The Krüppel-like zinc finger protein GLIS3 transactivates neurogenin 3 for proper fetal pancreatic islet differentiation in mice
- 863 Downloads
Mutations in GLIS3, which encodes a Krüppel-like zinc finger transcription factor, were found to underlie sporadic neonatal diabetes. Inactivation of Glis3 by gene targeting in mice was previously shown to lead to neonatal diabetes, but the underlying mechanism remains largely unknown. We aimed to elucidate the mechanism of action of GLIS family zinc finger 3 (GLIS3) in Glis3−/− mice and to further decipher its action in in-vitro systems.
We created Glis3−/− mice and monitored the morphological and biochemical phenotype of their pancreatic islets at different stages of embryonic development. We combined these observations with experiments on Glis3 expressed in cultured cells, as well as in in vitro systems in the presence of other reconstituted components.
In vivo and in vitro analyses placed Glis3 upstream of Neurog3, the endocrine pancreas lineage-defining transcription factor. We found that GLIS3 binds to specific GLIS3-response elements in the Neurog3 promoter, activating Neurog3 gene transcription both directly, and synergistically with hepatic nuclear factor 6 and forkhead box A2.
These results indicate that GLIS3 controls fetal islet differentiation via direct transactivation of Neurog3, a perturbation that causes neonatal diabetes in mice.
KeywordsGLIS3 Neonatal diabetes Neurog3 Pancreatic islet differentiation
Bacterial artificial chromosome
Electrophoretic mobility shift assay
Forkhead box A2
GLIS family zinc finger 3
GLIS3 response element
HNF1 homeobox B
Hepatic nuclear factor 6
v-Maf musculoaponeurotic fibrosarcoma oncogene family, protein A (avian)
Neonatal diabetes and congenital hypothyroidism
NK6 homeobox 1
Notch gene homologue
Pancreatic ductal cells
Pancreatic and duodenal homeobox 1
SRY-box containing gene 9
Yellow fluorescent protein
Monogenic neonatal diabetes is characterised by beta cell dysfunction within 6 months of age, rendering affected individuals transiently or permanently diabetic, and requiring insulin or other treatments for survival [1, 2]. Permanent neonatal diabetes usually results from mutations of genes encoding transcription factors or other proteins that regulate beta cell development or function . A novel autosomal recessive disorder featuring permanent neonatal diabetes was described recently by Taha et al. , and linkage analysis of affected families identified GLIS3, which encodes a member of the Krüppel-like family of transcription factors , as the mutant gene underlying this syndrome . The demonstration of mutant GLIS3 in families with neonatal diabetes is intriguing; however, the underlying pathogenetic mechanism linking the GLIS3 locus to abnormal glucose homeostasis in neonates is unknown.
GLIS3 is expressed in human and mouse pancreas from early developmental stages through to adulthood, with higher expression in beta cells than in other islet or exocrine cells . We have previously shown that Glis3 directly regulates insulin gene expression; we also identified a GLIS family zinc finger 3 (GLIS3) response element (GLIS3RE) in the insulin promoter . The recent description of neonatal diabetes in Glis3-deficient mice by two different groups [8, 9] supports a pivotal role of GLIS3 in beta cell function during fetal development. Watanabe et al. first reported that inactivation of Glis3 produced neonatal diabetes in mice . A subsequent study by Kang et al.  found a dramatic loss of beta and delta cells, with a more modest loss of alpha, pancreatic polypeptide and epsilon cells in the Glis3 mutant mouse pancreas. The same team also showed that the expression of several genes encoding transcription factors involved in the regulation of endocrine differentiation, including Pdx1, Neurog3, Nkx6-1, Pax4, Pax6, Isl1, Neurod1 and Mafa, was significantly decreased in the pancreas of Glis3 mutant mouse. However, neither of the two studies [8, 9] reported the precise function of Glis3, and particularly whether and how GLIS3 regulates Neurog3 expression in pancreatic islet development.
To gain insight into the physiological and pathophysiological roles of GLIS3, we created Glis3−/− mice, which die with severe hyperglycaemia and ketoacidosis within 4 to 6 days of birth. The pancreatic islets of these mice were much smaller and poorly organised as compared with controls. Neurogenin 3 (NEUROG3), a basic helix–loop–helix pancreatic islet lineage-defining transcription factor, is essential to pancreatic islet formation [10, 11, 12, 13]. Here we show that GLIS3 is involved in the differentiation of endocrine progenitor cells through direct and indirect transcriptional control of Neurog3 expression. The combination of in vivo and in vitro experiments identified GLIS3 as a key regulator of islet morphogenesis during embryonic development and provided the mechanistic basis for a crucial role of GLIS3 in fetal islet differentiation and neonatal diabetes.
Glis3 gene targeting and generation of global Glis3 targeted mice
We purchased a bacterial artificial chromosome (BAC) clone (RP23-358 M17) containing the mouse Glis3 gene from Invitrogen (Carlsbad, CA, USA). Two DNA fragments, 2.5 and 7.2 kb, were subcloned from this BAC by recombineering  and used for homologous recombination. A 1.4 kb DNA fragment containing the targeted exon 4 with its immediate 5′ and 3′ introns (partial) was amplified by PCR and inserted in between two loxP sites of the NeoFrtLoxP vector. Two TK cassettes were inserted into the 5′-end of the targeting vector.
We electrophorated R1 mouse embryonic stem cells  with a linearised targeting construct, and selected embryonic stem cells with G418 (Invitrogen) and ganciclovir. Blastocyst injection and germline transmission were done by standard techniques.
To generate global Glis3-deficient mice, we bred Glis3fl/fl mice with protamine-Cre transgenic mice (PrmCre+/0), which express Cre in sperm . We then bred the Glis3fl/+/PrmCre+/0 males with C57BL/6 J females to generate Glis3+/−/PrmCre0/0 mice, which are essentially heterozygous mutant mice (Glis3+/−). Cross-breeding of these mice produced homozygous deletion of Glis3 (Glis3−/−). Mice used in this study were maintained in a barrier facility. All mouse protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at Baylor College of Medicine.
Glucose, insulin and ketone body measurement
Glucose level was measured using a glucometer (One Touch; Lifescan, Milpitas, CA, USA). We measured plasma insulin using a mouse ELISA kit (Mercodia, Winston Salem NC; Millipore, Billerica, MA, USA) and β-hydroxybutyrate (ketone body) with a kit (Cayman Chemical, Ann Arbor, MI, USA). To measure pancreatic insulin content, we homogenised neonatal whole pancreas at postnatal day (P) 0 in 0.2 mmol/l HCl with 70% (vol./vol.) ethanol. After neutralisation with 1 mol/l Tris HCl (pH 7.5), insulin was measured using a mouse ELISA kit (Mercodia) with the appropriate dilution.
We performed immunostaining on paraffin-embedded, 5- to 7-μm sections of whole embryos collected at embryonic day (E)12.5 and E15.5, or of dissected pancreas, which were fixed in freshly made 4% (wt/vol.) paraformaldehyde in PBS. Primary antibodies were: guinea pig anti-insulin (Abcam, Cambridge, MA, USA); rabbit anti-glucagon, rabbit anti-pancreatic polypeptide and rabbit anti-somatostatin (Dako, Carpinteria, CA, USA); goat anti-ghrelin (Santa Cruz Biotechnology, Santa Cruz, CA, USA); goat anti-pancreatic and duodenal homeobox 1 (PDX1) antibody (gift of C. Wright, Vanderbilt University, Nashville, TN, USA); mouse anti-NEUROG3 and anti-NK6 homeobox 1 (NKX6-1) (Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA, USA); anti-v-maf musculoaponeurotic fibrosarcoma oncogene family, protein A (avian) (MAFA) antibody (Bethyl Laboratories, Montgomery, TX, USA); and anti-cleaved caspase-3 antibody (Cell Signaling, Danvers, MA, USA). The primary antibodies were detected with secondary antibodies conjugated to FITC, Cy5 or rhodamine (Invitrogen). We used an Axiovert (Carl Zeiss AG, Jena, Germany) microscope with Axiovision imaging software to capture fluorescent images and did post-acquisition image processing with Adobe Photoshop (San Jose, CA, USA).
Islet area and positive cell quantification
We estimated islet area and endocrine-positive cell area using ImageJ 1.4 (NIH, Bethesda, MD, USA) on four haematoxylin and eosin stained or fluorescent sections (approximately every tenth section) that had been processed from four independent pancreases. For quantification of NEUROG3-positive cells, we counted the absolute number of positive cells on five (Glis3+/+) and ten (Glis3−/−) embryonic sections from at least five different mice for each genotype.
In situ hybridisation
We performed non-radioactive in situ hybridisation with digoxigenin-UTP (Roche Diagnostics, Indianapolis, IN, USA)-labelled antisense RNA probes using a 645-bp full length mouse Neurog3 cDNA clone; this was done in collaboration with the Gene Expression Core at Baylor College of Medicine.
Cell culture studies
We obtained pancreatic ductal cells (PDCs) from A. K. Rustgi (University of Pennsylvania, School of Medicine, Philadelphia, PA, USA) and maintained them as described by Schreiber et al. . We transduced PDCs with pMSCV (Clontech, Mountain View, CA, USA)-Glis3 retroviral construct and maintained rat 832/13 insulinoma cells (gift of C. Newgard, Duke University, Durham, NC, USA) as described previously . We cultured HepG2 cells in RPMI 1640 with 10% (vol./vol.) FBS. We used Lipofectamine 2000 (Invitrogen) for transfection according to the manufacturer’s instructions.
Luciferase reporter constructs and assays
Using RT-PCR, we amplified the coding sequences of mouse Glis3, Pdx1, Hnf6 (also known as Onecut1), Sox9, Foxa2 and Hnf1b, and cloned them into a c-myc-tagged pBOS-MCS vector . A mutant Glis3 cDNA that corresponds to the sequence in a family with neonatal diabetes and congenital hypothyroidism (NDH) syndrome was constructed (Glis3-NDH1) as described previously . A 5.8 kb mouse Neurog3 promoter fragment (SacI/KpnI), modified from Neurog3-CreER plasmid (Addgene, Cambridge, MA, USA), was cloned into a pGluc-basic (New England Biolabs, Ipswich, MA, USA) vector to generate a Neurog3-Gluc reporter construct. Wild-type and mutant (GLIS3RE)5-Gluc constructs were made as described previously . All expression constructs were confirmed by DNA sequencing. At 48 h after transfection, gaussia luciferase assays were performed as described previously  and results normalised to the activity of β-galactosidase reporter (Sigma, Ronkonkoma, NY, USA).
RNA isolation and quantitative PCR
A kit (Mini RNA Isolation I; Zymo Research, Irvine, CA, USA) was used to extract RNA from the E13.5 pancreases. Reverse transcription and quantitative PCR were performed as described previously . The housekeeping gene cyclophilin A was used as an internal control. Primer sequences are shown in electronic supplementary material [ESM] Table 1.
We performed co-immunoprecipitation assays using a nuclear complex co-immunoprecipitation kit (Active Motif, Carlsbad, CA, USA) as described previously .
We performed chromatin immunoprecipitation (ChIP) assays in PDCs infected with the pMSCV-c-myc-yellow fluorescent protein (YFP) or -GLIS3 retroviral constructs as described previously . For E13.5 pancreas ChIP, 65 to 100 pancreases were isolated from C57BL/6 E13.5 embryos, snap-frozen in liquid nitrogen and stored at −80°C until use. Frozen pancreases were pooled and thawed on ice, and cross-linked immediately in 1.5% (wt/vol.) formaldehyde at room temperature for 15 min, followed by ChIP assays.
Electrophoretic mobility shift assay
Electrophoretic mobility shift assays (EMSA) were conducted using a biotin-labelled double-stranded oligonucleotide probe containing the recognition sequence for Glis3 (ESM Table 1) as described previously .
Rabbit anti-mouse GLIS3 peptide (LSAVDRCPSQLSSVYTEG) antibody was generated by Thermo Fisher Scientific (Waltham, MA, USA).
The standard Student’s two-tailed t test was used for comparisons. Results are presented as the mean ± SD unless otherwise specified.
Post-natal growth retardation and severe neonatal diabetes in Glis3−/− mice
The strategy for creation of Glis3−/− mice is shown online in ESM Fig. 1a. We targeted exon 4 of mouse Glis3 gene, resulting in a premature termination codon in exon 6 (exon 5 162 bp). Glis3 mRNA was undetectable by quantitative RT-PCR in RNA isolated from the pancreas of Glis3−/− mice (ESM Fig. 1b). Glis3−/− mice had no intrauterine growth retardation (ESM Fig. 1c), fed normally and were indistinguishable from their wild-type littermates at birth. However, they showed growth arrest and became smaller starting from P1 as compared with control mice. By P4, they were dehydrated and much smaller. All Glis3−/− mice died between P4 and P6 (ESM Fig. 2a, b).
Although Glis3−/− mice appeared normal at birth, their blood glucose level was mildly but significantly higher than that of Glis3+/− or Glis3+/+ littermates on P0 (day of birth) (ESM Fig. 2c). By P4, Glis3−/− mice exhibited marked hyperglycaemia (>33.3 mmol/l) while their littermate controls remained euglycaemic (ESM Fig. 2c). Hyperglycaemia in Glis3−/− mice was associated with severe hypoinsulinaemia, with plasma insulin (ESM Fig. 2d) and pancreatic insulin content (ESM Fig. 2e) markedly decreased as compared with control littermates. Glis3−/− mice also displayed severe ketonaemia on P4 (ESM Fig. 2f). Our data corroborate and extend those published in two previous studies on Glis3−/− mice [8, 9].
Defective islet cell differentiation in Glis3−/− mice
To investigate the defect associated with insulin deficiency in Glis3−/− mice, we performed histological and immunofluorescence analyses on mouse embryos. At E12.5, immunoreactive glucagon and insulin-positive cells were detected in the pancreas of Glis3+/+ but not in that of Glis3−/− mice (data not shown). By E15.5, although glucagon- and insulin-positive cells were detectable in Glis3+/+ and Glis3−/− mice, both types of cells were significantly decreased in the latter (ESM Fig. 3a–d). At E17.5, the abundance of immunoreactive insulin, glucagon, somatostatin and ghrelin was markedly decreased in the islets of Glis3−/− mice (ESM Fig. 3e–j) as compared with controls.
As the numbers of all types of endocrine cells were uniformly reduced in the neonatal Glis3−/− mice, we examined apoptosis by cleaved caspase-3 immunostaining and found that the number of apoptotic islet cells was similar in Glis3−/− mice and Glis3+/+ littermates (ESM Fig. 4a, b). Furthermore, the exocrine pancreas was normal histologically and levels of the exocrine enzyme, amylase, remained unchanged in the pancreas of Glis3−/− mice (ESM Fig. 4c, d).
Absence of Glis3 leads to reduced abundance of key pancreatic endocrine differentiation transcription factors
To further characterise the perturbed endocrine pancreas development in Glis3−/− embryos, we used immunofluorescence to analyse the abundance of key pancreatic transcription factors at various stages. PDX1 and NKX6-1 are two transcription factors whose production marks endodermal epithelium cells that have been specified to a pancreatic fate [10, 19, 20, 21]. At E12.5, the abundance of PDX1 and NKX6-1 was normal in the Glis3−/− pancreas (ESM Fig. 5a–d), suggesting that early pancreatic epithelium specification was unperturbed in the absence of Glis3. However, by E15.5, the number of positive cells for PDX1, NKX6-1 and NEUROG3 was markedly reduced in Glis3−/− mice compared with Glis3+/+ littermates (ESM Fig. 5e–h). At E17.5 and P0, numbers of positive cells for all four key pancreatic islet transcription factors, i.e. PDX1, NKX6-1, NEUROG3 and MAFA, were markedly decreased in the pancreas of Glis3−/− mice (ESM Fig. 5i–t).
Expression of the proendocrine gene Neurog3 is markedly reduced in Glis3−/− embryos
In agreement with the reduced Neurog3 mRNA expression, NEUROG3-positive cells were also markedly decreased at the protein level as detected by immunofluorescence in E12.5 Glis3−/− embryos (Fig. 2c). Quantification indicated that the number of the NEUROG3-positive cells was decreased by >95% (Fig. 2d).
To corroborate the loss-of-function findings in Glis3−/− mice and to determine whether expression of Neurog3 is regulated by GLIS3, we performed a gain-of-function experiment by overexpressing c-myc-tagged Glis3 in mouse PDCs, a well established cell type frequently used to study Neurog3 expression and regulation [26, 27, 28]. Compared with cells transduced with the c-myc-YFP, overexpression of c-myc-Glis3 was sufficient to stimulate endogenous Neurog3 mRNA expression in PDCs (Fig. 2e, f), suggesting that Neurog3 is a downstream target of Glis3.
Neurog3 gene expression is known to be upregulated by several transcription factors such as PDX1 [28, 29, 30], hepatic nuclear factor 6 (HNF6) , SRY-box containing gene 9 (SOX9) , forkhead box A2 (FOXA2)  and HNF1 homeobox B (HNF1B) [33, 34]. It is negatively regulated by the Notch gene homologue (NOTCH) [11, 35, 36] and TGFβ [37, 38] signalling pathways. To determine whether Glis3 controls Neurog3 expression indirectly through these modulators, we examined their expression in E13.5 pancreas by quantitative RT-PCR. We found that mRNA expression of Pdx1, Hnf6, Sox9, Foxa2, Hnf1b, Notch1 to Notch4, Hes1 and Gdf11 occurred at similar levels in Glis3−/− to those in controls (Fig. 2a, ESM Fig. 4g).
As Hes1 is a negative regulator of pancreatic endocrine cell fate determination [11, 36], we examined its expression in E12.5 embryos, a critical stage in endocrine cell fate determination. We found that hairy and enhancer of split 1 (Drosophila) (HES1) abundance was similar in E12.5 Glis3−/− and Glis3+/+ pancreases (ESM Fig. 4e, f).
Taken together, these data indicate that Glis3 inactivation leads to reduced Neurog3 expression. However, downregulation of Neurog3 expression in Glis3−/− islets is not mediated indirectly by alterations in the expression of Pdx1, Hnf6, Sox9, Foxa2, Hnf1b, Notch1 to Notch4, Gdf11 or Hes1.
GLIS3 binds to the Neurog3 promoter and activates Neurog3 gene transcription
To determine if GLIS3 does indeed bind to these five putative GLIS3REs in the endogenous Neurog3 gene in cells, we first performed ChIP assays in PDCs transduced with c-myc-Glis3. The results showed that c-Myc-GLIS3 binds to all five putative GLIS3REs, but not to a control DNA sequence that is located in intron 1 (+7,963) of Neurog3 (Fig. 3b). To further determine whether GLIS3 binds to any of these sites in vivo, we performed GLIS3 ChIP assays using wild-type fetal (E13.5) pancreases. We found that GLIS3 occupied all five putative GLIS3REs, being particularly enriched at the −2,718 and −1,117 sites, in the mouse Neurog3 promoter of E13.5 fetal pancreas (Fig. 3c).
Next we performed EMSA using the DNA binding motif of GLIS3, the zinc finger domain. Of these five putative GLIS3REs, GLIS3 (zinc finger domain) was found to bind to the GLIS3RE probes at −2,862, −2,718, −1,160 and −1,117; the complexes were out-competed by molar excess of the corresponding non-biotinylated GLIS3REs. No binding was shown for the putative GLIS3RE at −4,692 (Fig. 3d).
GLIS3 activates Neurog3 promoter synergistically with HNF6 and FOXA2
To establish whether GLIS3 physically interacts with HNF6 and FOXA2, we co-expressed Glis3 with c-myc-Hnf6 or c-myc-Foxa2 in HepG2 cells and used an antibody against GLIS3 to immunoprecipitate GLIS3-interacting protein complexes from the nuclear extracts. Anti-c-Myc antibody detected c-Myc-HNF6 and -FOXA2, indicating that both transcription factors were co-precipitated with GLIS3 (Fig. 5b), whereas pre-immune rabbit IgG failed to immunoprecipitate either complex. These results demonstrate that GLIS3 interacts with HNF6 and FOXA2 both physically and functionally to activate Neurog3 gene transcription.
In this study, we used in vivo and in vitro approaches to examine the function of GLIS3 in fetal mouse pancreas development. PDX1 is considered a master regulator of early pancreas specification [10, 21, 24] and NKX6-1 is a marker of early pancreatic epithelium specification [19, 20]. In Glis3-deficient mice, pancreatic levels of PDX1 and NKX6-1 were normal at E12.5, indicating that the pancreatic epithelium was normally specified in the absence of Glis3.
Once the pancreatic progenitor cells have been specified, the next important cell fate decision is whether the cells adopt an exocrine, ductal or endocrine fate. NEUROG3 promotes an islet fate within the domain of PDX1-positive cells [10, 11, 12, 13]; therefore, in the absence of Neurog3, none of the major endocrine cell lineages can be formed . Furthermore, ectopic expression of Neurog3 in endodermal progenitor cells has been shown to program these cells into endocrine islet-like cells [39, 40, 41]. In contrast to the normal abundance of PDX1 and NKX6-1, NEUROG3 levels were significantly reduced in Glis3-deficient pancreases at E12.5, a finding that is consistent with the marked reduction of all five types of islet endocrine cells in the pancreas of Glis3−/− mice. Interestingly, from E15.5 onwards, all endocrine-related hormones and transcription factors examined by us were reduced in the islets of Glis3−/− embryos. Therefore, our studies suggest that the Glis3 gene is functionally important in a critical time window in mouse endocrine pancreas development around E12.5 or E13.5, immediately before the secondary transition begins. This conclusion is supported by the significantly increased Glis3 expression found at E12.5 in the fetal pancreas . From E15.5 onwards, we found numerous defects in the islets of Glis3−/− mice, e.g. reduction of PDX1 and NKX6-1 abundance, which may be a secondary effect of the arrested normal islet development. Neurog3 has been demonstrated to be a direct downstream target of Pdx1 in fetal pancreas development . Our study using Glis3−/− mice also places Glis3 upstream of Neurog3 during early pancreas development.
We examined whether Neurog3 is a direct downstream target of Glis3 in experiments involving beta and non-beta cell lines. We found that c-Myc-GLIS3 was sufficient to stimulate endogenous Neurog3 mRNA expression in PDCs, suggesting that Neurog3 is a downstream target of Glis3. Next, using a combination of luciferase reporter analyses, ChIP assays (in PDCs and in E13.5 pancreases) and EMSA, we identified four GLIS3-binding sites in the mouse Neurog3 promoter region. The discrepancy between ChIP and EMSA for the −4692 site may have been caused by the different sensitivities of the ChIP and EMSA assays, or by the presence of a site that was covered by the primers used in the ChIP (covering 140 bp), but not by the EMSA assay with its smaller span (covering only 30 bp). It is noteworthy that the fetal pancreases analysed by ChIP showed the best conserved GLIS3RE among different species at −2718 to be the strongest among all elements tested. These data support the conclusion that Neurog3 is a direct downstream target of Glis3 in the transcriptional hierarchy of pancreas development.
Neurog3 gene expression is tightly regulated by a network of positive regulators such as PDX1 [28, 29, 30], HNF6 , SOX9 , FOXA2  and HNF1B [33, 34], as well as by negative regulators such as those in the Notch [11, 35, 36] and TGFβ [37, 38] signalling pathways. However, Glis3 does not seem to control Neurog3 expression through these positive and negative modulators of Neurog3 expression, because quantitative PCR showed similar expression levels of all these modulators in Glis3−/− and Glis3+/+ embryos.
In addition to directly controlling Neurog3 transcription, GLIS3 was shown here to physically interact with HNF6 and FOXA2 in co-precipitation experiments; it also synergistically activated Neurog3 promoter when co-expressed with each of the above transcription factors. Therefore, in addition to direct activation of Neurog3 transcription via GLIS3RE in the Neurog3 promoter, GLIS3 also functions in a combinatorial manner with other transcription factors, underscoring the multiple functions of GLIS3 within the regulatory network of Neurog3 gene expression.
In conclusion, we used genetic mouse models and in vitro experiments to define the molecular interactions and functions of GLIS3 in regulating the differentiation of pancreatic endocrine progenitors. We found that GLIS3 binds to the Neurog3 promoter and directly activates Neurog3 gene transcription; it also activates Neurog3 expression synergistically with two other Neurog3-activators, HNF6 and FOXA2. Therefore, loss of Glis3 function produces defective Neurog3 activation, which results in impaired fetal islet differentiation and neonatal diabetes.
We thank AK Rustgi (University of Pennsylvania, School of Medicine, Philadelphia, PA, USA) for PDCs, C. Newgard (Duke University, Durham, NC, USA) for 832/13 cells, C. Wright (Vanderbilt University, Nashville, TN, USA) for goat anti-PDX1 antibody; Developmental Studies Hybridoma Bank (University of Iowa, Iowa City, IA, USA) for anti-NEUROG3 and anti-NKX6-1 antibodies; W. Qian, A. Liang and Integrated Microscopy Core (Baylor College of Medicine) for technical support. This research was supported by the US National Institutes of Health (NIH) grant DK-68037 (to L. Chan), the Diabetes and Endocrinology Research Center (P30DK079638), the Betty Rutherford Chair from St Luke’s Episcopal Hospital and the T.T. & W.F. Chao Foundation.
YY, BH-JC and LC designed the study, analysed and interpreted the data, drafted and revised the manuscript. VY, WC and MJT participated in the study design, analysed and interpreted the data, and revised the manuscript. LL participated in the study design, analysed the data, and revised the manuscript. All authors have approved the final version of the manuscript.
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.