Glucose Transporters Are Key Components of the Human Glucostat

Mouse models play a major role for understanding key biochemical and physiological mechanisms of human disease. However, differences between humans and mice, such as metabolic measures (1), also pose significant hurdles for translating findings from the laboratory into the clinic. One prominent example is glucose homeostasis. Proper physiological functions of the body require the maintenance of blood glucose (BG) levels at a narrow range. Any deviation from this range triggers an immediate response that restores it back to normal. Glucose homeostasis is achieved primarily by two pancreatic endocrine hormones: insulin, secreted by β-cells, which lowers BG; and glucagon (GCG), secreted by α-cells, which elevates BG (2). Dysregulated insulin secretion can disrupt glucose homeostasis and cause severe metabolic disorders, including diabetes, a fast-growing disease characterized by high BG levels. Much of our knowledge about insulin secretion and glucose homeostasis stems from murine studies, but despite many similarities between humans and mice, there are also notable differences (3,4). One of those differences is their glycemic set point. The American Diabetes Association defines the normoglycemia range in humans to be 79–100 mg/dL (∼4–5.5 mmol/L) (5–7), lower than the range of 120–150 mg/dL (∼6.5–8.5 mmol/L) in mice (8,9). In other words, the normal glycemic range in mice surpasses the diabetic threshold in humans. This discrepancy must be considered when interpreting results from mice, starting with the identification of specific factors that contribute to the different glycemic set point in humans.

Glucose homeostasis involves the interplay between a network of organs, mostly the pancreas, and other metabolic organs such as the liver. These organs are collectively known as the glucostat, a mechanism responsible for glucose homeostasis that detects changes in BG and coordinates appropriate responses (10). Although many tissues contribute to glucose homeostasis, pancreatic islets have been suggested to be the primary glucostat, harboring the cellular machinery that controls the glycemic set point. When exposed to increasing extracellular glucose concentrations in vitro, murine islets showed enhanced insulin secretion at high glucose levels of 180–360 mg/dL (10–20 mmol/L), whereas human islets responded to lower glucose levels of 40–135 mg/dL (2–7.5 mmol/L) and reached a plateau in insulin secretion when glucose concentrations increased further (11). As such, the optimal glucose concentration range for insulin secretion from mouse and human islets matches the natural glycemic set point in each species. Further supporting a key role for islets as a glucostat, transplantation of human islets lowered the glycemic set point of the murine hosts to human BG levels (10,12,13). However, these studies were mostly performed in diabetic mouse models, where the endogenous murine islets were suboptimal or ablated, complicating the interpretation for the roles of islets in controlling species-specific glycemic set point. Therefore, whether human islets alone are sufficient to reset BG levels in nondiabetic conditions with intact mouse islets remains uncertain. Additionally, the specific cellular components within the islets that constitute the species-specific glucostat mechanism remain poorly characterized (14). Identification of these components will help elucidate the glucostat mechanisms and provide insight into the regulation of normoglycemia.

To comprehend the disparities in BG levels between humans and mice, we surveyed the proteins involved in glucose sensing in the insulin secretion pathway, building upon a previously established link between glucose sensors and the glucostat (15). Insulin secretion begins with the transport of extracellular glucose into β-cells through glucose transporters. Glucose is then phosphorylated by glucokinase (GCK), and further metabolized to generate ATP, which induces changes in the opening of ion channels, leading to insulin secretion (16). GCK is a major component in glucose sensing. This highly conserved enzyme (17) is the rate-limiting enzyme in glucose-stimulated insulin secretion (GSIS) due to its low affinity for glucose (K_m of 8–11.8 mmol/L in both humans and mice [18,19]). Its association with a monogenic type of diabetes, maturity-onset diabetes of the young 2, further emphasizes its crucial role in β-cell function and glucose homeostasis (7,17,20,21).

However, different BG levels between species cannot stem from a sensing mechanism that involves the same conserved proteins. Instead, we postulate that the difference could be due to differentially expressed glucose transporters in mouse and human β-cells. Glucose transporters are a large family of proteins consisting of 14 different transporters that differ in their affinity for the substrate (22). The primary glucose transporter expressed in murine islets is GLUT2, which has a relatively low binding affinity for glucose, with a K_m value of ∼15–17 mmol/L (23). Glut2 (Slc2a2)-null mice die shortly after birth from severe diabetes, supporting its role in glucose sensing (23,24). In contrast, the most highly expressed glucose transporter in human islets is GLUT1 (25–28), which has a much higher affinity for glucose, with a K_m of ∼6 mmol/L (2,29,30). GLUT2 is also present in human islets but is expressed at much lower levels compared with GLUT1 (29).

The K_m values of GLUT2 and GLUT1 correlate to the optimal extracellular glucose levels for insulin secretion from mouse and human islets, respectively (11), suggesting that the differences between GLUT1/2 expression could account for the species-specific glucose set point. However, despite this correlation, the precise requirements for GLUT1/2 in the human system remain unclear due to limited experimental research and inconclusive clinical data. The dominant human transporter, GLUT1, is widely expressed in many tissues and organs, including the brain. Heterozygous GLUT1 mutations are associated with GLUT1 deficiency syndrome, a neurological disorder resulting from impaired glucose transport to the brain (31). Biallelic mutations in GLUT1 (SLC2A1) lead to severe neurological developmental failures, and null mutations are likely to cause embryonic lethality (32–34). This pleiotropic gene requirement may complicate the evaluation of a postnatal diabetic phenotype. In contrast to GLUT1, GLUT2 shows a more restricted expression pattern, predominantly expressed in the digestive system, including liver, kidney, intestine, and pancreas. Consequently, biallelic mutations in GLUT2 are better tolerated, associated with Fanconi-Bickel syndrome, and characterized by glycogen accumulation in the kidney and the liver (35,36). GLUT2 mutations can also cause transient neonatal diabetes, but only in 4% of human carriers (37). Thus, the clinical significance for GLUT1/2 in BG control remains uncertain.

Here, we show that transplanting human primary islets or stem cells-derived islets (SC-islets) into mice decreases their BG levels to a human glycemic range, supporting a central role of islets in determining a species-specific normal glycemic range. We further conducted chemical and genetic perturbations to study the contribution of GLUT1/2 to human β-cell functionality and, consequently, BG regulation. Disruption of GLUT1 or GLUT2 led to reduced glucose transport and insulin secretion. Combined perturbation of both transporters had an additive effect, underscoring their distinct contributions. These findings, together with predictions from our mathematical model, emphasize the contribution of both GLUT1 and GLUT2 to glucose homeostasis and their impact on species-specific normal glycemic levels.

Human pancreatic islets were purchased from the National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases-funded Integrated Islet Distribution Program (IIDP) (RRID:SCR _014387) at City of Hope, grant 2UC4DK098085. Only islets from donors without diabetes were used for the study based on the criterion of HbA_1c <6% (Supplementary Table 1). HbA_1c is the preferred marker for evaluation of organ donors without a history of diabetes, because rigorous testing is typically not feasible in organ donations, and BG readings may be abnormal due to acute illness and/or medical interventions (38–40). Islets were transplanted into NSG mice on the day of shipment or were cultured in a ultra-low attachment six-well plate (Corning, 3471) or T75 flask (Thermo Fisher Scientific, 174952) in PIM(S) medium (Prodo Laboratories, PIM-S001GMP) supplemented with human AB serum supplements (Prodo Laboratories, PIM-ABS001GMP), Glutamine/glutathione supplement (Prodo Laboratories, PIM-G001GMP) and triple antibiotic supplement (Prodo Laboratories, PIM-3X001GMP) at 37°C with 5% CO₂. Islets were cultured for no longer than 3 days before undergoing experiments.

Mouse islets were purchased from the Joslin Diabetes Center, isolated from 90 male C57BL/6 mice aged 7–8 weeks.

Immunocompromised NSG mice (7-week-old males and females) were purchased from The Jackson Laboratory and housed under standard conditions in a controlled environment with a 12-h dark-light cycle and with free access to food and water ad libitum. Under an approved Institutional Animal Care and Use Committee protocol, mice were anesthetized via isoflurane inhalation and placed on a warm pad. Under aseptic conditions, a small longitudinal incision was made in the left lower back. Then, 3,000–5,000 islet equivalent (IEQ) of human pancreatic islets, SC-islets, or ∼3,000 IEQ of murine islets, mixed 1:1 with Matrigel (Corning, 354234), were injected with a 27-gauge needle into the renal capsule. In human islets, the proportion of β-cells is estimated to be ∼50% of the total cell population, whereas in murine islets, it is estimated to be ∼70% (41–43). Therefore, 3,000 IEQ of murine islets contain roughly the same number of β-cells as 4,000 IEQ of human islets. The number of β-like cells in SC-islets was estimated based on the percentage of INS⁺NKX6.1⁺ cells, which was ∼30% of the total cell population, as presented in Fig. 1A. Using this ratio, we matched the total number of SC–β-cells for transplantation to that of primary islets. Following wound closure, mice were monitored daily, and sutures were removed 7 days later.

Human fetal pancreas tissues (Supplementary Table 2) were obtained from secondary sources (Advanced Biosystems Resources [ABR]) under approved Material Transfer Agreements and with protocols approved by the University of Wisconsin’s Institutional Animal Care and Use Committee and Institutional Review Board (IRB) (IRB Study no. 2013-141). ABR obtains consent in accordance with Uniform Anatomical Gift Act (UAGA) and National Organ Transplant Act (NOTA) guidelines. ABR warrants that appropriate consent for tissue donation is obtained and that adequate records of such consents are maintained. In addition, tissues were obtained with local, state, and federal laws and regulations governing the procurement of human tissue. Within 24 h of recovery, the organs were received and cleaned of surrounding connective tissue, fixed with 4% paraformaldehyde, and processed for paraffin embedding.

Adult human pancreas tissues (Supplementary Table 2) were procured by the University of Wisconsin Organ and Tissue Donation Services from donors with no indication of diabetes or pancreatitis, with consent obtained for research from next of kin and authorization by the University of Wisconsin-Madison Health Sciences IRB, which granted an exempt from protocol approval for studies on postnatal tissue because research on deceased donors is not considered human subject research. IRB oversight of the project was not required because it did not involve human subjects as recognized by 45 CFR 46.102(f), which defines a “human subject” as “a living individual about whom an investigator (whether professional or student) conducting research obtains 1) data through intervention or interaction with the individual, or 2) identifiable private information.”

Mouse BG level was measured 1 day prior to surgery and on the day of surgery. Mice were left to recover for 2 weeks, and then BG was measured weekly for 91 days after surgery. Measurements were done using Contour next EZ glucometer and Contour next BG test strips.

Insulin was quantified 4–5 weeks following transplantation. Murine blood was collected in BD Microtainer tubes (BD, 365963) and was left to clot for 15–30 min. Tubes were then centrifuged to remove the clot at 2,000g for 10 min in a refrigerated centrifuge (4°C). Following centrifugation, supernatant was immediately transferred to a clean tube, and an ELISA kit was used to quantify insulin level (Mercodia, 10-1132-01 for human insulin, 10-1249-01 for mouse insulin).

Mice were deprived of food overnight for 16 h. Blood was collected to determine insulin level at t = 0. Mice were then injected intraperitoneally with 20% glucose stock solution for a final dose of 2 g of glucose/kg body mass. A second collection of blood was performed 30 min after glucose injection. Murine blood was collected in BD Microtainer tubes (BD, 365963) and was left to clot for 15–30 min. Tubes were then centrifuged to remove the clot at 2,000g for 10 min in a refrigerated centrifuge (4°C). Following centrifugation, supernatant was immediately transferred to a clean tube, and an ELISA kit was used to quantify insulin level (Mercodia, 10-1132-01 for human insulin, 10-1249-01 for mouse insulin).

H1 (NIHhESC-10-0043) human embryonic stem cells (hESCs) were cultured on vitronectin-coated plates (Thermo Fisher Scientific, A14700) in Essential 8 (E8) medium (Thermo Fisher Scientific, A1517001) at 37°C with 5% CO₂. Then, 5 µmol/L of Rho-associated protein kinase (ROCK) inhibitor Y-27632 (Selleck Chemicals, S1049) was supplemented to the medium on the first day after passaging or thawing. Cells were regularly confirmed to be free of mycoplasma by the Memorial Sloan Kettering Cancer Center Antibody & Bioresource Core Facility. Experiments with hESCs were conducted per National Institutes of Health guidelines and approved by the Tri-SCI Embryonic Stem Cell Research Oversight (ESCRO) Committee.

All differentiation experiments were performed on hESCs grown on vitronectin. hESCs were maintained in E8 medium for 2 days to reach ∼80% confluence. Cells were washed with PBS and differentiated to β-like cell stage, following previously described protocols (44,45). In brief, hESCs were washed with PBS and exposed to S1/2 medium supplemented with 100 ng/mL activin A (Bon Opus Biosciences), 5 µmol/L CHIR99021 (Stemgent, 04-0004-10) and 5 µmol/L Y-27632 (Selleck Chemicals, S1049) for 1 day of 3 days of definitive endoderm (DE) differentiation. On day 2, S1/2 medium was supplemented with 100 ng/mL activin A and 0.5 µmol/L CHIR99021. On day 3 medium was supplemented with 100 ng/mL activin A. On day 4, DE cells were rinsed with PBS and exposed to S1/2 medium supplemented with 50 ng/mL keratinocyte growth factor (fibroblast growth factor 7 [FGF7]; PeproTech, 100-19) and 0.25 mmol/L vitamin C (Sigma-Aldrich, A4544) for 2 days. For pancreatic progenitor 1 (PP1) stage, cells were switched to S3/4 medium (44) supplemented with 50 ng/mL FGF7, 0.25 mmol/L vitamin C, 1 µmol/L retinoic acid (Sigma-Aldrich, R2625), 5 µmol/L Y-27632 (Selleck Chemicals, S1049), and 1:200 insulin-transferrin-selenium-ethanolamine (ITS-X) for 2 days. For pancreatic progenitor 2 stage, cells were exposed to S3/4 medium supplemented with 2 ng/mL FGF7, 0.25 mmol/L vitamin C, 0.1 µmol/L retinoic acid, 200 nmol/L LDN, 0.25 µmol/L SANT-1, 100 nmol/L tryptose phosphate broth, and 1:200 ITS-X for 4 days. Cells were then dissociated into single cells with TrypLE Select (Thermo Scientific, A1285901). Cells were resuspended in S5-7 medium (44) and transferred into an ultra-low attachment six-well plate (Corning, 3471), each well contained 8 million cells. The plate was incubated on an orbital shaker at 100 rpm. Alternatively, 30 million cells were transferred to a 30 mL Bioreactor unit (REPROCELL, ABLE Bioreactor Magnetic Stir System). Cells were cultured until day 35.

Human pancreatic islets and SC–β-cell aggregates were washed with PBS and fixed in 4% paraformaldehyde (Thermo Fisher Scientific, 50980495) for 3 h at room temperature, washed with PBS, equilibrated in 30% sucrose overnight, and embedded in optimal cutting temperature compound (Tissue-Tek, 4583). Blocks were sectioned to 8-μm slices with a cryostat (Leica CM1950) onto slides (Fisherbrand, 12-550-15). Slides were washed twice with PBS and incubated in blocking solution (PBS with 0.1% Triton X-100, 0.1% Tween 20, 5% donkey serum) for 15 min at room temperature. Blocking solution was aspirated, and slides were incubated overnight at 4°C with primary antibodies, diluted in blocking solution. Slides were then washed 3 times with wash buffer (PBS with 0.1% Triton X-100, 1% donkey serum), and incubated for 1 h at room temperature with secondary antibodies diluted in wash buffer. Slides were washed once with wash buffer, stained with DAPI for 10 min at room temperature, and washed twice with wash buffer. Slides were mounted with Mowiol solution, covered in cover glass no. 1 (Corning, 2975-224), and imaged using confocal laser scanning platform (Leica TCS SP8). We quantified 20–50 different islets per donor from one to two nonadjacent sections. The files were loaded into ImageJ using the Bio-Formats Importer plugin. The DAPI channel was used to segment the nuclei through filtering, thresholding and the watershed segmentation algorithm. Next, the other fluorescent channels were thresholded, and the segmented nuclei were overlaid onto them. The percentage area of each nucleus was measured and used to determine positivity for that channel. All antibodies and dilution factors are listed in Supplementary Table 3.

Organs were received within 24 h of recovery and trimmed of extrapancreatic connective tissues, cut into 1 cm³ cubes, and fixed with 4% paraformaldehyde and processed for paraffin embedding. Then, 5-µm paraffin sections were deparaffinized using xylene and rehydrated. Antigen retrieval was performed by treatment with 10 mmol/L citrate buffer, pH 6.0, for 2 h at 80°C. Slides were blocked with 10% BSA/1× PBS for 1 h at room temperature, incubated with primary antibodies overnight at 4°C, washed, and incubated with secondary antibodies for 40 min at room temperature. For GLUT1 and GLUT2, signal was amplified using Tyramide SuperBoost Kits (rabbit Alexa Fluor 488, B40922, Invitrogen; mouse Alexa Fluor 594, B40942, Invitrogen), following the manufacturer’s protocol. Nuclei were counterstained with DAPI. Then, 20× and 40× images were collected using a Zeiss Axiovert 200M microscope. GLUT1 and GLUT2 were costained, while INS and GCG were costained on a serial section. Images of the serial sections were aligned using image registration on the DAPI channel in MATLAB. Then the aligned images were imported into FIJI. The DAPI channel was segmented using thresholding and watershedding. Then thresholding was used to determine positivity for INS, GLUT1, and GLUT2. All antibodies and dilution factors are listed in Supplementary Table 3.

GLUT2 homozygous knockout (KO) H1-iCas9 hESC (46) lines were generated using a guide RNA (gRNA) targeting around exon 1 of the GLUT2 gene. The gRNA targets sequence (AGCGGCAGCTGAATATTGAT) in the promoter region and we verified by Western blotting that the GLUT2 protein was absent in the GLUT2-KO cells differentiated to the SC–β-cell stage. gRNAs and tracer RNA were ordered from IDT and added at a 15 nmol/L final concentration. Doxycycline (2 μg/mL) was added to H1-iCas9 cells 1 day prior to transfection to induce Cas9 expression. On the day of transfection, cells were dissociated into single cells using TrypLE Select. Then, 0.5 million cells were plated in 12-well plate supplemented with doxycycline and 5 µmol/L Y-276325 (Selleck Chemicals, S1049). In parallel, Lipofectamine RNAiMAX (Thermo Fisher Scientific, 13778030) was mixed with Opti-MEM (Invitrogen, 31985070). In a separate tube tracer, RNA and gRNA were mixed with Opti-MEM. Both tubes were combined, incubated for 10 min at room temperature, and added dropwise to the freshly plated H1 iCas9 cells. Doxycycline was also supplemented the day after transfection. At 3–4 days after transfection, cells were dissociated into single cells, and ∼1,000 cells were plated into one 100-mm tissue culture dish for colony formation. After ∼10 days, single colonies were picked into a 96-well plate and were left to expand. Each well was then lysed, and genomic DNA was extracted and used for PCR genotyping. Primer sequences are listed in Supplementary Table 4.

Cells were dissociated into single cells using TrypLE Select and resuspended in FACS buffer (5% FBS in PBS) with Live-Dead Fixable Violet Dead cell stain (Invitrogen, L34955). Cells were incubated for 10 min at room temperature, followed by permeabilization/fixation at room temperature for 1 h. Primary and secondary antibodies were diluted in permeabilization buffer (eBioscience, 00-5523-00). Cells were incubated with primary antibodies at room temperature for 1 h and then incubated with secondary antibodies at room temperature for 1 h. Analysis was performed using BD LSRFortessa and FlowJo software. All antibodies and dilution factors are listed in Supplementary Table 3.

Aggregated SC–β-cells were dissociated with TrypLE Select and lysed using cell lysis buffer (Cell Signaling Technology, 9803) supplemented with proteinase/phosphatase inhibitors (Cell Signaling Technology, 5872) and 1 mmol/L phenylmethylsulfonyl fluoride (MP Biomedicals, ICN19538105). Cells were centrifuged for 10 min at 14,000g and 4°C, supernatant was transferred to a new vial, and protein concentration was quantified using Bradford Protein Assay (Bio-Rad, 500-0202). Protein samples and PageRuler Plus Prestained Protein Ladder (Thermo Fisher Scientific, 26619) were loaded into Bis-Tris 10% gel (Novex, NP0301BOX), and transferred to nitrocellulose membranes (Novex, LC2001). Membranes were incubated in blocking solution (5% milk in Tris-based saline with Tween 20, 0.1% TBST) for 1 h at room temperature. Primary antibodies were diluted in blocking buffer, and membranes were incubated overnight at 4°C. Horseradish peroxidase-conjugated secondary antibodies were diluted in TBST, and membranes were incubated at room temperature for 1 h. Enhanced chemiluminescence Western blotting detection reagent (Amersham, RPN2236) was used to visualize the protein bands. All antibodies and dilution factors are listed in Supplementary Table 3.

Glucose uptake in human pancreatic islets and SC-islets was evaluated using Glucose Uptake-Glo Assay (Promega, J1342) according to protocol and measured with GloMax Navigator Microplate Luminometer (Promega). In brief, equal numbers of IEQs of pancreatic islets or SC-islets were transferred to 96-well plates. A 2 nmol/L final concentration of GLUT1 inhibitor, BAY-876 (Millipore Sigma, SML1774-5MG), was added to appropriate wells and to each reagent mix applied to those wells subsequently in the protocol. The plate was incubated at 37°C for 1 h. 2DG6P Detection Reagent mix was prepared and left to equilibrate at room temperature for 1 h. Cells were washed with PBS, and 0, 5, or 17 mmol/L 2DG solution was added per well and incubated at room temperature for 10 min. Stop solution, followed by neutralization solution, was added per well. 2DG6P Detection Reagent mix was added, and the plate was incubated at room temperature for 1 h. Luminescence was then recorded. Total insulin was quantified using ELISA (Mercodia, 10-1113-01) and used for normalization.

Perifusion buffer was prepared fresh (24 mmol/L NaHCO₃, 120 mmol/L NaCl, 4.8 mmol/L KCl, 2.5 mmol/L CaCl₂ 2H₂O, 1.2 mmol/L MgCl₂ 6H₂O, 10 mmol/L HEPES, 0.25% BSA). Glucose was added to prepared 2.8, 5, and 17 mmol/L buffers, as well as 50 mmol/L KCl buffer. Each buffer was prepared with and without 2 nmol/L final concentration GLUT1 inhibitor BAY-876 (Millipore Sigma, SML1774-5MG). All subsequent steps were performed in both conditions. Equal numbers of IEQs of human pancreatic islets or SC-islets were washed in PBS and incubated in 2.8 mmol/L glucose buffer for 1.5 h at 37°C. Cells were loaded to the columns, and prewashed in 2.8 mmol/L glucose buffer. Columns were loaded to BioRep Perfusion System V5 instrument. Glucose buffer (2.8 mmol/L) was flowed at a rate of 100 μL/min for 30 min to allow for adjustment to flow pressure. Supernatant was then collected for 16 min every 90 s. Buffer was switched to 5 mmol/L glucose buffer for 30 min, and supernatant was collected every minute. Glucose buffer (2.8 mmol/L) was flowed for 15 min, and supernatant was then collected for 16 min every 90 s, followed by 17 mmol/L glucose buffer for 30 min, and supernatant was collected every minute. KCl buffer (50 mmol/L) was then flowed for 16 min, and supernatant was collected every 90 s, followed by a last step of 2.8 mmol/L glucose buffer, and supernatant was collected every 90 s. Insulin was then quantified using ELISA (Mercodia, 10-1113-01).

The model integrated GLUT1 and GLUT2 expression patterns based on the proportions of cells expressing GLUT1-only, GLUT2-only, or both, among cells that express at least one of transporters. The values were calculated for wild-type (WT) primary islets and SC-islets based on data shown in Fig. 1C and D. GLUT1/2 values were altered to reflect a lack of transporter function for GLUT2-KO SC-islets (GLUT2 = 0) and GLUT1 inhibition (GLUT1 = 0) in SC-islets and human primary islets. The kinetic parameters used in the model, except for n, which was determined in this work, were previously established (47). Calculation of predicted values for the input function and the dynamic equation were independently performed for all GLUT1/2 combinations for primary islets (WT, GLUT1 inhibition) and SC-islets (WT, GLUT1 inhibition, GLUT2 KO). MATLAB scripts for input and dynamic functions are shown in supplemental information.

All data generated or analyzed during this study are included in the published article (and in the Supplementary Material).

Human islets can be effectively modeled by differentiating hESCs into SC-islets (48–50) (Fig. 1A). Consistent with findings from previous studies using similar differentiation protocols (44,45,51–53), we were able to generate primarily monohormonal β-like cells expressing C-peptide but not GCG or somatostatin (Fig. 1A). These β-like cells expressed NKX6.1 (Fig. 1A), along with additional β-cell markers such as MAFB, NKX2.2, and NTPDase3 (54) (Supplementary Fig. 1). This SC-islets model provides a readily available source of insulin-producing cells for experimental interrogation, providing a practical way to study the contribution of GLUT1 and GLUT2 to insulin secretion and glucose homeostasis. On the other hand, SC-islets exhibit suboptimal insulin secretion in response to extracellular glucose when compared with adult β-cells, suggesting that they may correspond to a partially mature fetal developmental stage (55,56). Moreover, SC-islets differ from adult islets in both endocrine cell composition and overall architecture, which could impair intercellular interactions and affect overall islet function (52,57,58). To date, comprehensive human protein expression data of GLUT1 and GLUT2 at the single β-cell level has been lacking. To investigate the roles of GLUT1/2 in glucose regulation, we first evaluated their expression profiles in the insulin-expressing cell population of SC-islets from six independent differentiation experiments, as well as primary islets from six adult donors, by coimmunostaining with INS (Fig. 1B), followed by quantification using ImageJ (Fig. 1C and D and Supplementary Fig. 1G). β-Cells expressed GLUT1 and GLUT2 in both systems, but with different compositions. Most adult primary β-cells (77.6% ± 10.5) expressed GLUT1 but not GLUT2, and a small percentage (13.2% ± 7.8) coexpressed both transporters, consistent with previous studies demonstrating GLUT1 to be the primary glucose transporter expressed in human β-cells (22,29,30,58–60). In contrast, 64% of INS⁺ cells in SC-islets expressed GLUT2, approximately half of which also coexpressed GLUT1 (Fig. 1C).

To investigate whether the different GLUT1/2 expression profiles between primary β-cells and SC-islets could reflect their respective developmental stages, we examined GLUT1 and GLUT2 expression in pancreatic sections from adult and fetal donors. Pancreatic specimens from four adult donors and fetal pancreas specimens from four donors between 17 and 22 gestational weeks were examined (Fig. 1E). To accommodate the specific staining requirements for GLUT1/2 antibodies in paraffin blocks, we performed INS staining in adjacent serial sections in conjunction with GCG staining to facilitate section alignment. Despite these technical differences, the results from adult pancreas specimens were qualitatively similar to those observed for adult islets (Fig. 1D and G). We found 91.8% ± 2.8 INS⁺ cells expressed GLUT1, among which a small population (8.1% ± 5.9) coexpressed GLUT1 and GLUT2, and only a negligible population expressed GLUT2 but not GLUT1. In comparison, most fetal INS⁺ cells also expressed GLUT1, but 33.8% ± 2.1 coexpressed GLUT1 and GLUT2, and a small population (4.2% ± 2.8) expressed only GLUT2 (Fig. 1F). Overall, these results show that fetal β-cells and SC-islets have a higher percentage of cells expressing GLUT2 compared with adult β-cells. However, there are also notable differences between fetal β-cells and SC-islets: SC-islets have a higher proportion of GLUT2-only cells and fewer GLUT1-only cells.

Transplantation of human islets into mice has been a well-established practice in diabetes research. It is commonly used to test treatment approaches and evaluate the efficacy of islet transplantation in reversing diabetes in mouse models. As such, transplantations are typically performed in diabetic mouse models, where the endogenous islets are underperforming or ablated (10,61,62), and the potential of human islets to reset the recipient mice’s BG level to the human set point has not been thoroughly investigated. If differences in GLUT1/2 expression contribute to the species-specific glucose set point, transplanting human islets and SC-islets, both of which contain substantial numbers of cells expressing the low K_m glucose transporter GLUT1, should recalibrate mouse normoglycemia and set it to the human range. To test this prediction and to compare between primary adult islets and SC-islets, immunocompromised NSG mice with intact endogenous islets were transplanted with human islets (n = 21) or SC-islets (n = 10) under the kidney capsule (Fig. 2A). Transplanted mice were compared with three sets of controls: 15 nontreated (NT) mice, 15 mock-transplanted mice that underwent surgery but were injected with PBS instead of islets, and an additional 6 mice transplanted with murine islets containing a comparable number of β-cells to that of transplanted human islets. In vivo GSIS was performed 3 months after transplantation to assess both human and mouse β-cell functions through quantifying the species-specific insulin levels (Fig. 2B and C). Human insulin was only detected in mice transplanted with human islets and SC-islets, with a significant increase observed 30 min after glucose injection. In comparison, mouse insulin levels were significantly elevated in all groups at the 30-min time point, indicating that the transplantation did not affect mouse β-cell function. In addition, we performed intraperitoneal glucose tolerance test (IPGTT) to test the effect of the transplantation on glucose clearance (Fig. 2D and Supplementary Fig. 2A–C). All mice displayed a sharp increase in BG following glucose injection, which then gradually decreased and reverted to normal level. The area under the curve (AUC) was quantified (Fig. 2E), and no significant difference in BG regulation between the groups was observed, except for mice transplanted with SC-islets, which displayed a higher AUC value, suggesting a decreased ability to respond to changes in extracellular glucose, despite an overall successful restoration of BG levels.

We monitored murine BG weekly for 91 days (summarized in Fig. 2F and G, and results from individual mice are shown in Supplementary Fig. 2D and E). Throughout the experiment, the NT and PBS control mice both maintained a BG level averaging between 120 and 130 mg/dL. Mice that received mouse islet transplants showed no statistically significant difference from the control mice. In contrast, mice receiving primary human islets and SC-islets exhibited significantly lower BG levels compared with the control mice starting at 1 month after transplantation. This disparity increased further after 2 months and persisted until the conclusion of the experiments on day 91 (Fig. 2G). These results indicate that the transplantation of both human primary islets and SC-islets effectively lowers mouse BG levels to the human normoglycemia range. The subtle differences observed between primary islets and SC-islets may be attributed to the lower expression of GLUT1 in the latter.

We next examined the contribution of GLUT1 and GLUT2 to β-cell functionality using both primary islets and SC-islets (Fig. 3A). We first examined the roles of GLUT1 in human islets using a well characterized specific and potent inhibitor, BAY-876 (63), that blocks glucose transport by GLUT1. Glucose uptake was evaluated as a direct measure of transporter activity using a luminescence-based method. Because glucose undergoes phosphorylation and subsequent metabolism once it enters the cell, it cannot be quantified directly. Hence a glucose analog, 2-deoxyglucose (2DG), is used. 2DG is transported into the cell similarly to glucose, and then undergoes phosphorylation to generate deoxyglucose-6-phosphate (2DG6P). 2DG6P is not metabolized and is accumulated in the cell. Luciferin is used to generate a luminescent signal that is proportional to the concentration of 2DG6P. For the GLUT1 inhibition experiments, BAY-876 was added to the islet media 1 h prior to the start of the experiment and maintained throughout the protocol. Human islets were incubated with 5 mmol/L or 17 mmol/L 2DG, corresponding to the K_m value of GLUT1 or GLUT2, respectively (Fig. 3B). Inhibition of GLUT1 led to a significant decrease in glucose uptake in both 5 mmol/L and 17 mmol/L 2DG treatment conditions (Fig. 3B), indicating that GLUT1 is required for glucose uptake across a range of glucose concentrations.

We next conducted dynamic GSIS (dGSIS) to evaluate the responsiveness of primary β-cells to changes in external glucose levels. Human islets were fasted in a low glucose (2.8 mmol/L) solution for 2 h prior to perifusion. To examine the roles of GLUT1, primary islets were treated with the GLUT1 inhibitor during the fasting period and throughout the rest of the experiment. After the fasting period, we exposed the islets to varying glucose concentrations. We started with a low glucose step, followed by a high step of 5 mmol/L glucose to match the K_m of GLUT1. Next, we exposed the islets to another low glucose step, followed by the second high glucose step of 17 mmol/L, which corresponds to the K_m of GLUT2. To complete the dGSIS sequence, the islets were exposed to low glucose, followed by KCl treatment and a final low glucose step. The supernatant was continuously collected during the experiment and quantified for insulin content. The dGSIS plot displays the fold change in insulin secretion relative to the initial baseline level at 2.8 mmol/L glucose (Fig. 3C). To compare the change in insulin secretion during dGSIS, we quantified the AUC during the 5 mmol/L and 17 mmol/L treatments (Fig. 3D). Insulin secretion was significantly increased in both 5 mmol/L and 17 mmol/L glucose conditions compared with the baseline, with higher insulin secretion when islets were exposed to 17 mmol/L glucose. Moreover, inhibition of GLUT1 caused a significant decrease in insulin secretion at both glucose concentrations. Taken together, our findings demonstrate the requirements of GLUT1 for primary β-cell functionality, including glucose transport and insulin secretion in response to glucose stimulation.

To study the roles of GLUT1 and GLUT2 in SC-islets, we applied CRISPR-Cas9 to generate clonal KO hESC lines. We were not able to generate GLUT1-KO lines, which is consistent with previous findings showing that GLUT1 is essential for mouse ESC viability (64). Therefore, we used the GLUT1 inhibitor BAY-876 to assess its functional impact. We were able to generate two homozygous GLUT2-KO clones (Fig. 3E). Western blot analysis confirmed that GLUT2-KO SC-islets lacked GLUT2 protein expression but showed comparable expression of INS and GLUT1 (Fig. 3F). We first performed glucose uptake assays using 2DG. GLUT2-KO SC-islets exhibited a significantly lower level of glucose uptake compared with WT SC-islets when exposed to 17 mmol/L 2DG, whereas no significant difference was observed between KO and WT cells under the 5 mmol/L 2DG condition (Fig. 3G). These results indicate that GLUT2 is required for glucose transport at high glucose concentrations, which aligns with its high K_m for glucose. We further found that inhibiting GLUT1 resulted in significantly decreased glucose transport in both WT and GLUT2-KO SC-islets, regardless of the 2DG concentrations. These results indicate that GLUT1 is required for efficient glucose transport when exposed to both low and high external glucose concentrations and that its function is independent of the presence of GLUT2.

We next performed dGSIS on WT and GLUT2-KO SC-islets (Fig. 3H and Supplementary Fig. 3A–C) to investigate the role of GLUT2 on insulin secretion. Similar to human islets, WT SC-islets exhibited increased insulin secretion in response to both 5 mmol/L and 17 mmol/L glucose, with a greater response observed at 17 mmol/L glucose. Compared with WT SC-islets, GLUT2-KO SC-islets exhibited a reduced response to 17 mmol/L glucose, whereas no significant difference was observed between KO and WT at 5 mmol/L glucose. In addition, the increase in glucose concentration from 5 to 17 mmol/L did not result in a significant increase in insulin secretion in GLUT2-KO SC-islets. These results support the crucial role of GLUT2 in enabling SC-islets to effectively respond to high glucose levels. In addition, inhibiting GLUT1 resulted in a significant decrease in secreted insulin in both WT and GLUT2 KO SC-islets, regardless of the glucose concentrations (5 or 17 mmol/L). Together, these experiments establish that both GLUT1 and GLUT2 are required for SC-islets functionality, including glucose transport and insulin secretion in response to glucose stimulation.

Several mathematical models have been developed to study the regulation of glucose and insulin secretion in both healthy and diabetic conditions (65–71). However, most models did not specifically consider the effects of GLUT1 and GLUT2 expression on β-cell functionality. One model evaluated the difference in glucose uptake and its utilization in healthy versus type 2 diabetes (T2D) islets, taking into account their differential expression levels of GLUT1 and GLUT2 (47). However, this model did not incorporate how changing GLUT expression within otherwise healthy β-cells would alter functionality, an important aspect that could contribute to the onset of diabetes. Furthermore, the model did not account for the inherent heterogeneity within the β-cell population. Here, we developed a model that assesses how varying GLUT expression profiles can affect β-cell functionality and influence the normoglycemic range. The specific combination of expressed glucose transporters and the concentration of extracellular glucose determines the amount of glucose transported into β-cells, which in turn determines the level of insulin secretion (60,72). We generated a mathematical model that describes the process of glucose uptake, as this is the most immediate outcome of GLUT activity. It has been established that the relationship between glucose and insulin is well described by a sigmoidal curve, for which the Hill equation provides a good mathematical fit (73). Modeling the level of intracellular glucose (G_i) would allow for a simplistic model. We developed an equation (Fig. 4A) to describe the possible states of glucose. Extracellular glucose (G_o) is transported into the cell at a rate V_i. Once inside the cell, G_i can diffuse out of the cell at a rate V_o or undergo phosphorylation by GCK to produce glucose 6-phosphate (G_p) at rate V_p. We used the input function f(G_i) to describe the processes contributing to G_i (Fig. 4B). The function depicts the uptake of glucose via both GLUT1 and GLUT2 at their respective uptake rates. The concentrations of either transporter, [GLUT1] or [GLUT2], would be multiplied by the specific transport rate of the corresponding transporter (V_Glut1 and V_Glut2). The rate is determined using a Hill equation, which shows that glucose transport is dependent on three kinetic parameters: 1) V_max, the maximal rate of glucose transport, 2) K_m, the affinity coefficient, and 3) n, the Hill coefficient (47,74).

To construct the model, we first sought to determine the appropriate Hill coefficient values. When we calculated the glucose transport rate as a function of G_o for both GLUT1 and GLUT2 (Supplementary Fig. 4A), we observed a clear difference in response for the two transporters at different n values. When n was set between 2 and 3, the plots displayed a differential relationship between the two transporters depending on the levels of G_o. GLUT1 showed a greater uptake rate compared with GLUT2 in low G_o values. However, at higher G_o values, the two curves intersected, and we observed a shift in GLUT1/2 relationship. The glucose uptake rate of GLUT2 showed a sharp increase and superseded that of GLUT1, while the transport rate of GLUT1 progressed toward a steady state. This pattern is consistent with the response profile of adult human islets that predominantly express GLUT1 and mouse islets that predominantly express GLUT2 in GSIS experiments (11), suggesting to us that the Hill coefficient values for GLUT1 and GLUT2 are in the range of 2 to 3. This is consistent with published work that inferred the Hill coefficient from dGSIS data (73). Therefore, we used n = 2 for both transporters in all further calculations.

To test the model, the level of G_i was determined for G_o concentrations of 5 mmol/L and 17 mmol/L, which correspond to experimental 2DG concentrations in the glucose uptake assay (Fig. 3B and G). GLUT1 and GLUT2 expressions were shown to be heterogenous within the β-cell population (Fig. 1). To further incorporate heterogeneity into the model we established a population size of 1,000 “β-cells.” To indicate transporter expression in a cell, we generated two vectors assigning a value of 1 (expressed) or 0 (absent) per “β-cell” for a total vector length of 1,000 values. The heterogeneity vector was adjusted to reflect the difference in expression profile of the transporters in islets and SC-islets, as presented in Fig. 1B–D (Fig. 4C and Supplementary Fig. 4B). The predicted G_i values were compared with the measured glucose uptake data from human islets and SC-islets (Fig. 3B and G). A Pearson correlation coefficient of r = 0.69 was observed, strengthening the validity of the model.

To describe the change in insulin secretion, as represented in this model by the change in G_i over time, we developed a dynamic equation that displays the difference between G_i production and clearance from the system. Production is described by the input function f(G_i), whereas clearance depends on G_i concentration and the combined rates of glucose export and phosphorylation (Fig. 4D). Given the well-established correlation between measured levels of extracellular glucose and the degree of insulin secretion response, our model assumes a direct proportional relationship between the levels of secreted insulin versus G_i, which represents the output of glucose transport (14,75). We tested the model by examining the correlation between insulin secretion as measured by dGSIS experiments and insulin secretion as predicted by G_i in our model. To do so, we calculated the predicted steady-state values of G_i for G_o concentrations of 2.8, 5, and 17 mmol/L and compared them to the measured values of secreted insulin from Fig. 3D and H. The high correlations observed for human islets (r = 0.86) and SC-islets (r = 0.73) support the validity of the model. The stronger correlation observed between the predicted and measured values for human islets compared with SC-islets (Fig. 4E and F and Supplementary Fig. 4C) may reflect differences in their functional maturity (44,55,76). This is consistent with the more potent response to extracellular glucose elevation observed in human islets compared with SC-islets (Fig. 3D and H).

The expression profile of glucose transporters in human islets has shaped the perception of their contribution to glucose homeostasis. GLUT1 is considered the primary contributor, whereas GLUT2 is viewed as a minor player. For that reason, it was essential to first characterize GLUT1 and GLUT2 protein expression profiles in human β-cells and establish a comprehensive picture of the various subpopulations. During the neonatal stage, there is an upregulation in GLUT1 and GLUT2 expression. The number of GLUT1⁺ cells exhibits a rapid increase during development, with the majority of β-cells expressing GLUT1 by 16–20 weeks postconception (wpc) (24,77). Consistent with previous data (76), we found that ∼80% of INS⁺ cells in fetal islets (17–22 wpc) expressed GLUT1, as did ∼85–90% of adult β-cells. In SC-islets, however, GLUT1 was detected in a considerably smaller cell population, consistent with previously published data (58), and was primarily coexpressed with GLUT2 (∼30%). The difference in protein expression between SC-islets and primary islets might contribute to the lower functionality exhibited by SC-islets, as shown here and previously (50). Additionally, we found GLUT2 expression was higher in fetal islets relative to adult β-cells (Fig. 1E–G). Fetal β-cells and SC-islets both show a large proportion (∼30%) of cells coexpressing GLUT1 and GLUT2. The shared property of a large GLUT1/2 population between fetal islets and SC-islets could suggest a similar level of functional maturation, characterized by suboptimal function compared with primary islets. This is supported by the lower insulin content and reduced insulin secretion response in both cell types (24,52,56–58,77,78). Thus, the proportion of the GLUT1/2 double-positive population could serve as an indicator of maturation for both SC-islets and endogenous islets during pancreatic development. On the other hand, there are also distinctions between SC-islets and fetal islets. Fetal β-cells also display a notable percentage (44%) of cells expressing GLUT1 alone, whereas SC-islets have a higher proportion (∼30%) of cells expressing GLUT2 only, with very few cells expressing GLUT1 alone. These findings suggest potential disparities between SC–β-cells and fetal β-cells. Alternatively, SC–β-cells may resemble fetal stages not encompassed in this study, because our analyses were confined to fetal specimens from 17 to 22 gestational weeks.

GLUT1 and GLUT2 have not been considered essential components of the human glucostat mechanism. A main reason is that most individuals with mutations in either of these transporters do not exhibit a diabetic phenotype (23,60,72,79). Our study challenges this notion by revealing the crucial roles of GLUT1 and GLUT2 in β-cell responsiveness to glucose stimulation. Our findings demonstrate that GLUT1 is necessary in both human islets and SC-islets for proper glucose uptake and insulin secretion. We further establish that GLUT2 is required in SC-islets, suggesting a similar requirement for GLUT2 in fetal islets. It is plausible that as β-cells mature and exhibit higher levels of GLUT1 expression, they gradually become less reliant on GLUT2. This shift in glucose transporter expression may contribute to the transient nature of diabetes phenotypes observed in patients with GLUT2 homozygous mutations (37). Interestingly, we also detected a small but significant subpopulation of GLUT2-expressing β-cells in adult islets. We speculate that under conditions of relatively high glucose levels, when GLUT1 reaches saturation and is no longer able to respond to further increases in glucose, GLUT2 likely facilitates glucose responsiveness. Our findings with SC-islets support this hypothesis. However, further investigation is needed to elucidate the specific functions of GLUT2 in adult β-cells.

Although this work focused on GLUT1 and GLUT2, other glucose transporters were previously identified in human β-cells, such as GLUT3, that has a very low K_m value of 1–2 mmol/L (60). The dGSIS experiment in the current study (Fig. 3H) showed that in the absence of functional GLUT1 and GLUT2, increasing the glucose concentration from 2.8 mmol/L to 5 mmol/L did not induce additional secretion of insulin. Hence, GLUT3 is unlikely to have a significant impact on glucose homeostasis in physiological conditions. Single-cell RNA sequencing data obtained in human islets suggested the presence of additional glucose transporters, such as GLUT8 and GLUT13 (26,80). The data were not validated but warrant consideration in future experiments, since any change in the GLUT expression profile may affect glucose homeostasis and β-cell responsiveness.

Our work has focused on the glucose-insulin circuit as well as the roles of GLUT1/2 in human islets and their contribution to resetting BG to human levels. However, other factors may contribute to glucose regulation. For instance, there may be differences between human and murine insulin in regulating downstream targets. As shown in the IPGTT experiment (Fig. 2D), control mice and transplanted mice both exhibited effective BG regulation. This is supported by previous studies, where human islets were transplanted into diabetic mice and facilitated rescue of the diabetic phenotype (10,61,81,82). In another study (83), the murine Ins1 gene was replaced with the human INS gene in NOD mice, and the resulting knock-in mice displayed normal glucose tolerance and reduced the incidence of spontaneous diabetes. Given that mouse INS-1 is a known target of diabetes-causing T cells and the study was conducted in NOD mice, it is unclear whether there are differences between human and mouse insulin on BG regulation beyond their distinct susceptibility to T-cell targeting in NOD mice. Another aspect to consider is the effect of the other endocrine hormones, mainly GCG, on glucose homeostasis. It has been shown that the GCG input is needed to induce an insulin response from pancreatic β-cells (10). The proximity of both α- and β-cells in human islets suggests that their intercellular interaction may play an important role in glycemia regulation by the human pancreas. However, this was not the case in rodents, where the effect of α-cells on β-cells was shown to be negligible (84–86). The discrepancy in these results could stem from another difference between human and murine islets, which is the proportion and distribution of endocrine cells within the islets. The proportion of α-cells in human islets exceeds that in mice (30–40% in humans vs. 15–20% in murine), and the human islet structure is more complex, with many more α- and β-cell interactions compared with its murine counterpart (41–43,87).

The steady-state model presented here is a simplified representation of the complex biology of pancreatic islets based on several assumptions. Firstly, it assumes that all β-cells have equivalent functionality. However, we know that β-cells have intrinsic differences and are subjected to different signals based on their location within the islet, islet size, and interactions of β-cells with other cell types (43,88,89). In SC-islets, not all differentiated β-cells are considered mature and functional; hence, not all contribute equally to functionality (55,90). This may explain the better correlation achieved in primary islets compared with SC-islets when comparing observed versus predicted values (Fig. 4E and F and Supplementary Fig. 4C).

Secondly, the model assumes that the correlation between G_i and insulin secretion is linear, but their relationship is likely more complex, as additional factors need to be considered, such as the feedback loop insulin secretion can have on extracellular glucose levels, and the effect of other pancreatic paracrine factors (73,91). For simplicity, we used an open-loop model with a fixed concentration of G_o. This allowed us to compare predicted G_i values to experimental data, where extracellular glucose is in excess and does not change throughout the experiment. However, in physiological glucose-insulin regulation, a negative feedback loop is present, where insulin secretion is reduced as BG levels decrease in response to insulin (92,93). Incorporating this negative feedback loop into the model will improve its predictive accuracy for an in vivo system.

Despite the simplicity of the model, it provides a valuable tool for predicting β-cell performance based on the expression of GLUT1 and/or GLUT2, shedding light on the distinct impact of each glucose transporter on islet functionality. The different compositions of GLUTs possessing different K_m values help explain the disparities in BG levels between humans and mice. In humans, the presence of GLUT1 enables the detection of glucose at low concentrations, triggering insulin secretion and clearance of glucose, thus setting a relatively low baseline glucose level. In contrast, mice rely on GLUT2, which responds to a higher BG threshold (Fig. 4H).

Beyond contributing to our understanding of species-specific glycemic set points, our quantitative model also has practical implications for improving the functionality of SC-islets through manipulating GLUT1/2 expression. For instance, we speculate that overexpression of GLUT1 in SC-islets may enhance their responsiveness to glucose stimulations and ensure adequate insulin secretion when exposed to physiological glucose levels found in human settings. The quantitative models can serve as a predictive tool for diabetes progression as well. Dysregulation of GLUT1 and GLUT2 expression has been observed in patients with T2D (47). Although glucose transport is not rate limiting in the insulin secretory pathway of normal human β-cells (23,79,94), a decrease in transporter expression can make them a rate-limiting factor. Measuring GLUT1 and GLUT2 expression in patients with prediabetes and with T2D, combined with quantitative modeling, can determine the range of sufficient GLUT1/2 expressions and the threshold for disease onset. This approach also helps identify individuals with a susceptibility to develop diabetes, enabling the prediction of β-cell function deterioration before noticeable effects on glycemia occur. As such, these findings have implications for potential preventive treatments and lifestyle changes to improve clinical outcomes in diabetes management.

Acknowledgments. The authors thank Shuibing Chen and Todd Evans for their valuable advice on the project, Angie Chi Nok Chong for sharing her expertise with dGSIS, Eric Chen for his assistance with ImageJ analysis, and Doron Haviv for the insightful discussion about systems biology. Figures 1A, 2A, 3A, 4A, and 4H and the graphic abstract were created with BioRender.com. During the course of preparing this work, the authors used ChatGPT, developed by OpenAI, to improve the clarity of some sentences. Following the use of this tool, the authors formally reviewed the content for its accuracy and edited it as necessary. The authors take full responsibility for all the content of this publication.

Funding. This study was supported in part by grants to D.H. from National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases (R01DK096239), and the American Diabetes Association (1-19-IBS-125), grants to J.S.O. and D.H. from JDRF (3-SRA-2021-1060-S-B) and Department of Defense Peer Reviewed Medical Research Program (W81XWH-20-1-0670), and the University of Wisconsin-Madison, Office of the Vice Chancellor for Research and Graduate Education with funding from the Wisconsin Alumni Research Foundation, a Memorial Sloan Kettering Cancer Center Support Grant from the National Institutes of Health, National Cancer Institute (P30CA008748), a General Atlantic Graduate Research Fellowship (to I.C.), a Beatrice P. K. Palestin Fellowship and a Bruce Charles Forbes Fellowship (to D.L.), and a postdoctoral fellowship from a NYSTEM training grant from the Center for Stem Cell Biology of the Sloan Kettering Institute (DOH01-TRAIN3-2015-2016-00006, to D.Y.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. I.C. performed most experiments, collected data, and developed the mathematic model. I.C and D.H. wrote the manuscript, and all other authors provided editorial advice. I.C. and D.H. devised experiments and interpreted results. D.M.T. and J.S.O. obtained and processed adult and fetal pancreas specimen, and D.M.T. performed staining. J.P. assisted with cloning, experimental design, and additional experiments not included in the manuscript. D.Y., D.L., and J.Y. assisted with SC–β-cell differentiations. D.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.