Gene regulatory factors in pancreatic development


  • Jan Jensen

    Corresponding author
    1. Barbara Davis Center for Childhood Diabetes, University of Colorado Health Sciences Center, Denver, Colorado
    • Barbara Davis Center for Childhood Diabetes, University of Colorado, Health Sciences Center, 4200 E 9th Avenue, B140, 80262 Denver, CO
    Search for more papers by this author


The intensity of research on pancreatic development has increased markedly in the past 5 years, primarily for two reasons: we now know that the insulin-producing β-cells normally arise from an endodermally derived, pancreas-specified precursor cell, and successful transplants of islet cells have been performed, relieving patients with type I diabetes of symptoms for extended periods after transplantation. Combining in vitro β-cell formation from a pancreatic biopsy of a diabetic patient or from other stem-cell sources followed by endocrine cell transplantation may be the most beneficial route for a future diabetes therapy. However, to achieve this, a thorough understanding of the genetic components regulating the development of β-cells is required. The following review discusses our current understanding of the transcription factor networks necessary for pancreatic development and how several genetic interactions coming into play at the earliest stages of endodermal development gradually help to build the pancreatic organ. Developmental Dynamics 229:176–200, 2004. © 2003 Wiley-Liss, Inc.


Pancreatic Morphogenesis

Organogenesis of the pancreas is a highly coordinated process (Table 1). The first visible evidence of pancreas development in rodents is the formation of two independent thickenings, or “anlage,” each containing a few hundred cells, dorsally and ventrally, of the endodermal primitive gut tube at the 20- to 22-somite stage (Wessells and Cohen, 1967; Pictet and Rutter, 1972; Pictet et al., 1972). These regions are found immediately posterior to the anlagen of the stomach.

Table 1. Stages of Pancreatic Development in Rodentsa
Time pointEventMorphological appearanceGene expression
  • a

    E, embryonic day; P, postnatal day.

GastrulationGerm layer formationCell movement through the primitive streak generates mesoderm and endoderm.Foxa2 in endoderm
6–10 somitesDetermination, primary transitionPrimitive gut as flattened sheet, no visible pancreas. Intimate contact of endoderm to the notochord. Signaling necessary. Ventral pancreas defined by absence of cardiogenic mesenchyme/septum transversum instructionIsl1 expressed in dorsal mesenchyme.
E8–8.5 (mouse)
E10–10.5 (rat)
10–20 somitesEmbryo rotationClosure of the gut tube. Formation of dorsal aorta between the gut tube and the notochord. Endoderm moves ventrally, displaced by aorta. No demarcation between anlage of stomach and pancreas. Mesodermal cells, “splanchnic mesoderm,” accumulate laterally on the gut tube.Isl1 and Isl2 expression in mesoderm. Pdx1 activation (11–13 somites). HlxB9 expressed
E8.5–9.5 (mouse)
E10.5–11 (rat)
20–25 somitesBuddingThickening of dorsal/ventral walls of the gut, followed by epithelial evagination dorsally. Pancreatic mesoderm accumulates dorsally.Nkx6.1 activation. Glucagon-positive cell clusters. Expression of Isl1 and Pax6in endocrine cells. Ngn3 expressed.
E9.5–10 (mouse)
E11–11.5 (rat)
26–30 somites Accumulation of mesoderm primarily on the left side of the gut generates asymmetry. The base of the dorsal evagination narrows. Ventral bud appears adjacent to liver bud.Pitx2 involved in left/right handedness of primitive gut.
E10 (mouse)
E11.5 (rat)
30+ somites Continuous epithelial evagination. Branching morphogenesis initiates. Mitotic index highest in mesenchyme. Innervation commences (50 somites).NeuroD, Ptf1a expressed. Epithelium strongly expressing E-cadherin and β-catenin.
E12–12.5 (rat)
E12 mouseVisible branching morphogenesisLatest time point for mesenchymal–epithelial separation.Hnf6 homogeneously expressed in epithelium.
E14 rat
→ E14 (mouse)Epithelial expansion“Protodifferentiated” epithelium.Nkx6.1, Pdx1, Nkx2.2 homogenously expressed in epithelium.
→ E15.5 (rat) Very low-level presence of terminal markers (insulin, amylase, CPA, and others).Pax6 and NeuroD speckled.
  Fusion of dorsal and ventral pancreas.High-level Notch1, 2, Hes1 expressed in epithelium.
  Mesenchymal percentage high, but decreases gradually due to epithelial growth. Vascularization The expanding epithelium phase overlaps the onset of the secondary transition. 
Appx. E13.5 (mouse)Secondary transitionTerminal differentiation within β-cell and exocrine compartments.> 100 fold activation of insulin and exocrine genes.
Appx. E15 (rat) Appearance of zymogen granules. Scattered appearance of mature β-cells.Pdx1, Nkx6.1, Nkx2.2 become β-cell restricted.
   Pax4 activated. Ngn3 expression peaks.
E15–E19, prenatal (mouse)IsletogenesisEndocrine cells aggregate in islets of Langerhans. Exocrine acinary ultrastructure attained. Functional organ at birth.Delta-cells appear. (App. E15) PP cells appear.
E17–E21 (rat)
P1–P30GrowthTissue mass increase. High percentage of endocrine cells in the early postnatal phase is overtaken by massive growth of the exocrine mass, which condenses to form compact acini.Trypsin and CPB activated (P1).

The onset of pancreagenesis is known as the “primary transition,” referring to the change in shape of the pancreatic domains of the gut and the attained pancreatic cellular state. Wessells and colleagues demonstrated that the period of pancreatic determination occurs between the 6- and 10-somite stages in the rat (corresponding to embryonic day [E] 8.0 in the mouse; Wessells and Cohen, 1967; Wessells and Evans, 1968). More recent studies have revealed that the dorsal and ventral programs of the pancreas are not completely identical and slightly asynchronous, due to the two independent endodermal domains receiving distinct signals from their surrounding tissues.

The dorsal pancreatic bud develops in close proximity to the notochord. This proximity is essential for dorsal bud development. Slightly later, as the notochord is displaced by the dorsolateral splanchnic mesenchyme, which forms the dorsal aorta in between the dorsal epithelium and the notochord, a new set of signals are exchanged between the aorta and the pancreas. Recent studies of the molecular mechanisms that play a role in these processes have revealed the involvement of several morphogenetic signaling systems (including fibroblast growth factor [FGF], transforming growth factor-beta [TGFβ], vascular endothelial growth factor [VEGF], and hedgehog-type ligands; Kim et al., 1997; Hebrok et al., 1998, 2000; Wells and Melton, 2000). The early morphogenetic signals have been reviewed excellently previously (Melton et al., 1997; Wells and Melton, 1999; Grapin-Botton and Melton, 2000).

The ventral bud of the pancreas develops in close connection to the liver and bile duct epithelium that forms on the ventral face of the gut. The ventral pancreas originally develops as two independent ventral foregut endodermal regions, laterally positioned, which are brought together at the time of gut tube closing. Ventral pancreatic development is closely connected to organogenesis of the liver, with the latter being specified due to its proximity instructive signaling from the cardiac mesoderm. Concomitantly, the ventral pancreatic program is initiated due to the absence of this instructive signaling. The ventral pancreatic fate program appears to be a default state, because in the absence of cardiac induction cells of the prospective liver domain attain a ventral pancreatic fate (Zaret, 2000, 2001; Deutsch et al., 2001). Even in adult mouse pancreas, the plasticity between liver and pancreas fates can be observed (Shen et al., 2000) and often occurs naturally (Grompe, 2003). Such observations of plasticity between endodermal fates are not unique. Flexibility also exists between the pancreas and stomach fates (Hald et al., 2003), as well as between pancreatic vs. duodenal fate (Apelqvist et al., 1997; Kawaguchi et al., 2002). Such observations clearly point to the existence of shared genetic components amongst these endodermal derivatives.

After its initial specification, the epithelial thickening comprising the nascent pancreas changes over the next day into proper epithelial evaginations. These form perpendicular to the gut wall, destined to form the duodenum (Fig. 1). The epithelium is of a multilayered stratified type consisting of cells with a high nuclear/cytoplasmic ratio. This composition is very different from the later single-layered epithelium of the mature pancreatic ducts. At this stage, mesodermal cells (pancreatic mesenchyme) tightly associate with the pancreatic epithelium. They guide its development mainly through proliferative signaling (Wessells and Cohen, 1967). In both mouse and rat, the pancreatic epithelium expands markedly in the following days, during which the stratified epithelial nature is maintained, although a gradual thinning occurs and epithelial lobulations become pronounced.

Figure 1.

The primary transition. Embryonic day 10 mouse embryo. Initiation of pancreatic development is observed as two evaginations, dorsally and ventrally, form on the prospective duodenal region of the primitive gut. Pdx1 (green) is highly expressed both dorsally and ventrally. Note the lower level of Pdx1 immunoreactivity in both the primitive gut tube, extending up in the posterior stomach (future antropyloric domain). Glucagon cells (red) have started to form in the dorsal bud—similar cells will over the following 12–24 hr also start to differentiate in the ventral bud. Endocrine differentiation in the adjacent regions (stomach and intestine) is not initiated until approximately 5 days later. Reproduced with permission from Kluwer Academic Press (Madsen et al., 2001).

It is generally recognized that the pancreas undergoes a process of branching morphogenesis, while the epithelial expansion ensues. However, as different mechanistic solutions exist to coordinate such a process, care should be taken when using this nomenclature. As recently pointed out by Hogan and Kolodziej (2002), tubulogenesis differs structurally between organs generally recognized as undergoing branching morphogenesis. In the lung, a continuous expansion and folding of the epithelial layers is observed and this process is governed by mesenchymal–epithelial cross-talk. This particular process is of the classic branching morphogenesis type (Fig. 2A). In contrast, in both pancreas and salivary glands the formation of a ductal tree occurs through a process of intraepithelial lumen formation, wherein multiple microlumens eventually coalesce to form continuous lumens around which the ductular network is established (Fig. 2B). This is the first time point that cells having a typical basal–apical polarity signifying a cuboidal/rectangular epithelial are observed in the pancreas. Subsets of the endodermal cell population thereby form a ductal tree that, in the mature state, drains exocrine secretory products. Probably closely linked to the morphologic changes thus imposed, more differentiated characteristics appear, such as expression of ductal marker genes (e.g., CFTR, carbonic anhydrase II, mucins, and increased expression of certain cytokeratins; Githens, 1988). At present, we have only a descriptive understanding of pancreatic ductular development—no data have been derived that explain how the given cellular subsets of the pancreatic endoderm are allocated to a ductal fate, nor do we understand the mechanism underlying pancreatic tubulogenesis.

Figure 2.

Tubulogenesis. A: Schematic presentation of the classic branching type. Cells are at all time points of a single-layered epithelial type. Expansion occurs through facultative divisions within the increasingly folding sheet of cells, thereby generating tubules. B: Tubulogenesis due to fusion of microlumens. Cells are initially residing in a multilayered stratified epithelium and are not showing a typical basal/apical polarity. Polarity is induced as microlumens form due to fluid secretion by individual cells. The organization of cells around such lumens generates a single-layered epithelium. As the lumens coalesce, a tubular network is formed. In yellow, the formation of endocrine cells is outlined, as these initially are selected from epithelial cells, where after clustering occurs due to migration.

Our current appreciation of pancreatic development is strongly founded upon electron microscopic analyses addressing the presence of individual cell types in development (Pictet et al., 1972). From the presence of secretory granules, clearly distinguishable as being of β-cell, α-cell, and exocrine types, it was demonstrated that mature β-cells and exocrine cells were not detectable before embryonic day 15 in the rat (E15, corresponding to E13.5 in mice). However, glucagon-producing cells were detected at the early somite stages of development, concomitant with the budding process. On the basis of these data, Rutter and colleagues proposed a model of pancreatic development describing two major transitions in time. During the “primary transition,” cells at the appropriate regions of the gut become destined to form pancreas. These acquire molecular identities, which may be referred to as the “the pancreatic state” as described by Slack (1995). Only a few cells containing α-cell–like granules differentiate, and the remaining epithelial cells are characterized as being “protodifferentiated” (Pictet et al., 1972). The term “protodifferentiated” was used to reflect the presence of low levels of pancreas-specific terminal gene products, such as insulin and amylase in pancreatic extracts. This terminology may still be well suited to describe cells in “the pancreatic state.” At the time, no knowledge was available to support concepts of cell intrinsic regulators that might confer such identity.

During “the secondary transition”—a relatively short period between E13.5 and E15.5 in mouse (Fig. 3) —protein concentrations of β-cell and exocrine products increase between 100- and 1,000-fold (Han et al., 1986). A concomitant increase is observed for insulin, amylase, carboxypeptidase A, lipase, and chymotrypsin. At this time point, the structure of the adult organ begins to be recognizable. These descriptions of the primary and secondary transitions have been substantiated by a range of alternative techniques, including histologic approaches (immunofluorescence, in situ hybridization), and molecular techniques (Northern blotting, reverse transcriptase-polymerase chain reaction, DNA chip analyses). After the secondary transition, few cells of the pancreas progenitor type remain and expansion occurs by growth of fully differentiated cell types. Exocrine cells are strongly proliferating after their differentiation, and endocrine cells show strong mitotic activity closer to birth. In the interim, the endocrine cells aggregate by migration into clusters in a process that may be termed “isletogenesis” (Fig. 4).

Figure 3.

The secondary transition. Embryonic day (E) 13.5 mouse pancreas, double-probe in situ hybridized with insulin (red) and amylase (blue). At this stage, exocrine and β-cell terminal differentiation is concomitantly initiated. In the central core of the pancreas, β-cells arise with a scattered distribution and are exclusively nonepithelial at onset of insulin gene expression. In contrast, at the tips of the epithelial branches, exocrine differentiation is initiated, occurring in multiple neighboring cells simultaneously. At E14.5, all cells at the exterior tips have activated the amylase gene—as this occurs over such a short time frame that divisions of a presumed exocrine precursor cell could not have occurred, this shows that individual acini are not clonally derived. One possible exception to this argument is the existence of a clonal precursor before the terminal differentiation process that may have given rise to all cells in the particular group differentiating. Such a possibility has been noted (Gu et al., 2002). The mechanism for initiating exocrine differentiation is not known, but it is conceivable that spatial signals allow differentiation to proceed when a given threshold is met. This could explain why clusters of cells initiate differentiation at the same time. mes, mesenchyme; epi, epithelia.

Figure 4.

Isletogenesis. Embryonic day (E) 19 mouse pancreas double stained for Pdx1 and glucagon. This figure shows a quite interesting organization of the pancreas in between the adult structure with dispersed islets and the islet cell migration phase immediately after the secondary transition. Note the “ribbon-like” organization of the endocrine cells (endoc) in the central core of the pancreas—much reflecting the earlier central expression of the pro-endocrine gene Ngn3. At E19, most endocrine cells have aggregated, and now α-cells begin to organize around core structures of β-cells (strongly Pdx1-positive cells). Pdx1 is not expressed in α-cells and is also not expressed in the pancreatic duct cells. This finding may signify that the ductal cells may have lost the capacity of undergoing endocrine differentiation and are most likely developing toward a more mature and specialized ductal cell type. Low-level expression of Pdx1 is observed in the exocrine cells (exoc) that are completely filled with secretory granules. Proper islet formation appears to occur through an ensnaring process by the growth of the exocrine tissue, which leads to a cutting off of islets of the “ribbon” as pearls on a string.

Lineage Determination During Pancreagenesis

The appearance of the glucagon-containing cells within the primitive buds is the first evidence of cytodifferentiation of pancreatic cells. It had been suggested that these are precursors for all cells of the endocrine cell lineage. This notion dominated the pancreatic development field for almost two decades. However, lineage relation analyses have proven that this is not true. Herrera and colleagues (1994) showed, by targeting the diphtheria toxin A (Dip.ToxA) chain to α- or β-cells by using the glucagon and insulin promoters, respectively, that insulin and glucagon-producing cells were not related through a coexpressing precursor cell. Insulin promoter–driven Dip.ToxA only eliminated β-cells, and the glucagon promoter–driven Dip.ToxA selectively destroyed α-cells. Due to its inconsistency with the then-current notion, the study was not generally accepted, as it could be argued that the activity of the toxin was not sufficiently rapid, or effective, to kill the transient precursor cell.

Herrera later performed a lineage analysis using the Cre-Lox method of labeling a cell irreversibly. Asking whether an insulin cell ever expressed glucagon during an earlier stage, he crossed glucagon-promoter Cre-recombinase mice with a reporter mouse carrying a silenced loxP-STOP-loxP-flanked growth hormone allele. In the presence of the Cre-recombinase, the reporter would be activated, and in the presence of growth hormone/insulin, double-positive cells could be analyzed by histology. These were not found, showing that insulin-cells do not develop through a glucagon-expressing stage (Herrera, 2000). As the reciprocal experiment also showed that glucagon-cells do not develop through an insulin-expressing stage, it was concluded that these cell types develop independently. Immunohistochemical data, based on expression of transcription factors rather than the pancreatic hormones, are consistent with these results (Jensen et al., 2000a). Nevertheless, a common origin of endocrine cells does exist at a non–hormone-expressing precursor level where both α- and β-cells develop from PDX1-positive cells. This finding was demonstrated by using Pdx1-promoter CRE-mice, which generated an active reporter gene in both α- and β-cells (Herrera, 2000). Extending these observations, Gu and colleagues showed that all endodermally derived cells of the pancreas—endocrine, exocrine, and ducts—were originating from Pdx1-expressing precursors (Gu et al., 2002). Furthermore, by taking advantage of tamoxifen-inducible recombination by generating mice expressing the CreERT protein (a Cre-recombinase:estrogen receptor ligand binding domain [ER-LBD] chimera, with three point substitutions in the ER-LBD rendering the nuclear receptor responsive to tamoxifen) under control of the Pdx1 and Ngn3 promoters, the window of recombination in Pdx1-expressing cells could be controlled. Of interest, cells of the ductal lineage appeared to be specified within a short period from E9.5 to E12.5 in the mouse, as only during this time did tamoxifen-induced recombination lead to labeling of mature duct cells (Gu et al., 2002). Such temporal restriction was not observed for exocrine and endocrine cells. Contrasting the common Pdx1-expressing ancestry of all pancreatic cells, Gu observed that cells expressing Ngn3-Cre were almost exclusively of endocrine types, an observation in agreement with other studies (Gradwohl et al., 2000; Jensen et al., 2000a; Schwitzgebel et al., 2000). Finally, Kawaguchi et al., by performing a modified knock-out approach of the Ptf1a gene in which the Cre-recombinase was inserted into the locus in the process, revealed that pancreatic exocrine cells arise from Ptf1a-expressing precursors—but also discovered that a high percentage of all pancreatic cell types (except for the earliest glucagon-expressing cells) could be traced by this method, showing that Ptf1a, like Pdx1, is expressed within undifferentiated pancreatic precursors (Kawaguchi et al., 2002). Together, these studies demonstrate that the mature pancreatic cells develop through a common precursor cell stage, which express the genes Pdx1 and Ptf1a—and that differentiation toward the endocrine lineages occur through a transient Ngn3 expressing precursor cell, which is not shared with nonendocrine lineages.

In addition to the above genetic studies, tracing of cells can also be performed by marking individual cells at one given stage in development, followed by investigations of the phenotype of the progeny. Fishman and Melton recently obtained further evidence for the existence of a common endo/exocrine precursor cell using a replication-deficient retroviral labeling of pancreatic explants (Fishman and Melton, 2002). This study complements the Cre-LoxP methods and highlights the usefulness of such additional approaches, which has not been exploited to any large extent in the developing pancreas. Such experiments, as well as dedicated clonogenic assays should be capable of revealing the level of lineage relationships, lineage restrictions, and timing of developmental competence in the embryonic pancreas.


Overview of the Genetic Network of Pancreatic Development

Multiple genes have now been shown to play a role during pancreatic development and make for a complex and intricate interplay. Supplementary Figure 1 (which is available online at is a schematic linkage of some currently demonstrated interactions amongst genes involved in pancreatic development, and these are arranged by their actions at given points in the developmental programs. The first genes that come into play are involved in establishing an endodermal fate and will play a role during gastrulation as the endoderm forms (e.g., Sox, Mix genes). After endodermal establishment, certain genes are positioned in a “core endodermal program,” which includes several HNF-type genes (e.g., Hnf1β [tcf2], HNF1α [tcf1], Hnf4, Hnf3β [Foxa2], Hnf6 [Onecut1]), all of which become more or less restricted in the specialized endodermal cells at late stages. One of these, Foxa2, is positively autoregulated, suggesting an intrinsic stability of this particular state. Many of these genes play a role in tissue-specific gene regulation in adult cells, including pancreas, liver, lung, and intestinal gene expression. In those cases, these factors appear to cooperate with more tissue-specific regulators, such as Pdx1 (ipf1; pancreas), gata4 (intestine, liver), or c/ebp-α or COUP-TFII (liver). The specification of the more specialized compartments of the endoderm will be determined by spatial activation of certain genes in the network by—to a large extent—unknown morphogenetic inputs, themselves regulated by the intrinsic factors. Thus, pancreatic determination involves the activation of Pdx1 and HB9, immediately followed by further activation of other genes (e.g., Nkx2.2, Nkx6.1). In comparison, intestinal development relies on the activation of Cdx1 and Cdx2, which helps secure the intestine-specific gene expression pattern. The pancreatic determination state is not permanent—cells will acquire their more specialized terminal fates over time, and this specialization involves the further involvement of cell intrinsic regulators. In the case of the endocrine pancreas, this includes the activation of neurogenin3 (atoh5) followed by several genes that may be common between endocrine cells (e.g., Pax6, Isl1, NeuroD), or cell-specific (e.g., Pax4, Brn4), among endocrine types in the pancreatic islet. Analogously, exocrine development relies on the continued activation of the Ptf1a gene, followed by Mist1. Certain genes, such as the nuclear receptor 5A2 (nr5a2) or gata4 might play a role likewise, but these genes are not specific to the pancreas, again showing that the combination of a gene set is determining tissue specificity and not the single action of a highly specific master regulator.

Details for all genes described in this review can be found in Supplementary Table 1 (which is available online at This table contains hyperlinked information to various Web-based resources providing further information regarding the individual genes in the mouse. Genes encoding various terminal products of the endocrine, exocrine, and ductal cells of the pancreas are also included.

Endodermal Establishment and the Core Endodermal Program

Endoderm Formation.

Understanding pancreatic development, wherein genes “stand upon other genes,” requires an appreciation of epistatic relationships among genes of the common endodermal program and those more pancreas-specific. The subject of endoderm formation has been reviewed excellently by others (Wells and Melton, 1999; Hogan and Zaret, 2002; Zaret, 2002). Briefly, the endoderm forms during gastrulation from the set of cells in anterior proximity to the node, which take up position on the ventral surface of the egg cylinder in the mouse embryo after migration through the primitive streak. Most information regarding endoderm induction is available from studies of the developing Zebrafish, Danio rerio, and the African clawed frog, Xenopus laevis. These results also excellently demonstrate the value of first understanding endodermal development in nonmammalian species, as the results to a very large extent can be superimposed onto the mouse. In fish, a genetic network consisting of the Mix-like homeodomain gene bonnie-and-clyde (Bon) and the HMG-box gene Casanova (sox32, alike Sox17) helps induce the Sox-type gene Sox17, which acts to secure endoderm development (Alexander et al., 1999). Moreover, the TGFβ/bone morphogenetic protein [BMP]-type ligand Nodal, or a constitutively active type BMP receptor, establishes Casanova expression by means of Mixer, revealing the involvement of the BMP-signaling system (Alexander and Stainier, 1999). In Xenopus, high concentrations of the BMP/TGFβ-like ligand activin induce endoderm development, and XenopusSox17 and MIXER are both sufficient to induce an endocrine program (Henry and Melton, 1998). Of interest, activin has also been found to induce the Xlhbox8 gene (Gamer and Wright, 1995), which is the orthologue to mammalian Pdx1, perhaps suggesting that pancreatic formation occurs at high levels of activin-like signaling. However, activin might not be the true BMP/TGFβ-type ligand operating to establish the endodermal program. More likely, the inducing substance is a heterodimer consisting of members of the Xnr (Xenopus nodal-related) genes (Takahashi et al., 2000) or in mammals, orthologues of these. A close homologue to MIXER exist in mammals (Mix1) and is capable of inducing the endodermal program if expressed in Xenopus (Mohn et al., 2003). Similarly, the mammalian orthologue to Sox17 is also required for endoderm formation (Kanai-Azuma et al., 2002). Nodal/BMP-like signals may also be involved in mammalian endoderm induction, as mice deficient for the BMP signal transducing factor SMAD2 (madh2) fail to generate endoderm (Tremblay et al., 2000). Similarly, embryonic stem (ES) cultures deficient in the heterodimeric partner to SMAD2, SMAD4 (madh4), do not generate endoderm (Yang et al., 1998). The BMP/TGFβ ligand (or combination) that may induce mammalian endoderm has yet to be found.

Gata family genes appear to link the early endodermal establishing genes (Mix-type, Sox-type) to the later endodermal-specific genes. In both Zebrafish and Xenopus, Gata5 is involved in endodermal development. Zebrafish mutations in faust (Gata5) result in a strongly reduced endodermal compartment, with decreased expression of the axial gene (ortholog of Foxa2) as well as sox17 (Reiter et al., 2001). In frogs, Gata5 expression is directly stimulated by activin, and overexpression of Gata5 is capable of inducing endoderm in the animal cap assay (Weber et al., 2000). The mouse and human genome contain close homologues to gata5 in zebrafish and frog. A specific role for mouse Gata5 in endodermal development has not been reported. Mice mutant in Gata5 display late defects in genitourinary function and are viable (Molkentin et al., 2000b). It is possible that the mouse Gata6 gene (described later) could be redundant to mouse Gata5. An alternative possibility is that Gata6, although displaying weaker similarity to frog and fish Gata5, is the true functional orthologue and that mouse Gata5 plays no role in endodermal development.

A key gene in endodermal development in most species is the forkhead gene Foxa2 or its orthologues like axial (Ang et al., 1993; Ang and Rossant, 1994). By using Foxa2 as a marker for endodermal development, Guo et al. discovered that, for endodermal differentiation, an absence of repression of the endodermal fate might be as important as positively acting instructive signals. By studying endoderm induction in ES cells, it was found that Foxa2 expression was promoted by Foxd3 (genesis) and inhibited by the POU-domain factor Oct4 (Guo et al., 2002). Thus, endodermal differentiation could be regulated by a reduction, or loss, of Oct4 (pou5f1) expression. In relation to the role of Foxd3, this is an interesting observation as Oct4 (pou5f1) and Foxd3 are normally both expressed in the undifferentiated ES cell state, and Foxd3 has been shown to play a role in the developing neural crest (Kos et al., 2001) but is not expressed in the developing endoderm. Most of the conclusions of the above study were based on transfection analyses, and it seems likely that overexpression of Foxd3 might mimic yet another member of the Fox family in its endoderm-inducing capability. Nonetheless, the results underscore the importance of a threshold level of Fox family member activity in the establishment of an endodermal fate. In the normal embryo, Foxa2 may be the gene of importance, as those cells becoming endodermal arise from the population of Foxa2-expressing cells of the preendodermal domain of the epiblast (Ang et al., 1993).

A “Core Endodermal Program”—and Variations Thereof.

In the mouse, the time from the late-streak gastrula (E7.5) to the onset of visible organogenesis (E8.5, 6–10 somites) is only 24 hr, and during this time, prepatterning events divide the endoderm into independent compartments. As examples, in the anterior position, TTF/Nkx2.1(Nkx2.1) is involved in thyroid and lung development (Kimura et al., 1999; Yuan et al., 2000), and more posteriorly, cdx2 governs intestinal development (Beck et al., 2003). In between, the gene hex regulates liver and thyroid development (Martinez Barbera et al., 2000; Keng et al., 2000). Genes belonging to the 4 Hox-clusters become extensively divided in endodermal compartments (Sekimoto et al., 1998; Kawazoe et al., 2002; Beck, 2002). However, these genes are not specific to the endoderm and are equally important in patterning the other germ layers, where they have been studied in more detail (Hunt and Krumlauf, 1991; Tam and Trainor, 1994; Trainor and Krumlauf, 2001). A smaller network of more restricted genes cross-interact in the early endodermal cells and are extensively involved in tissue-specific regulation also in multiple fully differentiated endodermally derived cell types, including the pancreas. I will refer to this set of genes as a “core endodermal program” (Supp. Fig. S1, which is available online at with the caveat that not all members are expressed by all endodermally derived populations later, and some are even expressed in nonendodermal cells. It should also be noted that smaller variations in expression and changes in posttranslational modifications of the various genes and their products might be important in fate allocation amongst endodermal cells, but this question has largely not been studied.

The Foxa2 gene is key regulator and participates in lung- (Costa et al., 2001), liver- (Cirillo et al., 2002), and pancreas-specific (Cockell et al., 1995; Wu et al., 1997) gene expression in the adult. Mice lacking Foxa2 fail to generate fore- and midgut endoderm (Ang and Rossant, 1994). Foxa2, together with Gata4, is involved in chromatin opening of the albumin promoter and is thus important in initiating expression of liver-specific genes (Cirillo et al., 2002). Surprisingly, Foxa2 is not required for maintenance of liver gene expression, as expression of only few genes change in liver-specific mutants of Foxa2 (Sund et al., 2000). This result may be due to redundant activities of other Fox family members such as Hnf3α (Foxa1), and/or Hnf3γ (Foxa3) (Sund et al., 2000). Foxa2 is regulated by several other genes in a more upstream position, in addition to the very likely inputs from the intercellular signals necessary for endoderm establishment during gastrulation. In the liver and pancreas, the nuclear receptor nr5a2 may help maintain Foxa2 expression (Pare et al., 2001). Also, the homeodomain factor Hex may be important, because in its absence, Foxa2 expression in liver cells but not in foregut endoderm is lost. However, it is uncertain whether this finding is due to a loss of hepatic precursors or to a loss of positive regulation of the Foxa2 promoter (Martinez Barbera et al., 2000).

A possible downstream target of Foxa2 is Hnf4, another integral member of the core endodermal program (Bailly et al., 2001). This steroid receptor is initially required for development of the visceral endoderm (Chen et al., 1994). Hnf4 is subsequently required for normal liver cell differentiation (Li et al., 2000). This role is probably exerted by means of regulation of Hnf1α expression by HNF4 (Hayhurst et al., 2001) as Hnf1α (Tcf1) expression is absent in Hnf4-/- livers. In contrast, other members of the core endodermal program (Foxa1 [Hnf3α], Foxa2 [Hnf3β], Foxa3 [Hnf3γ], Hnf1β [Tcf2], Hnf6 [onecut1], Gata4) are unaffected in Hnf4-/- embryonic liver (Li et al., 2000). HNF4 and HNF1α mutually support the expression of each other in hepatoma cells, and either one promotes the liver program (Bailly et al., 1998). In addition to the liver, HNF4 participates in tissue-specific regulation in the adult intestine and pancreas (Sladek et al., 1990; Hayhurst et al., 2001). It is expressed throughout pancreatic development, but its role in this process has not been investigated. However, the recent availability of the LoxP-flanked Hnf4-/- mice will allow for such a study (Parviz et al., 2002). The target of HNF4, Hnf1α (Tcf1), is a variant homeodomain-type factor that is involved in both liver- and pancreas-specific gene expression. It is expressed in undifferentiated pancreatic precursors, as well as in mature exocrine and endocrine cells (Nammo et al., 2002). In the adult pancreas, both Hnf1α (Tcf1) and Hnf4 are important for adult β-cell function. Hypomorphic mutations of Hnf4 are responsible for MODY1 (maturity onset diabetes of the young) (Yamagata et al., 1996a) and Hnf1β (Tcf1) mutations are responsible for MODY3 (Yamagata et al., 1996b). The finding that compound double-heterozygous mutants of Hnf1α (Tcf1), Hnf4, or Foxa2 with Pdx1 display profound defects in β-cell function (Shih et al., 2002) supports a hypothesis that these genes operate as a network in the fully developed β-cell.

Two genes, Gata6 and Hnf1β (Tcf2), are positioned upstream of the Hnf4/Hnf1α pair. GATA6 and HNF1β activate the Hnf4 promoter (Morrisey et al., 1998a). Loss of Gata6 leads to a loss of extraembryonic development, which could suggest that the GATA6-HNF4 interaction at that stage is critical (Morrisey et al., 1998a; Koutsourakis et al., 1999). Gata6 is not required for the establishment of the endoderm but is involved in its differentiation (Morrisey et al., 1998b). Embryoid bodies deficient in Gata6 fail to activate multiple endodermal markers, including Hnf4, Foxa2, and Gata4 (Morrisey et al., 1998b). Gata6 is not limited to a role in the early endoderm, and is expressed by multiple nonendodermal cell types (Narita et al., 1996). In the lung, gata6 is required for branching morphogenesis (Keijzer et al., 2001; Yang et al., 2002), and in the heart, gata6 is involved in cardiogenic mesoderm development (Molkentin et al., 2000a). Also, gata6 is highly expressed in tissues such as ovary, brain, and mammary glands. HNF1β, on the other hand, is much more restricted to endodermally derived tissues such as liver, lung, and pancreas, although transcripts have been identified from kidney and uterus as well. HNF1β has also been assigned as a MODY-gene, MODY5 (Horikawa et al., 1997), again highlighting the conservation of the network in the β-cells. In kidneys, mutations of HNF1β are responsible for glomerulocystic kidney disease (Bingham et al., 2001; Sun and Hopkins, 2001), and Hnf1β was recently shown by Sun and colleagues to play a role in several aspects of zebrafish endodermal development, including the proper establishment of Pdx1 and Shh expression in the gut (Sun and Hopkins, 2001). In mice, loss of Hnf1β causes loss of visceral as well as definitive endodermal development, including the absence of Hnf4, Hnf1α, and Hnf3γ expression (Coffinier et al., 1999; Barbacci et al., 1999). Liver-specific loss of Hnf1β leads to a loss of the development of intrahepatic bile ducts and causes dysplasia of gall bladder epithelium (Coffinier et al., 2002). Thus, Hnf1β seems to play a role in both establishment of the endodermal program, the specialization of fates beyond this, and also occupies a central position in adult cells.

The last two gene products that can presently be assigned at least subroles within the endodermal program are Hnf6 and nr5a2. Both HNF6 (onecut1, a CUT-domain homeodomain transcription factor) and nr5a2 (LRH/PHR/FTF) are expressed predominantly in liver and pancreas and not in other endodermal derivatives. Thus, these genes interact with the core program within the domain of the liver and pancreas and not outside it. Hnf6 appears to play a role in pancreatic determination, and may be one of the important bridging components between the genes of the core endodermal program and those more pancreas-specific. Hnf6-/- mice are deficient in endocrine pancreatic development, where the gene is required to establish Ngn3 activation within pancreatic progenitor cells (Jacquemin et al., 2000). However, Hnf6 plays an even earlier role, as Hnf6 mutants display severe ventral pancreatic hypoplasia and a deficiency in bile duct development (Clotman et al., 2002). It was recently shown that Hnf6 is both expressed before Pdx1 activation in the prospective pancreatic domains and the protein is capable of activating the Pdx1 promoter in transfection studies (Jacquemin et al., 2003). HNF6 also activates the Foxa2 gene (Rausa et al., 1997). Thus, parallel stimulation of the Foxa2 promoter may help establish different levels of FoxA2 protein. This might be important if Foxa2 also acts by means of a threshold mechanism later in endodermal development. Hnf6 gene expression itself is activated by both HNF1β and HNF4 (Lahuna et al., 2000). Together, these observations might suggest that a gradual pancreas specification mechanism seem to come about by means of a refinement in expression due to cross-regulatory activities of select members of the core endodermal program.

The role of Nr5a2 is less well understood. Nr5a2 is expressed early in endodermal development in the liver and pancreas domain (Rausa et al., 1999). Certain observations have cast the Nr5a2 gene onto center stage in liver transcription. Nr5a2 regulates α-fetoprotein expression (Galarneau et al., 1996) but is also capable of activating the Hnf1α, Hnf4, and Foxa2 promoters through binding sites in the upstream regions (Pare et al., 2001). Therefore, even if restricted to the liver and pancreas, Nr5a2 interacts intimately with the central members of the core endodermal program. Analysis of the Nr5a2 promoter region also revealed potential GATA binding sites, suggesting that the early endodermal factors (probably GATA6 or GATA4) may help establish Nr5a2 expression (Pare et al., 2001). Additionally, Hnf1β and Foxa2 may activate the Nr5a2 promoter (Zhang et al., 2001). Little is known regarding the involvement of Nr5a2 in pancreatic gene expression, but it would be expected that the factor could regulate Hnf1α, Hnf4, and Foxa2 there. Notably, Nr5a2 is only expressed in pancreatic ductal and exocrine cells in the mature organ (Rausa et al., 1999; Zhang et al., 2001), demonstrating that Nr5a2 does not participate in the Hnf1α, Hnf4, Foxa2 expression network in adult β-cells. On the other hand, it is possible that Nr5a2 participates directly in exocrine gene transcription.

In summary, individual endodermal cells integrate the use of a common set of genes throughout their development. Some of these factors are more restricted than others, but to offer true organ specificity another layer of genes are needed, super-positioned on the common program. For the pancreas, such specificity begins with Pdx1.

Pancreatic Determination


The Pdx1 (Ipf1) gene was the first gene shown to be cell-autonomously required for formation of the pancreas in mice (Jonsson et al., 1994) and humans (Stoffers et al., 1997b), and its identification served to trigger the modern era of molecular pancreatic developmental research. It was initially isolated by three independent groups as the A1-element binding protein of the insulin promoter (Ohlsson et al., 1993) and the TAAT-binding factor of the somatostatin upstream enhancer (Leonard et al., 1993; Miller et al., 1994) and only subsequently emerged in a developmental context when the gene was deleted. Expression of Pdx1 begins in epithelial cells of the dorsal and ventral pancreatic anlage at the 10-somite stage (E8.5). Lower expression is observed in adjacent prospective duodenal and posterior stomach cells (Fig. 1). Pdx1 persists as homogeneously expressed in the pancreatic epithelium as this subsequently evaginates and branches over the following period. The early glucagon-producing endocrine cells do not express Pdx1. At the onset of the secondary transition, heterogeneous expression of Pdx1 becomes clear. Insulin-expressing β-cells up-regulate Pdx1 expression, and differentiating exocrine cells gradually lose Pdx1, although continuing to express the protein at a low level. Ductal-type cells also down-regulate Pdx1, and eventually only β-cells and a subset of δ-cells express Pdx1 at levels readily detectable using standard histochemical techniques (Øster et al., 1998). Pdx1 is known to be involved in several aspects of adult β-cell function, including activation of the insulin (Ins), IAPP (Iapp), and glucokinase (Gck) promoters, but these roles will not be discussed here (Peers et al., 1994; Petersen et al., 1994; Serup et al., 1996; Chakrabarti et al., 2002). Importantly, Pdx1 haploinsufficiency has been demonstrated to affect adult β-cell function, where heterozygous Pdx1+/- mice display reduced insulin secretion, and reduced Glut2 expression (Brissova et al., 2002). Thus, the protein level of Pdx1 dictates important aspects of β-cell function.

Pdx1 null embryos display strong dorsal and ventral pancreatic hypoplasia, with limited glucagon-cell differentiation, lack of development of the Brunners glands in the duodenum, and failed specification of gastrin-producing stomach endocrine cells (Jonsson et al., 1994; Offield et al., 1996). Initial bud formation occurs, showing that Pdx1 is not essential for this aspect of pancreas development. The presence of endocrine cells is noteworthy, as the cre-lox lineage tracing studies reveal that endocrine cells derive from Pdx1+ precursors (Herrera, 2000; Gu et al., 2002). However, as these early endocrine cells do not require Pdx1 activity for formation, we can conclude that at least the early endocrine program can function independently of Pdx1. A requirement for Pdx1 is evident soon after pancreatic budding, which is manifest as a failure of the undifferentiated cells of the pancreas to expand. Several possibilities could account for this effect. For example, (1) Pdx1 is cell-autonomously required within the precursor cells allowing these to maintain their dividing state; (2) Pdx1 is required for production of a secreted factor, which could be involved in a feedback signaling mechanism to the adjacent pancreatic mesenchyme; or (3) a decreased number of endodermal precursors become specified to the pancreatic lineage. The third possibility does not seem to be the case, as the initial bud region is not different between WT and Pdx1 nulls (Offield et al., 1996). Instead, the first hypothesis has received support through studies of mice lacking Fgf10 (Bhushan et al., 2001). These mice display impaired pancreatic epithelial growth, which must be caused by a deficiency in mesenchymal/epithelial signaling as Fgf10 is only expressed in the mesenchyme (Bhushan et al., 2001). Importantly, Pdx1 expression is lost in the pancreatic epithelium of Fgf10 nulls (Bhushan et al., 2001), positioning Pdx1 as a downstream target of the FGF-signaling pathway. In Pdx1 mutants, the overall structure of the dorsal pancreatic mesenchyme and the formation of the spleen is normal (Ahlgren et al., 1996). Therefore, the dorsal pancreatic mesenchyme is capable of sensing local patterning mechanisms, suggesting that it is within the pancreatic mesenchyme that the genes governing the early pancreatic geometry of the pancreas are expressed.

The activity of Pdx1 is modulated by Pbx1. Pbx1 is capable of forming heterodimers in vitro with Pdx1, and is a member of the TALE family of homeodomain factors, which are known to confer binding specificity of other homeodomain partners. The Pbx1/Pdx1 complex may be important in discriminating the activity of Pdx1 in endocrine and exocrine cells, as the binding of Pbx1/Pdx1 and Pdx1 alone is not identical between the Pdx1-binding sites of the exocrine-specific elastase promoter and the A1 element of the insulin gene (Swift et al., 1994). Pbx1 is important for early pancreas development, as Pbx1-/- display a pancreatic hypoplasia, and reduced exo- and endocrine cell differentiation (Kim et al., 2002). This phenotype may be due to a lack of Pbx1/Pdx1 heterodimers in the pancreatic precursor cells, as Pdx1-/- mice rescued with an Pbx1-interaction–defective mutant of Pdx1 also display a strongly reduced precursor cell mass (Dutta et al., 2001). The heterodimeric Pdx1/Pbx1 complex is also important in the adult β-cell, as mice double heterozygous for Pdx1 and Pbx1 mutant alleles develop a more severe diabetes and stronger hypoinsulinemia than single mutants of the genes (Kim et al., 2002).

Several factors of the core endodermal program activate Pdx1, including Foxa2, Hnf1α, and Hnf1β (Wu et al., 1997; Gerrish et al., 2001; Lee et al., 2002b; Samaras et al., 2002). Most likely, Pdx1 is acting together with these factors as well as HNF4 in the adult β-cell, as Pdx1 itself is a known MODY gene (MODY4; Stoffers et al., 1997a). Pdx1 is autoregulated (Gerrish et al., 2001; Chakrabarti et al., 2002), suggesting that Pdx1 expression imparts intrinsic stability to the pancreatic state. The definition of the Pdx1 expression domain is influenced by the local environment at the primary transition, which include signals from the notochord (Kim et al., 1997; Hebrok et al., 1998), that help to repress endodermal shh expression in the prospective pancreas domain. However, at early time points the notochord is aligned with the nonpancreatic endoderm, so other signals are also required to define pancreatic specificity. A recent suggestion is that contact with the vascular tissue of the dorsal and ventral aortas, which gets into proximity with the pancreatic domains at E9.0–E9.5, is important. On the dorsal side, this occurs due to the midline fusion of the two dorsal aortas, which forces a dorsal movement of the notochord away from the endoderm (Lammert et al., 2003). Indeed, Lammert and colleagues demonstrated that endothelial tissue, including nonpancreatic types, such as the umbilical artery would induce Pdx1 expression in prepancreatic endoderm (Lammert et al., 2003). Also, mice transgenic for a pPdx1-VEGF DNA fragment were found to contain increased numbers of endocrine cells in the pancreas, and insulin cells were developing in the stomach within the domain in which the pdx1 promoter was active. In addition to the progressive signaling by notochord, and endothelial tissue, the pancreatic mesenchyme itself may play a role in stabilizing the pancreatic program. For instance, the expression of FGF10 in the distal-most pancreatic mesenchyme (Bhushan et al., 2001) may help to increase expression of Pdx1, thereby strengthening a progressive pancreatic fate assignment. Our finding that ectopic FGF10 signaling in the pancreas is capable of maintaining Pdx1-expression by arresting cells in the precursor cell state supports such a possibility (Norgaard, Jensen et al., in press).


The NK family member Nkx2.2 is a likely target for Pdx1. Nkx2.2 is expressed in the precursor epithelium slightly after Pdx1 activation. It persists here during precursor cell expansion, but is later restricted to endocrine α, β, and PP cells. As neither α, nor PP cells express Pdx1, it is clear that Pdx1 does not play a direct role in Nkx2.2 gene regulation in those cells. Targeted deletion of Nkx2.2 does not lead to obvious defects in the expansion, or phenotype of the precursor cells. Nkx2.2 deficiency results in impaired maturation of endocrine cell types at the secondary transition. Sussel and coworkers found that in Nkx2.2-/- mice endocrine cells differentiate in comparable numbers to that of wild-type littermates (Sussel et al., 1998). However, in absence of Nkx2.2, β-cells fail to activate the insulin gene and display a loss of Nkx6.1, suggesting that Nkx2.2 is essential for specification of the mature β-cell phenotype (Sussel et al., 1998). Nkx2.2 has a complex genomic structure containing three independent promoters. The islet-specific 1a promoter contains binding sites for Foxa2 and NeuroD. Like Pdx1, Nkx.2.2 immunoreactivity is strongly increased after β-cell differentiation. It is conceivable that Pdx1 and Foxa2 activate Nkx2.2 in the pancreatic precursor cells, and after β-cell differentiation, Nkx2.2 is further boosted by the presence of NeuroD. Nkx2.2 is capable of binding the insulin promoter in vivo (Cissell et al., 2003), and binding to the Pax4 promoter has also been noted (Cissell et al., 2003).

Nkx6.1 and Nkx6.2.

Nkx6.1, which is specific for β-cells in the adult pancreas (Jensen et al., 1996), is another potential target for Pdx1. Analysis of the Nkx6.1 promoter suggested that the gene is activated by both Nkx2.2 and Pdx1 (Watada et al., 2000) and conditional mutation of Pdx1 in differentiated β-cells lead to loss of differentiated characteristics, alongside loss of Nkx6.1 (Ahlgren et al., 1998). Nkx6.1 shares a large overlap in early endodermal expression with Pdx1 (Øster et al., 1998; Watada et al., 2000). It is expressed as the dorsal and ventral buds evaginate, and can be detected shortly after Pdx1 activation. Nkx6.1 is also expressed in the developing neural system, where it is involved in neuronal fate specification (Vallstedt et al., 2001). In the endoderm, Nkx6.1 is completely restricted to cells developing as pancreas, absent from adjacent endodermal tissues. As a gradient of Pdx1 immunoreactivity is observed between duodenal/stomach cells (low levels) and the pancreatic precursors (high levels), it is conceivable that Nkx6.1 is only activated where Pdx1 protein levels exceed a given threshold. Two overexpression studies during embryogenesis support such a relationship. By electroporating developing chick endoderm with an expression construct for Pdx1, Grapin-Botton and colleagues observed that pancreatic budding and Nkx6.1 activation could be initiated in endodermal domains outside the pancreas region (Grapin-Botton et al., 2001). However, no endocrine development was observed, again showing that initiation of this genetic program is dependent on other factors besides Pdx1 and Nkx6.1. Elevated expression of constitutively active Notch-1 (NICD) using the Pdx1 promoter, leads to increased endogenous Pdx1 expression and ectopic activation of Nkx6.1 within the posterior-most stomach epithelium, which leads to a pancreatic fate of cells in that endodermal region (Hald et al., 2003). Also, transgenic rescue experiments in Pdx1-/- mice have revealed that the proper physiological function of β-cells require a specific threshold level of Pdx1 (Gannon et al., 2001). In the same study, it was additionally shown that independent transgenic lines carrying Pdx1resc transgenes failed to fully rescue the developing pancreas, and notably the growth of the dorsal pancreas. Therefore, Pdx1 may also help pattern the endoderm by means of a threshold mechanism.

Mutants of Nkx6.1 do not exert any obvious effects within the pancreatic precursor cells, suggesting that Nkx6.1, like Nkx2.2, is dispensable for embryonic pancreatic growth (Sander et al., 2000). However, Nkx2.2 and Nkx6.1 are not redundant during this period, as the pancreas phenotype of double mutants of these two genes is indistinguishable from that of Nkx2.2 mutants (Sander et al., 2000). Nkx6.1, nevertheless, is essential for β-cell formation. A complete absence of mature β-cells but normal development of α-cells is observed in Nkx6.1-/- pancreas (Sander et al., 2000). Ngn3 expression is not reduced in Nkx6.1 mutants showing that Nkx6.1 is required downstream of endocrine cell selection. Thus, Nkx.6.1 is essential to stabilize the β-cell phenotype, rather than in the induction of the endocrine program, despite its expression during that transition. In the neural tube, Nkx6.1 acts as a repressor (Vallstedt et al., 2001). In the pancreas, no direct targets for Nkx6.1 has been observed; hypothetically, genes conferring α-cell specificity, such as Brn4 or Arx1 (see note added in proof), are good candidates. Another less-studied member of the NK family, Nkx6.2 (also known as Gtx), is also expressed in the developing pancreas (Øster et al., 1998). Here, Nkx6.2 is restricted to the differentiating glucagon cells that do not express Nkx6.1. It is conceivable that reciprocal activity of Nkx6.2 and Nkx6.1 may be decisive in the α- vs. β-cell fate.


HlxB9 is a homeodomain factor expressed in mature β-cells. Developmentally, expression of Hlxb9 is observed along the dorsal endoderm at the 8-somite stage extending beyond the dorsal pancreas and in the ventral pancreatic domain but not outside this (Harrison et al., 1999; Li et al., 1999). Hlxb9 is also expressed in the notochord. Endodermal expression of Hlxb9 is transient, but remains expressed in both pancreatic buds at E10.5. Mice lacking Hlxb9 display dorsal bud agenesis with an even stronger phenotype than that of Pdx1-/- mice, as no dorsal bud is initiated. The ventral bud develops normally (Harrison et al., 1999; Li et al., 1999). The lack of dorsal bud formation in Hlxb9 mutant mice is not completely understood. Due to the spatial alignment along the anteroposterior axis, it seems likely that the notochord may help establish Hlxb9 expression in the dorsal endoderm along its entire length, which in turn sets a permissive state for pancreatic development, whereas other signals help to establish the proper spatial position for budding. On the opposite side of the notochord at E8.5, Hlxb9 marks developing motoneurons, a cell type induced by shh originating from the notochord and the floor plate, raising the possibility that shh might also be critical for Hlxb9 expression in the pancreas, although this requirement has not been tested. It thus appears that Hlxb9 might be upstream of Pdx1 in the dorsal pancreatic program. As Pdx1 is not activated in the entire Hlxb9 domain, other signals must be present to restrict Pdx1 expression, and such an epistatic relationship does not seem to exist in the ventral pancreas, as Hlxb9 expression does not precede Pdx1 there.

Therefore, a likely possibility is that several modes of Hlxb9 function, temporally defined, exist. Such a temporal separation of functions is also observed for Pdx1, as previously described. First, HlxB9 is initially activated by notochord signaling, and only initiated on the dorsal side, followed by a second effect where Pdx1, or other pancreas-specific genes, help to maintain Hlxb9 expression in both pancreatic domains. This could explain the concomitant expression of Hlxb9 and Pdx1 in the ventral pancreas and help explain the expression of Hlxb9 in the dorsal domain a full day after the notochord has moved away from the dorsal endoderm. Later, HlxB9 has a role in the adult β-cell, as those β-cells that form in the ventrally derived pancreas, fail to express Glut2 at normal levels. Importantly, the expression of Hlxb9 needs to be controlled temporally. Overexpression of Hlxb9 from the Pdx1 promoter, which extends the expression of Hlxb9 through the early period, is detrimental to pancreatic development (Li and Edlund, 2001), resulting in a decrease of endocrine and exocrine cell differentiation and causing cells within the pancreas to adopt intestinal fates (Li and Edlund, 2001). As this phenotype reflects that of the Ptf1a mutant mice (Kawaguchi et al., 2002), it raises the possibility that HlxB9 is capable of negatively regulating Ptf1a expression in the pancreatic domains. Such a possibility is supported by the observation that HlxB9 can function as a repressor in the developing neural tube (Thaler et al., 1999; William et al., 2003).

Exocrine Differentiation


Studies of transcriptional control of exocrine pancreatic genes led to the identification of an exocrine-specific promoter binding complex named PTF1 (Petrucco et al., 1990; Rose et al., 1994). The PTF1 complex consists of possibly two different basic helix-loop-helix (bHLH) A/B-class heterodimers formed by the B-class bHLH member p48, and the ubiquitous A-member E2A or HEB (Rose et al., 2001). The Ptf1a gene encodes the tissue-specific component (p48) of this complex and was initially found to be essential for exocrine pancreatic development. Mice deficient in Ptf1a failed to form an exocrine pancreas (Krapp et al., 1998). p48 has been shown to be expressed as early as E9.5, several days before the onset of exocrine development and suggestive of a role of Ptf1a beyond that of exocrine specification. By generating a knock-in of the Cre-recombinase into the Ptf1a gene, disrupting the Ptf1a gene in the process, Kawaguchi et al. found that Ptf1a was required for maintaining cells in a pancreatic state (Kawaguchi et al., 2002). Taking advantage of the Cre-recombinase, which allowed for concomitant lineage tracing, the Ptf1a null cells were shown to follow a duodenal fate choice, contributing to all lineages in the gut (Kawaguchi et al., 2002). Such a specification role was unexpected and Ptf1a, therefore, should not be viewed as a simple exocrine pancreatic specification gene. This hypothesis was corroborated by the finding that in heterozygous Ptf1a-CRE mice, which are not defective in exocrine development, the contribution of Ptf1a-expressing cells extended to both endocrine- and ductal cell types (Kawaguchi et al., 2002). Thus, p48 plays a role in uncommitted pancreatic precursors and does not program the exocrine fate simply by its expression. Reciprocally, endocrine or ductal development does not seem to rely on specific absence of Ptf1a expression either, as some of these cells had clearly expressed the gene previously. Arguments questioning the efficiency of recombination, which could explain why not all pancreatic cell types appeared to go through a Ptf1a-expressing precursor, should not be disregarded. However, a plausible scenario is that Ptf1a-expression is finely regulated by morphogenetic signals (yet unknown) in the developing pancreas and that precursor cells may be guided by a dynamic balance of expression of a pro-exocrine (such as Ptf1a) gene vs. a pro-endocrine gene (such as Ngn3). In such a situation, some pancreatic precursor cells would be expected to express both genes simultaneously, and only those exceeding a given threshold of one vs. the other would commit to a fate choice. A colocalization study of these factors have not been made, although it is striking that a histologic analysis of these two proteins at the late stages of the secondary transition reveal only minimal overlap in their expression (Fig. 5). It will be highly interesting to define whether such mutually exclusive expression also exists at earlier stages, when Ptf1a is expressed in undifferentiated precursor cells. As cells displaying a mixed phenotype (i.e., amphicrine) of exo- and endocrine characteristics have not been observed in the nontransformed pancreas, one possibility is that a reciprocal repression mechanism of the two fates exists during development. Such a mechanism would probably be indirect, as neither Ptf1a nor Ngn3 normally act as repressors.

Figure 5.

Expression of Ngn3 and Ptf1a/p48 in early pancreatic development. Immediately adjacent sections (embryonic day 15.5 mouse pancreas) stained for Ngn3 and Ptf1a (p48). Ngn3-expressing cells are centrally localized (A), and their expression domain does not overlap that of p48 (B). p48 is expressed in the exocrine cells.


MIST1 is another member of the B-class bHLH factors and is strongly expressed in pancreas (Lemercier et al., 1997). MIST1 can heterodimerize with E2A factors and bind typical E-boxes (Lemercier et al., 1997). The recruitment of E2A factors suggests that its role is to promote gene activation. Mist1 is expressed in nonpancreatic tissues, including salivary glands and stomach, where it is expressed by the exocrine cells and chief cells, respectively (Pin et al., 2000). In the pancreas, MIST1 is likewise confined to exocrine cells, and absent in ducts (Yoshida et al., 2001). Although Mist1-LacZ heterozygous embryos display expression lacZ activity in some pancreatic precursor cells at E10.5 (Pin et al., 2001), the significance of this is unknown. The fate of these cells has not been determined. At E12.5, Mist1 expression has become heterogeneous, where expression is observed only in the peripheral epithelial regions most likely destined to develop as exocrine cells. This finding could suggest that Mist1 plays a role in the stabilization of exocrine fate. Mist1-/- mice display a loss of differentiated characteristics of the exocrine cells, leading to exocrine lesions subsequently undergoing what might be viewed as some form of a regenerative process (Pin et al., 2001). The cells in these lesions coexpress ductal and exocrine marker genes, revealing that Mist1 is required for maintaining a stable exocrine cell identity (Pin et al., 2001). Also, the exocrine cells in Mist1-/- mice do not have gap-junctions, which can be ascribed to a loss of connexin32 expression (Rukstalis et al., 2003). It is noteworthy that loss of Mist1 does not abrogate exocrine transcription. Perhaps redundant activities between Ptf1a and Mist1 ensure a continued activity of the exocrine terminally expressed genes. Of the genes known to play a role in the pancreas, Mist1 is the only one completely restricted to exocrine cells, and it can be hypothesized that Mist1 may be autoregulated to stabilize the exocrine state. Very likely, p48 will play a role in the establishment of mist1 expression, perhaps alike the cascade of Ngn3 and NeuroD operating in endocrine development. It will be interesting to see to what extent Mist1 interacts with members of the core endodermal program in exocrine function.

Endocrine Differentiation


The involvement of bHLH factors in pancreatic cell function was initially revealed by determination of the insulin promoter configuration. Adjacent to the Pdx1-binding A1-elements of the human and rat insulin promoters is a conserved E-box, E1, with the sequence CATCTG. This sequence element is critical for insulin gene activity, because, when mutated, the promoter activity drops to less than 20% (Whelan et al., 1990). In the rat insulin I promoter, a second E-box (E2) is found, and this element binds the same complex as the E1 element. A combined mutation of both E-boxes eliminates promoter activity (Karlsson et al., 1987). Although the full rat insulin I promoter was examined in this linker-scanning analysis, no other elements were observed to reduce activity this dramatically. Thus, the factors binding the E-boxes appear to be essential for the activity of other elements. Initially, the E-box binding complex was named IEF-1 (insulin enhancer factor-1) and thought only to be present in islet-cell lines (Ohlsson et al., 1988). It was later characterized as a heterodimeric complex, consisting of the ubiquitous E2A/Pan-factor (Nelson et al., 1990) and the tissue-specific Beta-2/NeuroD factor (Naya et al., 1995; Lee et al., 1995). NeuroD is expressed by several neuronal cell types, where it plays a role in inducing the neuronal program. This finding was initially shown by injection of NeuroD mRNA Xenopus oocytes, resulting in excessive and premature neurogenesis (Lee et al., 1995). In the pancreas, NeuroD is expressed in all endocrine cells, but only after these have undergone differentiation and cell cycle arrest. NeuroD is not absolutely critical for endocrine differentiation, as NeuroD-/- mice generate all islet cell types. However, the islet cell number is gradually reduced, and β-cells are found to undergo significant apoptosis prior to birth, the reason for which is unknown. The reduced β-cell mass leads to diabetes with a variable penetrance, and on certain genetic backgrounds, this phenotype does not manifest at all. As with several genes of the core endodermal program that operate in islets, mutations in NeuroD are associated with MODY (MODY6; Kristinsson et al., 2001). NeuroD mutations also associate with human type II diabetes. The downstream functions of NeuroD are poorly understood, other than its involvement in insulin gene transcription. NeuroD is autoregulated (Yoon et al., 1998), suggesting a role in stabilization of the endocrine fate. Although NeuroD is not required for endocrine differentiation, Adenovirus-mediated NeuroD transfection can initiate an endocrine program in human ductal cells (Heremans et al., 2002), and betacellulin treatment (a TGFα family member) combined with adenovirus-mediated NeuroD expression initiates an endocrine program in liver cells in adult mice (Kojima et al., 2003).

The expression of NeuroD precedes that of other endocrine postmitotic markers such as Pax6 and Isl1 (Jensen et al., 2000a). Therefore, NeuroD may be involved in promoting a cell cycle exit. Studies of other bHLH factors in, e.g., neuronal and muscle development, point to a role in cell cycle withdrawal. In the retina, forced expression of neuroD forces cell cycle arrest (Morrow et al., 1999). In the intestine, NeuroD induces cell cycle withdrawal of secretin cells (Mutoh et al., 1997), and this effect is inhibited by CyclinD1, which is normally expressed only in proliferating cells of the intestinal crypt (Ratineau et al., 2002). Mitotic silencing is possibly linked to the activation of CDK-inhibitors such as p16, p21, p27, and p57. For instance, Nex, a close homologue of NeuroD, activates p21 in PC12 cells (Uittenbogaard and Chiaramello, 2002), leading to mitotic arrest.


The establishment of NeuroD expression in the developing nervous system is dependent on another subclass of bHLH factors, the neurogenins. Sommer and colleagues were the first to describe the expression of a neurogenin family member, Ngn3 (Atoh5) (referred to as Ngn3 in the following), in the developing pancreas (Sommer et al., 1996). Ngn3 is expressed in epithelial pancreatic precursor cells prior to endocrine differentiation (Fig. 6; Jensen et al., 2000a; Schwitzgebel et al., 2000). This timing contrasts with NeuroD, as NeuroD is always expressed in nonepithelial endocrine cells (Jensen et al., 2000a). Ngn3 induces the endocrine fate choice as seen in pPdx1-Ngn3 mice, which display a massive conversion of pancreatic cells into glucagon-producing endocrine cells (Apelqvist et al., 1999). Also, adenovirus-mediated introduction of Ngn3 (Ad-Ngn3) into human pancreatic ductal cells can induce an endocrine program, whereby NeuroD, and other endocrine-specific markers are activated (Heremans et al., 2002). This is likely mediated by the activation of NeuroD, as Ad-NeuroD has the same effect. Accordingly, a Ngn3/E47 heterodimeric complex is capable of activating NeuroD expression through E-boxes residing in the proximal 1 kb of the NeuroD promoter (Huang et al., 2000). In developing chicken gut, Ngn3 is capable of inducing an endocrine program in endodermal cells outside the pancreas-specific domain (Grapin-Botton et al., 2001). Recently, Ngn3 transfection of endodermal ES-cells derived by retinoic acid treatment was shown to induce insulin gene transcription, and the activation of several other endocrine factors (Vetere et al., 2003).

Figure 6.

Ngn3 expression in epithelial precursor cells. Coexpression analysis of Ngn3 with the epithelial marker β-catenin. β-Catenin allows easy identification of the morphologic shape of the pancreatic cells, here clearly revealing that Ngn3-expressing cells are of a typical epithelial morphology. β-Catenin is strongly expressed in pancreatic progenitor cells. Note the highly variable intensity of Ngn3 expression. This finding could be the outcome of pro-endocrine morphogenetic signaling with a superimposed lateral specification mechanism.

In the mouse, Ngn3 expression is initiated at the pancreatic budding step, when glucagon cells first differentiate (Apelqvist et al., 1999; Jensen et al., 2000a). Hereafter, relatively low levels of Ngn3 expression are observed until E13.5, when the major peak of Ngn3 gene activation occurs, concomitant with the secondary transition. Absence of Ngn3 leads to a complete loss of all pancreatic cell types (Gradwohl et al., 2000). Ngn3 is also required for enteroendocrine cell development in the intestine, and in part for gastric endocrine development (Jenny et al., 2002; Lee et al., 2002a). At the secondary transition, expression of Ngn3 is clearly restricted to the central core, i.e., within a pro-endocrine domain. After differentiation, endocrine cells cease to express Ngn3. The activity and requirement of Ngn3 in endocrine differentiation, combined with its limited temporal expression, allows us to categorize this gene as pro-endocrine, in a very similar manner as the pro-neural genes in both Drosophila and vertebrates. In those cases, the Notch signaling system is integral to both fate assignment and precursor cell maintenance, exerting control over the activity of bHLH class members.

Notch Signaling

Notch Signaling During Pancreatic Development: Lateral Specification.

Notch signaling is the caretaker of binary cell fate decisions in most multicellular organisms. This particular process is referred to as lateral specification (Artavanis-Tsakonas et al., 1999; Fig. 7) and is operating in the developing vertebrate neural tube, where lateral specification occurs between neural progenitor cells (Fortini and Artavanistsakonas, 1993). This system also operates in the developing pancreas and the mechanism can explain the initially scattered distribution of endocrine cells within the pancreatic epithelium. Loss-of-function of various Notch pathway genes (Hes1, Dll1, Rbp-jκ) leads to up-regulation of Ngn3 and consequent increased endocrine formation (Apelqvist et al., 1999; Jensen et al., 2000b). Similarly, the increase in expression of Dll1 and Dll3 in Hes1 mutants (Jensen et al., 2000b), and the activation of Dll1 and Dll4 in pancreatic duct cells transfected with Adenovirus-Ngn3 (Heremans et al., 2002) also supports the lateral specification model as does the expression patterns of ngn3 and Hes1 at the secondary transition (Figs. 6, 9).

Figure 7.

Lateral specification. The selection process of a pro-endocrine precursor is accomplished by cell–cell signaling between epithelial neighbors. This signaling occurs through a progressive strengthening of the expression levels of Ngn3 in the differentiating cell (left), with a weakening in neighbors (right). The lateral signaling is mediated through the activation of a notch ligand (Delta) controlled by Ngn3. Ligand expression on the differentiating cell activates the notch receptor in the neighbor, in turn activating the bHLH-type Hes1 repressor gene. Hes1 effectively reduces Ngn3 expression in these cells, silencing other pro-endocrine inputs that the cell may receive. Possibly, as the differentiating endocrine cell delaminates, the previously suppressed cells may initiate a second round of fate selection, whereby such cells may become endocrine if these are still situated in a pro-endocrine field.

Figure 9.

Basic helix-loop-helix (bHLH) factors in transcriptional control. bHLH factors are pivotal in positive and negative promoter regulation, and control can be exerted at various levels. Here, a single E-box is shown, through which both negative and positive gene regulation is conferred. Often, promoters contain independent E-box–like sequences to which positively and negatively acting bHLH heterodimers bind. 1. Passive suppression of gene expression can occur through binding of B- and A-class members to proteins of the id family. Id-type bHLH members lack basic-domain residues and generate a non-DNA binding complex. 2. Active suppression is exerted by, e.g., members of the Hes family. These proteins are capable of DNA binding and recruit a histone deacetylase activity (e.g., HDACs) to target promoters. Long-range repression may be exerted due to multimerization of the groucho-type adaptor molecule. By deacetylation, adjacent nucleosomes (blue) remain closed at the target promoter. The actual DNA-binding complex may consist of a heterodimer of Hes and HERP family members (Iso et al., 2003). 3. In contrast, the opening of a target promoter may be conferred by binding of a heterodimeric complex consisting of a B-class tissue-specific bHLH member and a more ubiquitous A-class member. A-class, as well as some B-class members, are capable of recruiting coactivators of the CBP/p300 family, which act as histone acetylases facilitating promoter opening.

Notch Signaling During Pancreatic Development: Suppressive Maintenance.

Some observations suggest that Notch signaling is not limited to lateral specification in the pancreas. First, Notch1 and Notch2, both capable of activating hes1 gene expression, are ubiquitously expressed in pancreatic precursor cells between the first and second transitions (e.g., between E11 and E12). These cells also express Hes1. At this time point, there are only few endocrine cells developing, signified by few ngn3-expressing cells. Importantly, in reduced presence (Jacquemin et al., 2000), or complete absence of endocrine cells (Ngn3-/- pancreas, J. Jensen, unpublished observations), Hes1 expression continues. Therefore, a mechanism explaining the maintenance of Notch signaling needs to be discriminated from lateral specification.

This alternative role is probably connected to the maintenance of the precursor population in the pancreas, as it is in these Notch signaling is active. An enticing concept is the possibility that this mode of Notch activation is controlled by the mesenchyme. The pancreatic mesenchyme is required for precursor pool maintenance, as in its absence, endocrine differentiation occurs, and the epithelium fails to expand. FGF10, normally produced by the mesenchyme, is critical for this process (Bhushan et al., 2001). Loss of fgf10 leads to a loss of precursor expansion, but not pancreatic specification. This finding could suggest that FGF10 is a simple trophic factor for the pancreatic progenitor cells; however, some of our recent observations suggest that the role of FGF10 may not be so simple. Mice overexpressing FGF10 within the pancreatic epithelial cells display pancreatic hyperplasia and arrested development signified by a block in exocrine, ductal, and endocrine cell differentiation in addition to a continuing Pdx1 expression (Norgaard et al., in press). Importantly, the arrested epithelial cells also remain strongly positive for Notch1 and Notch2 as well as Hes1, whereas Ngn3 expression is abrogated. Thus, ectopic FGF10 signaling is capable of maintaining the notch-system in an active state, which in turn may repress differentiation. The mechanism by which increased FGF10 signaling leads to maintenance of Notch signaling is not fully understood. However, we have observed that the notch ligand genes encoding Jagged1 (Jag1) and Jagged2 (Jag2) are expressed throughout the FGF10-expressing arrested epithelium. Therefore, Jagged-mediated activation of Notch might be critical during precursor cell expansion prior to the secondary transition and can help explain the expression of Jag1 and Jag2 in normal undifferentiated pancreatic progenitors (Mitsiadis et al., 1997; Jensen et al., 2000a; Norgaard et al., in press). To distinguish between the two modes of notch signaling, we decided to refer to this latter mechanism as “suppressive maintenance” (Norgaard et al., in press).

Establishing a Pro-Endocrine Field During Notch Signaling.

Acknowledging two such modes of notch signaling, an important aspect of Ngn3 gene activation surfaces: its initial activation will have to operate against a Notch/Hes1 threshold, and its establishment must be independent of this. Therefore, we can hypothesize the existence of a morphogenetic signaling mechanism(s?) that will activate the Ngn3 promoter independently of Notch signaling, which also helps to restrict the Ngn3 activation spatially to the central core of the pancreas at the secondary transition. A similar argument pertains to the initial activation of Ngn3 at the primary transition, but the signal(s?) needs not to be identical to that/those of the secondary transition. One observation that suggests two modes of Ngn3 activation is the result of the Hnf6-/- mice. Hnf6-/- pancreas is almost devoid of endocrine cells, with an almost complete loss of Ngn3 expression (Jacquemin et al., 2000). The Ngn3 promoter contains binding sites for Hnf6 in its distal enhancer and Hnf6 can activate a Ngn3 promoter fragment in heterologous cells (Jacquemin et al., 2000). However, the early glucagon cells (these are dependent on Ngn3), still form in Hnf6-/- pancreas. Therefore, Hnf6 seems to play a role only in the secondary transition activation of Ngn3 and not earlier.

Several lines of evidence suggest that endocrine formation, and thus perhaps Ngn3 activation, may be regulated by the BMP/TGFβ/activin signaling system. Thus, endocrine differentiation of the amphicrine AR4-2J cell line is induced by addition of activin, together with betacellulin or HGF, and this effect was recently shown to operate directly by means of elements in the Ngn3 promoter (Ogihara et al., 2003). Similarly, culturing pancreatic explants with the TGFβμ activin-antagonist follistatin leads to a decline in endocrine development, whereas exocrine development is increased (Miralles et al., 1998). Also, transgenic mice expressing a dominant negative form of the type II activin receptor in β-cells (Yamaoka et al., 1998) display a strongly hypoplastic islet mass at late embryogenesis as do compound heterozygous mice of the Activin type IIA and B receptors (Kim et al., 2000).


Notch-target genes include members of the Hes family (hairy and enhancer-of-split-like), which are dominant repressors (Jarriault et al., 1995; Iso et al., 2002). Multiple mammalian homologs of these proteins exist together comprising the HES and HERP families (Iso et al., 2003). They are very strong antagonists of the differentiation cues by B-class bHLH factors. For example, Hes1 blocks both myogenesis, as well as neurogenesis, when expressed in uncommitted cells (Sasai et al., 1992; Ishibashi et al., 1994; Ohtsuka et al., 2001). Accordingly, Hes1-deficient mice display a “neurogenic” phenotype that lead to premature generation of neurons and up-regulation of a mammalian achaete/scute homologue (MASH-1; Ishibashi et al., 1995).

The HES family proteins have an N-terminal bHLH-domain (Dawson et al., 1995; Iso et al., 2003) and a WRPW four amino acid motif in the distal C-terminal region, which mediates recruitment of strong corepressor molecules (groucho-like factors; Paroush et al., 1994; Fisher et al., 1996; Grbavec and Stifani, 1996; Giebel and Camposortega, 1997). Groucho proteins are able to multimerize and can bind histone deacetylases (HDACs; Chen et al., 1999; Brantjes et al., 2001). It is believed that the recruitment of a Groucho/HDAC complex by means of HES-factors to target promoters effectively maintains a closed promoter configuration, preventing the formation of positively acting complexes (Fig. 8). Such complexes can exert long-range repression (Courey and Jia, 2001).

Figure 8.

Hes1 expression in epithelial precursor cells. Hes1 (green nuclear staining) is dynamically expressed within pancreatic epithelial cells, some mesenchymal cells, but completely absent in differentiated endocrine cells (insulin+ cells shown, red). Weak Hes1 expression is observed in cells in the centroacinar position and absent in differentiating exocrine cells. Certain epithelial cells corresponding to ductal-like precursors do not express Hes1 and are immediately neighboring others that are strongly expressing Hes1. This pattern is expected as lateral inhibition may help single out endocrine precursors. Very likely, although not proven, the Hes1- cells may correspond to Ngn3+ cells.

The Hes1 gene is a critical Hes-type gene involved in regulation of pancreatic endocrine differentiation. Hes1 is expressed in the pancreatic bud, and later by the pancreatic epithelial precursors together with Notch1 and 2 (Apelqvist et al., 1999; Lammert et al., 2000; Jensen et al., 2000a, b). Endocrine cells do not express Hes1. The expression of Hes1 protein varies strongly between epithelial cells, suggesting that it is highly dynamic (Fig. 9) and could well reflect the process of lateral inhibition. Hes1 itself is autorepressed (Hirata et al., 2002), and both Hes1 mRNA and protein have a very short half-life in multiple cell types (Hirata et al., 2002). To date, no studies have addressed the possibility of cycling Hes-gene expression in the pancreas, but it seems likely. The expression pattern of Hes1 (Fig. 9) is compatible with such a possibility, if Hes1 activation is asynchronous.

The absence of Hes1 in endocrine cells, and the presence of the protein in the undifferentiated precursors, strongly suggests it could be involved in the repression of endocrine fate determination, presumably through antagonizing effects on the Ngn3-NeuroD cascade. Hes1-deficient mice support this hypothesis, showing increased endocrine formation at the expense of the pancreatic precursors (Jensen et al., 2000b). This effect is not limited to the endodermal region of the pancreas, as lung, stomach, and intestinal tissues at later stages all show a strong bias toward enteroendocrine fate allocation. Furthermore, an up-regulation of Ngn3, NeuroD, Math1, and Mash1 gene expression is observed (Jensen et al., 2000b). All these genes are members of the B-Class of bHLH activators, suggesting that Hes1 antagonizes a wide set of bHLH genes in the gastrointestinal tract. This antagonism is direct as the Ngn3 upstream promoter region contains several Hes1-binding sites, from which hes1 may repress Ngn3 promoter activity (Lee et al., 2001).

Endocrine Cell Specification


Concomitant with establishment of the endocrine fate, and mitotic arrest, a new group of genes become activated, including Isl1. Isl1 was the first homeodomain factor identified from β-cells and a founding member of the LIM-domain containing protein family. Isl1-/- mice demonstrate an early requirement for Isl1 and die around E10 due to lack of formation of the dorsal aorta (Pfaff et al., 1996). Isl1 expressed by cells derived from lineages of all the three major germ layers (Thor et al., 1991), including neurons, neuroendocrine cell types, and mesodermal cells, in addition to the four types of islet cells in the pancreas. The Isl1-expressing cells in the neural tube are motoneurons, and these are absent in Isl1-/- mice (Pfaff et al., 1996). Furthermore, Isl1 is expressed by those mesodermal precursors normally condensing to form the dorsal aorta showing the cell-autonomous requirement of Isl1 there.

In pancreatic development, the mesoderm condensing to form the dorsal aorta is developmentally related to the mesoderm associating with the dorsal pancreatic bud. A detailed study of how Isl1 deficiency influences pancreatic development was performed by Ahlgren et al. (1997). Isl1 is expressed by the dorsal pancreatic mesenchyme at early stages and not in the ventral mesenchyme. Inspection of the early dorsal bud region in Isl1-/- mice revealed a complete dorsal agenesis. This result reveals the role of the surrounding mesenchyme in the initiation of dorsal pancreatic budding. In addition, Isl1 also plays a cell-autonomous role in endodermal pancreatic development, which could be revealed by analysis of the ventral bud forming in Isl1 null animals (Ahlgren et al., 1997). Contrasting wild type, no glucagon-positive cells were formed, proving that Isl1 is a necessary cell-autonomous component for endocrine cell differentiation in the pancreas.

At what level does isl1 play a role in endocrine formation? Although amongst the first genes identified as playing a role in pancreatic development, it still remains an elusive component. No target genes for Isl1 has been unequivocally identified, although it has been suggested that IAPP and glucagon are candidates (Wang and Drucker, 1995, 1996). Similarly, Isl1 has not been shown to be a target gene for any of the other genes studied. Isl1 expression is restricted to postmitotic endocrine cells. Therefore, it cannot be involved in the selection process of the endocrine precursor cell. This conclusion is corroborated by the finding that no Isl1 expression is observed in Ngn3-/- animals. That endocrine cells do not form at all in isl1 mutants is noteworthy, as it would be assumed that the endocrine selection process is still occurring. This process takes place in nonresting epithelial precursors, clearly upstream of Isl1. Ngn3 or NeuroD were not identified at the time of analysis of the Isl1-/- embryos, but unpublished information from Ahlgreen and Edlund conclude that NeuroD is expressed in Isl1-/- pancreas (Ahlgren and Edlund, 2001). While it needs to be shown directly, a currently logical assumption would be that Ngn3 expression is also unaffected in absence of Isl1. Thus, Isl1 appears to play an early role in postmitotic endocrine cells, which might be linked to survival. However, studies determining whether Isl1-/- cells are rapidly eliminated or re-routed into alternative nonendocrine fates have not been undertaken.

Pax4 and Pax6.

Pax4 and Pax6 contain both a paired domain and a homeodomain and are independent members of a common subclass of the Pax-gene family (Noll, 1993; Mansouri et al., 1996; Dahl et al., 1997). The involvement of Pax4 and Pax6 in pancreatic endocrine development has been demonstrated by gene inactivation (St-Onge et al., 1997; Sosa-Pineda et al., 1997). Both targeting constructs were designed to insert a lacZ-gene marker within the Pax loci, thereby allowing detailed expression analysis in heterozygous animals. Pax4 homozygous animals die within 3 days of birth (Sosa-Pineda et al., 1997). Immunohistochemical analysis of the pancreas in the newborn mice revealed a complete absence of both β- and δ-cells. α-Cells were numerous but disorganized. Although the authors argued that the lack of Pax4 would cause the multipotent endocrine precursor cells that are normally destined to form β-cells to diverge toward the α-cell phenotype, this cannot be true, as such multipotential precursors do not exist. Rather, the main role of Pax4 is to control β- and δ-cell development when this is initiated several days later by the action of ngn3. However, this does not rule out that pax4 may be able to suppress α-cell differentiation cues in those cells. Our knowledge of Pax4 expression in normal development is poor. No study has yet related the expression of Pax4 to most of the important gene regulatory factors. We do not know if Pax4 operates together within Ngn3 in the pro-endocrine, pre–β-cell type, or at later stage in β-cell development, when these cells have become postmitotic. However, it appears likely that it serves an early role as the expression kinetics of Pax4 closely follows that of Ngn3 (J. Jensen, unpublished observations). Both genes peak at the secondary transition, and thereafter, expression quickly diminishes.

The immediate impression of the phenotype of mice deficient in Pax6 is that these display a somewhat reciprocal phenotype to Pax4 mutant mice (St-Onge et al., 1997). In the pancreas, very few α-cells were detected, whereas insulin and somatostatin cells were found. Based on costaining for the inserted lacZ-gene in heterozygotes, all hormone-positive cells expressed Pax6. These observations led to the conclusion that Pax6 was necessary for α-cell development but not required for development of the β- and δ-cell lineages. Combined mutation of Pax4 and Pax6 completely abrogates development of all pancreatic endocrine cell types (St-Onge et al., 1997). Thus, Pax4 and Pax6 could occupy nonredundant functions in branches of the lineage trees of the endocrine cells of the pancreas. This conclusion was not justified by later studies. Sander et al. investigated the pancreatic development the Small eyeNeu mutation mouse strain and showed that homozygous mice (SeuNeu/SeyNeu) had a reduced number of all islet cells (Sander et al., 1997). Importantly, insulin gene transcription in homozygous mutants was less than 10% of wild-type littermates, revealing a requirement for Pax-6 in insulin gene activation.

In conclusion, Pax4 operates specifically in a β/δ-cell lineage, and this role might be linked to suppression of α-cell differentiation cues. In contrast, Pax6 is needed in all the endocrine cell types but at different levels. It is possible that absence of Pax6 leads to a respecification of the α-cell type, although it is not fully supported by evidence. β-Cells can develop in absence of Pax6 but are not capable of maintaining their differentiated functions, such as the high-level insulin production. It may be hypothesized that Pax4 could compensate for some activity of the Pax6 gene in the β-cells. To clarify this, a knock-in of the Pax4 coding region into the Pax6 loci would be highly informative, and such an experiment would also clarify whether Pax4 is capable of specifying a β-cell program, as one outcome could be that the earliest glucagon-producing cells may develop as β-cells instead.


The elusive RIPE3b1 β-cell–specific binding complex of the insulin promoter was identified as the mammalian MafA factor by Olbrot et al. (Olbrot et al., 2002). MafA is a member of the Maf subfamily of leucine zipper factors and is capable of strongly activating the insulin promoter dependent on the RIPE3b1 sequence element. MafA is restricted to β-cells (Olbrot et al., 2002) and might be involved in specification of this cell in addition to its role in insulin transcription. The roles of other Maf family members suggest a role of this family in cell differentiation. MafB is involved in myeloid differentiation during hematopoiesis (Kelly et al., 2000) and also helps to pattern the hindbrain (Theil et al., 2002). MafB is synonymous with Kreisler, which has been shown to be extremely restricted in expression to rhombomere R5 and R6 and where it is required during establishment of the rhombomeric identity in hindbrain segmental patterning (McKay et al., 1994; Cordes and Barsh, 1994). MafA is not the only Maf-member expressed in pancreatic cells—MafB and c-maf are also expressed by endocrine cells (Olbrot et al., 2002). Taking into consideration the impact of the identification of the other transcription factors of the insulin gene Pdx1 and NeuroD in the area of pancreatic development, it will be highly unlikely if Maf-factors are not playing major roles as well.


More Factors, More Markers, More Cell Types

Although a significant number of transcriptional regulatory components playing a role in pancreatic development have been identified, the total number is not impressive. Moreover, their disclosure have often been a serendipitous finding of a phenotype manifestation in a mutant animal generated not with pancreatic development in mind. Fortunate as this might be, as it has brought in several strong groups to work on pancreatic development, there is a still a substantial need for more genome-wide searches for developmental regulatory genes specifically operating in the pancreas. With the current advent of microarray technology, where for instance a large fraction of the mouse transcriptome is residing on, e.g., the latest Unigene-designed Affymetrix chipset (MOE-430 chips), we are now in a position to devise experiments that bring the full (or almost complete) transcriptome to a test. By releasing some constrictions in the data set that otherwise limits the result of a sound hypothesis-based approach, the same experiment can now address genome-wide changes, or patterning variations, that never would have been brought into the initial hypothesis to be tested. Also, such experiments will bring in the identification of not only key regulatory components, but may be crucial in identifying a more elaborate set of markers that, in turn, may allow us to better define cellular subsets in the pancreas; stable as well as transient phenotypes. The custom array experiments recently performed by Chiang and Melton (2003) have shown how valuable such methods can be in defining independent cell types.

Some Important but Unresolved Issues

Despite the recent progress in lineage relationships and gene functions, we still lack substantial information as to how several aspects of pancreatic development are coordinated. Some of the more prominent areas are those of endocrine cell subtype specification, ductal cell specification, mesenchymal/epithelial signaling, and probably most important, the issue of morphogenetic patterning of cell fates. The issue of endocrine subtype specification is difficult, as we presently do not know if there is more than one type of each of the major endocrine cells in the pancreas. For instance, there is evidence to suggest that the earliest developing glucagon-producing cells are significantly different from the normal pancreatic α-cell. Expression of PC1/PC3, as well as IAPP (Wilson et al., 2002), suggest that this cell type expresses otherwise β-cell–specific genes—genes that also characterize the intestinal L-cell—a pro-glucagon–expressing cell that uses PC1/PC3 in the production of Glp1 rather than glucagon from the pro-glucagon polyprotein precursor. Perhaps we need to discriminate the earliest α-cells as a separate lineage, with no connection to the adult α-cell pool? At present, no lineage models have specifically addressed this issue, but this problem could be addressed using a glucagon-promoter Cre-ERT mouse line. Also, if true, a series of questions as to the development of the normal α-cell mass arise. Do these cells then develop concomitantly with the β-cells at the secondary transition? Alternatively, do normal α-cells develop closer to birth? — at a time point when PP cells also appear, several of which coexpress glucagon. Also, what genes may be involved in such a later α-cell specification? Apparently, a balance between the Pax4 and Pax6 factors may be determining such subtype specification, where Brn4 (Jensen et al., 2000a; Schwitzgebel et al., 2000; Hussain et al., 2002) could also play a role.

A clarification of the issue of ductal cell differentiation mechanisms is highly needed. Ductal cells are by far the most overlooked cell type in pancreatic development, and at the same time recognized as the most promising cell type as a source for possible neoformation of endocrine cell types in the adult pancreas. Generally, ductal cells are expected to harbor stem cell properties even in the adult organ, and certain studies have noted that neoformation of endocrine cells occurring from pancreatic ducts (Bouwens, 1998). Such changes require genetic networks that well could reflect that of undifferentiated precursor cells. However, a strict comparison between the cells of the adult duct to that of the expanding epithelial progenitor pool prior to the secondary transition in embryogenesis has not been made. What are the similarities between these cells, and what are the differences in the expression of cell intrinsic factors that allow duct cells to initiate a duct-specific program of bicarbonate production, mucin secretion, and a typical ductal morphology? Much is yet to be learned unraveling the secrets of the ducts.

Even though studies of the role of FGF10 in pancreatic development has provided long-sought information about a critical component during mesenchymal/epithelial signaling in the pancreas, this interplay is still far from understood. Cell intrinsic components, the patterning, and cell differentiation is largely uncharacterized in the pancreatic mesenchyme, and we already know that several factors are secreted from the mesenchyme that regulates growth and patterning of the epithelium (Fig. 10). At the budding stages, the dorsal pancreatic mesenchyme expresses Isl1, whereas the ventral does not. Isl1 expression also distinguishes the dorsal mesenchyme from, e.g., the mesenchyme of the gut. On the other hand, Gut mesenchyme expresses Bmp4, whereas the pancreatic does not (J. Jensen, unpublished observations). Knowing the importance of the mesenchyme in early organ patterning that occurs even in the absence of epithelium, it appears that mesenchymal cells exert significant control of pancreagenesis. How exactly is this orchestrated? The contribution of mesenchymal cells to the adult organ function should not be overlooked either. Vasculogenesis occurs as cells of the pancreatic mesenchyme initiate an endothelial program. The temporal onset of this, and its initial mechanism, has not been studied in detail. Not all pancreatic mesenchymal cells may become endothelial cells; interstitial cells associate tightly on the surface of individual acini in the adult. Most likely, these cells play a crucial role in adult pancreas regenerative processes, as in many cases, a very brisk expansion of this population is observed after pancreatic destruction.

Figure 10.

Schematic presentation of mesenchymal/epithelial signaling. Pancreatic mesenchyme signals to the epithelium by secretion of several growth factor components, such as FGF10, HGF, EGF family factors. During vasculogenesis, cells of the mesenchyme differentiate into intrapancreatic vessels and their signaling may be either attenuated or changed in nature. Concomitantly, pancreatic epithelial cells differentiate into their respective cell types. The level of control by the pancreatic mesenchyme upon terminal differentiation of the different pancreatic cell types is only poorly understood.

A Morphogenetic Perspective

The research in the factors regulating pancreatic development has come a long way in the recent decade. The results of the studies of these factors has led to a revised view on pancreatic lineage relationships and also spawned new ideas of how different genes in combination serve to specify cellular identities in the developing organ. However, despite the plethora of genes identified, it has not yet been possible to exploit these to trigger a full pancreatic program in nonpancreatic cells, nor are we able to specify an insulin-producing cell from a pancreatic precursor, and we are even further from specifying that cell type from adult pancreatic tissue.

Why has this not been possible? Perhaps rather than expecting the existence of master regulating genes, we now understand that genes act in concert and in defined periods. Thus, we may need to test these genes in proper context, before they reveal their true capabilities. Developmental biologists have for almost a century known the importance in questioning whether a region for grafting is permissive or the grafted tissue competent for a given change when performing grafting experiments. These phenomena exist due to the presence, or absence, of genetic networks that may alter the cell types in a graft, and it is these genetic networks we begin to unravel as we investigate the cell intrinsic factors present in a tissue. Cell intrinsic factors cannot by themselves explain regulative development, but they are needed to understand how changes in gene activation and, thus, a phenotype change occur. However, when we also understand how these factors are controlled by cell extrinsic factors, i.e., morphogenetic signals, this combined knowledge can provide the understanding of program of pancreagenesis, as well as other aspects of regulative development.

Although many present investigations focus upon which morphogenetic pathways are active in pancreatic development, revealing significant information as to the importance of these in pancreagenesis, we largely lack understanding of which cell intrinsic factors are targets. However, as these genetic links gradually are uncovered, we may expect to see fruitful attempts in controlling β-cell development from either stem cell sources, ES cells, or adult pancreatic tissue, thus giving new hope to diabetic individuals.


It was recently published (Collombat et al., Genes Dev 2003, 15:2591–2603) that the gene Arx plays an important role in the specification of alpha-versus beta-cells in a complementary fashion to Pax4. Thus, Arx represent the first gene identified that selectively is required for pancreatic alpha cell development.


I thank Drs. John Hutton, Howard Davidson, David Stenger, and members of my laboratory, Jan Nygaard Jensen, Juan Casanova, and Alecksandr Kutchma for critical comments and valuable suggestions to the present manuscript. Figures 4 and 8 were originally performed at the Hagedorn Research Institute, Gentofte, Denmark, with the technical expertise of Erna E. Pedersen in the laboratory of Dr. Ole D. Madsen. Alecksandr Kutchma is thanked for help in preparing Supplementary Table 1. Dr. Helena Edlund, Umea, Sweden, is thanked for generously providing antibodies against Ptf1a and Ngn3.