Mesenchymal Hox6 function is required for mouse pancreatic endocrine cell differentiation

Hox genes play a well-established role in axial and appendicular skeletal patterning, and knowledge of their importance in organogenesis is expanding. Hox genes have important roles in the development of organs that involve Hox expression along the anterior-posterior axis. Examples include the Hox3 paralogous group genes, which are crucial for thymus, thyroid and parathyroid development; Hox5 genes in lung development; Hox9, Hox10 and Hox11 genes in the reproductive tract; Hox10 and Hox11 genes in kidney development; and Hoxb13 for prostate development (Dolle et al., 1991; Benson et al., 1996; Gendron et al., 1997; Taylor et al., 1997; Manley and Capecchi, 1998; Podlasek et al., 1999; Economides and Capecchi, 2003; Schwab et al., 2006; Yallowitz et al., 2011; Boucherat et al., 2013; Raines et al., 2013; Hrycaj et al., 2015).

The Hox6 paralogous group includes three genes: Hoxa6, Hoxb6 and Hoxc6. Paralogous Hox genes have been shown to exhibit a high degree of functional redundancy owing to sequence similarity and significant overlap in expression. Loss of a single gene within a paralogous group often results in little to no observable phenotype, but disruption of all members of a given paralogous group results in dramatic patterning phenotypes (Davis and Capecchi, 1994; Horan et al., 1995; Fromental-Ramain et al., 1996a,b; Manley and Capecchi, 1998; Rossel and Capecchi, 1999; van den Akker et al., 2001; Wellik et al., 2002; Wellik and Capecchi, 2003; McIntyre et al., 2007; Yallowitz et al., 2009, 2011; Xu and Wellik, 2011; Xu et al., 2013). Single-mutant animals for each of the Hox6 paralogs have undetectable or very mild defects, but collectively this group has been demonstrated to play important roles in patterning the rib cage and in determination of neuronal cell fate (Kostic and Capecchi, 1994; Rancourt et al., 1995; McIntyre et al., 2007; Mallo et al., 2010; Lacombe et al., 2013). Hox6 genes are also expressed in the pancreatic mesoderm, suggesting a possible role for Hox6 in pancreatic organogenesis.

In the mouse, the pancreas is specified at approximately embryonic day (E) 9.5 and expands as a dorsal and ventral bud from the endoderm-derived gut tube into the surrounding mesoderm (Wells and Melton, 2000; Gittes, 2009; Puri and Hebrok, 2010). At E11.5, these buds fuse to become the dorsal and ventral regions of the single pancreas. The pancreas has two main components: an exocrine and an endocrine component. The exocrine component is composed of digestive enzyme-secreting acinar cells and ductal cells. Ductal cells form a complex branching network that transports the digestive enzymes into the small intestine. The endocrine component is composed of five distinct types of endocrine cells, each of which secretes a single hormone: insulin, glucagon, somatostatin, ghrelin or pancreatic polypeptide. Endocrine cell differentiation is initiated in the ductal epithelium by the expression of Ngn3 (Neurog3 – Mouse Genome Informatics) in a subpopulation of the epithelial cells (Gu et al., 2002). These Ngn3⁺ cells subsequently delaminate from the ductal epithelium into the surrounding mesenchyme and differentiate further to the specific endocrine lineages. The bulk of endocrine cell differentiation occurs from E12.5 to E15.5, which is termed the secondary transition. Although all of the major functional components of the pancreas are derived from the endoderm, the surrounding mesodermally derived mesenchyme is crucial for the growth and development of these cell types (Golosow and Grobstein, 1962; Gittes et al., 1996; Miralles et al., 1998; Bhushan et al., 2001; Attali et al., 2007; Gittes, 2009; Puri and Hebrok, 2010).

Explant studies were first used to demonstrate the importance of the mesenchyme. When pancreatic epithelium was cultured in the absence of its surrounding mesenchyme, both endocrine and exocrine development arrested with defects in growth and differentiation. These defects were rescued by recombination with pancreatic mesenchyme (Golosow and Grobstein, 1962; Wessells and Cohen, 1967). A more recent study, in which the pancreatic mesenchyme was genetically ablated in vivo, showed a similar failure of all components of the pancreas to develop (Landsman et al., 2011).

Wnt signaling is crucial for multiple aspects of pancreatic development. A multitude of Wnt ligands, receptors, modifiers and inhibitors have reported expression in both the epithelium and the mesenchyme of the pancreas, with gene expression being highest early in development and declining with organ maturation (Heller et al., 2002). Wnt has largely been studied in the pancreas through manipulation of required canonical Wnt signaling components, such as β-catenin. Others have shown specific roles for individual Wnt ligands and, taken together, there are significant roles for both Wnt/β-catenin signaling and the non-canonical Wnt planar cell polarity pathway for the proper development of both endocrine and exocrine pancreas (Heller et al., 2002; Kim et al., 2005; Murtaugh et al., 2005; Heiser et al., 2006; Attali et al., 2007; Wells et al., 2007; Jonckheere et al., 2008; Murtaugh, 2008; Baumgartner et al., 2014; Afelik et al., 2015).

Although Hox genes have been shown to be important for the development of many organs of endodermal and mesodermal origin, a role for Hox genes in development of the pancreas has not been reported. Herein, we report that mouse Hox6 genes function in pancreatic organogenesis. Hox6 genes are expressed only in the mesoderm of the developing pancreas and not in the endoderm. The Hox6 mutant pancreas buds normally and Ngn3⁺ endocrine progenitors are specified, but there is a >90% reduction of mature endocrine cells in the Hox6 mutant pancreas compared with controls. Loss of Hox6 function results in a decrease in Wnt5a expression in the pancreatic mesenchyme (although epithelial Wnt5a expression is unperturbed). This leads to a subsequent loss of expression of two Wnt inhibitors, Sfrp3 (Frzb – Mouse Genome Informatics) and Dkk1, in endocrine progenitor cells. The addition of recombinant Wnt5a protein to pancreatic explant cultures is sufficient to rescue endocrine cell differentiation in Hox6 mutant pancreata. Thus, regional mesodermal patterning factors are crucial for establishing the mesenchymal-epithelial crosstalk required for proper endocrine cell differentiation in pancreatic development, highlighting the potential importance of considering the mesodermal niche in ex vivo β-cell differentiation protocols.

Hox6 aabbcc (lower case letters represent null alleles) mice do not survive postnatally, but mutant embryos are indistinguishable in appearance (Fig. S1A) and mass (Fig. S1B) from littermate controls. Examination of internal organ defects reveals somewhat abnormal pancreatic morphology in Hox6 mutants compared with controls (Fig. 1A). The ventral pancreas (black arrowheads) and trunk of the dorsal pancreas (yellow arrowheads) are more compact in Hox6 mutant pancreata compared with controls. Despite mildly perturbed pancreatic morphology, a normal expression pattern of acinar cell marker amylase (Amy) is observed in the mutant pancreas (Fig. 1B).

Early stages of pancreatic initiation were examined by immunofluorescent staining for the pancreatic epithelial marker Pdx1. Quantification of immunofluorescent staining of the entire early pancreas (dorsal and ventral bud) showed that both pancreatic buds initiate normally and the volume of Pdx1-positive epithelium is unchanged between the control and Hox6 mutant at E10.5 (Fig. S2A,B). A small decrease in epithelial volume was observed at E11.5 (Fig. S2A,D). Minor decreases in both epithelial and mesenchymal proliferation were measured at E11.5 and E14.5 (Fig. S2C,G), but no overall differences in mesenchyme volume were measured (Fig. S2E), and there was no overall change in pancreatic mass at E18.5 regardless of Hox6 genotype (Fig. S2H). There were no differences in apoptosis measured by cleaved caspase-3 staining (Fig. S2I).

Examination of morphological defects at E14.5 by staining with cadherin 1 (Ecad) reveals typical loose, lobular branching epithelium in the control pancreas, but more compact clusters of epithelial cells with less branching in the mutant pancreas (Fig. 1C). Branch pattern was further analyzed at E14.5 using Muc1 antibody staining in whole pancreata (Villasenor et al., 2010). Hox6 null pancreata exhibit impaired branching and apparent defects in the remodeling of the early ductal plexus (Fig. 1D). Many thin, single lumens have resolved throughout the control pancreas, whereas these lumens are less apparent in the mutant (Fig. 1D, red arrowheads). Bright, dense staining for Muc1, indicating acinar clusters, is found readily throughout both control and Hox6 null pancreata (Fig. 1D, white arrowheads).

Before E13.5, the ‘tip’ acinar cells of the pancreas are multipotent progenitor cells and can produce exocrine, ductal and endocrine cells. All epithelial cells at these early stages express Sox9 and Hnf1b. After E13.5, as Sox9 and Hnf1b are downregulated in tip cells, these cell types become unipotent and give rise only to acinar cells (Zhou et al., 2007; Kopp et al., 2011). In contrast to controls, Hox6 mutant pancreata demonstrate continued Sox9 and Hnf1b in Cpa1-positive tip cells at E14.5 (Fig. 1E,F). This has resolved by E16.5, consistent with the eventual differentiation and maturation of multipotent progenitor cells into mature exocrine acinar cells. This defect might be related to the perturbed morphology and branch pattern observed in mutants.

The mild defects in branching and in exocrine differentiation are in contrast to a dramatic reduction of all five mature endocrine cell hormones in Hox6 mutant pancreata (Fig. 2A,B). Antibody staining for insulin (Ins) and glucagon (Gcg) reveals a dramatic reduction of both endocrine protein and mRNA expression of all five endocrine hormones at E14.5 (Fig. 2A). This decrease is more pronounced by newborn stages, with Ins and Gcg mRNA expression reduced to less than 5% of control values (Fig. 2B).

All five endocrine cell types derive from Ngn3⁺ cells that arise from the ductal epithelium. During development of the pancreas, sporadic ductal cells express Ngn3, initiating a delamination process and allowing these cells to migrate into the surrounding mesenchyme, where they mature into the five types of hormone-producing cells. We investigated the cellular defects leading to the dramatic decrease of endocrine cells in the mutant by examining the initiation and differentiation of this cell type. There is no change in the amount of Ngn3⁺ immunofluorescent staining (Fig. 3A) or expression of Ngn3 mRNA by qRT-PCR (Fig. 3D) in the mutant pancreas compared with controls. There are also no measured changes in expression of Nkx6-1 or Nkx2-2 (Fig. 3D); however, these genes are expressed more broadly in the epithelium (Sussel et al., 1998; Schaffer et al., 2010). There are significant reductions of pan-endocrine markers Chga and Isl1 protein and mRNA at E14.5 (Fig. 3B-D). The mRNA expression of endocrine lineage genes Mafa, Mafb and Neurod1 are also significantly reduced (Fig. 3D). Despite the apparent loss of endocrine differentiation, Ngn3 staining in controls and mutants is comparable, suggesting that Ngn3⁺ progenitor cells do not accumulate in Hox6 mutants up to E18.5 (Fig. S3).

In the first phase of endocrine differentiation, sporadic Sox9⁺ ductal cells begin additionally to express Ngn3, allowing these cells to delaminate from the epithelium while concomitantly turning off Sox9 expression (Gouzi et al., 2011; Shih et al., 2012). We examined this process by antibody staining control and Hox6 mutant pancreata for Ngn3 and Sox9. Immunostaining reveals that about 25% of Ngn3⁺ cells are also Sox9⁺ in control pancreata at E14.5 (Fig. 3E, yellow arrows). This ratio is identical in the Hox6 mutant pancreas, providing evidence that endocrine progenitor cells are able to transition adequately from Sox9⁺/Ngn3⁺ ductal cells into Sox9^–/Ngn3⁺ early endocrine cells (Fig. 3E). Ngn3⁺ cells that have successfully migrated from the duct are observed throughout the control and in the Hox6 mutant pancreas (Fig. 3E, white arrows), further supporting a differentiation defect after specification and delamination.

All three Hox6 genes are expressed in the embryonic pancreas, with expression highest early in development and persisting until E16.5 (Fig. 4A). Expression was below our limit of detection for all three Hox6 genes at E18.5 (Fig. 4A). In situ hybridization (ISH) analyses of Hox6 mRNA reveal expression exclusively in the pancreatic mesoderm at E11.5 and E12.5 (Fig. 4B). Mesodermally restricted expression was confirmed using a previously reported Hoxb6-inducible Cre reporter line (Hoxb6CreER^T;Rosa26-tdTomato; Nguyen et al., 2009; Madisen et al., 2010). Administration of tamoxifen at early stages (E9.5 and E10.5) reveals an anterior expression limit at the level of the budding dorsal and ventral pancreas, and also marks the majority of mesoderm posterior to this in the embryo (Fig. 4C).

Expression from the ROSA locus in these mice is observed throughout the pancreatic mesenchyme, but is completely excluded from early Ins-positive endocrine cells, Pdx1-positive pancreatic endoderm and PECAM (Pecam1)-positive vasculature (Fig. 4C). As endocrine cells undergo epithelial-to-mesenchymal transition before differentiation, it was important to exclude possible initiation of Hox6 expression in endocrine cells during the secondary transition. To test this, tamoxifen was administered to pregnant dams at E12.5, E13.5 and E14.5 and embryos were examined at E15.5. We found no Cre activity in Ngn3⁺ endocrine progenitors (Fig. S4A), with the pan-endocrine marker Chga (Fig. S4B) or with epithelium (Fig. S4C). Similar results were obtained with tamoxifen administration daily from E10.5 to E17.5. There was no overlap of tdTomato with Chga-positive endocrine cells, pancreatic epithelium, endothelial, neuronal or smooth muscle cells (Fig. 4D; Fig. S4D,E). We observed extensive and complete co-labeling of tdTomato and vimentin-positive mesenchyme (Fig. 4D), confirming that Hox6 expression is exclusive to the pancreatic mesoderm-derived mesenchyme.

It has been reported previously that hypervascularization affects organogenesis of the pancreas by reducing branching and differentiation of epithelial cells, and signals from the vasculature are also important for the proper formation of endocrine cells of the pancreas (Lammert et al., 2001; Magenheim et al., 2011). Immunofluorescence staining for PECAM at E14.5 (Fig. S5A) and E18.5 (Fig. S5B) reveals no difference in vasculature between the control and mutant pancreas. There are also no differences in the association of endothelial and insulin-positive endocrine cells between control and Hox6 null E18.5 pancreas (Fig. S5C).

Microarray analysis was performed on E12.5 control and Hox6 mutant pancreata to probe for the molecular mechanism(s) responsible for the Hox6 mutant phenotypes (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE68390). ToppFun (https://toppgene.cchmc.org/enrichment.jsp) analysis identified changes in members of the Wnt receptor-signaling pathway as significantly enriched in the Hox6 mutant pancreas, but no significant changes in other major signaling pathways.

To confirm our microarray results, we used qRT-PCR analysis to measure the expression levels of genes from major signaling pathways with reported or measured expression in the pancreas. No changes in the levels of Fgf, Bmp or Notch pathway members were measured in E12.5 mutant pancreata (Fig. 5A). There were also no differences in the measured expression of Wnt receptors between control and Hox6 mutant pancreata (Fig. 5B). There are no differences in expression of Wnt2b, Wnt4, Wnt7b or Wnt11, other Wnt ligands expressed in pancreas at E12.5, between mutant and control pancreata (Fig. 5B; Heller et al., 2002), and we were unable to detect expression of Wnt1 and Wnt8b, genes reported to be expressed in the mesenchyme later in development, in either the control or the Hox6 mutant pancreas at E12.5 (data not shown; Heller et al., 2002). However, we observed significantly reduced expression levels of mesodermally expressed Wnt ligand, Wnt5a, at both E12.5 and E14.5 (Fig. 5B,C).

Hox6 genes are expressed solely in the mesoderm and therefore we examined mesodermally expressed Wnt genes as possible direct targets of Hox6. Wnt5a is expressed in both the pancreatic epithelium and mesenchyme (Heller et al., 2002). ISH analyses of Wnt5a revealed much weaker expression in the Hox6 mutant pancreatic mesenchyme at E12.5 compared with controls (Fig. 5D). At E14.5, reduced Wnt5a expression is even more apparent in the Hox6 mutant pancreatic mesenchyme compared with controls, whereas expression in the epithelium appears identical (Fig. 5E). At E14.5 and beyond, we measured a significant reduction of more Wnt receptors and ligands as well, suggesting a secondary perturbation of Wnt signaling more globally in the Hox6 mutant pancreas at later stages (Fig. 5C). Wnt2b, Wnt4, Wnt7b and Wnt11 were also examined by ISH; however, no obvious changes in expression pattern were detected between the control and Hox6 mutant pancreas (Fig. S6A-D).

We reasoned that this disruption in Wnt signaling in the mesoderm might lead directly to changes in the Ngn3-expressing endocrine precursors after delamination from the ductal epithelium. The expression of a number of Wnt inhibitors has been reported in the pancreatic epithelium and in endocrine cells throughout development, including Sfrp1-4, Dkk1-3 and Wif1 (Heller et al., 2002; Gu et al., 2004). Of the Wnt inhibitors examined in the pancreas, only two (Sfrp3 and Dkk1) showed significantly reduced expression levels in the Hox6 mutant at E12.5 (Fig. 6A). Using immunofluorescent staining with antibodies to Sfrp3 and Dkk1, we observe extensive overlap of Sfrp3 and Dkk1 with endocrine cells in wild-type pancreata (Fig. 6B). Overall, Hox6 mutant pancreata exhibited drastically reduced levels of Sfrp3 and Dkk1 (Fig. 6C,D). Close examination reveals colocalization of Sfrp3 and Dkk1 in a subset of Ngn3⁺ endocrine progenitor cells in the control and none in the Hox6 mutant pancreas (Fig. 6E,F). This reduction of Wnt inhibitors continues through later stages of pancreatic development (Fig. S7).

Wnt5a is the first mesodermally expressed Wnt ligand with disrupted expression in the Hox6 mutant pancreas. If the lack of endocrine cell differentiation in the Hox6 mutant pancreas stems from the lack of a sufficient amount of Wnt5a signaling from the mesoderm, we reasoned that it might be possible to rescue this phenotype with the addition of exogenous Wnt5a to the Hox6 mutant pancreata. To test this, we dissected control and Hox6 mutant pancreata at E12.5 and cultured them for 6 days in the presence or absence of exogenous Wnt5a (Fig. 7A). Without exogenous Wnt5a in the culture, we observed a significant reduction of mature endocrine cells in the Hox6 mutant pancreata compared with controls, similar to the in vivo phenotype (Fig. 7B,C). When recombinant Wnt5a protein was added to the culture media of Hox6 mutant pancreata, the quantity of mature endocrine hormone staining was rescued to the same level as control pancreata (Fig. 7B,C).

Although the importance of the mesenchyme in pancreatic development was demonstrated decades ago, the molecular mechanisms involved in the crosstalk between the epithelium and the mesenchyme are poorly understood. Here we report a novel role for Hox6 genes in development of the pancreas as illustrated in the model in Fig. 8. Loss of Hox6 function in the pancreatic mesoderm leads to mild defects in branching and delayed exocrine cell differentiation; however, the pancreas achieves its normal size by newborn stages with extensive acinar cell formation. Endocrine cells are specified normally, but immature endocrine cells do not mature, resulting in a dramatic reduction of all types of fully differentiated endocrine cells, including >90% decreases in insulin and glucagon expression.

Specific components of the Wnt signaling pathway are disrupted by E12.5, whereas no changes in other crucial developmental signaling pathways, including Fgf, Bmp and Notch, are observed. Mesenchymal ligand expression of Wnt5a is significantly downregulated (or absent), whereas epithelial expression of this ligand appears unperturbed. Wnt5a is expressed from E11 to the end of gestation, with peak levels of expression at E12, similar to our measured Hox6 gene expression profile (Heller et al., 2002). A study examining morpholino knockdown of wnt5a and fz2 in zebrafish and global loss of function of Wnt5a in mice reported defects in islet formation in both organisms (Kim et al., 2005), consistent with loss of Wnt5a expression in the mesoderm contributing to the endocrine phenotype in our Hox6 mutant animals.

In addition to loss of Wnt5a signaling in the pancreatic mesoderm, which might result directly from loss of Hox6 function, there is a subsequent loss of expression of Wnt inhibitors Sfrp3 and Dkk1, specifically in delaminated Ngn3⁺ endocrine precursor cells as they are entering the mesoderm. As previous studies have demonstrated the importance of repressing Wnt signaling in developing endocrine cells (Pedersen and Heller, 2005), it is likely that loss of Wnt inhibitor induction in Ngn3⁺ progenitors leads directly to loss of further endocrine cell differentiation in Hox6 mutant pancreata. Pharmacological treatment of Hox6 mutant pancreata with exogenous Wnt5a protein is sufficient to restore endocrine cell differentiation, and demonstrates that Wnt5a is a crucial mediator downstream of Hox6 genes in the pancreatic mesenchyme during development. Collectively, our results suggest that Hox6 genes are crucial for the establishment of the Wnt mesenchymal-epithelial crosstalk necessary for development of the pancreas and endocrine cell specification.

Hox genes are important for many aspects of development and organogenesis, but no disruption of Hox function has previously been implicated in the development of the pancreas. This work contributes to the growing understanding of pancreatic mesoderm signaling and the important roles that the mesoderm plays in the development of both the endocrine and the exocrine components of the pancreas. Here we show a direct link between Hox function and Wnt signaling; a theme that is reminiscent of a recent report of loss of Wnt2/2b expression in the mesenchyme of Hox5 triple-mutant lungs during development (Hrycaj et al., 2015). Another recent study examining Hoxd13 function in digit development reported that Hoxd13 promotes expression of Wnt5a in vitro (Kuss et al., 2014). All Hox proteins bind to the same (ATTA) binding sequence, and therefore it is plausible that Wnt5a is a direct target of Hox proteins during development. Future work will be required to establish possible direct regulation of Wnt5a by Hox6.

This study adds to growing evidence that Hox function in the mesoderm of several organ systems plays region-specific roles associated with the establishment of proper Wnt signaling crosstalk during organogenesis (Kuss et al., 2014; Hrycaj et al., 2015). A more complete elucidation of mesodermal-endodermal crosstalk during development of the pancreas is crucial to the enhancement of ex vivo protocols for generating functional β-cells as a cellular therapy for the treatment of diabetes.

Mice mutant for all three Hox6 paralogous genes were generated using standard genetic crosses (Kostic and Capecchi, 1994; Rancourt et al., 1995; Garcia-Gasca and Spyropoulos, 2000). Hoxb6CreER^T mice were contributed by Dr Susan Mackem (Nguyen et al., 2009). Rosa26-tdTomato mice were obtained from The Jackson Laboratory (Madisen et al., 2010). All experiments were performed following protocols approved by the University of Michigan's Institutional Committee on the Use and Care of Animals.

Tamoxifen and progesterone were dissolved in 100% ethanol and diluted in corn oil. Pregnant dams were given intraperitoneal injections with 1.5 mg of tamoxifen and 0.75 mg of progesterone on the days noted in the Results.

For section ISH, embryos were collected in PBS and fixed overnight in 4% PFA in PBS at 4°C. Embryos were then rinsed in PBS and immersed in 30% sucrose at 4°C overnight before embedding into optimal cutting temperature (OCT) media. Frozen sections 20 µm in thickness were cut, and slides were stored at −80°C. Section ISH was performed as previously described (Mendelsohn et al., 1999; Di Giacomo et al., 2006). Detection of Hox6 mRNA was done using probes generated against the 3′ untranslated region of Hoxa6, Hoxb6 and Hoxc6 or against the Neo^r cassette as previously described and with indistinguishable results from probes against Hox6 gene mRNA (McIntyre et al., 2007). Wnt5a cDNA was ligated into PCR4-TOPO vector and reverse transcribed with T3 RNA polymerase. The sequenced plasmid aligns to mouse Wnt5a: GenBank gi 46909566/NM 009524.2 from bp 406 to 1440. Wnt2b, Wnt4, Wnt7b and Wnt11 ISH was performed with previously published riboprobes (Miller et al., 2012; Soofi et al., 2012; Ranghini and Dressler, 2015).

Mouse embryos were collected as described above. Frozen sections 12 µm in thickness were cut, and slides were stored at −80°C. Slides were blocked for 1 h at room temperature in 0.1% or 0.5% Triton X-100 in PBS (PBS-T) with 1% donkey serum and treated with primary antibody overnight at 4°C. On day 2, slides were washed in PBS-T, incubated with secondary antibody for 2 h at room temperature, followed by a 10 min wash in PBS-T with DAPI (Sigma-Aldrich). Coverslips were added to the slides using Prolong Gold Antifade Reagent (Invitrogen). The primary and secondary antibodies and dilutions used are listed in Table S1. Slides were imaged using either an Olympus BX-51 or a Leica SP5X 2-Photon confocal microscope.

Tissue was fixed in 4% paraformaldehyde (PFA) for 3 h at 4°C, dehydrated in MeOH and stored at −20°C until stained. Fixed tissue was treated with Dent's bleach (MeOH:DMSO:H₂O₂, 4:1:1) for 2 h at room temperature, blocked with TNB (Perkin Elmer) and incubated overnight with anti-Muc1 antibody in TNB at 4°C. Pancreata were then washed with PBS and incubated with secondary antibody in TNB overnight at 4°C. Pancreata were imaged in BABB (benzyl alcohol:benzyl benzoate, 1:2) on a Leica SP5X Inverted 2-Photon FLIM Confocal microscope. Confocal z-stacks were reconstructed using ImageJ. Details of antibodies used and dilutions are listed in Table S1.

RNA was isolated from mouse pancreata with the Qiagen RNeasy Micro Kit. Quantitative RT-PCR (qRT-PCR) was carried out using Roche FastStart SYBR Green Master Mix and the Applied Biosystems StepOnePlus Real-time PCR system (Life Technologies). Relative expression values were calculated as 2^−ΔΔCt, and values of controls were normalized to 1. Rn18s served as an internal control for normalization in all qRT-PCR experiments. All data are shown as the mean of at least three independent biological replicates; error bars represent s.e.m. Calculations and P-values (Student's two-tailed, unpaired t-test) were generated in Microsoft Excel. Results were considered statistically significant at P<0.05. Graphs were generated using Prism 6 (GraphPad). Primer sequences are listed in Table S2.

Pancreata were dissected at E12.5 and cultured at the air-media interface on a Nucleopore Track-Etched Membrane (Whatman) in DMEM/F12 with l-glutamine and 15 mM HEPES (Gibco) supplemented with penicillin-streptomycin (Gibco). For rescue, the medium was supplemented with 500 ng/ml Recombinant Human/Mouse Wnt5a (R&D Systems) and refreshed with new media and Wnt5a protein on day 2 of culture. After 6 days in culture, pancreata were fixed at room temperature in 4% PFA for 3 h, immersed in 30% sucrose at 4°C overnight and frozen in OCT the next day. Pancreatic explants were cryosectioned at 12 µm and stained as described above for Ins/Gcg. ImageJ software was used to calculate the area of signal for DAPI and Ins/Gcg. Quantification is displayed as total Ins and Gcg signal/total DAPI signal for each explant.

We thank Dr Susan Mackem for contributing Hoxb6CreER^T mice. We would also like to thank Dr Ondine Cleaver for sharing the Muc1 whole-mount staining protocol and Drs D. Gumucio, Å. Kolterud and Kate Walton for the Wnt5a ISH probe. Dr Gregory Dressler provided riboprobes for Wnt4 and Wnt11. Dr J. Spence provided several qRT-PCR primers. We thank Holly Fischer for model artwork. Bright Kim provided technical assistance on parts of this work. The PECAM antibody developed by Steven A. Bogen was obtained from the Developmental Studies Hybridoma Bank, created by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health and maintained at The University of Iowa, Department of Biology, Iowa City, IA, USA.