Role of nutrient-driven O-GlcNAc-post-translational modification in pancreatic exocrine and endocrine islet development

O-GlcNAc Transferase (Ogt) is the sole enzyme capable of adding an O-GlcNAc post-translational modification, known as O-GlcNAcylation, onto target proteins in response to nutrient flux. The nutrient-sensing hexosamine biosynthetic pathway (HBP), which is sensitive to nutrient flux of carbohydrates, amino acids and lipids, produces the substrate UDP-GlcNAc used by Ogt. Through O-GlcNAcylation, Ogt is uniquely poised to orchestrate cell signaling pathways that impact growth and survival in response to nutrient availability. Whole-body deletion of Ogt is embryonically lethal (Shafi et al., 2000). As a nutrient sensor, Ogt is expressed in all mammalian tissues but is most abundant in the pancreas (Lubas et al., 1997). Ogt is well positioned to regulate proper cellular function and development in a spatiotemporal manner (Harwood and Hanover, 2014), and many proteins essential for pancreas development are also potential Ogt targets. However, the role of O-GlcNAcylation during crucial developmental time windows, such as pancreatic organogenesis, is unexplored.

The pancreatic epithelium gives rise to all endocrine and exocrine cells (Mastracci and Sussel, 2012). The epithelial cells are exquisitely sensitive to nutrient signals and receive essential messages from the surrounding tissue (Edlund, 2002; Gittes et al., 1996; Kim and Hebrok, 2001; Murtaugh and Melton, 2003) and intrauterine environment (Elghazi et al., 2017; Filhoulaud et al., 2009; Guillemain et al., 2007; Mohan et al., 2018). Pancreas development is regulated by the hierarchy of transcriptional activation (Arda et al., 2013; Wilson et al., 2003), among which Pdx1 is required for the differentiation, survival and function of mature β-cells (Gannon et al., 2008; Johnson et al., 2003; Spaeth et al., 2017). Pdx1 and Ptf1a define the pool of cells that become pancreatic progenitors (Burlison et al., 2008; Yoshida et al., 2009). A subset of these progenitors go on to become endocrine progenitors, expressing Ngn3 and differentiating into endocrine islet cell types (Gu et al., 2002). Proteins crucial for exocrine and endocrine cell development and function, such as Pdx1 and Neurod1, are O-GlcNAc-modified by Ogt (Andrali et al., 2007; Gao et al., 2003; Özcan et al., 2010). O-GlcNAcylation of Pdx1 and Neurod1 has been correlated to DNA-binding activity/transcription and localization (Andrali et al., 2007; Gao et al., 2003; Kebede et al., 2012). Therefore, we hypothesized that O-GlcNAc modification of key developmental regulatory proteins is essential for pancreas development, based on the ability of Ogt to integrate signaling networks required for the survival of pancreatic progenitors in response to nutrient availability.

Here, we show clear evidence that Ogt is expressed in the embryonic pancreas, and by genetically manipulating Ogt in pancreatic epithelial progenitors we identified that Ogt is required for pancreas organogenesis. Through single cell transcriptome-based approaches, we reveal distinct cell types between Ogt-deficient and normal cells and show that Ogt regulates apoptosis in pancreatic progenitors via multiple protein targets, including Pdx1, Ptf1a and p53.

Several studies have previously demonstrated that Ogt expression is highly enriched in the pancreas (Akimoto et al., 1999; Lubas et al., 1997); however, it is not clear whether there is high O-GlcNAcylation level in the embryonic pancreas during development. Therefore, we assessed O-GlcNAcylation levels via the specific O-GlcNAc antibody RL2 in various embryonic tissues, including the pancreas from wild-type animals. We observed the highest level of O-GlcNAcylation in the head and pancreas, and the lowest level in the liver (Fig. S1A-B), consistent with previous studies (Lubas et al., 1997). Next, we sought to assess O-GlcNAcylation and total Ogt protein levels in male and female control embryonic pancreases. We detected strong Ogt protein expression and relatively equal levels of O-GlcNAcylation in both male and female control embryos (Fig. 1A,B).

Using genetic manipulation, we investigated the requirement of O-GlcNAcylation in pancreas development by deleting Ogt in the pancreatic epithelium of Pdx1Cre^Late mice, where it is expressed at embryonic day (E) 12.5 (Herrera, 2000). OgtKO^Panc newborns at postnatal day (P) 0, exhibited a lower bodyweight (Fig. 1C), which can be explained in part by reduced pancreas development. Insulin is a fetal growth factor (Milner and Hill, 1984); therefore, we assessed average litter size. We observed decreased litter size between P0 and P7 in OgtKO^Panc compared with the control (Fig. S1C). Bodyweight at E14.5 and E15.5 between OgtKO^Panc and control mice was comparable (Fig. S1D). At P0, OgtKO^Panc pancreas mass was diminished compared with the control (∼8.7 versus 4 mg, Fig. 1D). Approximately 93% of OgtKO^Panc pancreata (15 out of 16) were below the average pancreas weight of the control (Fig. 1D). One-time measurement of glucose at sacrifice showed a trend toward increased glucose levels in OgtKO^Panc mice at P0 (Fig. 1E). The nonsignificant change in random glucose levels can be partly explained by the requirement of at least an 80% reduction of β-cell mass to induce hyperglycemia (Leahy et al., 1984). It is also important to note that maternal blood glucose can contribute to fetal blood glucose at P0, and dams used were normoglycemic.

Homozygous deletion of Ogt resulted in either mild (Fig. 1F) or severe (Fig. 1G) pancreatic hypoplasia: Fig. 1F (right) illustrates mild hypoplasia versus near-complete pancreatic agenesis, as shown in Fig. 1G. This can be partly explained by the reported inherent mosaicism in Pdx1Cre recombination (Elghazi et al., 2017). Normal gross pancreas morphology was observed in OgtHET^Panc (Fig. 1G), suggesting that one allele of Ogt was sufficient to maintain pancreas mass. By using CAG-EGFP as a reporter of Cre activity, we observed endogenous EGFP expression as expected throughout the Cre⁺ pancreas of P0 and P60 control mice. No EGFP was observed in control mice without the Cre reporter (Fig. S1E), but only in Cre⁺ mice (Fig. S1F). Within the pancreas at P0, EGFP was expressed in ∼70% of cells in control and ∼15% of cells in the OgtKO^Panc (Fig. 1H). Adult male P60 control pancreata showed EGFP expression at ∼64% (both exocrine and endocrine compartments) but, remarkably, OgtKO^Panc mice only showed ∼7% (Fig. 1I,J). Within the islets, EGFP expression was observed (Fig. 1K,L) to be comparable between control and OgtKO^Panc mice (Fig. S1G). In summary, these data support previous studies showing that PdxCre^Late is intrinsically mosaic (Elghazi et al., 2017), thus resulting in cells that have escaped recombination (EGFP-negative cells) in both control and OgtKO^Panc mice. Thus, the mosaicism, in this case, allowed us the opportunity to study the pancreas of OgtKO^Panc mice. We speculate that using a more-efficient Cre line would have resulted in full pancreas agenesis in the OgtKO^Panc mice.

We next focused on characterizing the impact of Ogt deletion in the formation of β-cells, α-cells and acinar cells at birth (P0). Morphometric analyses of five different regions of neonatal pancreata at P0 showed significant reductions in both endocrine and exocrine mass (Fig. 2A-D). A significant reduction in β-cell mass was evident in P0 OgtKO^Panc mice (Fig. 2A,B,D,E). The deletion of one Ogt allele (heterozygous, OgtHET^Panc) also resulted in a significant decrease in the ratio of β-cells to whole-pancreas area (Fig. S2A), suggesting one allele of Ogt was sufficient to alter β-cell development. Moreover, the number of cells expressing both insulin and Pdx1 was reduced (Fig. 2A,B). α-Cell mass was diminished in OgtKO^Panc mice (Fig. 2C,D,F). Differentiated acinar tissue, assessed by amylase staining (Fig. 2C,D), was reduced (Fig. 2G). The overall mass of synaptophysin-expressing endocrine cells was also diminished (Fig. 2H-J), suggesting that all types of endocrine cells were affected. After Hematoxylin and Eosin (H&E) staining, the acini appeared to be grossly comparable with the control (Fig. S2B, arrow). After pancreas development, adult P60 OgtKO^Panc mice displayed reduced pancreas mass and β-cell mass (Fig. S2C,D). The decrease in both endocrine and exocrine mass indicated a reduction in the pancreatic progenitor population.

To understand the mechanism of pancreatic hypoplasia in OgtKO^Panc, we subsequently investigated E15.5 pancreata of OgtKO^Panc and littermate control mice. The estimated number of pancreatic cells showed a nonsignificant trend toward reduction in the OgtKO^Panc mice (Fig. 3A,B). No difference in the total number of mesenchymal cells was observed between the OgtKO^Panc mice and controls (Fig. 3A,C) The absolute number of cells labeling positive for E-cadherin (E-Cad⁺) was reduced in OgtKO^Panc compared with control mice (Fig. 3A,D). However, when the number of E-Cad⁺ cells was corrected to the number of DAPI⁺ cells, there was a decreasing trend only in the OgtKO^Panc mice (Fig. 3E), suggesting that the proportion of E-Cad⁺ to DAPI⁺ cells, or pancreas area, was not significantly altered. Western blotting for E-Cad protein itself also yielded a nonsignificant decreasing trend (Fig. 3F,G).

The estimated number of pancreatic progenitors (Pdx1⁺) during development is thought to determine the ultimate size of the pancreas (Stanger et al., 2007). Thus, we assessed the absolute number of Pdx1⁺ cells that could account for the smaller pancreas in OgtKO^Panc. Indeed, the absolute number of Pdx1⁺ cells was significantly reduced in OgtKO^Panc (Fig. 3H). Together, these data suggest that Ogt is required for the survival of Pdx1⁺ cells.

We next investigated the mechanisms underpinning the loss of Pdx1⁺ cells post Ogt deletion at E12.5. Ogt is known to regulate the survival of multiple cell types, and ablation of Ogt in β-cells reduces prosurvival Akt/Pdx1 signaling (Alejandro et al., 2015). First, we performed immunofluorescence staining for Pdx1 protein. As shown in Fig. 4A, we observed EGFP⁺ cells lacking Pdx1 protein (thin arrow) and some non-EGFP-expressing cells (thick arrow, wild-type Ogt-expressing cells) showing nuclear Pdx1 staining. These data suggest a reduction of Pdx1 protein in Ogt-deficient cells. We assessed total Pdx1 protein levels by western blotting in whole pancreas. Despite containing a mixture of cells with or without Ogt, Pdx1 protein level was significantly reduced (Fig. 4B,C, P=0.05) in E15.5 OgtKO^Panc pancreas. These data suggest that the reduction of Pdx1 protein contributed to the hypoplasia phenotype in the OgtKO^Panc pancreas.

We assessed Ogt protein level, as well as Ogt activity, by assessing levels of global O-GlcNAc modification using the specific antibody RL2 in whole pancreas at E15.5. We observed a reduction of Ogt protein in OgtKO^Panc mice (Fig. 4D, P=0.05). A mild, but significant reduction of global O-GlcNAc modification using RL2, a surrogate for Ogt activity, was also observed (Fig. 4E, P=0.035). These data suggest that a significant reduction in global O-GlcNAc modification and decreased Pdx1 protein levels contributed to the increased cell death in Pdx1⁺ cells.

Pdx1 protein has been previously shown to regulate islet cell survival (Johnson et al., 2003), and Ogt loss was associated with increased apoptosis in β-cells (Alejandro et al., 2015). Next, we investigated the decreased number of Pdx1⁺ cells in OgtKO^Panc by measuring apoptosis in pancreatic epithelial cells (EGFP⁺ cells). We found a significant increase in apoptosis in the pancreatic epithelial cells in the OgtKO^Panc mice (Fig. 4F,G). Within the E15.5 pancreas, the proportion of cells dying was increased in EGFP⁺ compared with EGFP⁻ (non Cre-expressing) cells, suggesting that cells lacking Ogt were more prone to apoptosis (Fig. 4H). After pancreas development, we determined that pancreas and β-cell mass were reduced but apoptosis rate was unchanged in β-cells and non-β-cells of P60 OgtKO^Panc mice compared with littermate controls (Fig. S2E-H). We conclude that increased cell death in pancreatic epithelium cells was one of the causes of severe hypoplasia in OgtKO^Panc animals.

To assess cell population clustering, we conducted single-cell RNA sequencing in the E15.5 pancreas. Owing to the mosaic expression of EGFP in the OgtKO^Panc mice, we sorted live cells using fluorescent EGFP in OgtKO^Panc mice and E-cad-EGFP in littermate control mice to properly account for equal age. As we have shown previously (Gregg et al., 2014), in the E15.5 pancreas, E-cad and Pdx1 are both expressed in pancreatic epithelial cells (Fig. 3A) and both colocalize with EGFP (Cre-reporter of Pdx1⁺, Fig. S3A). Zero inflated factor analysis (ZIFA), a dimensionality-reduction method for single-cell data that improves modeling accuracy in biological data sets (Pierson and Yau, 2015), was employed to categorize cells in OgtKO^Panc and control pancreata. A nontransformed raw count table of these cells was inputted to Seurat for unsupervised cell-subtype analysis between control and OgtKO^Panc (Fig. 5A). We identified seven distinct clusters of cells among the OgtKO^Panc and control cells (Fig. 5A). Cluster identity was inferred using known marker genes, which are the top three (differentially expressed genes, DEG) per nominal identity reported by Krentz et al. in the E15.5 pancreas (Krentz et al., 2018). Analysis of (marker) gene expression within each cluster was represented in a dot plot (Fig. 5B). Interestingly, some level of Ins1, Ins2 and Gcg expression appeared to show in multiple populations. As expected, the β/endocrine cell population showed high levels of Ins2 (Fig. 5B). Looking at our data more closely, for Ins2, we noted 416 out of 579 cells have a unique molecular identifier (UMI) count of 0, 105 out of 579 cells have a UMI count of 1, and 58 cells have a UMI count greater than 1, which suggests that ∼10% of cells have high insulin levels. Likewise, Ggc expression appeared in multiple populations but was strongest in the endocrine population. For Ggc, 553/579 have a UMI count of 0 and 26/579 have a UMI count equal to or greater than 1, which was ∼4.5% of cells tested. The expression of Ggc in multiple populations has also been shown by Byrnes et al. (2018). We examined the expression of an additional set of genes (Amy1, Arx, Pax6, Neurod1 and Sst; Fig. S4B-F). As one would expect, the expression of these genes was more restricted to specific cell populations. For example, Amy1 was highly expressed in acinar cells (Fig. S4B). Arx, Pax6 and Neurod1 (Fig. S4C-E) were expressed highly in endocrine cells. Not surprisingly, Sst was expressed only in a few cells (Fig. S4F). As expected in a continuum of cell states, specific clustering of specific genes was observed with some overlapping. Together, these data suggest that at the single-cell level, there was differential population clustering in OgtKO^Panc and control cells.

We identified gene-expression patterns in OgtKO^Panc and control cells (Fig. 5C; Table S1; Table S2). As shown in Fig. 5C, Neurog3- or Ins and Pdx1-expressing cells were reduced in OgtKO^Panc compared with the control cells. The number of Pdx1- or Ins and Spp1-expressing cells were also reduced in OgtKO^Panc compared with the control cells. However, Spp1 and Ghrl-expressing cells were increased in OgtKO^Panc compared with control cells (Fig. 5C). We speculated an increase in the total number of Ghrl-expressing cells at P0, but they were significantly reduced (Fig. 5D,E).

We determined 785 DEGs associated with Ogt loss (Table S3). Based on ingenuity pathway analysis (IPA), the top molecular and cellular functions were cellular movement, cellular function and maintenance, cell death and survival, and cellular development (Fig. S5A). The top canonical pathways that emerged also included Fc receptor-mediated phagocytosis in macrophages and monocytes, production of nitric oxide and reactive oxygen species in macrophages, and phagosome formation (Fig. S5A). As O-GlcNAcylation is linked to transcriptional regulation (Hanover et al., 2012), we used our list of DEGs to predict upstream regulators (UPRs) that could be targeted for Ogt-mediated post-translational modifications. UPRs are predicted to govern the identified set of DEGs. To address the limitation of the scRNAseq as a ‘non-deep sequencing analysis’, for rigor, we overlapped UPRs identified from two independent studies (Lockridge et al., 2020) of deep RNA sequencing using islets of P30 male or female OgtKO^Ins mice (Fig. S5B). We identified 80 common UPRs (Fig. S6C). Some of them, including p53 and Pdx1, have been previously suggested to be O-GlcNAc-modified (Yang et al., 2006; Gao et al., 2003). To show that a protein is O-GlcNAc-modified, the most widely used method is immunoprecipitation (IP) of the protein of interest with the O-GlcNAc-specific antibody RL2. Owing to the large requirement of lysates for IP, we used MIN6 cells and confirmed that p53 was O-GlcNAc-modified via the RL2 antibody in β-cells (Fig. 6A). An immunoblot of isolated primary acini showed a novel interaction between Ptf1a and RL2, providing the first evidence that Ptf1a is O-GlcNAc-modified by Ogt in acini. As shown in Fig. 6B, Ptf1a was successfully blotted in RL2 pull-down lysates. In a complementary experiment, we observed an RL2 signal at 50 kDa (the expected size for Ptf1a) in Ptf1a pull-down lysates (Fig. 6C), suggesting that Ptf1a is O-GlcNAc-modified. We confirmed that Pdx1 was glycosylated, as demonstrated by sWGA IP pulldown (Fig. 6D), and subsequently showed specific O-GlcNAc-modification with RL2 IP pulldown (Fig. 6E). Moreover, we demonstrated the inhibition of OGT by using a small molecule inhibitor Osmi-1; a significant reduction of RL2 was observed in MIN6 cells treated with 40 µM of Osmi-1 for 8 h (Fig. 6F,G). Next, we showed reduced Pdx1 O-GlcNAc-modification when OGT was significantly blocked with 40 μM Osmi-1 (Fig. 6F-H). Together, we conclude that O-GlcNAcylation plays an important role in pancreas development and provides a mechanistic link between external nutrient stimuli and pancreas development.

In this study, we demonstrate the first instance of genetic ablation of Ogt in the pancreatic epithelium (Pdx1⁺ cells) in vivo and show that Ogt is required for both pancreas and endocrine cell development. These findings are consistent with studies carried outwith the rat pancreas in vitro, where HBP activators increased β-cell development and inhibitors (i.e. Ogt small-molecule inhibitor) repressed β-cell development (Filhoulaud et al., 2009). These data underscore the importance of nutrient flux-mediated O-GlcNAcylation in the regulation of pancreas growth and support the role of external stimuli (i.e. nutrients from the intrauterine environment) in the regulation of organ development.

In this study, we show moderate to severe loss of exocrine and endocrine mass at P0 when Ogt is ablated on E12.5. The loss of acinar cells, β-, α- and ε-cells, and likely pancreatic polypeptide (PP) and δ-cells (based on the reduced number of cells expressing synaptophysin) in the P0 OgtKO^Panc pancreas suggests that Ogt regulates pancreas development. A reduction of GFP-label in acinar and endocrine cells in OgtKO^Panc suggests that Ogt is important in both the exocrine and endocrine compartment of the pancreas. Thus, future studies may be directed to specifically ablate Ogt in acini tissue. Ablation of Ogt specifically in insulin-positive cells (OgtKO^Ins) results in normal β-cell mass at P30 or P60, suggesting that OGT is not required for β-cell development and mass, and thus points to a temporal regulatory role for Ogt in β-cell mass (Alejandro et al., 2015). The temporal and regulatory role of Ogt in pancreas development may arise in part because of the specific protein targets of Ogt (Özcan et al., 2010) that might play a relevant role in the development and survival of pancreatic epithelium cells at a specific crucial window of development.

The increase in apoptosis in the pancreatic epithelium can impact final pancreas mass and endocrine islet cell number. Stanger et al. (2007) demonstrated that the size of the murine pancreas is limited by the total number of Pdx1⁺ progenitor cells. The general reduction of endocrine cells and acinar tissue in OgtKO^Panc mice supports the notion of reduced survival of pancreatic epithelium cells or Pdx1⁺ progenitor cells. This is further supported by our data showing reduced pancreas weight and β-cell mass in adult OgtKO^Panc mice after pancreas development. Hyper-O-GlcNAcylation is a general feature of evading apoptosis in cancer cells (Ferrer et al., 2014), and Ogt loss has been reported to regulate apoptosis in various cell types, including neurons (Wang et al., 2016) and intestinal epithelial cells (Zhao et al., 2018). Ogt deletion in pancreatic β-cells causes diabetes with remarked reduction in β-cell mass associated with increased apoptosis at P90 (Alejandro et al., 2015; Ida et al., 2017). Adult hyperglycemic OgtKO^Ins mice display reduced β-cell mass and an increased rate of apoptosis in part because of increased endoplasmic reticulum stress, and reduced prosurvival Akt and Pdx1 protein levels (Alejandro et al., 2015). Ida et al. (2017) also deleted Ogt in mature β-cells using the Pdx1ERTM mice that express Cre in pancreatic α-cells and β-cells, and the hypothalamus (Wicksteed et al., 2010). Likewise, adult OgtKO^PdxERTM mice display hyperglycemia and β-cell loss associated with apoptosis (Ida et al., 2017). However, deletion of Ogt in mature β-cell (inducible OgtKO^Ins), using the β-cellspecific MIP CreERT, lead to normal β-cell mass (Alejandro et al., 2015) and normoglycemia in vivo. Therefore, hyperglycemia may contribute independently to the increased apoptosis in the OgtKO^Ins (Alejandro et al., 2015) and OgtKO^PdxERTM mice (Ida et al., 2017). In the current study, adult OgtKO^Panc mice show reduced pancreas weight and β-mass not associated with increased apoptosis rate. These data are consistent with our previous findings that Ogt loss does not induce cell death once β-cells are mature (Alejandro et al., 2015). Ogt deletion in α-cells does not appear to impair glucagon levels, suggesting that α-cell mass may not be altered (Ida et al., 2017).

In the current paper, we identified that one mechanism that can contribute to increased apoptosis in pancreatic epithelium cells is the loss of Pdx1 protein. Indeed, full deletion of Pdx1 causes pancreas hypoplasia (Jonsson et al., 1994) and partial ablation increases β-cell apoptosis in mice (Ahlgren et al., 1998; Johnson et al., 2003). Pdx1 is also important in exocrine tissue development and acinar maturation (Kodama et al., 2016). OgtKO^Panc mice with severe phenotypes phenocopy the Pdx1 null mice (Ahlgren et al., 1998; Jonsson et al., 1994). Interestingly, deletion of either Ogt or Pdx1 in β-cells does not prevent the development of insulin-producing cells (Ahlgren et al., 1998; Alejandro et al., 2015), but is required for the maintenance of β-cell identity in adulthood (Ahlgren et al., 1998). Thus, temporal O-GlcNacylation of Pdx1 may have different biological outcomes. Although Pdx1 is post-translationally modified by other modalities, such as phosphorylation, the principal phosphorylation site of Pdx1 at Serine 61 was determined to be non-essential for Pdx1 function in pancreas development (Frogne et al., 2012). Future studies can be directed toward the identification of O-GlcNacylation sites on Pdx1, and subsequent mutation studies can assess functional impact on pancreas development. Moreover, additional targets should be identified, as it is likely that multiple proteins are involved downstream of Ogt.

As an initial step to identify potential Ogt targets, we turned to our single cell and bulk (islet) RNAseq datasets. The scRNAseq analysis revealed distinct populations of OgtKO^Panc cells, and the identified altered pathways, based on differentially expressed genes (DEGs), support the notion that Ogt regulates cell survival. From the list of DEGs, IPA identified upstream regulators (UPRs) and a significant number of these proteins (e.g. p53, Pdx1, Bax and Bak1) are known regulators of cell survival. In the scRNAseq data, the reduction of cell types highly expressing Ngn3, and Ins1/2 and Pdx1 further support the reduced number of differentiated endocrine cells (α, β, ε) observed at P0 in OgtKO^Panc mice. As mentioned above, Ogt does not appear to be required for α or β-cell development post-expression of Ins1/2 (Alejandro et al., 2015) or Gcg (Ida et al., 2017 ). Thus, future studies involving time-course lineage tracing in mice lacking Ogt earlier than E12.5 will be relevant to assess the impact of Ogt in the expansion of multipotent pancreatic progenitors at the end of the primary transition of pancreas development (around E11.5). The robust list of UPR generated from the RNAseq and bulk RNAseq provides potential Ogt targets to pursue further. On this list, we validated survival proteins Pdx1 and p53. Additionally, we showed that Ptf1a is O-GlcNAc-modified in acinar tissue. Future studies will be directed toward identifying the cellular consequences of this modification on Ptfa1 and additional Ogt targets, and performing mechanistic studies to show causal effect.

We have previously reported a role for Ogt in glucose homeostasis and β-cell function using constitutive or inducible β-cell-specific OgtKO^Ins male and female mice (Alejandro et al., 2015), in which Ogt levels in the hypothalamus were not altered. In this study, we do not present our glucose homeostasis data on adult OgtKO^Panc mice. These mice develop glucose intolerance and hyperglycemia but the extra-pancreatic and mosaic expression of Pdx1Cre complicates the interpretation of the data (Wicksteed et al., 2010). Therefore, we focused on the outcome of genetically ablating Ogt in E12.5 and provided the first model showing valuable first insights in to the role of Ogt in pancreas development. In summary, we provide the first evidence that O-GlcNAc modifications of proteins are essential for pancreatic progenitor cell survival, and thus pancreas development. These findings underscore the requirement of O-GlcNAcylation during pancreas development and provides a mechanistic link between external nutrient stimuli and pancreas development.

Late Pdx1Cre line (L-Pdx1Cre) and RIP-Cre mice were provided by Dr Pedro Herrera (University of Geneva, Switzerland), and Ogt^flox/flox (Alejandro et al., 2015) and CAG-EGFP reporter transgenic animals were purchased from the Jackson Laboratory. Pdx1Cre; Ogt^flox/flox or Pdx1Cre; Ogt^flox/y, herein referred to as OgtKO^Panc, Pdx1Cre; Ogt^flox/x or OgtHET^Panc, and littermates Pdx1Cre;Ogt^wt or Ogt^flox/flox as controls. All mice were generated on a C57Bl/6J background and group-housed on a 14:10 light-dark cycle. All procedures were performed in accordance with the University of Minnesota Institutional Animal Care and Use Committee (protocol number 1806-36072A).

Litter size was determined by counting the number of pups present at pancreas harvest on E15.5, E17.5, P0 and P7. For timed pregnancy, the morning of vaginal plug was taken to be E0.5. Dissected pancreases were stained blue (Mark-It, Thermo Fisher Scientific) for 10 s prior to fixation in 4% paraformaldehyde for 30 min then encapsulated in specimen processing histogel for 10 min followed by overnight incubation in 70% ethanol (1× PBS) and immediate processing/embedding. Sex as a biological factor was not considered in the morphological analysis of E15.5 and P0 OgtKO^Panc pancreases or in our RNAseq analysis. OgtKO^Ins mice do not reveal any sexual dimorphism phenotype (Alejandro et al., 2015).

For adult paraffin-embedded pancreases, 5 μm tissue sections from five different regions spaced 200 μm apart were prepared. Neonatal tissue sections (5 μm) were collected from top to bottom of tissue. Pancreases analyzed were randomly selected. Deparaffination and tissue preparation for immunofluorescence imaging were performed as described previously (Akhaphong et al., 2018). Tissues were treated with primary antibodies against guinea pig insulin, mouse glucagon, rabbit amylase, mouse ghrelin, rabbit synaptophysin, mouse E-cadherin and rabbit Pdx1, and incubated at 4°C for 12-16 h. Secondary antibodies conjugated to FITC, Cy3 or AMCA, were used. TUNEL staining was completed using the Millipore ApopTag Red in situ Apoptosis Detection Kit following the manufacturer's protocol. See Table S4 for additional information.

All morphometric analyses and cell counts were conducted with ImageJ2 (Rueden et al., 2017) as previously described (Akhaphong et al., 2018). Cell numbers (∼4000 per animal) for E-cadherin and Pdx1 were counted manually and labeled with numbers using the Multi-point function in ImageJ2. E-cad-/DAPI⁺ or Pdx1-/DAPI⁺ cells were counted as mesenchymal cells. Fluorescent images for endocrine/exocrine cell mass were captured at 4× or 10× magnification, and individual islet pictures at 20× and 40× magnification, with a Nikon Eclipse NI-E microscope equipped with a motorized stage. Cell types and areas were identified by hormone-positive immunofluoresence staining and normalized to total pancreatic area, as determined by tissue morphology. For cell mass calculations, the average of five regions (∼50 μM apart for newborns, and 200 μM apart for adults) of area ratio per pancreas (e.g. insulin⁺ area /pancreas area, glucagon⁺ area/pancreas area and ghrelin⁺/pancreas, for β-cell, α-cell and ε-cell mass, respectively, or amylase⁺/pancreas for exocrine cell mass) was multiplied by pancreas weight (mg). For apoptotic analysis, TUNEL⁺ cell nuclei were normalized to the number of E-cadherin or GFP⁺ cells. Total cell counts were averaged across 3-5 histological sections per animal as described previously (Akhaphong et al., 2018). For Fig. 2C,D, raw images did not produce the same canvas size, even though they were imaged at the same magnification. Using Fiji, we defined the canvas size as length/width in μm and adjusted both images to equal the same canvas size. Any extra space developed by increasing the canvas size was filled or ‘tiled’ in black.

The percentage efficiency of Cre was determined in the control or OgtKO^Panc mice (P0 or P60) based on endogenous GFP signal in prepared histological sections. Using Fiji, the area of GFP⁺ pancreas was normalized to total pancreatic area (background).

Pancreatic buds were isolated on E15.5 and digested in DMEM media with 0.05% trypsin in 0.5 mM EDTA at 37°C for 5 min. Cells were dissociated into a single-cell suspension in 2% fetal bovine serum in PBS on ice. Cell suspensions of each type were pooled based on control or OgtKO^Panc from the same dam to have equivalent developmental time points. Control cells were incubated with FITC anti-E-cadherin for 1 h. Cells were washed then sorted for GFP or FITC positivity on a BD FACSAria II, ‘barcoded’ in a 10× Chromium instrument, and cDNA from each was loaded into an Illumina HiSeq 2500 for sequencing.

Single-cell RNA sequencing was carried out at the University of Minnesota Genomic Center. Briefly, raw sequencing images from the Illumina HiSeq 2500 were processed through CellRanger v. 2.0.0. For each sample, three fastq files (forward R1 reads, reverse R2 reads and barcode/index reads) and a raw gene UMI count table were produced. R (v. 3.3.3) was used to read in the raw counts per gene of all samples into a data matrix as per the 10× Chromium manufacturer's protocols. The count matrix was log transformed with base 2 and prior count of 1, as required by ZIFA (Pierson and Yau, 2015), which implemented a factor analysis to separate cell groups. Python function DBSCAN, from package scikit-learn (Pedregosa et al., 2011), was used to assign each cell to one of the following four groups: Ctrl-main, Ctrl-minor, KO-main and NA. A nontransformed raw count table of pancreatic progenitor cells was inputted into Seurat (Satija et al., 2015) for further cell-subtype analysis and data visualization. Subsequently, two methods were used to test for differential gene expression: the beta-Poisson model of analysis (Vu et al., 2016); and a method that grouped single cells into six ‘pseudo’-samples – three for KO and three for control. Then DESeq2 (Love et al., 2014) was used to analyze the ‘pseudo’ groups as standard ‘bulk’ RNAseq data. Genes that showed at least a 1.5-fold change in expression (KO versus control), with a raw P-value of ≤0.05 in all the tests, were considered as differentially expressed and collated. This list of genes was later imported into IPA to identify enriched pathways and potential upstream regulators that are responsive to Ogt deletion. Curated marker genes were evaluated for their expression in each cell. The cell type was determined by the gene(s) with the highest expression if two or three genes were simultaneously expressed at a similar level.

For lectin-based precipitation studies, the lysates were incubated with succinylated wheat germ agglutinin (sWGA) agarose beads as previously described (Jo et al., 2019). MIN6 cells (a gift from Dr Peter Arvan, University of Michigan, MI, USA) were authenticated to secrete insulin in response to glucose and tested negative for mycoplasma. For western blotting, we probed with antibodies against Pdx1, E-Cad, RL2,Ogt, p53 and vinculin, and with a pan-O-GlcNAc antibody. RL2 and Pdx1 antibodies were validated in our laboratory (Jo et al., 2019). See Table S4 for more information, such as antibody dilutions. Densitometry analysis was performed with ImageJ2.

All values are expressed as mean±s.e.m. Analyses were conducted using GraphPad Prism (v.7.0d) using an unpaired two-tailed Student's t-test. Results were deemed significant when P≤0.05.

We thank M. Gannon, C. Cras-Méneur and M. Doyle for feedback and discussion, and C. Weaver for assistance with pancreas morphological assessment. Dr Alejandro is the guarantor of this work