Neurog3-dependent pancreas dysgenesis causes ectopic pancreas in Hes1 mutant mice

Vertebrate organs form in a stereotypical manner that is based on patterning of embryonic tissues along the three axes of the body. Organ dysfunction can occur if the initial patterning events or the later growth, morphogenesis and cell differentiation processes go awry. Congenital malformations of the pancreas include ectopic pancreas, defined as pancreatic tissue lacking anatomic and vascular continuity with the main body of the gland, annular pancreas, where the ventral pancreas encircles the duodenum and occasionally causes duodenal obstruction, and pancreas divisum, where the ventral and dorsal pancreas fail to fuse (Tadokoro et al., 2011). Ectopic pancreas can develop at multiple sites including the stomach, duodenum and jejunum (Türkvatan et al., 2013), but little is known about the underlying genetic and developmental causes.

The mouse pancreas develops from two primordia visible as thickenings of the posterior foregut epithelium around embryonic day (E) 9.0. The pancreatic buds are composed of multipotent pancreatic progenitor cells (MPCs) that express a cohort of transcription factors, including Hnf1β, Pdx1, Ptf1a, Nkx6-1, Sox9, Gata4 and Gata6, many of which have segregated to distinct tip and trunk compartments by ∼E14 (Bastidas-Ponce et al., 2017). The two buds grow independently until ∼E12.5, when they fuse to form a single organ connected to the digestive tract via the common bile duct. The ventral pancreas becomes associated with the duodenal loop, whereas the dorsal pancreas grows along the side of the stomach towards the spleen. At E14.5, the gastric lobe grows out from the base of the dorsal pancreas and wraps around the pylorus reaching towards the other side of the stomach (Hörnblad et al., 2011; Villasenor et al., 2010).

Some pancreatic malformations are coupled to loss-of-function mutations in developmental signalling pathways. For example, the Hedgehog family of signalling molecules play an important role in pancreas morphogenesis. Two genes of the Hedgehog family, sonic hedgehog (Shh) and indian hedgehog (Ihh), are expressed in partially overlapping patterns in the gut tube (Bitgood and McMahon, 1995; Echelard et al., 1993). Shh is expressed in endoderm adjacent to the pancreas but is excluded from pancreatic endoderm (Ahlgren et al., 1997; Apelqvist et al., 1997; Hebrok et al., 2000), and although Ihh is co-expressed with Shh in many endodermal domains, it is also expressed at a low level in the pancreas (Hebrok et al., 2000). Notably, null mutations in both genes cause annular pancreas, a ventral pancreas malformation, albeit with variable penetrance, depending on the genetic background (Hebrok et al., 2000; Ramalho-Santos et al., 2000). Comparably, loss of the transcription factor Hes1, the gene for which is a target and effector of Notch signalling in the pancreas, causes ectopic pancreas formation (Fukuda et al., 2006; Sumazaki et al., 2004). The Notch pathway is best known for its role in regulating differentiation of both endocrine and exocrine cell types in the pancreas. Loss-of-function mutations in the Notch pathway cause accelerated and excessive differentiation of pancreatic endocrine cells, by de-repression of the proendocrine gene Neurog3, which combined with reduced proliferation of MPCs results in a hypoplastic pancreas (Apelqvist et al., 1999; Fujikura et al., 2006, 2007; Jensen et al., 2000). More recently, loss of the Notch signalling components Mib1 and Hes1, as well as expression of dominant-negative Maml1, was shown to also cause the loss of the Nkx6-1⁺ bipotent duct/endocrine progenitors and an excess of Ptf1a⁺ pre-acinar cells at the secondary transition (Afelik et al., 2012; Horn et al., 2012). Conversely, overexpression of constitutively active Notch1 in the pancreas epithelium promotes an Nkx6-1⁺ fate, but prevents differentiation of endocrine and acinar cells (Hald et al., 2003; Murtaugh et al., 2003; Schaffer et al., 2010). A Hes1 null mutation is notable among the different Notch pathway mutations, as the only mutation reported to cause gall bladder agenesis along with ectopic pancreas in the common bile duct (Sumazaki et al., 2004), and a later study also revealed ectopic pancreas tissue in the stomach and duodenum (Fukuda et al., 2006). Further investigations of the malformations in the common bile duct revealed that Hes1 acts in conjunction with the transcription factor Sox17 to define the pancreato-biliary boundary in the ventral posterior foregut (Spence et al., 2009). The patches of ectopic pancreas located outside the biliary system in Hes1 nulls were suggested to arise from re-specified stomach or intestinal endoderm, after somatic loss of additional Hes family genes, but this idea was not tested (Fukuda et al., 2006).

Here, we show that ectopic pancreas tissue positioned outside the biliary system in Hes1 mutants is located dorsally and anterior to the pancreas proper and, in extreme cases, approaches the level of the oesophagotracheal septum. We show that conditional deletion of Hes1 in the endoderm recapitulates the ectopic pancreas phenotype seen in Hes1 nulls, suggesting that Hes1 is specifically required in the endoderm. Significantly, we show by lineage tracing that the anterior ectopic pancreas tissue is derived from the E8.5 dorsal pancreas primordium, which is specified normally in Hes1 mutants but becomes dysgenic at later stages. Remarkably, we found that dorsal bud morphogenesis was normalized, and ectopic pancreas formation suppressed, in Hes1^−/−Neurog3^−/− double mutants. We find that mutations in two other Notch pathway genes, Dll1 and Mib1, caused either no sign of ectopic pancreas (Dll1) or an anterior extension of the dorsal pancreas made up entirely by endocrine cells and devoid of progenitors (Mib1). Because Dll1 and Mib1 mutants have respective weaker and stronger endocrinogenic phenotypes than Hes1 mutants, our results suggest that the ratio between MPCs and cells of the endocrine lineage is important for normal morphogenesis. Thus, a huge excess of endocrine-fated cells without complete progenitor depletion, as seen exclusively in Hes1 mutants, appears to provoke ectopic pancreas formation by disrupting normal morphogenesis.

To examine how ectopic pancreas formation is manifested in Hes1 mutants we first evaluated the structure of the developing pancreas and surrounding gut regions from E8.5 to E12.5 by whole-mount immunofluorescence (IF). The dorsal and ventral pancreas anlage, marked by expression of Pdx1, appeared normal in E8.5 Hes1^+/+ and Hes1^−/− embryos, whereas Sox9⁺ cells were confined to the dorsal anlage in both genotypes (Fig. S1). In contrast, the dorsal anlage of E9.5 Hes1^−/− embryos displayed an irregular, saddle-like shape with Pdx1^Hi cells extending in both anterior and posterior directions. The bud itself appeared flattened and, as expected (Ahnfelt-Rønne et al., 2012; Jensen et al., 2000), a vast excess of Neurog3⁺ and glucagon⁺ cells were present compared with wild type (Fig. 1A,B). A flattened bud was also seen in E10.5 Hes1^−/− embryos but, remarkably, we also found ectopic tissue along the entire dorsal aspect of the stomach (Fig. 1C,D). The ectopic tissue contained Pdx1⁺Ptf1a⁺Nkx6-1⁺ cells, indicating a pancreatic MPC fate, as well as Neurog3⁺ endocrine precursors and endocrine cells expressing glucagon and PYY, as is typical for E10.5 embryos (Upchurch et al., 1994) (Fig. 1C,D, Fig. 2I, Fig. 5C, Fig. S2). At E11.5, the ectopic pancreas tissue is extending more anteriorly, forming a continuous cord from the flattened dorsal pancreas bud along the dorsal stomach endoderm and the posterior oesophagus (Fig. 1E,F). At E12.5, the ectopic pancreas tissue was typically found as a series of discrete patches along the dorsal stomach (Fig. 1G,H). However, in extreme cases, we noted ectopic pancreas tissue running anteriorly along the oesophagus and reaching the level of the oesophagotracheal septum anteriorly (Fig. S3). Thus, ectopic pancreas tissue is progressively found at more rostral locations as development proceeds. The penetrance of the ectopic pancreas phenotype in Hes1^−/− embryos was 100% at all stages examined (>30 embryos in total), whereas it was never seen in wild types (>50 embryos in total).

To get a more detailed view of the boundary between the stomach and ectopic pancreas epithelia, we analysed transverse sections of the posterior foregut region by IF. At E10.5, the epithelium of the prospective forestomach expresses E-cadherin and encircles a simple lumen (Fig. 2B). This structure was preserved in Hes1^−/− embryos, except that the dorsal part of the stomach epithelium was connected to an ectopic E-cadherin⁺ epithelium that extended in a dorsal direction (Fig. 2H). The ectopic dorsal epithelium contained Pdx1^Hi, Nkx6-1⁺ and glucagon⁺ cells, all markers of a pancreas fate, and was devoid of Sox2⁺ cells, which otherwise were found in the stomach epithelium proper in both Hes1^+/+ and Hes1^−/− embryos (Fig. 2C-E,I-K). Pancreatic markers were absent at the level of prospective forestomach in wild types (Fig. 2C,D), while Nkx6-1 was found in ventral stomach mesenchyme (Fig. 2C,I) and Pdx1 in the ventral part of the posterior stomach epithelium (Fig. 2E,K), in both wild types and mutants.

Specification of the dorsal pancreas from posterior foregut endoderm is dependent on signals from the notochord (Kim et al., 1997) and the vascular endothelium (Lammert et al., 2001; Yoshitomi and Zaret, 2004). Because Hes1 is expressed in both the endoderm and the inducing tissues (Klinck et al., 2011), the ectopic pancreas phenotype could be the result of loss of Hes1 in either of these tissues. To distinguish between these possibilities, we conditionally deleted Hes1 using Cre lines that are active in different combinations of these tissues. Foxa2-2A-iCre and Sox17-2A-iCre both delete efficiently in definitive endoderm but are also active outside the endoderm (Fig. 3A-C). However, the nonendodermal domains do not overlap between these two genes, so while Foxa2-2A-iCre is also active in the node, notochord and floor plate (Fig. 3B) (Horn et al., 2012), Sox17-2A-iCre is additionally active in vascular progenitor cells (Fig. 3C) (Engert et al., 2009). Accordingly, we generated Hes1^f/−; Foxa2^2A-iCre/+ (Hes1^ΔFoxa2) and Hes1^f/−; Sox17^2A-iCre/+ (Hes1^ΔSox17) embryos and analysed the appearance of the posterior foregut region by whole-mount IF. E9.5 Hes1^ΔFoxa2 and Hes1^ΔSox17 embryos both presented the anterior and posterior extensions of the Pdx1^Hi domain also seen in Hes1^−/− embryos (Fig. 3D-F, compare with Fig. 1B). At E10.5, Hes1^ΔFoxa2 and Hes1^ΔSox17 embryos both displayed pancreatic tissue along the entire dorsal aspect of the stomach in a manner indistinguishable from that observed in Hes1^−/− embryos (Fig. 3G-I, compare with Fig. 1D). E12.5 Hes1^ΔFoxa2 and Hes1^ΔSox17 embryos both had hypoplasia of the pancreas proper and ectopic pancreas tissue located in the dorsal part of the stomach as seen in Hes1^−/− embryos (Fig. 3J-L, compare with Fig. 1H). The penetrance of the ectopic pancreas phenotype was 100% at all stages examined (18 embryos in total for the two conditional mutants). Together, these results indicate that Hes1 is required specifically in the endoderm, the only tissue in which both Cre lines are active.

We next wanted to determine the origin of the ectopic pancreas tissue. We considered two possibilities: first, it is possible that Hes1-deficient dorsal endoderm responds abnormally to inductive signals, e.g. from the notochord, resulting in the re-specification of prospective dorsal stomach endoderm into pancreas; second, it is possible that Hes1-deficient dorsal pancreas epithelium extends parallel to the gut tube, along the dorsal stomach endoderm, instead of extending perpendicularly to the gut tube, into its associated mesenchyme, as seen in wild types. To examine which of these alternative, but not necessarily mutually exclusive, possibilities contributes to ectopic pancreas formation, we performed lineage tracing with Sox9-CreER^T2 and R26R-YFP mice (Kopp et al., 2011; Srinivas et al., 2001) on a Hes1^+/+ or Hes1^−/− background (Fig. 4A). As shown above, Sox9 expression is confined to the dorsal pancreas anlage in both wild-type and Hes1^−/− embryos (Fig. S1) at E8.0-E8.5, but soon thereafter it expands into the ventral pancreas as well as the endoderm of the prospective duodenum and stomach (Shih et al., 2015). Accordingly, 4-hydroxytamoxifen (4-OHT) administration at E9.5 resulted in embryos that showed extensive labelling of both ventral and dorsal pancreas, as well as stomach and intestine at E10.5 (Fig. S4A-C). Such embryos are uninformative for our purpose and were not pursued further. We therefore administered 4-OHT at E8.5 and quantified the number of labelled cells in the different midgut regions at E10.5 (Fig. 4A-D). On average, 94.8±7.0% (n=5) of the total YFP⁺ cells were found in the dorsoanterior ectopic pancreas of Hes1 mutants, which was similar to the 91.2±5.1% (mean±s.d., n=6) of the total YFP⁺ cells found in the dorsal pancreas of controls (P=0.39). Correspondingly, no difference was seen between controls and mutants in the stomach (0.5±1.2% vs 2.2±3.6% YFP⁺ cells, P=0.36) or intestine (8.0±5.9% vs 1.9±3.7% YFP⁺ cells, P=0.10). Notably, the majority of the embryos in both groups showed no stomach labelling at all (Fig. 4D). Enlarged views of individual optical sections, focused on regions in which the ectopic pancreas tissue was juxtaposed to, or fused with, the dorsal stomach epithelium, revealed that labelled cells were confined to the ectopic pancreas tissue (Fig. S4D-N). These results demonstrate that at least part of the ectopic pancreas tissue found in stomach and oesophageal regions of Hes1^−/− embryos is derived from an outwardly normal E8.5 dorsal pancreas primordium that subsequently undergoes a dysmorphic development. Nevertheless, these results do not refute that transdifferentiation of Hes1-deficient endoderm, induced by signals from dorsal mesoderm, might contribute to the ectopic pancreas tissue.

We therefore next explanted E9.5 Hes1^+/+Pdx1-GFP and Hes1^−/−Pdx1-GFP midguts and performed time-lapse imaging of the GFP signal to directly visualize dorsal bud formation. As can be seen in Movies 1 and 2, the dorsal buds of Hes1^+/+ and Hes1^+/− explants display a relatively normal growth pattern, with the buds expanding in a dorsal direction with no extension along the rostrocaudal axis. In contrast, the dorsal bud of the Hes1^−/− mutant fails to expand dorsally, but instead initiates a clear rostral extension (Movie 3). Moreover, no ectopic GFP signal appears in the stomach region of Hes1^−/− explants. These data support the notion of aberrant morphogenesis as the cause of ectopic pancreas formation in Hes1 mutants and argue against a cell-autonomous induction of transdifferentiation of stomach epithelium caused by loss of Hes1.

To begin to understand the mechanism underlying ectopic pancreas formation we sought to identify changes in gene expression in Hes1-deficient pancreas cells. We therefore performed RNA sequencing (RNA-seq) on fluorescence-activated cell sorted (FACS) GFP⁺ cells from individual E10.5 Pdx1-GFP; Hes1^+/+ (n=7) or Pdx1-GFP; Hes1^−/− (n=4) embryos. Differential expression analysis revealed 75 upregulated and 47 downregulated genes [false discovery rate (FDR)-adjusted P<0.1] in the Hes1^−/− compared with Hes1^+/+ embryos. As expected, many of the top upregulated transcripts, including Arx, Cacna2d1, Cbfa2t3, Cck, Chga, Chgb, Cpe, Fbxl16, Fev, Gadd45a, Gfra3, Ghrl, Hmgn3, Ids, Lrp11, Neurod1, Neurog3, Pax6, Pclo, Pcsk1, Rab15, Slc38a5, Tmem27 (Cltrn – Mouse Genome Informatics) and Upp1, are enriched in endocrine cells and/or their precursors (Fig. 5A) (de Lichtenberg et al., 2018). Conversely, the top downregulated transcripts, which included – in addition to Hes1 – Adamts1, Chst2, Cpa1, Gc, Jag1, Matn4, Myc, Nr5a2 and Wnt7b, are enriched in progenitor cells (Fig. 5A). Accordingly, the most significant gene ontology (GO) term for the upregulated genes is ‘Endocrine systems development’ (P=0.00004), but other significant terms, most notably ‘Cell migration’ (P=0.004) were also found high on the list. Among the genes associated with this GO term were many classical endocrine markers – Chga, Arx, Pax6, Cck and Lhx1 – but also genes encoding GDNF receptor alpha 3 (Gfra3) and podocalyxin-like (Podxl), which have been associated with enhanced migration and invasion of non-small cell lung carcinoma cells and invasiveness in pancreatic cancer, respectively (Tang et al., 2010; Taniuchi et al., 2016). We therefore performed IF for these markers to uncover any potential correlation between these proteins and the ectopic pancreas phenotype. However, as shown in Fig. S5, Gfra3 was largely confined to endocrine cells as previously noted (Nivlet et al., 2016), and no difference between Hes1^−/− and controls was observed. Podocalyxin-like was found at the apical side of most epithelial cells, wherever a microlumen had formed, but again without a noticeable difference between Hes1^−/− and controls.

Thus, the RNA-seq data largely identified genes that were expected from the excessive endocrine differentiation known to occur in Hes1 mutants.

Accordingly, we decided to directly test whether the endocrinogenic phenotype could be causally related to the dysmorphic development of the dorsal bud. We therefore generated Hes1^−/−Neurog3^−/− double mutants with the aim of preventing endocrine differentiation (Gradwohl et al., 2000). As expected, the dorsal bud of E10.5 Hes1^−/−Neurog3^+/+ embryos from these crosses was dysmorphic, and Pdx1⁺Ptf1a⁺ progenitors, as well as glucagon⁺ endocrine cells, extended along the entire dorsal part of the stomach compared with wild-type controls (Fig. 5B,C). Remarkably, the dorsal bud of Hes1^−/−Neurog3^−/− embryos displayed the typical spherical appearance, albeit without signs of endocrine differentiation, comparable to Neurog3^−/− embryos (Fig. 5D,E). Consistently, the prominent patches of ectopic pancreas along the stomach and oesophagus typically seen in E12.5 Hes1^−/− embryos were absent from E12.5 Hes1^−/−Neurog3^−/− embryos, although a few scattered glucagon⁺ cells could be found in both stomach and pancreatic regions (Fig. 5F-I). The appearance of these cells is remarkable in light of the lack of Neurog3, but is not without precedent. Scattered glucagon⁺ cells have previously been found in Dll1^−/−Neurog3^−/− embryos (Ahnfelt-Rønne et al., 2012) and in certain Neurog3^−/− strains (Wang et al., 2008). Nevertheless, these results demonstrate that the anterior extension of the dorsal pancreas at E10.5, and the resulting formation of ectopic pancreas tissue, depends on at least a part of the endocrine differentiation programme.

Ectopic pancreas has so far not been described in mice with mutations in other Notch pathway genes, but not all have been analysed by methods that allow visualizing the GI tract in whole embryos. We therefore examined the appearance of the pancreatic buds in Dll1^ΔFoxa2 and Mib1^ΔFoxa2 embryos, as mutations in these two genes cause a weaker and stronger endocrinogenic phenotype, respectively (Ahnfelt-Rønne et al., 2012; Horn et al., 2012), compared with Hes1^−/− and Hes1^ΔFoxa2 (Fig. 6A-O). E9.5 dorsal buds showed some excessive endocrine differentiation in Dll1^ΔFoxa2 and Mib1^ΔFoxa2 embryos, but the MPC epithelium appeared relatively normal in morphology (Fig. 6A-C,G-I). E10.5 ventral buds were all relatively normal in appearance in these mutants, but the dorsal buds showed a distinctly different appearance, depending on which gene was mutated. Dll1^ΔFoxa2 embryos displayed a hypoplastic, but otherwise well-formed, spherical bud with a cap of glucagon⁺ cells, no sign of an anterior extension and hypoplasia at E12.5, which is identical to that seen in Dll1^−/− embryos (Fig. 6E,K,N) (Ahnfelt-Rønne et al., 2012; Horn et al., 2012). Remarkably, the dorsal bud of Mib1^ΔFoxa2 embryos was severely hypoplastic, with the bud appearing completely converted into anteriorly extending glucagon⁺ cells (Fig. 6F,L). However, in contrast to Hes1^ΔFoxa2 embryos, no ectopic pancreas is evident in E12.5 Mib1^ΔFoxa2 embryos, although both dorsal and ventral buds are hypoplastic (Fig. 6M,O). These results show that the weak endocrinogenic phenotype of Dll1 mutants correlates with a relatively normal dorsal bud morphogenesis, whereas the ‘hyperendocrinogenic’ Mib1 mutant phenotype is associated with an anteriorly extended, ‘dorsal bud’-like endocrine structure, entirely depleted of progenitor cells, that eventually results in dorsal pancreas agenesis (Horn et al., 2012). These phenotypes contrast the overt ectopic pancreas phenotype seen in Hes1 mutants (Fig. 6P).

Here, we show that the Notch signalling effector Hes1 is crucial for normal morphogenesis of the dorsal pancreas through its repression of Neurog3 and, at least partly, the endocrine differentiation program. In the absence of Hes1, the dorsal pancreas anlage fails to form the stalk during outgrowth from the gut tube and undergoes aberrant morphogenesis, initially leading to an anterior displacement of most of the dorsal pancreas epithelium to a position along the dorsal stomach epithelium. Parts of these two epithelia, both of which express E-cadherin, often appear merged at E10.5, suggesting that the anteriorly dislocated pancreas epithelium fuses with the dorsal stomach epithelium. Between E10.5 and E12.5, the stomach undergoes extensive morphogenetic transformation from an initially almost linear tube to a complex structure, in which the oesophagus moves towards the ventral side concomitant with a massive expansion of the dorsal part of the organ. We speculate that the initially contiguous cord of ectopic pancreas tissue seen at E10.5-E11.5 cannot expand as fast as the dorsal stomach, when the latter undergoes its major expansion between E11.5 and E12.5. Thus, the ectopic pancreas epithelium, which is likely adhering to the stomach endoderm through E-cadherin interactions, gets shredded, and this results in the formation of discrete patches of pancreatic tissue located adjacent to the dorsal stomach epithelium. However, in extreme cases, the anteriorly dislocated pancreatic tissue can reach an even more anterior position along the oesophagus.

Hes1 is considered to regulate endocrine differentiation by directly suppressing Neurog3 expression in the endoderm; however, Hes1 is also expressed in adjacent tissues, such as notochord and vascular endothelium (Klinck et al., 2011), that provide signals to the endoderm that are crucial for dorsal pancreas development (Kim et al., 1997; Lammert et al., 2001; Yoshitomi and Zaret, 2004). Thus, in spite of the notion that Hes1 acts in endoderm to regulate endocrine differentiation, the formation of ectopic pancreas tissue could be the result of loss of Hes1 in any of these tissues. Our finding that Hes1^ΔFoxa2 embryos recapitulate the ectopic pancreas phenotype shows that Hes1 is required in Foxa2-expressing cells or their progeny, mainly the endoderm, node and notochord, and the floor plate of the neural tube. Likewise, our finding that Hes1^ΔSox17 embryos recapitulate the ectopic pancreas phenotype shows that Hes1 is required in Sox17-expressing cells or their progeny, chiefly the endoderm and the vascular system. The most parsimonious explanation of these findings is that the loss of Hes1 in the endoderm is causing the formation of ectopic pancreas tissue. A mouse strain with strictly endoderm- or pancreas-specific Cre expression combined with highly efficient recombination prior to, or at the onset of, pancreas specification would have been the optimal choice to test this notion, but in spite of testing several lines we have not been able to secure a Cre deleter line that fulfils these requirements. Nevertheless, as discussed below, our finding that the additional deletion of Neurog3 prevented the formation of ectopic pancreas tissue supports our conclusion regarding an endodermal requirement for Hes1.

The dorsoanterior location of the ectopic pancreas tissue analysed in this work suggests a couple of potential explanations for the underlying mechanism. Dorsal endoderm is patterned by the overlying notochord that, prior to embryo turning, is in direct contact with the dorsal midline endoderm along most of the rostrocaudal axis (Hebrok et al., 1998; Jørgensen et al., 2007; Kim et al., 1997). Nevertheless, only endoderm located at the pancreatic level responds to notochord signals by initiation of a pancreatic program (Hebrok et al., 1998; Kim et al., 1997). We first hypothesized that the Hes1-deficient prospective dorsal stomach endoderm might acquire the competence to initiate a pancreatic program; for example, in response to notochord signals. However, our lineage-tracing experiment showed that Sox9⁺ cells located in the E8.5 dorsal pancreas primordium could contribute to the ectopic pancreas tissue without contributing to the stomach epithelium, which argues against transdifferentiation of stomach endoderm, at least as a pure cell-autonomous mechanism. Instead, these data support a model in which the morphogenesis of the dorsal pancreas bud is disturbed such that it extends parallel to the gut tube, along the dorsal stomach endoderm in Hes1 mutants, rather than perpendicularly into its associated mesenchyme, as seen in wild types.

This finding appears to contradict the study of Fukuda and co-workers, who suggested that ectopic pancreas was caused by a transcommitment of the prospective stomach to pancreatic lineage by misexpression of Ptf1a (Fukuda et al., 2006). The authors of that study concluded that de-repression of Ptf1a in the absence of Hes1 was sufficient to induce this transcommitment. It should be noted though that the conclusion of Fukuda and co-workers was based primarily on two central observations. First, β-galactosidase labelling of ectopic pancreas tissue in Hes1^−/−Ptf1a^Cre/+; R26R animals. However, owing to its lack of time-dependent (i.e. tamoxifen-regulated) recombination, Ptf1a^Cre/+-based labelling does not allow one to distinguish ectopic, de novo expression of Ptf1a in stomach epithelium, from aberrant morphogenesis, resulting in Ptf1a-expressing pancreatic tissue being dislocated along the dorsal aspect of the stomach. Second, on transverse sections of stomach/ectopic pancreas tissue, Fukuda et al. found laminin immunoreactivity surrounding both tissues. This continuous basal lamina was interpreted as delamination of transdifferentiated ectopic pancreas tissue from the rest of the stomach. Yet, such a continuous basal lamina does not necessarily indicate a process of delamination. Instead, we propose that the dorsal pancreas tissue extends anteriorly between, or intercalating with, the stomach epithelium and the basal lamina in Hes1 mutants, displacing the latter in a dorsal direction. However, even though we disagree with the interpretation, we think that the observations of Fukuda et al. are fully consistent with our proposal of aberrant morphogenesis causing ectopic pancreas. Conversely, we find it difficult to reconcile our observations with transdifferentiation of stomach endoderm. As mentioned above, our lineage-tracing data argue that transdifferentiation is not required for ectopic pancreas formation and the fact that concurrent deletion of Neurog3 prevents ectopic pancreas formation in Hes1 mutants is difficult to attribute to an effect exerted in the stomach epithelium, as Neurog3 does not become expressed in main stomach epithelium until E12.5 in the Hes1 mutants (E13.5 in wild types), two full days after overt ectopic pancreas formation is observed. If transdifferentiation of stomach endoderm, perhaps induced by signals from adjacent mesoderm, was occurring, one would expect that this would still occur in Hes1^−/−Neurog3^−/− embryos.

Mutation of Hes1 is the only mutation in the Notch pathway that, to date, has been observed to cause ectopic pancreas. Remarkably, when we examined pancreas development in Mib1 mutants, the most prominent morphological feature of the E10.5 dorsal bud was an anterior extension reminiscent of that seen in Hes1 mutants at the same stage. However, whereas this extension is composed of a mix of Pdx1⁺Ptf1a⁺ progenitors and glucagon⁺ endocrine cells in the Hes1 mutants, it is composed entirely of postmitotic glucagon⁺ cells in the Mib1 mutants. In contrast, Dll1 mutants, which display a rather weak endocrinogenic phenotype, displayed a hypoplastic, but otherwise well-formed spherical bud with a cap of glucagon⁺ cells and no sign of an anterior extension. Thus, the rather weak endocrinogenic phenotype of Dll1 mutants, which declines before E9.5 (Ahnfelt-Rønne et al., 2012), allows for a relatively normal dorsal bud morphogenesis. In contrast, the stronger endocrinogenic phenotypes of Hes1 and Mib1 mutants are associated with a highly abnormal dorsal bud morphogenesis. However, the anteriorly extended dorsal bud of the ‘hyperendocrinogenic’ Mib1 mutants is much less pronounced, perhaps because the extension in the Mib1 mutant does not contain proliferating progenitor cells. We hypothesize that the disproportionate mix of Pdx1⁺Ptf1a⁺ progenitors, Neurog3⁺ endocrine precursors and glucagon⁺ endocrine cells in the early dorsal pancreas of Hes1 mutant embryos, perhaps in combination with loss of Hes1 in MPCs, provokes an aberrant morphogenesis, ultimately leading to ectopic pancreas (Fig. 6P). It would be interesting to test this notion by inducible expression of Neurog3 in either wild-type or e.g. Dll1 mutant backgrounds.

Irrespective, the rescue of early dorsal bud morphogenesis seen in Hes1^−/−Neurog3^−/− double-knockout embryos points to the endocrine cells and/or the Neurog3⁺ precursors as being involved in the formation of the anteriorly extended dorsal bud. One can imagine that excessive production of glucagon or PYY from the overabundant endocrine cells could affect cell motility. However, treatment of pregnant females with glucagon- and Y2-receptor antagonists yielded no evidence to suggest that these hormones played a role in the abnormal morphogenesis. Instead, we favour a hypothesis that it is the inherently motile phenotype of differentiating endocrine cells that is driving the anterior extension of the dorsal bud. When a vast excess of endocrine precursors and mature endocrine cells are delaminating from the epithelium, it is possible that they embark on an abnormal migratory route and that the disproportionality between endocrine lineage cells and MPCs makes it likely that the latter become dragged along in a passive process. Alternatively, the MPCs might contribute actively to the aberrant morphogenesis; for example, if their proliferation rate was increased. However, re-examination of our previously published confocal analysis of 5-bromo-2'-deoxyuridine (BrdU) incorporation (Ahnfelt-Rønne et al., 2012) showed that Pdx1⁺ cells in the E10.5 ectopic pancreas tissue had a marked reduction in BrdU incorporation compared with Pdx1⁺ dorsal bud cells from wild-type controls, arguing against this idea. However, it is noteworthy that the MPCs of the ectopic tissue are expected to maintain (or even increase) active Notch signalling in spite of (or due to) the loss of Hes1. Indeed, preliminary data indicate that such MPCs display strong expression of a BAC-transgenic Hes1-GFP reporter construct. Future experiments that conditionally inactivate genes required for motility of Neurog3⁺ cells or other Notch pathway genes could test these hypotheses.

Ectopic pancreas is occasionally found in humans and our results raise the possibility that HES1 or other genes that affect HES1 expression could be involved in the aetiology. Assuming that HES1 function is conserved in humans, it is unlikely that global homozygosity for null alleles, i.e. through germ line transmission of HES1 mutations, is involved, owing to the pleiotropic effects of global loss of Hes1 (Ishibashi et al., 1995). However, somatic mutations, perhaps via loss of heterozygosity, could potentially lead to ectopic pancreas in humans. Genomic sequencing of resected ectopic pancreas tissue could help to resolve this issue.

The generation, breeding and genotyping of Hes1^+/−, Hes1^f/+, Dll1^f/+, Mib1^f/+, Neurog3^tTA, Foxa2^T2AiCre, Sox17^T2AiCre, Sox9-CreER^T2, R26R, R26R-EYFP and Pdx1-GFP mouse lines have been reported previously (Engert et al., 2009; Horn et al., 2012; Imayoshi et al., 2008; Ishibashi et al., 1995; Koo et al., 2007; Kopp et al., 2011; Micallef et al., 2005; Soriano, 1999; Srinivas et al., 2001; Wang et al., 2009). Animals were maintained on a C57Bl/6J background. Littermate embryos were used as controls in all experiments. In lineage-tracing experiments, littermate Hes1^+/+; Sox9-CreER^T2 mice were used as controls. Animals were maintained, and experiments were conducted, in adherence to guidelines issued by the European Convention for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes (ETS 123).

E10.5 embryos were dissected and fixed for 1 h in 0.2% glutaraldehyde/5 mM EGTA/100 mM MgCl₂/1× PBS. Embryos were washed for 3×5 min in 1× PBS and then incubated at 30°C overnight in 1 mg/ml X-gal (5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside)/20 mM K₃Fe(CN)₆/20 mM K₄Fe(CN)₆*3H₂O/2 mM MgCl₂/0.01% NaDOC/0.02% Nonidet-P40/1× PBS. Embryos were washed for 3×5 min in 1× PBS and postfixed overnight at 4°C in 0.2% glutaraldehyde/1× PBS. Embryos were dehydrated stepwise in 1× PBS:MeOH (3:1, 2:2, 1:3) to 100% MeOH and cleared in BABB (1:2 benzyl alcohol:benzyl benzoate) before imaging.

Sox9-CreER^T2, R26R-EYFP and Hes1^+/− animals were interbred to produce embryos carrying Sox9-CreER^T2 and R26R-EYFP on Hes1^+/+ and Hes1^−/− backgrounds. To label cells in the dorsal pancreas the pregnant females were injected intraperitoneally gestational days E8.5 and E9.5 with 0.05 mg/g body weight 4-OHT (Sigma-Aldrich, H6278) from a 10 mg/ml solution in corn oil (Sigma-Aldrich, C8267). Embryos were harvested at E10.5 and processed for whole-mount IF. From six litters, receiving 4-OHT at E8.5, we obtained seven Sox9-CreER^T2+R26R^Yfp/+Hes1^−/− embryos that were analysed by whole-mount IF, together with a corresponding number of Sox9-CreER^T2+R26R^Yfp/+Hes1^+/+ and Sox9-CreER^T2+R26R^Yfp/+Hes1^+/− littermate controls.

Statistical analysis was performed with two-tailed Student's t-test for unpaired samples. Sample size was determined post hoc. Three embryos with prominent labelling in ventral pancreas (≥18 YFP⁺ cells) were excluded from the analysis. Embryos receiving 4-OHT at E9.5 showed extensive labelling of ventral and dorsal pancreas, as well as stomach and intestine at E10.5, and were not analysed further. The exclusion criteria were established post hoc.

Embryos dissected at E10.5 and smaller stages were processed and stained as whole embryos, whereas at E11.5 and larger stages, the stomach-pancreas-duodenum block was dissected out. The tissue was fixed overnight at 4°C in 10% formalin in phosphate buffer, pH 7.2, washed for 3×30 min in 1× PBS, and then gently dehydrated stepwise in 1× PBS:MeOH (3:1, 2:2, 1:3) to 100% MeOH. The tissue was permeabilized for at least 2 h at room temperature (RT) in MeOH:DMSO:H₂O₂ (4:1:1) and then transferred back to 100% MeOH. The tissue was gently rehydrated in MeOH:1× PBS (3:1, 2:2, 1:3) to 1× PBS, blocked in 0.5% TNB (Perkin Elmer, NEL704A001KT)/1× PBS for 2 h at RT, and incubated with primary antibodies diluted in 0.5% TNB/1× PBS for 48 h at 4°C with gentle nutation. The tissue was washed 3× for 1 h in 1× PBS and incubated with Cy2-, Cy3- and Cy5-conjugated secondary antibodies raised in donkey (Jackson ImmunoResearch), diluted 1:500 in TNB/1× PBS for 48 h at 4°C with gentle nutation. For some of the rabbit antisera, a tyramide signal amplification (TSA) step was added to the protocol. Here, a biotinylated secondary antibody raised in donkey against rabbit IgG (Jackson ImmunoResearch) was applied together with the other secondary antibodies. After washing 3× for 1 h in 1× PBS, the tissue was incubated with peroxidase-conjugated streptavidin (PerkinElmer) for 48 h at 4°C with gentle nutation. The tissue was washed 3× for 1 h in 1× PBS and the signal developed with Cy3-TSA substrate, diluted 1:100 in amplification diluent (PerkinElmer) for 60-75 min at RT with gentle nutation and protected from light. The tissue was washed 3× for 1 h in 1× PBS, dehydrated stepwise in 1× PBS:MeOH (3:1, 2:2, 1:3) to 100% MeOH, and cleared in BABB before confocal imaging as described (Ahnfelt-Rønne et al., 2007).

Embryos dissected at E10.5 were fixed overnight in 10% formalin in phosphate buffer, pH 7.2, washed 3× in 1× PBS and cryoprotected overnight in 30% sucrose/1× PBS. Embryos were then embedded in OCT (Sakura) and cryosectioned at 10 μm. Sections were rinsed in 1× PBS, blocked for 30 min in TNB/1× PBS, and incubated overnight with the primary antibodies diluted in TNB/1× PBS. Sections were washed for 3×5 min in 1× PBS and incubated for 30 min with A488-, Cy2-, Cy3- and Cy5-conjugated secondary antibodies raised in donkey (Jackson ImmunoResearch), diluted 1:500 in TNB/1× PBS. Sections were washed for 3×5 min in 1× PBS and mounted with 20% glycerol in 40 mM Tris base, pH 8.4. All steps in the staining procedure were performed at RT. Imaging was performed essentially as described in Klinck et al. (2011).

Primary antibodies used were as follows: goat anti-Foxa2 (Santa Cruz Biotechnology, sc-6554, 1:1000), rabbit anti-Nkx6-1 (Jensen et al., 1996) (1:500), goat anti-Pdx1 (BCBC, ab2027, 1:10,000), guinea pig anti-glucagon (Millipore, 4031-01F, 1:10,000), rabbit anti-GFP (Clontech, 632460, 1:2000, TSA), rabbit anti-Neurog3 (BCBC, ab2011, 1:10,000, TSA), rabbit anti-Ptf1a (BCBC, ab2153, 1:2000), rabbit anti-PYY (kind gift from Ole D. Madsen, 1:1000), rabbit anti-Sox2 (Chemicon, AB5603, 1:1000), rabbit anti-Sox9 (Millipore, AB5535, 1:1000), rat anti-Cdh1 (Novo Nordisk A/S, ECCD2, 1:2000), goat anti-Gfra3 (R&D Systems, AF2645, 1:500) and goat anti-Podxl (R&D Systems, AF1556, 1:500).

Pdx1^GFP/+; Hes1^+/− mice (Ishibashi et al., 1995; Micallef et al., 2005) were time mated with Hes1^+/− mice and E10.5 embryos were isolated. The head and tail regions of each embryo were removed and used for genotyping. Each sample then consisted of the midbody, from the forelimb buds to the hindlimb buds. Based on previous cell sorting experiments, the total number of Pdx1⁺ cells in the E10.5 pancreas is estimated to be 500-1000. To be able to sort out such a low number of GFP⁺ cells, we were advised to spike the samples with a significant number of GFP⁻ cells. As Pdx1, and Pdx1-GFP, is expressed strongly only in the pancreatic endoderm (GFP^Hi cells) and weaker in neighbouring duodenal and posterior stomach endoderm (GFP^Lo cells), we therefore did not dissect away nonpancreatic tissue, which is impossible to do in the mutant embryos anyway, but allowed this tissue to serve as a spike in each sample. We then used gating on the FACS to isolate GFP^Hi cells from GFP^Lo cells and GFP⁻ cells. The samples were treated individually with Liberase TL (Roche) and DNAse I (Thermo Fisher Scientific) in Dulbecco's modified eagle medium (Thermo Fisher Scientific) for 20 min at 37°C, with shaking (1100 rpm) on a thermomixer to dissociate the cells, and subsequently spun down at 500 g for 5 min. The cells were suspended in 0.125% Trypsin-EDTA (Thermo Fisher Scientific) and, after 10 min at 37°C, the trypsinization was stopped with fetal bovine serum (FBS). Then, the cells were spun down at 500 g for 5 min and suspended in FACS buffer [2% FBS in PBS with 4′,6-diamidino-2-phenylindole (DAPI)]. Each embryo sample was sorted separately on a Sony SH800 cell sorter directly into RLT Plus (Qiagen). We retrieved 200-500 GFP⁺ cells per embryo and RNA was purified with RNeasy Plus Micro (Qiagen) according to the manufacturer's instructions. RNA quality was verified using a Bioanalyser Pico Kit (Agilent) and the libraries were generated using the Ovation SoLo RNA-seq system (NuGen) according to the manufacturer's instructions. The libraries were quantified with Qubit high sensitivity DNA kit (Thermo Fisher Scientific) and sequenced with either HiSeq or NextSeq 500 sequencing systems (Illumina). The reads were mapped with STAR aligner (Dobin et al., 2013), and differential expression analysis between Hes1^+/+ and Hes1^−/− samples was performed by DeSeq2 (Love et al., 2014), using independent filtering and an adjusted P-value of 0.1. GO term analysis was performed using PANTHER software (Mi et al., 2013); REVIGO software (Supek et al., 2011) was used to cluster GO terms in semantic space. The RNA-seq data have been deposited to ArrayExpress under the accession number E-MTAB-6890.

All IF images were captured using point scanning confocal Zeiss Axioimager systems (LSM510, LSM700 and LSM710). Z-stacks and tiled z-stacks from whole-mount-IF-stained tissue were stitched and projected to 3D using Zeiss Zen software. Alignment of two images to produce the composite image in Fig. 1H was performed manually with Adobe Photoshop.

Eight-well culture slides with glass bottom (Ibidi, 80827) were coated with 1 μg/ml fibronectin (Sigma-Aldrich, F1141) in 100 mM NaCl the day before explant isolation and stored at 4°C. Pdx1^GFP/+; Hes1^+/− mice were time mated with Hes1^+/− mice and E9.5 embryos were isolated. The foregut/midgut region of the primitive gut tube containing prospective stomach, pancreas and duodenum was microdissected in ice-cold dissection medium [1× minimum essential medium (MEM) with Hank's Salts and L-glutamine (Gibco, 21575-022), 10% FBS, 50 μg/ml gentamicin (Gibco, 15750-037)] from each embryo. We observed that 17/37 isolated gut tubes expressed GFP in the pancreas primordium. The GFP⁺ explants were placed left side down on the glass bottom of the culture wells in 37°C pre-warmed culture medium [1× MEM with Earle's Salts and L-glutamine (Gibco, 11095-080), 20% FBS, 50 μg/ml gentamicin (Gibco, 15750-037), 0.1 μg/ml FGF10 (R&D Systems, 6224-FG)]. The explants were allowed to attach overnight at 37°C before imaging. Live recordings of the Pdx1-GFP signal in dorsal pancreatic bud development were captured on a confocal Zeiss LSM880 with an Airyscan module using a Plan Apochromat 20×/0.80 NA objective (420650-9901-000). Z-stacks of 24 slices of 2 μm optical sections with minimal overlap were recorded every 20 min for 16 h, after which viability was affected in most explants. Movies were recorded in Imaris 9.0.2 software (Bitplane).

We thank Ryoichiro Kageyama for mice; Karsten Marckstrøm for technical assistance; Helen Neil, Gelo de la Cruz, Jutta Bulkescher, Gopal Karemore and the Center for Integrated Microscopy for help with cell sorting, RNA-seq, imaging and image data analysis; and Gerard Gradwohl, Phil Seymour and Thue W. Schwartz for advice.