Vertebrate Endoderm Development and Organ Formation

Xenopus

In the Xenopus blastula, prospective germ layers are arranged in a gradient along the animal-vegetal (top-bottom) axis (Figure 2a). The endoderm originates from the yolky vegetal cells, ectoderm is from the animal region on top of the bastocoel cavity, and the presumptive mesoderm is induced in a ring of equatorial tissue by signals from the vegetal cells (Dale & Slack 1987, Moody 1987, Nieuwkoop 1969). Initially, the presumptive endoderm and mesoderm territories overlap, and equatorial cells express both mesoderm and endoderm genes. However, by the late-gastrula stage, the domains are mutually exclusive (Lemaire et al. 1998, Wardle & Smith 2004). Single-cell transplantation experiments confirm that germ layer commitment is progressive; at the blastula stage, individual vegetal cells can contribute to all three germ layers, but in the gastrula they are committed to the endoderm lineage (Wylie et al. 1987).

In Xenopus, gastrulation initiates with the formation of the blastopore lip in the dorsal-anterior mesendoderm. The dorsal blastopore lip is the site of the Spemann organizer, which is analogous to the anterior primitive streak in amniotes and the shield in fish. The organizer has important roles in patterning all three germ layers along the dorsal-ventral and anterior-posterior axes. As gastrulation proceeds, the endoderm and mesoderm become internalized, and the leading edge of the dorsal-anterior endoderm migrates to the position of the future ventral foregut (Figure 3a) (Keller 1991). In the post-gastrula Xenopus embryo, the majority of the endoderm mass is located ventrally, and the presumptive organ domains are arranged along the A-P axis consistent with their final position in the gut (Figure 3a). During somite stages of development (stages 15–30), the mass of endodermal tissue elongates by radial cell intercalation and eventually cavitates to form a primitive gut tube (Chalmers & Slack 2000).

Zebrafish

In pregastrula zebrafish embryos, the epiblast is an inverted cup of cells that sits on top of a large yolk cell. The prospective germ layers are arranged along the animal-vegetal axis of the epiblast, with the endoderm progenitors restricted to the four rows of marginal cells closest to the yolk (Figure 2b). Mesoderm precursors are intermingled with the endoderm progenitors, but extend up to 8-cells from the yolk margin, and the ectoderm is derived from the animal portion of the epiblast (Kimmel et al. 1990, Warga & Nusslein-Volhard 1999). Single-cell labeling and transplantation experiments indicate that prior to gastrulation, single marginal cells can populate both endoderm and mesoderm, suggesting a common bi-potential mesendoderm progenitor (Ober et al. 2003, Schier & Talbot 2005).

During gastrulation, the marginal cells involute under the epiblast to form an internal layer between the epiblast and the yolk cell. Endoderm cells are the first to involute and migrate anteriorly to form a dispersed monolayer. The intervening cells between endoderm and the noninvoluting epiblast become mesoderm (Kimmel et al. 1990). Fate-mapping studies indicate that the presumptive organ domains are arranged in broad, overlapping regions that roughly correspond to their final anterior-posterior position in the GI tract (Warga & Nusslein-Volhard 1999). During early-somite stages of development, the single cell-layered endoderm sheet converges on the dorsal midline to form an endodermal rod, from which organ buds will emerge, and which will later cavitate to form a gut tube (Ober et al. 2003).

Mice and Chicks

In amniote embryos (mammals and birds), gastrulation initiates in an epithelial layer called the epiblast. Cells in the epiblast undergo an epithelial to mesenchymal transition (EMT), migrate through the primitive streak (PS), and incorporate in the middle (mesoderm) or outer (endoderm) layer. The presumptive definitive endoderm (DE) cells invade and displace an outer layer of extraembryonic tissue cells, the hypoblast in chick and the visceral endoderm (VE) in mouse, which form supporting structures such as the yolk sac. Lineage tracing and later studies with embryonic stem cells suggest that some endoderm derives from a transient mesendoderm progenitor in the streak. The first cells to exit the primitive streak migrate toward the anterior (Kimura et al. 2006, Lawson & Schoenwolf 2003, Lawson & Pedersen 1987), giving rise to medial and lateral anterior definitive endoderm (ADE) and axial mesoderm (Figure 2c). Endoderm that exits the PS later contributes to more posterior endoderm. At the end of gastrulation, the DE is a sheet of cells that surrounds the external surface of the mouse embryo. Early cell labeling studies, combined with molecular marker analysis indicated that DE from the primitive streak displaces the VE cells into the extraembryonic domain. However recent evidence suggests that some VE cells persist in the DE layer even during primitive gut tube formation at E8.5–9.5 (Kwon et al. 2008).

Lineage tracing of the late gastrula stage endoderm has provided insight on how a two-dimensional sheet of cells forms the primitive gut tube (E7.5 in mouse, HH4 in chick) (Kimura et al. 2006, Lawson & Schoenwolf 2003, Lawson et al. 1986, Rosenquist 1971, Tam et al. 2007, Tremblay & Zaret 2005). Morphogenesis begins when the epithelial sheet folds over at the anterior and posterior ends, forming foregut and hindgut pockets. Anterior axial and lateral endoderm fold toward the ventral midline forming the foregut, which then moves caudally like the cresting of a wave (Figure 3); until it hits the hindgut wave, which is expanding rostrally (Lawson et al. 1986). Remarkably, foregut organs such as the pancreas and liver are derived from different populations of medial and lateral endoderm that converge during foregut morphogenesis (Franklin et al. 2008, Tremblay & Zaret 2005). Similar bilateral organ domains have also been described in zebrafish (Ober et al. 2003).

Mesendoderm Induction by Nodal Signaling

In all vertebrates, the Nodal signaling pathway is necessary and sufficient to initiate endoderm and mesoderm development, and it is required for proper gastrulation and axial patterning. Nodal ligands are members of the TGFβ family of secreted growth factors. Whereas there is a single Nodal gene in mice, there are five nodal-related genes in Xenopus (xnr1,2,4,5,6); and two zebrafish genes, squint (sqt/ndr1) and cyclops (cyc/ndr2), that are all implicated in mesendoderm development. Nodal ligands signal via a complex of type I (Alk4 or Alk7) and type II (ActRIIA or ActRIIB) transmembrane serine-threonine kinase receptors; and an EGF-CFC family coreceptor (Cripto or Cryptic in mouse, Oep in zebrafish, and FRL/Xcr1–3 in Xenopus). The activated Alk4/7 phosphorylates the cytosolic proteins Smad2 or Smad3. Phosphorylated Smad2/3 then binds to Smad4, and translocates to the nucleus, where it associates with DNA-binding transcription factors; such as the Fork head box h1 (Foxh1/FAST1) or Mix-like homeodomain proteins; to stimulate the transcription of mesendoderm genes. In all species, examined loss-of-function in any core pathway component results in compromised mesendoderm development (Shen 2007).

Many studies support a model in which high levels of Nodal signaling promote endoderm development, whereas lower doses specify mesoderm identity. Although it was first demonstrated in Xenopus explant experiments (Clements et al. 1999, Green & Smith 1990), genetic and biochemical manipulations that progressively reduce endogenous Nodal activity demonstrate that this dose-dependent role is conserved in fish, mouse, and humans (D’Amour et al. 2005, Shen 2007, Zorn & Wells 2007). In all species, endoderm cells develop in close proximity to the Nodal source and require a sustained period to be specified (Aoki et al. 2002, Ben-Haim et al. 2006, Hagos & Dougan 2007). In Xenopus and zebrafish embryos, the endoderm arises from nodal-expressing cells, and the mesoderm is induced in the adjacent tissue. In mouse, DE comes from the anterior primitive streak, the source of Nodal (Figure 4).

Nodal activity is positively controlled by an autoregulatory loop that maintains and amplifies its own transcription, via evolutionarily conserved Foxh1 DNA-binding sites in the first intron of Nodal, xnr1, and sqt genes (Fan et al. 2007, Norris et al. 2002, Osada et al. 2000). As a result, paracrine signaling between cells sustains the high nodal activity required to maintain endoderm gene expression. Negative feedback of Nodal activity occurs via its transcriptional target Lefty, which inhibits Nodal ligand-receptor interactions. Thus, in a classical reaction-diffusion model, Nodal autoregulation and Lefty negative feedback defines the mesendoderm territory and prevents its expansion into the ectoderm (Shen 2007).

There is considerable divergence in the way different species activate the nodal pathway (Figure 4). In Xenopus, the maternally inherited T-box transcription factor VegT is localized to the vegetal region of the oocyte and early blastula where it directly stimulates the transcription of many mesendoderm genes including xnr1 and xnr5 (Hyde & Old 2000, Xanthos et al. 2001). This role of VegT appears to be unique to frogs. In zebrafish, an unidentified maternal signal from the extraembryonic yolk syncytial layer (YSL) is thought to promote mesendoderm development, and the expression of zygotic cyc and sqt in the blastoderm margin (Fan et al. 2007, Schier & Talbot 2005). In the pregastrula mouse embryo, Nodal is expressed radially around the proximal epiblast; but then it becomes restricted to the presumptive posterior by the combined action of the antagonists Cerberus and Lefty in the anterior visceral endoderm (AVE), and the proteases that process Nodal ligands. Nodal signaling then initiates gastrulation in the posterior epiblast by regulating the Wnt, FGF, and BMP pathways (reviewed in Grapin-Botton & Constam 2007, Tam & Loebel 2007).

One similarity between frogs, fish, and mice is that canonical Wnt signaling via β-catenin and Tcf transcription factors stimulates high levels of Nodal gene transcription in the pregastrula embryo (Figure 4a). (see Wnt Signaling, sidebar below) In Xenopus and zebrafish, a maternal Wnt/β-catenin pathway that is active on the dorsal side of the embryo cooperates with VegT and the maternal YSL signal respectively, to promote high levels of nodal-related transcription in the dorsal-anterior mesendoderm (Fan et al. 2007, Hilton et al. 2003, Xanthos et al. 2002). Similarly in mouse, Wnt3 signaling appears to maintain Nodal expression in the streak, via evolutionary conserved Tcf DNA-binding sites in the Nodal promoter (Ben-Haim et al. 2006). Consistent with this, deletion of β-catenin in the early mouse gastrula impairs DE formation (Lickert et al. 2002).

Endoderm TranscriptionFactor Network

Nodal signaling promotes the expression of a conserved network of transcription factors within the endodermal lineage, which include Mix-like proteins: Foxa2, Sox17, Eomesodermin, and Gata4–6. Although the precise roles of these factors vary between species, together they activate a cascade of gene expression that functions to: (a) segregate the endoderm and mesoderm lineages, (b) commit cells to an endodermal fate, and (c) integrate signaling events that regionalize the nascent endoderm (Stainier 2002, Zorn & Wells 2007).

There are seven Mix-like homeobox genes in Xenopus (mix1, mix2, bix1, bix2, bix3, bix4, and mixer); four in zebrafish (bon, mezzo, mxt1, and mxt2); and only one in mice, Mixl1. Several Mix-like proteins physically interact with Smad2, and can mediate Nodal-dependent transcription (Germain et al. 2000). In mouse, Mixl1 is required for DE formation and for suppressing mesoderm (Hart et al. 2002, Tam et al. 2007). Of the three forkhead family members Foxa1–3 (Hnf3α, β, and γ), Foxa2 is particularly critical; and Foxa2^−/− null embryos lack foregut endoderm (Dufort et al. 1998). The HMG box factor Sox17 is required for normal endoderm development in Xenopus and mice (Figure 5) (Hudson et al. 1997, Kanai-Azuma et al. 2002). In zebrafish, the Sox17-related gene casanova (cas/sox32) is essential for endoderm specification and cooperates with Pou51/Oct4 to activate sox17 and foxa2 transcription (Lunde et al. 2004, Reim et al. 2004, Stainier 2002).

Whereas the functions of Mix, Sox17, and Foxa are highly conserved; other transcription factors appear to have species-specific functions. For example, in zebrafish and Xenopus; the zinc-finger factors Gata4, 5, and 6 are also involved in endoderm specification downstream of nodal signaling (Stainier 2002, Zorn & Wells 2007). In mice, Gata4–6 regulate extraembryonic endoderm lineages and do not appear to participate in DE specification, although they have essential, conserved roles in later endoderm organ development (Watt et al. 2007). The T-box transcription factor Eomesodermin (Eomes) is also essential for endoderm specification downstream of nodal signaling in zebrafish and mouse (Arnold et al. 2008, Bjornson et al. 2005), but its precise function in the frog endoderm remains to be determined.

A number of studies have begun to define the inter-relationships between the endoderm-specifying transcription factors. Epistatic analyses in zebrafish suggest a pathway in which Nodal activates the expression of Bon and Gata5; and they form a complex with Eomes to activate Cas, which in turn stimulates sox17 and foxa2 transcription (Alexander & Stainier 1999, Bjornson et al. 2005, Ober et al. 2003). Recent genome-wide expression analyses in various species indicate that these transcription factors probably do not act in a simple linear fashion, but rather form a complex gene regulatory network (GRN). The transcription factors influence each other expression through multiple feedback loops, and they regulate both distinct and overlapping transcriptional targets (Brown et al. 2008, Dickinson et al. 2006, Sinner et al. 2006, Tamplin et al. 2008) (Figure 4d). Several Web sites now provide interactive models of the increasingly complex GRN; integrating expression data, genomic information, and DNA-binding evidence along with functional data (see related resources section below) (Chan et al. 2008, Loose & Patient 2004, Oliveri & Davidson 2004). Interestingly, some components of the endoderm-GRN, such as Foxa and Gata factors, are evolutionarily ancient and known to regulate endoderm development in C. elegans, Drosophila, sea urchins, and ascidians (Zorn & Wells 2007).

A key function of the endoderm–GRN is to segregate the mesoderm and endoderm lineages. This is partly mediated by mutual antagonism with an analogous mesoderm-GRN, in which low doses of Nodal induce the expression of FGF ligands and the T-box transcription factor Brachyury. An autoregulatory loop between FGF and Brachyury constitutes the core of a mesoderm-GRN (Latinkic et al. 1997), which promotes mesoderm fate and represses endoderm gene expression (Mizoguchi et al. 2006, Poulain et al. 2006). This may act on several levels, because FGF stimulated MAP kinase can phosphorylate Smad2 and Sox32 in fish, which reduces their activity (Poulain et al. 2006, Zorn & Wells 2007). In addition, Mix-like proteins and Brachyury appear to repress each other’s expression (Hart et al. 2002, Kofron et al. 2004, Lemaire et al. 1998). Thus, although cells initially express both endoderm and mesoderm genes in response to Nodal signaling, these mutually repressive pathways resolve the lineages by mid-gastrula (Figure 4d).

Initial Patterning Coincides with Endoderm Formation

The first regional differences can be detected in the early gastrula when the endoderm germ layer is being specified. In Xenopus and zebrafish, maternal Wnt/β-catenin signals that are active on the future dorsal side of the blastula (Figure 4a,b) cooperate with Nodal signaling to activate the expression of genes, such as the homeobox Hhex in anterior endoderm (Ho et al. 1999, Zorn et al. 1999). In mice, the high-nodal environment of the early primitive streak promotes anterior endoderm fate and expression of Hhex (Thomas et al. 1998). At later gastrula stages, cells in the posterior primitive streak are in a lower nodal environment and tend to contribute to posterior endoderm. The endoderm-GRN transcription factors also participate in endoderm patterning; Foxa2 and Mixl1 are preferentially required for anterior endoderm; and Sox17 is required for posterior endoderm (Dufort et al. 1998, Hart et al. 2002, Kanai-Azuma et al. 2002). Their different activities correlate with their timing of expression; with Foxa2 being is expressed first and Sox17 expressed slightly later, when the posterior DE emerges from the streak. Crosstalk with other signaling pathways may account for the different activities, because Sox17 physically interacts with β-catenin to promote the transcription of some target genes, whereas it represses others (Sinner et al. 2004).

A-P Patterning of the Primitive Gut Tube

By the end of gastrulation, the endoderm is partitioned into broad A-P domains that can be identified by the expression of Hhex, Sox2, and Foxa2 transcription factors in the anterior half of the embryo; and Caudal type homeobox genes Cdx1, 2, and 4 in the posterior half (Figure 6). These transcription factors are crucial for regional identity: Hhex, Foxa2, and Sox2 are required for foregut development, and the Cdx genes are required for the hindgut and positioning of the foregut-hindgut boundary (Chawengsaksophak et al. 2004, Dufort et al. 1998, Kinkel et al. 2008, Martinez Barbera et al. 2000.

Despite this differential gene expression, endoderm fate is still labile at this stage (Horb & Slack 2001, Kimura et al. 2007, Le Douarin 1968, Wells & Melton 2000). Dynamic tissue movements during gastrula and early somite stages bring the endoderm into proximity with different mesodermal tissues that secrete patterning factors (Chalmers & Slack 2000, Tam et al. 2007, Tremblay & Zaret 2005). These include FGF, Wnt, and BMP ligands; all of which appear to maintain hindgut identity and actively repress foregut fate in the posterior (Figure 6).

In mouse and chick, mesodermal FGF4 (but not FGF2, 5, or 8), signaling promotes Cdx expression in the presumptive hindgut endoderm and represses the expression of anterior genes Hhex and Foxa2 (Dessimoz et al. 2006, Haremaki et al. 2003, Wells & Melton 2000). In frog and zebrafish, Wnt/β-catenin signaling, similar to FGF4, is necessary and sufficient to promote hindgut development and inhibit foregut fate (Goessling et al. 2008, McLin et al. 2007). Moreover, in Xenopus, β-catenin activity must be repressed in the anterior endoderm by secreted Wnt-antagonists such as Sfrp5 to maintain the foregut progenitors, and for subsequent liver and pancreas development (Li et al. 2008). It is currently unclear if differential Wnt/β-catenin signaling also patterns the late gastrula amniote endoderm. Experiments in zebrafish, Xenopus, and chick indicate that, during these early stages of development, BMP signaling also promotes posterior endoderm development (Tiso et al. 2002, Wills et al. 2008) and that RA is important in establishing the foregut-hindgut boundary (Chen et al. 2004, Kumar et al. 2003, Stafford & Prince 2002).

Although foregut and hindgut progenitors are initially uncommitted, there is evidence that they have distinct developmental potentials and respond differently to subsequent inductive events. For example, in transplantation experiments, the foregut has a greater intrinsic hepatic potential than the hindgut (Fukuda-Taira 1981, Le Douarin 1968), which may be due to foregut enriched transcription factors such as Foxa2. In vivo DNA-binding studies suggest that Foxa2 binds to enhancer elements of the Albumin gene and opens chromatin, allowing it to be transcribed (Zaret 2002). Consistent with this, the conditional deletion of both Foxa1 and Foxa2 from the mouse foregut prevents liver induction (Lee et al. 2005). In another example, Cdx4-expressing hindgut endoderm is not competent to respond to subsequent pancreatic induction by RA signaling in zebrafish (Kinkel et al. 2008).

It is unclear how the effects of FGF, Wnt, BMP, and RA are integrated during endoderm patterning. However, there is evidence that these same signals coordinately regulate the transcription of Cdx and Hox genes in the mesoderm and neuro-ectoderm (Benahmed et al. 2008, Pilon et al. 2006, Shimizu et al. 2006); this suggests that these signaling factors synchronize A-P identity in all three germ layers. Indeed, a recent study found that RA and FGF4 cooperate to posteriorize the chick gastrula endoderm (Bayha et al. 2009).

Endoderm Migration in the Gastrula

In addition to their role in endoderm specification and patterning, Nodal, FGF, and Wnt signaling are all required for the coordinated series of cell movements that drive mesendoderm morphogenesis (reviewed in Keller 2005). During gastrulation, the dorsal mesendoderm elongates via polarized cell intercalations in a process known as convergent-extension, which is coordinately regulated by FGF and non-canonical Wnt/planar cell polarity (PCP) signaling (Keller 2005). The anterior endoderm cells, on the other hand, exhibit directional migration, which requires Nodal signaling and is mediated by dynamic cell-cell adhesion and integrin-fibronectin interactions with the extra cellular matrix (Davidson et al. 2002, Pezeron et al. 2008, Yamamoto et al. 2004).

In addition to specifying endodermal fate, Mix-like factors; Eomes, Lim1, Foxa2, and Gata4–6; also promote mesendoderm migration behavior (Arnold et al. 2008; Fletcher et al. 2006; Kofron et al. 2004; Luu et al. 2008; Tam et al. 2004, 2007); thus linking Nodal-induced endoderm identity with morphogenesis. Downstream targets of Nodal signaling and Mixer include the chemokine receptor Cxcr4 (Dickinson et al. 2006, Fukui et al. 2007, Sinner et al. 2006) and the ligand Sdf1, which acts as a chemoattractant for Cxcr4-expressing endoderm cells (Fukui et al. 2007, Mizoguchi et al. 2008, Nair & Schilling 2008). The disruption of Cxcr4/Sdf1 in zebrafish results in disrupted endoderm migration and gut-tube duplications (Mizoguchi et al. 2008, Nair & Schilling 2008). Cxcr4/Sdf1, as well as PDGF signaling, appear to act by regulating integrin fibronectin-mediated endoderm migration (Keller 2005). In Xenopus, the nodal target fibronectin-leucine rich transmembrane protein (FLRT3) also regulates cadherin-dependent cell adhesion and mesendoderm migration via the small GTPase Rand1 (Ogata et al. 2007). FLRT3^−/− mouse embryos also have DE migration defects (Egea et al. 2008, Maretto et al. 2008).

In the mouse, gastrula FGF signaling, MAP kinase, and Eomes are required to downregulate E-cadherin and allow epiblast cells to undergo an EMT and ingress through the PS (Arnold et al. 2008, Ciruna & Rossant 2001, Zohn et al. 2006). However, less is known about what regulates the subsequent migration of DE cells and their incorporation into the visceral endoderm. Genetic analysis suggest that DE migration requires continued FGF signaling; the activity of Mixl1, Lim1, Foxa2; and the protein Nap1 (Burtscher & Lickert 2009, Garcia-Garcia & Anderson 2003, Rakeman & Anderson 2006, Sun et al. 1999, Tam et al. 2007). However, these mutants often have disruptions in the anterior visceral endoderm migration and the PS, which makes it difficult to study DE migration (Garcia-Garcia & Anderson 2003, Rakeman & Anderson 2006).

Early Gut-Tube Morphogenesis

Gut-tube morphogenesis occurs very differently across vertebrate species. In amniotes, it begins with the invagination, and rostral-to-caudal movement, of the foregut and hindgut pockets (Figure 3) (Franklin et al. 2008, Tremblay & Zaret 2005). Recent forward genetic screens in mice have revealed that this process requires a novel zinc finger protein Chato (Garcia-Garcia et al. 2008) and Lulu, a FERM domain cytoskeletal protein thought to regulate epithelial organization (Lee et al. 2007).

In zebrafish, vascular endothelial growth factor (VEGF) and Wnt/PCP pathways are required for convergent-extension of the dispersed endodermal sheet into a solid endodermal rod at the dorsal midline. Disruptions in this process result in foregut duplications with bilateral liver and pancreas (Matsui et al. 2005, Ober et al. 2004). Formation of a gut tube lumen within the solid endodermal rod is regulated by the transcription factor Hnf1β/Tcf2, through ion channel mediated fluid accumulation (Bagnat et al. 2007). Initial looping of the zebrafish gut appears to be controlled by asymmetric migration of the lateral plate mesoderm on one side of the embryo in response to global left-right asymmetry cues in the embryo (Horne-Badovinac et al. 2003).

The Wnt/PCP pathway also regulated gut elongation in Xenopus (Figure 3a) (Li et al. 2008). Moreover, the secreted Wnt-antagonist Sfrp5 appears to simultaneously restrict both Wnt11/PCP and Wnt11/β-catenin signaling in the anterior endoderm to coordinate foregut fate and morphogenesis (Li et al. 2008). Repression of Wnt11/PCP signaling is necessary to allow the foregut epithelium to form over the deep endoderm cells that are undergoing the convergent-extension movements that drive gut elongation. It is unclear if this mechanism is conserved in mammals, because Sfrp5^−/− null mice do not exhibit foregut defects, although redundantly expressed Wnt-antagonists may be involved.

Subdividing the Foregut: Organ Induction

The foregut epithelium gives rise to the esophagus, trachea, lungs, thyroid, stomach, liver, pancreas, and hepatobiliary system (e.g., intrahepatic and extrahepatic bile ducts, common duct, gall bladder, cystic duct). Human congenital malformations often occur simultaneously in several of these organs, consistent with experimental evidence that they are derived from a common progenitor cell population in the ventral foregut (Zaret 2008).

Foregut explants and mouse genetics suggest a model in which varying concentrations of FGF2 act through MAP Kinase to induce different lineages from a common region of the ventral foregut (Calmont et al. 2006, Jung et al. 1999, Zaret 2008). High levels of FGF signaling induce expression of the homeobox gene Nkx2.1 in the lung and thyroid progenitors; moderate doses activate Albumin expression in liver progenitors; and low levels promote expression of the homeobox gene Pdx1 in the ventral pancreas and duodenum (Figure 7a) (Serls et al. 2005). It is unclear whether the proximity or duration of contact with the cardiac mesoderm controls the FGF dose in vivo. However, this may be regulated in part by Hhex-mediated cell proliferation, which displaces the presumptive ventral pancreatic epithelium away from the cardiac mesoderm (Bort et al. 2004).

BMP signaling from the septum transversum mesenchyme also promotes hepatic fate, and at high levels, it appears to repress pancreas development (Figure 7) (Zaret 2008). BMP and FGF signaling similarly regulate liver and pancreas induction in zebrafish, chicken, and Xenopus (Shin et al. 2007, Zaret 2008). Although in zebrafish these factors (and hepatic promoting Wnt2b) originate from the lateral plate mesoderm and not cardiogenic and septum transversum mesenchyme (Ober et al. 2003, Zaret 2008). Recent single cell labeling experiments in zebrafish provided direct evidence that the liver, pancreas, and intestine indeed arise from a common pool of progenitors; and that proximity to the source of BMP2 signaling influences cell fate (Chung et al. 2008).

RA has multiple roles regulating cell identity in the pharyngeal arches, the anterior and posterior ends of the foregut, as well as a posteriorizing role in the hindgut (Duester 2008). Mouse embryos deficient for retinaldehyde dehydrogenase 2 (Raldh2) lack RA signaling in the foregut and fail to develop lungs, stomach, and dorsal pancreas; and have impaired liver growth (Molotkov et al. 2005, Wang et al. 2006). Experiments in chick, Xenopus, and zebrafish suggest that RA regulates the Pdx1-Cdx expression boundary and the position of the dorsal pancreas (Bayha et al. 2009, Chen et al. 2004, Kinkel et al. 2008, Stafford & Prince 2002, Kumar et al. 2003). RA signaling appears to act in part by directly activating the transcription of homeobox genes such as Hoxb1 and Hoxa5, which contain retinoic acid responsive enhancers (Grapin-Botton 2005). However, other effects of RA may be indirect owing to the activation of downstream factors including FGF10, Shh, and Tbx1; which regulate Raldh2 expression in a feedback loop (Ivins et al. 2005, Wang et al. 2006).

Dorso-ventral patterning of the gut tube is poorly understood, but critical for separating the dorsal esophagus from the ventral trachea, and it impacts the distinct ventral and dorsal pancreatic primordia. D-V patterning in the gut appears to be similar to neural tube patterning, in which mutual antagonism between Shh and BMP establishes D-V identity (Petryk et al. 2004, Que et al. 2006, Que et al. 2007, van den Brink 2007). Mouse knockouts of Shh, or its transcriptional effectors Gli2 and Gli3, cause a range of defects from lung agenesis to tracheo-esophageal fistulas, similar to a human congenital condition in which the trachea and esophagus fail to separate. Mutant mice that lack the BMP antagonists Noggin or Twisted-Gastrulation exhibit similar defects, with reduced Shh and Hhex expression in the ventral foregut. The separation of the trachea and esophagus also appears to involve the transcription factors Foxf1 in the mesenchyme (Mahlapuu et al. 2001) and mutual repression between dorsal Sox2 and ventral Nkx2.1 (Que et al. 2006, Que et al. 2007).

Translating Patterns into Organ Domains

Foregut organs are specified around the 6–7-somite stage of development in mouse and chick, but not until 20+ somite stages in Xenopus and zebrafish. Although we have begun to understand the cell signals that induce different foregut organ fates, it is unclear how the resulting gene expression patterns are actually translated into distinct organ primordia. Increasingly sophisticated maps of transcription factor expression domains suggest that there is not a one-to-one correspondence between any individual factor and a given cell lineage, but rather a complex combinatorial code of overlapping factors along the GI tract (Figure 8). A recent study of 15 transcription factors in the mouse foregut (Sherwood et al. 2009) identified over a dozen domains; with unique combination of factors that roughly correspond to the presumptive liver, biliary system, stomach, duodenum, and pancreas.

Recent studies have begun to determine how combinations of transcription factors specify distinct lineages. For example, a domain in the ventral foregut that coexpresses Pdx1 and Sox17 was shown to give rise to the pancreas, gall-bladder, cystic duct, and common duct. Segregation of these extrahepatobiliary/pancreatic progenitors was Sox17-dependent. The deletion of Sox17 resulted in the loss of biliary structures and ectopic pancreas in the common duct, whereas maintained Sox17 expression suppressed pancreas development and induced ectopic biliary tissue (Spence et al. 2009). These and other studies support a model in which the common foregut progenitors first separate into Hhex+ liver cells and a multipotential Pdx1+/Sox17+ progenitor population (Figure 7b). Sox17 then acts to segregate the biliary from the pancreas lineage, possibly in concert with Hnf6 and Hnf1β, and the Notch effector Hes1 (Clotman et al. 2002, Coffinier et al. 2002, Hunter et al. 2007). Genetic analysis suggests that Foxf1 and FGF10 in the surrounding mesenchyme also regulate proliferation and correct positioning of the liver, pancreas, and various hepatobiliary ducts (Dong et al. 2007, Kalinichenko et al. 2002).

Below, we briefly summarize early liver, pancreas, and lung bud development (~e9–11 in the mouse), and for a review of pharyngeal organogenesis, including the thyroid and thymus, we refer readers to Graham (2008).

Liver

Once specified, the hepatic endoderm thickens into a columnar epithelium, the surrounding basement membrane breaks down, and the hepatic precursors (hepatoblasts) delaminate and migrate into the adjacent stroma and form the liver bud. This process requires the transcription factors Hhex, Gata4/6, Hnf6/Onecut1, Onecut2, and Prox1; and appears to be mediated by matrix remodeling metalloproteases. Continued signaling, which include FGF, BMP, Wnt, and HGF, from hepatic mesenchyme and endothelial cells, as well as hepatic transcription factors such as Hnf4α, Tbx3, and Hlx, promote liver bud growth and differentiation. The hepatoblasts are bi-potential, and those residing next to the portal vein mesenchyme are induced by a Notch-mediated process to form biliary epithelial cells, whereas those deeper in the parenchyma become hepatocytes. For a detailed review of liver differentiation, see Zaret (2008), Zhao & Duncan (2005), and Zorn (2008).

Pancreas

The pancreas arises from distinct dorsal and ventral buds, which are induced by different mesodermal structures. As described above, the ventral pancreas fate is induced in the portion of the ventral foregut that has low levels of cardiac FGF signaling. Dorsal pancreatic development requires RA signaling (Molotkov et al. 2005, Stafford & Prince 2002); as well as secreted factors from the notochord and dorsal aorta, including Activin and Fgf2, that repress Shh expression in the presumptive dorsal pancreatic epithelium (Figure 9). In both the dorsal and ventral buds, pancreatic progenitors express the essential transcription factor Pdx1 and Ptf1a/p48, but only the dorsal bud expresses Hoxb9 (Sherwood et al. 2009). Mesenchymal FGF10 then promotes the proliferation of pancreatic progenitors (Bhushan et al. 2001). Wnt signaling is required for the expansion of the exocrine pancreas (Murtaugh et al. 2005, Wells et al. 2007), whereas Notch-Delta signaling via the Ngn3 transcription factor is required for segregating the endocrine lineages. We refer readers to Gittes (2009) and Murtaugh (2007) for a detailed review on pancreas development and the differentiation of the five endocrine lineages (α-cells, β-cells, δ-cells, PP cells, and grelin cells).

Lung

In response to high doses of cardiac FGF signaling, presumptive lung cells in ventral foregut upregulate the homeobox gene Nkx2.1. RA and Hedgehog signaling then represses TGFβ signaling and promotes the expression of Hoxa4/5 and Tbx4/5, as well as FGF10, in the mesenchyme (Chen et al. 2007). FGF10 in turn induces the proliferation and chemotaxis of the Fgfr2-expressing pulmonary epithelium, which is required for lung bud formation (Cardoso & Lu 2006). By e10.5 in the mouse, the growing lung bud undergoes a highly stereotyped branching morphogenesis to generate the tree-like structure of the lung. This is regulated by a series of negative feedback interactions between FGF10 and Tbx4/5 in the distal tip mesenchyme; with Shh, BMP4, Fgfr2, and the FGF-antagonist Sprouty in the growing distal tip epithelium (Cardoso & Lu 2006). The adult respiratory epithelium contains a diversity of cell types including ciliated, goblet, basal, and Clara cells. For a detailed review on the lung development and the specification of these mature cell types, see Cardoso & Lu (2006) and Maeda et al. (2007).

Establishing Domains in the Intestinal Tract

The available data support a model in which posteriorizing factors such as Wnt, FGF4, and Cdx progressively allocate more posterior regions of the gut; initially they are broadly active in the posterior and suppressed in the presumptive foregut but then become down-regulated in the midgut by early somite stages. In the hindgut, persistent FGF/Wnt signaling is required to maintain Cdx expression and define the anterior boundary of the intestine. Experimentally increasing FGF4 signaling in the gut shifts the Pdx-Cdx boundary and transforms the duodenum into intestine (Benahmed et al. 2008, Dessimoz et al. 2006), whereas deletion of FGF-Receptor 2b in mice results in duodenal atreasia (Rubin 2007). Gene targeting of various Wnt-pathway molecules, including Tcf1/Tcf4^−/− or the noncanonical Wnt5a ligand in mice; results in a range of phenotypes including truncated intestinal development, downregulation of intestinal Cdx2, and a posterior expansion of the foregut marker Sox2 (Cervantes et al. 2009, Gregorieff et al. 2004). Conversely, activating Wnt or Cdx2 more anteriorly can cause intestinal metaplasia of the stomach, pancreas, or lung similar to several human conditions (Grapin-Botton 2005). Recent findings indicate that Cdx2 function in the endoderm is required for initiating the intestinal program and supressing anterior cell fate (Gao et al. 2009), suggesting that regulation of Cdx2 expression is a point of convergence by which signaling pathways regulate intestinal development.

Reciprocal signaling between Indian and Sonic hedgehog in the epithelium and the surrounding mesenchyme also regulates many aspects of GI tract development (Figure 9) (Kumar et al. 2003, van den Brink 2007). For example, hedgehog signaling promotes the expression of Foxf1 transcription factors in the intestinal mesenchyme, which in turn regulates Wnt signaling and BMP expression (Ormestad et al. 2006). In the stomach, mesenchymal FGF10 regulates epithelial Hedgehog expression, which in turn influences BMP and Nkx2.5 expression in the pyloric mesenchyme (De Santa Barbara et al. 2005, van den Brink 2007). Shh in the stomach/duodenal domain is also regulated by Activin, and mouse embryos with Activin receptor mutations have a posterior shift in Shh expression (Kim et al. 2000). Finally, epithelial Shh signaling regulates the cocentric architecture of the gut by restricting smooth muscle differentiation in the immediately adjacent mesoderm, thus allowing connective tissue to develop (van den Brink 2007).

One mechanism by which FGF, Wnt, RA; and the Parahox factors Pdx1 and Cdx; control GI segment identity is by directly regulating Hox gene expression in precise domains along the A-P axis (Figure 8) (Grapin-Botton 2005, Kawazoe et al. 2002, Shimizu et al. 2006). Hox gene mutations indicate that they are required for GI tract development, and anterior misexpression of posterior Hoxa13 resulted in a homeotic transformed of stomach into intestinal epithelium (Grapin-Botton 2005). The distribution of Hox genes regulates the regional expression of downstream factors such as Fgf10, which impacts regional proliferation in the intestine (Zacchetti et al. 2007).

Stomach and Duodenum

In the anterior part of the GI tract, Shh and Sox2 are required for proper esophagus development and to establish the boundary with the stomach (Que et al. 2007). Specification of the stomach requires the homeodomain transcription factor Barx1, which in mice is localized to the presumptive stomach mesenchyme (e15), in which it activates the expression of secreted Wnt-antagonists Sfrp1 and Sfrp2. These in turn locally inhibit Wnt signaling to repress Cdx in the epithelium, and enable stomach differentiation (Kim et al. 2005). Thus, mutual antagonism between an intestinal promoting Wnt/Cdx2 pathway and a Barx1/Sfrp/Sox2 pathway determines the stomach-intestine boundary (Figure 9) (Stringer et al. 2008). Barx1 also regulates mesenchymal expression of the homeodomain factor Bapx1 (Nkx3.2), which is required for (Verzi et al. 2009) pyloric sphincter formation. Bapx1 appears to function in a complex regulatory loop with FGF10, Shh, Nkx2.5, and Sox9 to locally repress mesenchymal BMP expression (Figure 9) (De Santa Barbara et al. 2005, Moniot et al. 2004). For details on the further differentiation of stomach epithelium into mature squamous, parietal, chief, and gastric endocrine cells see (Yasugi and Mizuno 2008).

Intestine

The transformation of the simple pseudostratified epithelium into mature intestinal mucosa with a cryptvillus axis begins at approximately e14 in the mouse, and proceeds in an A-P wave along the GI tract. Cyto-differentiation and villi formation require Wnt, BMP, and Cdx2 (Rubin 2007), resulting in the four main cell types: absorptive enterocytes (colonocytes in the colon), the three secretory lineages the goblet cells, enteroendocrine cells, and Paneth cells. Notch and its effector Hes1 control segregation of the absorptive and secretory lineages by repressing expression of the bHLH transcription factor Atoh1 (Math1), thus specifying absorptive lineages over secretory lineages. Finally, Ngn3, as in the pancreas, is required for differentiation of the endocrine lineage (Sancho et al. 2004, Shroyer et al. 2007).

Making Pancreatic β-cells and Hepatocytes for Therapy

Human ESC-derived endoderm has been further differentiated into the pancreatic endocrine lineage (Figure 10) (D’Amour et al. 2005, 2006; Kroon et al. 2008; reviewed in Spence and Wells 2007). Using a stepwise differentiation protocol designed to mimic pancreatic development, hESCs are first differentiated into DE with Activin; then into posterior foregut and pancreatic endoderm (PE) with a combination of retinoic acid, FGF10, and cyclopamine (a hedgehog inhibitor). PE cultures express foregut and pancreatic markers such as Hlxb9 and Pdx1. These PE cells were then cultured in the presence of Notch pathway inhibitors to promote endocrine fate and insulin, such as growth factor (IGF1) and hepatocyte growth factor (HGF), to promote maturation. The resulting cells had high levels of insulin, but were not responsive to glucose, suggesting that they were not fully mature β-cells. However, following engraftment into adult animals, PE/endocrine progenitor cells acquired features of mature pancreatic endocrine cells and restored normoglycemia in diabetic mice (Kroon et al. 2008).

In attempts to generate hepatocytes, researchers have treated Activin-induced HESC-derived endoderm with FGF2 and/or BMP4 to mimic hepatic induction. This has been followed by different combinations of HGF, EGF, FGF2, TGFα, VEGF, and Dexamothasone to mature the hepatic precursors and generate cells with hepatocyte-like properties (Gouon-Evans et al. 2006, Soto-Gutierrez et al. 2006, Zaret & Grompe 2008). Therapeutically, hepatocytes generated from both mouse and human ES cells have been shown to engraft in the liver (to a modest extent) and have restored some liver function in an animal model of liver failure (Zaret & Grompe 2008).

Although these reports are promising, in most cases the cells generated do not appear to be entirely mature and exhibit some fetal characteristics, suggesting that there are still gaps in our understanding. There is every reason to believe that further studies of liver and pancreatic development will continue to increase both the efficiency and functionality of HESC-derived cell types. It is hopeful that similar differentiation protocols can be used to direct induced pluripotent stem cells (iPSCs) into patient-specific therapeutic cell types.