixed in 4% paraformaldehyde at 4°C for 5 min and then incubated in 30% sucrose solution for 2 h. The tissues were embedded in OCT (TissueTek) on dry ice and cut into sections (14 μm thickness) with a Leica cryostat and placed on Superfrost plus slides. The sections were rinsed with PBST for 5 min and blocked with 5% donkey serum in PBST for 30 min at room temperature. Primary antibodies were added at appropriate dilutions overnight at 4°C. Then the sections were rinsed with PBST three times, and secondary antibodies were added for 1 h at room temperature followed by rinsing with PBST 3X and adding Vector Laboratories mounting reagent with DAPI. The primary antibodies are listed as follows: guinea pig insulin (Dako; 1:1200), rabbit C-peptide (Cell Signaling Technology; 1:500), goat gastrin (Santa Cruz; 1:100), goat ghrelin (Santa Cruz; 1:500), rabbit serotonin (5-HT) (Immunostar; 1:2000), goat somatostatin (Santa Cruz; 1:500), guinea pig glucagon (Linco; 1:2000), rabbit PCSK1 (Millipore; 1:500), mouse NKX2-2 (DSHB; 1:50), mouse NKX6-1 (DSHB; 1:20), rabbit Pax6 (Millipore; 1:200), rat PECAM1 (Pharmingen; 1:200), goat CDH1 (R&D; 1:800), rabbit MKI67 (Abcam; 1:500), rabbit MAFA (Bethyl; 1:400), rabbit MAFB crossed with MAFA (Sigma; 1:400), rabbit MAFB (Bethyl; 1:200), rat Cherry (Chromotek; 1:1,000), chicken NEUROG3 (BCBC; 1:200), guinea pig PDX1 (Abcam; 1:1,000), rabbit PDX1 (Abcam; 1:2,000), and rabbit MAFA (Bethyl; 1:400). Secondary antibodies were purchased from the Jackson ImmunoResearch Laboratories. Pictures were taken with a Nikon Ti-S microscope. For quantification of marker⁺ cells such as insulin⁺ cells, a total of at least 1,000 cells were analyzed from tissues harvested from three different animals. Typically, at least 10 randomly selected sections were counted per animal.

FACS for RNA-seq or scRNA-seq

NPM-hGOs were washed with DPBS without Mg²⁺ and Ca²⁺ before being incubated in 700 μL of pre-chilled cell recovery solution for 3 min. The Matrigel dome was detached, and the organoids were placed on ice for 20 min. Following centrifugation at 300g for 3 min, the cell pellet was washed with DPBS and resuspended in pre-warmed Accutase, with pipetting every 5 min for 11–12 cycles. After allowing larger fragments to settle, the supernatant (fraction #1) was collected and further dissociated using a 27G needle. The larger tissue fragments (fraction #2) were similarly treated with Accutase, pooled with fraction #1, and filtered through a 40 μm strainer. The final cell suspension was centrifuged at 500g, the supernatant was discarded, and the pellet resuspended fluorescence-activated cell sorting (FACS) buffer (PBS Ca²⁺ Mg²⁺-free, 2% FBS, 25 mM HEPES, and 2 mM EDTA). Alexa Fluor 488 anti-human CD326 (EPCAM) antibody (BioLegend # 324209) was added and incubated with the cells on ice for 15 min. Cells were then washed with FACS buffer three times. DAPI was added before the cells were loaded into the flow cytometer. EPCAM⁺ DAPI⁻ cells (negative control) or EPCAM⁺ mCherry⁺ DAPI⁻ cells (Dox treated) were used for sequencing.

RNA-seq and scRNA-seq analysis

Bulk RNA-seq was analyzed as previously described (Gu et al., 2022). Sequencing data were processed using the 10× Genomics human reference. Outputs from Cell Ranger served as inputs for the creation of Seurat objects (v.5.1.0) (Butler et al., 2018). Low-quality cells were filtered out based on specific criteria: cells were deemed low quality if the number of detectable genes or read counts fell below the 2nd percentile or exceeded the 98th percentile of the dataset, or if the percentage of mitochondrial genes was greater than 15%. The NormalizeData function was applied for normalization using default parameters. Putative doublet cells were identified and removed using DoubletFinder (v.2.0.4) (McGinnis et al., 2019). Ambient RNA in the human islet datasets was eliminated with SoupX (v.1.6.2) (Young and Behjati, 2020). The normalized data were then scaled using the ScaleData function. Principal-component analysis (PCA) was conducted on the scaled data via the runPCA function. To group similar cells in two-dimensional space, selected top 18 principal components (PCs) were utilized in non-linear dimensional reduction techniques, including t-distributed stochastic neighbor embedding (t-SNE). The same PCs from each sample were employed to construct a K-nearest neighbor graph using the FindNeighbors function. The FindClusters function was used to identify cell clusters at a resolution of 0.7. FindAllMarkers was then applied to determine marker genes for each cluster. Clusters with overlapping marker profiles were subsequently merged into a single cluster. Seurat object was converted to monocle 2 (2.30.1) CellDataSet object for pseudotime trajectory analysis (Trapnell et al., 2014). Data dimension was then reduced by reduceDimension function with the following arguments: max_components = 2, method = “DDRTree”. Cells were then ordered along the trajectory by orderCells function. GO analysis was done by enrichGO function in ClusterProfiler (4.4.4) as previously described (Wu et al., 2021).

GSIS

Human islets (10 medium islets in each group, Prodo Laboratories) and NPM-hGOs derived from two independent clones (10 organoids in each group) were sampled. The hGOs were mechanically disrupted to expose the epithelial layer by first cutting with a feather blade followed by gentle pipetting. GSIS was performed as previously described (Li et al., 2014a, 2014b). Tissues were washed with Krebs buffer (129 mM NaCl, 4.8 mM KCl, 2.5 mM CaCl₂, 1.2 mM MgSO₂, 1 mM Na₂HPO₄, 1.2 mM KH₂PO₄, 5 mM NaHCO₃, 10 mM HEPES, and 0.1% BSA, in deionized water) several times to remove any debris. Tissue clusters were then collected and pre-incubated in low (2 mM) glucose Krebs for 4 h in 37°C cell incubator to remove residual insulin. Clusters were washed two times in Krebs, incubated in low-glucose Krebs for 60 min in incubator, and supernatant was collected. Then clusters were washed two times in Krebs, incubated in high (20 mM) glucose Krebs for 60 min in incubator, and supernatant was collected. Supernatant samples containing secreted insulin were processed using the Human Ultrasensitive Insulin ELISA (ALPCO, 80-INSHUU-E01.1).

Transplantation studies of hGOs

NPM-hGOs were taken out from 3D culture by dissolving Matrigel with Cell Recovery Solution (Corning) and washed once with PBS. RPMI 1640 media was used to resuspend the hGOs. Approximately 50–80 hGOs were delivered into the epididymal fat pads of SCID-Beige mice at 8 weeks of age.

Half of the transplanted mice were administered Dox (2 mg/mL) in drinking water. The medicated water was changed once a week. At selected time points, human insulin in mouse blood samples was evaluated. Plasma was collected by centrifugation at 1,000g to remove blood cells. Plasma was stored at −80°C and later assayed using a human ultrasensitive insulin ELISA (ALPCO, 80-INSHUU-E01.1). Intraperitoneal glucose tolerance test was performed after an overnight fast (16 h) and injection of glucose (2 g/kg; 30% glucose solution in saline). SCID-Beige mice without hGO transplantation were subjected to glucose tolerance test as the control group. For in vivo GSIS, transplanted mice were fasted overnight and injected with 2 g/kg glucose intraperitoneally.

Diabetic mice were generated via intraperitoneal injection of low-dose STZ (50 mg/kg) for 3 times in 5 days after 6-h fast in both transplanted and control SCID-Beige animals. Mice that displayed >200 mg/dL blood glucose levels for 2 consecutive days after STZ administration were used for experiments. Blood glucose was measured with an Ascensia Elite glucometer (Bayer) after overnight fasting. 7 weeks after transplantation, fat pads containing the grafts were dissected from mice and analyzed.

RNA isolation and quantitative reverse-transcription PCR

Total RNA was isolated from tissues using E.Z.N.A. RNA isolation micro-elute kit (Omega). Reverse transcription was performed from 100 ng RNA using iScript cDNA synthesis kit (Bio-Rad) according to manufacturer’s protocol. qPCR was done using SYBR Green Master Mix (Bio-Rad) on a Real-time PCR Detection System (Bio-Rad). Gene expression was analyzed using the ΔΔCT method. PCR primers are listed in Table S1.

Statistical analysis

All statistics were performed using GraphPad Prism software 6.0. Specific statistical tests for each experiment are described in the figure legends. Briefly, qPCR data for NPM-hGOs with Dox treatment (+Dox group) or without Dox treatment (−Dox group) were analyzed with the unpaired Student’s t test. For glucose-stimulated insulin secretion experiments, paired one-tailed t tests were used to compare low glucose versus high glucose (in vitro GSIS). The Student’s t test was used to compare the NPM-hGO-transplanted group with control group at each time point of the blood glucose-tracking curve. For all analyses, a minimum of p < 0.05 was considered statistically significant.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Xiaofeng Huang (xih4001@med.cornell.edu).

Materials availability

All materials used in this study will be available upon request.

Data and code availability

Sequencing data that support the findings of this study have been deposited in the GEO under accession code GSE279772 and FigShare https://doi.org/10.6084/m9.figshare.30290740.v2.

Acknowledgments

We thank Q.Z. for his major contribution in this study as a senior author. We thank Shahin Rafii for his general support after Q.Z.’s passing. We thank the WCM Epigenetics Core (Jenny Xiang) and David Redmond for the assistance with scRNA-seq data generation and processing. We thank WCM Flow Cytometry Core (Jason McCormick and Tom Baumgartner) for performing FACS sorting. We thank Jennifer Hyoje-Ryu Kenty and Deanne R. Watson for expert help with tissue transplantation and insulin measurement, Joslin Specialized Assay Core for insulin measurement, Jim lab at Cincinnati Children’s Hospital for sharing expertise at organoid cultures, the WCM Human Therapeutic Organoid Core (RRID:SCR_027251) for guidance with organoid cultures, and members of the Zhou lab for advice and feedback. J.L. acknowledges financial support from the China Scholarship Council. J. Zhu thanks Jiangsu provincial government scholarship program for study abroad. This study was supported by awards from the National Institutes of Health and the Harvard Stem Cell Institute (R01DK106253, R01DK133701, R01DK133332, and DP-0144-14-00) for Q.Z.

Author contributions

Q.Z. and Q.X. conceived the project and designed the experiments. Q.Z. supervised all aspects of the work and wrote the manuscript. X.H. co-supervised the project, performed RNA-seq and scRNA-seq data analyses, and contributed to manuscript writing. J.L. and H.K. carried out most of the experiments. J. Zhu analyzed data and performed animal transplantation experiments. J. Zhang and C.M. conducted stem cell differentiation. V.P. performed cryosectioning and immunostaining. L.A.L. contributed to manuscript editing. C.A. provided valuable advice and performed preliminary animal experiments. A.A.D. carried out flow cytometry. T.L. performed animal surgeries. X.C. conducted preliminary experiments.

Declaration of interests

H.K. is presently an employee of Simple Planet. X.H. is currently employed by Johnson & Johnson Innovative Medicine.

Published: November 6, 2025