Transcriptional dynamics of endodermal organ formation



Although endodermal organs including the liver, pancreas, and intestine are of significant therapeutic interest, the mechanism by which the endoderm is divided into organ domains during embryogenesis is not well understood. To better understand this process, global gene expression profiling was performed on early endodermal organ domains. This global analysis was followed up by dynamic immunofluorescence analysis of key transcription factors, uncovering novel expression patterns as well as cell surface proteins that allow prospective isolation of specific endodermal organ domains. Additionally, a repressive interaction between Cdx2 and Sox2 was found to occur at the prospective stomach–intestine border, with the hepatic and pancreatic domains forming at this boundary, and Hlxb9 was revealed to have graded expression along the dorsal–ventral axis. These results contribute to understanding the mechanism of endodermal organogenesis and should assist efforts to replicate this process using pluripotent stem cells. Developmental Dynamics 238:29–42, 2009. © 2008 Wiley-Liss, Inc.


One of the first cell fate decisions made during vertebrate embryogenesis is the separation into germ layers: ectoderm, mesoderm, and endoderm. The endoderm, which first forms on the outside of the embryo and is later internalized, gives rise to the digestive and respiratory tracts and their associated organs. Endodermally derived organs, including the liver, pancreas, lungs, intestine, and stomach, are of considerable therapeutic interest, but much remains to be learned about how they first emerge from the primitive gut tube.

In the mouse, the definitive endoderm forms during gastrulation as a single-cell thick sheet of cells on the outside of the cup-shaped embryo. As gastrulation proceeds, this definitive endoderm migrates anteriorly, displacing the extraembryonic visceral endoderm (Lawson and Pedersen,1987). The endodermal sheet subsequently begins to fold in at the anterior and posterior ends, forming the anterior and caudal intestinal portals (AIP and CIP, respectively). The ventral lips of these tubes extend toward the center of the embryo, causing the sheet of endoderm to fold progressively into a tube, so that by the 13- to 15-somite stage (embryonic day [E] 8.75 in the mouse), the entire endoderm has been internalized as a tube. Localized increased proliferation and/or migration instigated by mesodermal cues causes budding which is apparent by E9.5 in what will become the lung, liver, and dorsal and ventral pancreata. After E9.5, the specified endodermal organ domains as well as the distinct sections of the gut tube (esophagus, stomach, intestine) undergo morphogenesis and lineage diversification to create the functional architecture specific to each organ.

There is evidence suggesting that when endoderm is initially formed (between E6.5 and E7.5 in the mouse), it is not yet committed to specific organ domains. When explanted mouse E7.5 anterior endoderm is placed in contact with posterior mesoderm, posterior endodermal genes can be activated and vice versa (Wells and Melton,2000). Explants of mouse E8.25 ventral endoderm have the potential to activate pancreas-, liver-, or lung-specific genes depending on their proximity to cardiac mesoderm or the concentration of fibroblast growth factors (Deutsch et al.,2001; Serls et al.,2005). Endoderm differentiation potential has been investigated further in chick embryos using heterotopic transplantation. In these experiments, 5- to 12-somite (equivalent to E8.25–E8.5) endoderm alters positional marker expression when transplanted more posteriorly yet retains positional markers when transplanted anteriorly (Kumar et al.,2003), and positional marker expression is committed by the 10- to 12-somite stage (Kimura et al.,2007).

Clues to how the endoderm is patterned into organ domains have been uncovered but have not been pieced together to yield a coherent mechanism. In the mouse, genes that are eventually specific to the lungs, liver, and pancreas can be detected in small numbers of ventral endodermal cells at E8.5 (Serls et al.,2005), although detectable budding of these organ rudiments from the endodermal tube does not begin until one day later. Detailed transcription factor expression during this critical time period is lacking. Several signaling pathways, including those activated by retinoic acid (Stafford and Prince,2002; Molotkov et al.,2005; Pan et al.,2007), bone morphogenetic protein (BMP; Tiso et al.,2002; Shin et al.,2007), fibroblast growth factor (FGF; Jung et al.,1999; Wells and Melton,2000; Serls et al.,2005; Dessimoz et al.,2006), Wnt (Ober et al.,2007; McLin et al.,2007), Hedgehog (Hebrok et al.,1998) and Activin (Hebrok et al.,1998; Kim et al.,2000) families have been implicated in endoderm organ formation, yet little is known about direct transcriptional targets of these pathways. Additionally, adjacent tissues such as notochord (Kim et al.,1997), endothelium (Lammert et al.,2001; Matsumoto et al.,2001), cardiac mesoderm (Deutsch et al.,2001), septum transversum mesenchyme (Rossi et al.,2001), and lateral plate mesoderm (Kumar et al.,2003) are known to be important for endoderm organogenesis. Much work remains to fit these observations into a mechanistic model. One of the most basic facets of the organogenesis process that is poorly understood is the dynamics of expression of key transcription factors. Such knowledge could help to connect information about signaling pathways and inductive tissues to their specific transcriptional targets.

To begin to comprehend how the embryonic gut tube is subdivided into organ forming regions, we have focused on the expression of key transcription factors. These genes are, at a minimum, useful geographical markers and, in some cases, demonstrably play a role in specifying cell fates. We first performed microarrays of the endodermal progenitors in six organ domains to obtain an inclusive view of the factors expressed during organogenesis. We have followed up these microarrays by characterizing the dynamic cellular expression pattern for numerous genes that have restricted expression within the endodermal organs. These experiments provide a descriptive view of how the endoderm is patterned along its axes and where the organ domains form.


Microarray Analysis of Endodermal Organ Progenitors

To study gene expression during endodermal organogenesis, we sought to identify genes expressed in restricted domains during organogenesis. By E9.5, the endoderm is a tube that spans the embryo from anterior to posterior. All endodermal cells, as well as the notochord and floor plate of the neural tube, express Foxa2 (Ang et al.,1993; Monaghan et al.,1993; Ruiz i Altaba et al.,1993; Fig. 1a). Nascent organs are apparent as slight protrusions from the endoderm (Fig. 1b). These protrusions bud outward and gain a defined shape between E9.5 and E11.5 (Fig. 1c). The cell surface protein EpCAM is expressed most strongly in endoderm in the early mouse embryo (Sherwood et al.,2007), and EpCAM is expressed throughout the endoderm after organogenesis as well (Fig. 1d). Between E9.5 and E10.5, the hepatic endoderm loses expression of EpCAM (data not shown) but can be identified by expression of the cell surface protein Liv2 (Nierhoff et al.,2005).

Figure 1.

Diagrams and isolation strategy of endodermal organ domains. A: Embryonic day (E) 9.5 embryo whole-mount immunofluorescence image of Foxa2 antibody staining. B: Boxed region is magnified in B, and the nascent organ domains within the continuous endodermal epithelium are demarcated and labeled. C: Schematic of the gut tube at E11.5 with the distinct organ domains labeled. D: E9.5 embryo whole-mount immunofluorescence image of Foxa2 (red) and EpCAM (green). E,F: Flow cytometric analysis of live, dissected (E) E11.5 lung stained with EpCAM (X-axis) and (F) E11.5 liver stained with Liv2 (X-axis), plotted against side scatter (Y-axis). The percentage of cells within the boxed region is displayed.

For gene expression analysis, six morphologically distinct endodermal domains were dissected at E11.5: the esophageal region; the lung and distal tracheal region; the stomach region; the hepatic region; the dorsal and ventral pancreatic region; and the intestinal region. Through flow cytometric separation using EpCAM expression to distinguish endoderm from surrounding mesenchyme, pure populations of endoderm progenitors from the esophageal, lung, stomach, pancreatic, and intestinal regions were isolated (Fig. 1e, data not shown). Expression of Liv2 was used to isolate a pure population of hepatic endoderm progenitors (Fig. 1f).

Microarray analysis was performed on the six endodermal organ progenitor populations using Illumina whole genome arrays. Three biological replicates of each sample were obtained, and average expression among replicates was used to compare expression in tissues. To confirm the validity of the dissection and flow cytometric isolation of each domain, expression values of genes reported to be expressed in specific endodermal organs were analyzed. The esophagus-enriched Foxe1 (Dathan et al.,2002), lung-enriched Titf1 (Lazzaro et al.,1991), hepatic-enriched Afp (Shiojiri,1981), pancreas-enriched Ptf1a (Krapp et al.,1998), and intestine-enriched Cdx2 (Beck et al.,1995) were highly enriched in the appropriate microarray samples (Fig. 2a). Definitive endoderm from E7.75, E8.25 (4- to 6-somite stage), and E8.75 (13- to 15-somite stage) were also collected and profiled using high expression of EpCAM and lack of DBA expression (Sherwood et al.,2007). These data are available online at and at GEO (reference GSE13040).

Figure 2.

Results of microarray analysis of endodermal organs. A: Graph of microarray expression values of five genes reported to be specific to endodermal organ domains. Arbitrary expression values are normalized to the average expression among the six organ domains profiled and are displayed on a log scale with standard deviation of three replicates. B–D: Flow cytometric analysis of live, dissected (B) E11.5 stomach (left) and intestine (right) stained with Dpp4 (X-axis); (C) E11.5 stomach (left), pancreas (middle), and liver (right) stained with Dlk1 (X-axis); and (D) E11.5 stomach (left) and pancreas (right) stained with Rae1 (X-axis), all plotted against pan-endodermal Cdcp1 (Y-axis). The percentage of cells within each quadrant is displayed.

Statistical criteria were used to identify transcription factors with a restricted expression pattern in the endoderm. A total of 93 transcription factors were found to be expressed nonubiquitously in E11.5 endoderm organs (Table 1). Several known and novel transcription factors were specifically expressed in a single organ domain (Table 1). For example, the pancreas specifically expresses the known transcription factors Myt1, Neurog3, Nkx2-2, Nkx6-1, and Ptf1a as well as the novel factors Rxrg and Tcf15. The majority (55/93) of the organ-expressed transcription factors are expressed in multiple domains. Some are expressed in anterior–posterior segments. Hoxa1, Hoxa2, Irx3, Irx5, Sox2, and Sox21 are expressed in the anterior three organs (esophagus, lung, and stomach), and Dmrt3 and Tbx1 are expressed in the esophagus and lung. Conversely, Foxa3 and Gata6 are expressed in the five most posterior organs (lung, stomach, liver, pancreas, and intestine), Hnf4a and Onecut2 in the stomach, liver, pancreas, and intestine, and Nr5a2 in the liver, pancreas, and intestine (Table 2). However, there are a significant number of distinct expression patterns such as those of Osr2, which is expressed in the esophagus and intestine, and Mrg1, which is expressed in the esophagus, stomach and pancreas (Supp. Table S1, which is available online). Thus, while not unexpected, there is no simple relationship of a transcription factor or set of factors that accurately and uniquely marks the emergence of an organ in the endoderm.

Table 1. Transcription Factors Exclusive to One Organ Within the Endoderma
  • a

    Expression of these transcription factors is more than threefold higher in one organ than in all other organs, and expression is not detected significantly above background in other organs.

ErfIrx1 Creb3l3Neurog3Evx1
Foxe1Titf1 Klf2Nkx2-2Hoxa7
Nfix   Nkx6-1Hoxa9
Nrl   Ptf1aHoxb6
Otx1   RxrgHoxb7
Pitx1   Tcf15Hoxb9
Tcfab2c    Hoxd10
Twist1    Hoxd3
Zfpn1a2    Hoxd4
Table 2. Transcription Factors Expressed in Anterior–Posterior Blocks of the Endoderma
  • a

    Expression of these transcription factors is more than threefold higher in the organs indicated than in all other organs, and expression is not detected significantly above background in other organs. Gray shading represents expression of the noted transcription factors.

inline image

The Hox family of transcription factors has been implicated in a wide array of embryonic patterning events. Endodermal expression of Hox genes has been reported (Manley and Capecchi,1995; Beck et al.,2000); however, a systematic understanding of their expression and role in endoderm development is lacking. We found 19 of the 39 Hox genes to be expressed in the endoderm. Because the anterior boundary of Hox expression has been demonstrated to be of major functional significance in other germ layers (Duboule and Morata,1994), this boundary was analyzed in the endoderm. Of the 19 endoderm-expressed Hox genes, nine have an anterior boundary within the esophageal endoderm, although their actual anterior boundary may be in the more anterior pharyngeal endoderm, and 10 are exclusively expressed in the intestine (Supp. Table S1), suggesting that Hox genes may play more essential roles in pharyngeal and intestinal development than in the formation of the other major endodermal organs.

While they are not believed to be as integral to cell fate specification as transcription factors, cell surface proteins are useful in allowing for isolation of specific populations by flow cytometry, so differentially expressed cell surface proteins were sought. Several cell surface proteins expressed specifically in a single organ domain were identified (Supp. Table S2). Only a small subset of these proteins have commercially available antibodies suitable for flow cytometry, so these antibodies as well as antibodies to some cell surface proteins expressed in combinations of organs were tested by flow cytometry. By flow cytometric co-staining with the pan-endodermal Cdcp1 (Sherwood et al.,2007), it was found that Dpp4 is specifically expressed in intestinal endoderm (Fig. 2b), Dlk1 is expressed in pancreatic and hepatic endoderm (Fig. 2c), and Rae1 is expressed most strongly in stomach endoderm (Fig. 2d).

An Endoderm Transcription Factor Map at E9.5

To examine the expression of key genes throughout the endoderm at cellular resolution, immunofluorescence staining was performed on whole embryos at E9.5 and analyzed using confocal microscopy to achieve three-dimensional cellular resolution. This method allows annotation of expression of transcription factors such as the intestine-specific Cdx2 in the embryo as a whole and with sufficient resolution to analyze expression in individual endodermal cells (Fig. 3a).

Figure 3.

A high-resolution transcriptional map of the embryonic day (E) 9.5 stomach–intestine border. A: E9.5 embryo whole-mount immunofluorescence image of Cdx2 antibody staining (left). Boxed region is magnified in right panel, a whole-mount confocal immufluorescence image of the stomach–intestine border co-stained with Cdx2 (green) and Foxa2 (red). B–E: E9.5 embryo whole-mount confocal immunofluorescence images. B: Dorsal pancreatic region co-stained with Pdx1 (left and blue in merged image on right), Prox1 (middle, red), and Hlxb9 (abbreviated as Hb9, green). White lines demarcate regions of single expression of Hlxb9, co-expression of Hlxb9 and Pdx1, and co-expression of Hlxb9, Pdx1 and Prox1. C: Stomach–intestine border co-stained with Cdx2 (left, red), Onecut1 (middle, green), and Sox2 (right, blue) and merged image (far right). D: Stomach–intestine border co-stained with Cdx2 (left, red), Pdx1 (middle, green), and Sox2 (right, blue) and merged image (far right). E: Stomach–intestine border stained with Foxa2. White lines demarcate regions of expression of transcription factors listed in each region. Dotted white lines demarcate a subregion where, in addition to all other factors listed in the two solid-line regions, Pdx1 (in parentheses) is expressed weakly and is intermixed with Pdx1 cells. In the dorsal Pdx1-expressing region, Sox2 and Cdx2 (in parentheses) are expressed at E8.75, but they recede anteriorly and posteriorly, respectively, and become entirely excluded by E9.5. In thumbnail images, boxes highlight displayed region, and axes of embryo are labeled. Ant, anterior; Pos, posterior; Dor, dorsal; Ven, ventral.

Endodermal expression patterns of 15 transcription factors were investigated at E9.5. Our analysis was limited by the commercial availability of specific antibodies suitable for immunofluorescence. The region in which the liver and pancreas bud was of special interest, and as indicated by the microarray analysis, restricted expression of Cdx2, Hlxb9 (Hb9), Hnf4a, Nkx2-2, Onecut1, Pdx1, Prox1, and Sox2 was detected in this region. While expression of these genes in this region has been reported previously, the whole-mount confocal microscopy technique allowed resolution of their domains at significantly higher resolution and allowed analysis of co-expression, which has not been rigorously examined. Although this region will eventually become the stomach, pancreas, liver, and intestine, comparative analysis revealed multiple distinct expression patterns such as in the dorsal pancreatic region where Hlxb9 is expressed throughout the dorsal endoderm, and Pdx1 extends more broadly than Prox1 (Fig. 3b), yet all are co-expressed in much of the dorsal pancreatic region and are necessary for normal pancreatic development (Jonsson et al.,1994; Offield et al.,1996; Harrison et al.,1999; Li et al.,1999; Wang et al.,2005). Sox2 is expressed in most anterior endodermal cells, although it is excluded from certain regions including the nascent lung (data not shown; Que et al.,2007). At the anterior–posterior level of the dorsal and ventral pancreatic buds, Sox2 expression ceases and forms a border with Cdx2, which is expressed in all posterior endoderm cells (Fig. 3c). Neither Cdx2 nor Sox2 is expressed on the dorsal or ventral edges of the endoderm in this region. Onecut1 expression spans this Cdx2–Sox2 boundary and is additionally expressed in the dorsal and ventral hepatopancreatic regions (Fig. 3c), consistent with results from in situ hybridization (Landry et al.,1997; Rausa et al.,1997). Pdx1 expression is mostly exclusive with Sox2 and Cdx2, as it is expressed dorsally and in the small ventral pancreatic region (Fig. 3d). From this analysis, a cellular map of domains of transcription factor expression within the region spanning the nascent stomach, pancreas, liver, and intestinal endoderm at E9.5 can be constructed (Fig. 3e).

Anterior–Posterior Patterning Dynamics

Because the transcription factor expression patterns hint at developmental mechanism, the dynamic expression of transcription factors with distinct anterior–posterior boundaries was analyzed in detail. Endodermal Sox2 and Cdx2 can be detected at E7.75 in the most anterior and posterior cells, respectively, significantly before Onecut1, Pdx1, or Prox1 (data not shown). Between E8.0 and E8.5, the expression domain of Sox2 extends progressively posteriorly, and Cdx2 extends progressively anteriorly (Fig. 4a,b). By E8.75 (13- to 15-somite stage), the Sox2 and Cdx2 expression boundaries meet slightly posterior to the foregut–midgut junction, and there is co-expression of these genes in cells at the boundary (Fig. 4c). This co-expression recedes between the 13- and 15-somite stage and E9.0, and as noted above, the reciprocal Sox2-Cdx2 expression pattern remains at E9.5, except that some dorsal and ventral cells expressing pancreatic and hepatic genes express neither Sox2 nor Cdx2 (Fig. 3c,d). To provide a quantitative view of this border formation, average nuclear intensities of Sox2 and Cdx2 were calculated from cells spanning this border at several time points. At the six- to eight-somite stage, before Cdx2 and Sox2 form a visible border, the expected distribution of cells expressing either Sox2 alone, Cdx2 alone or neither is found (Fig. 4d). At the 13- to 15-somite stage, there is a population of cells expressing both factors (Fig. 4e). At E9.0, there are still a few cells that co-express Cdx2 and Sox2 yet most have resolved to express either Cdx2 or Sox2 (Fig. 4f), and by E9.5, the border is sharp with little to no co-expression (Fig. 4g).

Figure 4.

Anterior–posterior transcriptional dynamics at the stomach–intestine border. A: Embryonic day (E) 8.25 (four- to six-somite stage) embryo whole-mount confocal immunofluorescence image of Sox2 (green), Cdx2 (red), and Foxa2 (blue, right only). White lines demarcate region of endoderm expressing neither Cdx2 nor Sox2. B: E8.5 (9- to 11-somite stage) embryo whole-mount confocal immunofluorescence image of Sox2 (green) and Cdx2 (red). C: E8.75 (13- to 15-somite stage) embryo whole-mount confocal immunofluorescence image of Sox2 (green), Cdx2 (red), and Pdx1 (blue, right only). D–G: Graphs plotting mean nuclear intensity ratio of Cdx2 (X-axis) to Sox2 (Y-axis) in populations of cells spanning the stomach–intestine border at E8.25 six- to eight-somite stage (D), E8.75 (E), E9.0 (F), and E9.5 (G). Each dot represents one nucleus, and intensity values of Cdx2 and Sox2 are normalized as ratios to Foxa2 expression. H: E8.5 (six- to eight-somite stage) embryo whole-mount confocal immunofluorescence image of Sox2 (green, right only), Cdx2 (red, right only), and Onecut1 (blue). I: E8.75 (11- to 13-somite stage) embryo whole-mount confocal immunofluorescence image of Pdx1 (green), Prox1 (red), and Onecut1 (blue). In thumbnail images, boxes highlight displayed region, and axes of embryo are labeled. Ant, anterior; Pos, posterior; Med, medial; Lat, lateral; Dor, dorsal; Ven, ventral.

Because Cdx2 and Sox2 form a border in the middle of the pancreatic and hepatic domains at E9.5, the dynamics of pancreatic and hepatic transcription factors were studied in conjunction with Cdx2 and Sox2. Of the pancreatic and hepatic transcription factors analyzed, Onecut1 is expressed earliest, with expression starting in a stripe of foregut and midgut cells that covers the entire dorsal–ventral endoderm at the six- to eight-somite stage and spans both the Sox2 and Cdx2 expression domains (Fig. 4h). While the Sox2 and Cdx2 expression domains converge, Onecut1 remains expressed in a stripe spanning these two domains through E9.5 (Fig. 3c). Prox1, which has been shown by in situ hybridization to be expressed in the nascent pancreas and liver (Burke and Oliver,2002) becomes expressed in the ventral Onecut1-expressing cells by the 9- to 11-somite stage, and a subset of these cells ventral cells begins to express Pdx1 at the 11- to 13-somite stage (Fig. 4i). Dorsal Prox1 and Pdx1 expression do not appear until the 13- to 15-somite stage, and these cells co-express Onecut1 and either Sox2 or Cdx2 (Fig. 4c, data not shown), although Sox2 and Cdx2 are lost in this dorsal Pdx1-expressing domain by E9.5 (Fig. 3d).

To begin to link these expression patterns with the mechanism of anterior–posterior patterning in the endoderm, the effect of ectopic expression of transcription factors was studied. Plasmids carrying Cdx2 or Pdx1 DNA driven by a constitutive chicken beta-actin promoter were electroporated into the E8.25 foregut (anterior) endoderm (Pierreux et al.,2005). After electroporation, embryos were cultured for 24 hr and analyzed by whole-mount immunofluorescence and confocal microscopy. Mean nuclear intensity of Sox2 in cells ectopically expressing Cdx2 or Pdx1 to their neighbors was determined. Cells ectopically expressing Cdx2 have on average 50% as much Sox2 as their untransfected neighbors (Fig. 5a,b), whereas Pdx1 has no effect on Sox2 expression (Fig. 5c,d). Conversely, Sox2 was transfected into the posterior endoderm, and cells ectopically expressing Sox2 have 80% as much Cdx2 as their untransfected neighbors (Fig. 5e,f). Thus, Cdx2 appears capable of down-regulating Sox2 expression, providing a feasible explanation for stomach–intestine border formation.

Figure 5.

Dominance of Cdx2 over Sox2 in anterior–posterior patterning. A,C,E: Whole-mount confocal immunofluorescence of E8.25 embryos electroporated with (A) pCAGGS Cdx2 IRES GFP, (B) pCAGGS Pdx1 IRES GFP, or (C) pCAGGS Sox2 IRES GFP and cultured for 24 hr in 1:1 DMEM-F12:rat serum. A–C: In A, Cdx2 is in green and Sox2 in red; in B, Pdx1 is in green and Sox2 in red; and in C, Sox2 is in green and Cdx2 in red. The endogenous gene is typically expressed in all nuclei within the imaged territory, so green nuclei are indicative of ectopic gene down-regulation of the endogenous gene, whereas yellow nuclei are indicative of lack of down-regulation. B,D,F: Graphs plotting mean nuclear intensity ratio of the transfected factor (X-axis) to the endogenous gene (Y-axis). The average ratio of endogenous gene expression in transfected vs. untransfected cells is displayed to the right of each graph.

Hlxb9 Forms a Dorsal–Ventral Gradient at E9.5

While a large number of transcription factors have restricted anterior–posterior expression patterns, only one transcription factor, Hlxb9, which is expressed throughout the anterior–posterior axis of the endoderm, shows a dorsal–ventral asymmetry, as has been reported (Harrison et al.,1999; Li et al.,1999). Hlxb9 begins expression within the endoderm at E8.0 (data not shown). By E8.25, the endoderm is a single-cell thick layer with around 25 cells from left to right. Hlxb9 is expressed in the medial-most half of the endoderm surrounding the strongly Hlxb9-expressing notochord (Fig. 6a). At E8.5, after the anterior endoderm has undergone tubulogenesis to form the AIP, Hlxb9 retains expression in the dorsal-most 10–15 cell layers, presumably the descendants of the medial-most cells (Fig. 6b). At E9.5, expression is still detected in the dorsal half of the endoderm; however, expression appears to be strongest in the most dorsal endoderm and progressively weaker in more ventral cells (Fig. 6c).

Figure 6.

Hlxb9 forms a dorsal–ventral gradient within the endoderm. A: Embryonic day (E) 8.25 (four- to six-somite stage) embryo whole-mount confocal immunofluorescence image of Hlxb9 (abbreviated as Hb9, green) and Foxa2 (red, right only). White line demarcates region of endoderm expressing Hlxb9. B: E8.5 (9- to 11-somite stage) embryo whole-mount confocal immunofluorescence image of Hlxb9 (green) and Foxa2 (red, right only). Solid white line demarcates region of endoderm expressing Hlxb9 strongly and dashed white line demarcates region expressing Hlxb9 weakly. C: E9.5 embryo whole-mount confocal immunofluorescence image of Hlxb9 (green) and Foxa2 (red, right only). White line demarcates region of endoderm expressing Hlxb9, although levels of Hlxb9 expression decrease from dorsal to ventral. D–G: Graphs plotting ratio of mean nuclear intensity of Hlxb9 to Foxa2 (Y-axis) in endodermal cells at E9.5 (D), E8.5 (E), and E8.75 (F,G). Cells are plotted by their position along the dorsal–ventral axis (X-axis) from dorsal to medial (D,F,G) or medial to lateral (E). Equation and R2 value are shown for each graph. In F,G, graphs are of posterior (F) and anterior (G) regions of the same embryo. In thumbnail images, boxes highlight displayed region, and axes of embryo are labeled. Ant, anterior; Pos, posterior; Med, medial; Lat, lateral; Dor, dorsal; Ven, ventral.

As Hlxb9 appears to form a dorsal–ventral gradient of expression, statistical tools were used to determine whether this expression is indeed graded. The uniform endodermal expression of Foxa2 was used to normalize differences in intensity based on position within the epithelium. Thus, the average ratio of nuclear intensity of Hlxb9:Foxa2 was calculated for a dorsal–ventral line of endoderm in several anterior–posterior positions. When plotted, normalized Hlxb9 expression at E9.5 forms a linear gradient of intensity (Fig. 6d). This gradient forms between E8.5 and E9.5, as no linear decay of Hlxb9 can be observed along the dorsal–ventral axis at E8.5 (Fig. 6e).

At E8.75, some detectable dorsal–ventral asymmetry appears but without a clear gradient. In more posterior regions, the dorsal-most half of the Hlxb9-expressing endoderm (which is itself the dorsal half of the endoderm) has significantly higher expression than the ventral Hlxb9-expressing half (Fig. 6f). There is a sharp difference between the two halves, and there is no gradient or dorsal–ventral asymmetry within each half (Supplemental Fig. 1). More anteriorly, the two discernable halves are still present, but a gradient begins to form in the ventral Hlxb9-expressing cells (Fig. 6g, Supp. Fig. S1). Thus, Hlxb9 begins to form a gradient at E8.75, starting at the anterior end of the endoderm.

To determine whether a gradient forms by strengthening of Hlxb9 expression dorsally or weakening ventrally, the ratio of Hlxb9:Foxa2 expression was spatially and temporally compared. At all time points, regions of endoderm lacking visual Hlxb9 expression have a Hlxb9:Foxa2 ratio of 0.3 (data not shown), due to nonnuclear background. At E8.5, the ratio between Hlxb9 and Foxa2 averages 1, varying randomly with respect to the medial–lateral axis between 0.6 and 1.4. At E8.75, the dorsal half of the Hlxb9-positive region maintains an average expression ratio of 1, varying between 0.6 and 1.4, while the average ratio in the ventral half decreases to 0.7, varying between 0.4 and 1. While average Hlxb9 expression for each half at E8.75 does not change along the anterior–posterior axis, the ventral Hlxb9-positive half begins to take on a gradient, as noted above. At E9.5, at all anterior–posterior regions, the ratio decreases linearly from 1.4 to 0.3, suggesting that increased Hlxb9 expression in the most dorsal cells as well as decreases more ventrally contribute to gradient formation.


During a 24-hr period during mouse organogenesis between E8.0 and E9.0, the endoderm transforms from a flat sheet with little known heterogeneity into a tube in which the domains for all of the major organs have been defined. It has been appreciated that molecular heterogeneity precedes morphological differentiation, yet the transcriptional dynamics of endoderm organogenesis have not been well characterized. To begin to shed light on endodermal organogenesis, a combination of global transcriptional profiling and confocal imaging at cellular resolution has been applied.

Transcriptional profiling reveals the complexities of the organogenesis process. Few transcription factors are unique to individual organs. Organogenesis is likely to rely on multiple distinct signaling and transcriptional inputs, as is underlined by the fact that embryos lacking both Pdx1 and Ptf1a, the two transcription factors most vital to pancreatic development, still form a rudimentary pancreatic bud (Burlison et al.,2008) and ectopic expression of Pdx1 in the endoderm is insufficient to drive complete pancreatic morphogenesis (Grapin-Botton et al.,2001). Thus, these few organ-specific transcription factors are unlikely to be the sole effectors of organ morphogenesis. Instead, the large cohort of transcription factors expressed in multiple, but not all organ domains, are likely to make important, combinatorial inputs to organogenesis. Expression domains alone do not yield simple clues to mechanisms of organogenesis, as there are transcription factors expressed in many different combinations of organs; however, an increased understanding of mechanisms driving expression of these factors as well as their network interactions will be valuable.

Transcriptional profiling also allowed for the identification of cell surface proteins that can be used to isolate specific endoderm organ populations. Each organ has a signature set of cell surface proteins (Supp. Table S2) that can be exploited to isolate organ progenitors as well as subsets of cells within each organ. Using commercially available antibodies, intestinal endoderm can be isolated by its expression of Dpp4, hepatic and pancreatic endoderm by their expression of Dlk1, and stomach endoderm by its strong expression of Rae1. The ability to isolate specific populations of endoderm organ progenitors will be useful for embryonic stem cell differentiation experiments, in which cell types are not identifiable by morphology.

As a complementary approach to the global transcriptional profiling, whole-mount immunofluorescent staining followed by confocal imaging allowed elucidation of expression patterns at cellular resolution. This approach allows the inspection of transcriptional borders, which is difficult to do with whole-mount in situ hybridization because of a lack of cellular resolution, or with section immunofluorescence, in which only a small region can be analyzed. Such analysis is vital to comprehending the mechanism of organogenesis, as the molecular markers analyzed are likely the downstream effectors of organ-forming signals and transcriptional pathways.

Observing expression of Cdx2 and Sox2 between E8.0 and E9.5 reveals an interesting reciprocal relationship. Both genes become expressed in the endoderm before somitogenesis, when the entire endoderm is a flat sheet (summarized in Fig. 7a). Cdx2 expression begins in the most posterior endoderm and extends progressively anteriorly, while Sox2 expression begins in the most anterior cells and extends progressively posterior during early somite stages as the endoderm beings to form a tube at both ends. By the 13- to 15-somite stage, the Sox2 and Cdx2 boundaries meet in the midgut and some cells at this border co-express these two genes. This co-expression resolves, as by E9.5, while the cellular border remains, cells at the border express either Sox2 or Cdx2.

Figure 7.

Schematization of anterior–posterior and dorsal–ventral transcriptional dynamics in the endoderm. A: Endodermal Sox2 (green) and Cdx2 (red) expression start at the anterior and posterior ends of the endoderm, respectively. Between embryonic day (E) 8.0 and E8.5, the domains of these transcription factors expand toward each other, most likely as a result of endodermal cells that do not express either factor initiating expression of Sox2 or Cdx2. At E8.5 (∼11- to 13-somite stage), the Sox2 and Cdx2 expression domains meet, and between E8.75 and E9.25, these domains overlap slightly, although by E9.25, co-expression becomes rare. Pdx1 expression begins on the ventral and slightly later the dorsal side of the endoderm at E8.5 at the border of the Cdx2 and Sox2 domains, and the initial Pdx1-expressing cells co-express either Sox2 or Cdx2. By E9.25, dorsal and ventral Pdx1-expressing cells lose expression of Sox2 and Cdx2, although weaker Pdx1 expression is still detected in Sox2-expressing cells immediately anterior to the border of Sox2 and Cdx2. B: Endodermal Hlxb9 expression (blue) is present at E8.25 at uniform levels in the medial half of the endoderm. Through tubulogenesis, the medial endoderm stays dorsal while the Hlxb9 lateral endoderm wraps ventrally. By E8.75, the most dorsal quarter of the endoderm expresses Hlxb9 strongly, while the next quarter of the dorsal endoderm expresses Hlxb9 more weakly. A gradient of Hlxb9 expression emerges first in the anterior endoderm, which undergoes tubulogenesis earliest, and progressively more posteriorly, such that by E9.25, Hlxb9 is expressed in a gradient in the dorsal half of the endoderm throughout the anterior–posterior axis of the endoderm.

The extension of Cdx2 and Sox2 expression could occur by the migration of cells already expressing these genes, by oriented cell division of expressing cells, or by de novo expression in previously nonexpressing cells. Lineage tracing of endoderm suggests that the latter possibility is the most likely. Several lineage tracing studies of mouse somite stage endoderm have been performed by injecting dye into endodermal cells at specific locations and analyzing their descendants after further development (Lawson and Pedersen,1987; Tremblay and Zaret,2005; Franklin et al.,2008), and these studies do not suggest an extensive anterior–posterior migration that would be required for pre-existing Sox2 and Cdx2-expressing cells to colonize the entire endoderm. Thus, anteriorizing and posteriorizing signaling mechanisms must exist that spread expression of these two genes. The nature of these mechanisms is unclear and could consist of signaling molecules spreading toward the center of the embryo from the anterior and posterior poles or of timing-based mechanisms that allow cells that previously were exposed to morphogenic signals to begin to express these genes.

Once Cdx2 and Sox2 meet, a repressive interaction can be posited to stop the intermingling of these two expression domains. In fact, introduction of ectopic Cdx2 into Sox2-expressing foregut cells represses endogenous Sox2, while ectopic Sox2 does not repress Cdx2, suggesting that Cdx2 repression of Sox2 may cause the abrupt border between these genes. On the dorsal and ventral sides, Cdx2 also forms a border with Pdx1. While ectopic expression analysis was not performed to determine the effect of Cdx2 on Pdx1 expression, expression of Cdx4, a homolog of Cdx2, has been shown to limit pancreatic domain expansion in zebrafish (Kinkel et al.,2008). The ability of a more posteriorly expressed transcription factor to down-regulate an anterior transcription factor and the lack of the reciprocal effect provides a molecular basis for a commonly invoked concept known as posterior dominance (reviewed in McGinnis and Krumlauf,1992; Duboule and Morata,1994). Posterior dominance has been invoked in endoderm patterning (Kumar et al.,2003; Kimura et al.,2007), and this transcriptional repressive activity may help to explain this phenomenon.

Slightly before Sox2 and Cdx2 meet, at the six- to eight-somite stage, the first detectable hepatopancreatic regional marker, Onecut1, begins to be expressed in an anterior–posterior stripe spanning the Sox2 and Cdx2 domains. Onecut1 expression spans the entire dorsal–ventral axis, whereas Prox1 and Pdx1, which initiate expression at the 9- to 11-somite stage, are initially expressed first ventrally and then dorsally by the 13- to 15-somite stage but not in the medial endoderm (although by E9.0, Pdx1 is expressed weakly in medial cells that will become the posterior stomach and duodenum), and Hnf4a is expressed solely ventrally. The notochord is known to play a vital role in induction of Pdx1 in the dorsal endoderm (Kim et al.,1997; Hebrok et al.,1998), and the notochord becomes separated from the endoderm in the pancreatic region by the dorsal aorta between the 12- to 15-somite stage (Yoshitomi and Zaret,2004). Thus, Pdx1 can first be detected in the dorsal endoderm immediately after the notochord loses contact with the endoderm. This delay may be caused by the time taken for notochord-induced transcription of Pdx1 to be evidenced as protein expression and highlights the precise coordination of morphogenesis necessary for organogenesis, as the displacement of the notochord by the dorsal aorta is necessary to induce Ptf1a expression (Lammert et al.,2001; Yoshitomi and Zaret,2004).

At their inception, Onecut1, Prox1, Pdx1, and Hnf4a do not share borders, and even at E9.5, Pdx1 and Prox1 do not share an anterior border in the pancreatic domain. The presence of more expression domains than organs in the hepatopancreatic region suggests that multiple signaling mechanisms could be responsible for the initiation of expression of these genes. These additional hepatopancreatic domains could be vital for the formation of distinct structures within this region such as the pancreatic and bile ducts.

While anterior–posterior heterogeneity can be easily tied to the formation of organs at different positions along the anterior–posterior axis, there are fewer clues to the significance of the dorsal–ventral heterogeneity of Hlxb9. Embryos deficient in Hlxb9 have defects in dorsal but not ventral pancreas formation (Harrison et al.,1999; Li et al.,1999), suggesting a role for Hlxb9 in facilitating patterning of dorsal endoderm in the prepancreatic region. However, Hlxb9 is expressed throughout the dorsal endoderm, so it is feasible that its absence causes more subtle defects in dorsal–ventral patterning in other regions of the endoderm. The separation of the dorsal esophagus from the ventral trachea and the dorsal curvature and cell type disparity in the stomach are two events that could receive patterning input from Hlxb9.

While the functional significance of Hlxb9 in dorsal–ventral endoderm patterning is murky, its expression pattern clearly suggests that Hlxb9 is interpreting one or multiple dorsal–ventrally asymmetrical cues. Hlxb9 expression is expressed in the medial half of the endoderm by E8.0, and by E8.75, when the medial endoderm has become the dorsal endoderm due to ventral folding of the lateral endoderm during tubulogenesis, its expression level declines in the most ventral Hlxb9-expressing cells (summarized in Fig. 7b). Subsequently, slightly increased expression in the most dorsal Hlxb9-expressing cells and graded declines more ventrally create a linear gradient of expression in which each cell has less Hlxb9 than its dorsal neighbor. The linearity and slope of this gradient are within range of the values obtained for Drosophila Bicoid, whose concentration decreases ∼sevenfold between the anterior pole and the midpoint at a similar concentration/cell slope (Gregor et al.,2007). The data additionally suggest that all cells expressing Hlxb9 at E9.5 are descended from E8.5 Hlxb9-positive cells, barring significant cell movement, because the dorsal 50% of the endoderm consistently expresses Hlxb9.

Thus, the endoderm appears to be receiving graded morphogenic input along the dorsal–ventral axis, the nature of which is uncertain. Hlxb9 could directly or indirectly be positively regulated by signals deriving from medially/dorsally located tissues such as the notochord or negatively regulated by signals from the lateral plate mesoderm, or inputs could be combinatorial. Identifying these signals will be crucial to understanding endoderm organogenesis. Additionally, as Hlxb9 was the only transcription factor found to have dorsal–ventral polarity, it will be important to flush out the other transcription factors involved in dorsal–ventral patterning. One enticing candidate is Hhex, against which we have not found a working commercial antibody but which seems to be preferentially expressed in early ventral endoderm and is necessary for proper formation of the ventral pancreas and liver (Thomas et al.,1998; Martinez Barbera et al.,2000).

This work has provided a large volume of information about the genes that underlie organogenesis and has provided detailed characterization of the dynamics of anterior–posterior and dorsal–ventral patterning that guides the process. Linking these expression patterns and dynamics to causal mechanisms will be the next challenge. Meeting this challenge will likely be essential if one is to achieve efficient directed differentiation of embryonic stem cells to endoderm organ progenitors.


Animals and Flow Cytometry

For all experiments, outbred ICR mice were bred and maintained at the Harvard Biomedical Research Infrastructure. Embryos were considered to be E0.5 at noon of the day the plug was detected.

Dissected tissues pooled from 10 to 12 embryos per sample were dissociated in 0.25% Trypsin (Invitrogen, Carlsbad, CA) for 2–5 min at 37°C and, after neutralization and centrifugation, were stained on ice for 15 min in DMEM/F12 (Invitrogen) with 2% fetal bovine serum (FBS; Hyclone, Logan, UT) and 2 mM ethylenediaminetetraacetic acid (EDTA). Before flow cytometric sorting, cells were resuspended in staining media with calcein blue AM (Invitrogen) and sorted using a FACSAria (Becton Dickinson, San Jose, CA).

Electroporation experiments were performed according to published protocols (Pierreux et al.,2005). Full-length transcription factor cDNA was cloned into the pCIG vector, which has a chicken beta actin (CAGGS) promoter driving expression of a gene followed by an IRES driving GFP expression. Embryos were bathed in 1 μg/μl plasmid solution and electroporated using a T820 Electro Square Porator (BTX, Holliston, MA) using three 12-Volt pulses with 50-μsec interval. Embryos were cultured for 24 hr in 1:1 DMEM:F12 (Invitrogen): rat serum (Valley Biomedical, Winchester, VA).


Sorted cell populations were collected in Trizol (Invitrogen), and RNA was isolated according to the manufacturer's instructions. RNA was amplified and biotinylated cRNA probe generated using the Ambion Illumina TotalPrep RNA Amplification Kit (Applied Biosystems/Ambion, Austin, TX). Biotinylated cRNA was hybridized onto Illumina MouseRef-8 v2 microarrays (Illumina, San Diego, CA) according to the manufacturer's instructions. Microarrays were scanned using Beadstation (Illumina), and data were analyzed using BeadStudio (Illumina). Three biological replicates were performed for each tissue analyzed.

Antibodies and Immunostaining

The following primary antibodies were used: G8.8 (anti-EpCAM), F6A11 (anti-Pdx1), 74.5A5 (anti-Nkx2-2; Developmental Studies Hybridoma Bank, Iowa City, IA); goat anti-Foxa2 (M-20), goat anti-Sox2 (Y-17), rabbit anti-Hnf6 (H-100), rabbit anti-Hnf4a (H-171; Santa Cruz Biotechnology, Santa Cruz, CA); rat anti-Liv2, rat anti-Dlk1 (MBL International, Woburn, MA); biotinylated DBA (Vector Labs, Burlingame, CA); mouse anti-Prox1, rabbit anti-Otx2 (Millipore, Billerica, MA); mouse anti-Cdx2 (Biogenex, San Ramon, CA); rabbit anti-Hlxb9, rat anti-Pax9 (Abcam, Cambridge, MA); guinea pig anti-Pdx1 (gift from Christopher Wright); rabbit anti-Tbx1, mouse anti-Titf1 (Invitrogen); sheep anti-Cdcp1, rat anti-Rae1 (R&D Systems, Minneapolis, MN); mouse anti-Tcf2 and FITC-conjugated rat anti-CD26 (BD Biosciences, San Jose, CA). Alexa Fluor 488, 594, 647 and Pacific Blue secondary antibody conjugates (Invitrogen) as well as biotin, PE and APC conjugates (Jackson ImmunoResearch, West Grove, PA) were used for secondary detection.

For whole-mount immunostaining, whole embryos were blocked for 1 hr in PBS with 1% Tween-20, 20% donkey serum (Jackson ImmunoResearch), and when mouse primary antibodies were used, Vector M.O.M. Blocking Reagent (Vector, Burlingame, CA). Primary and secondary antibody stainings were performed rocking overnight at 4° in PBS with 1% Tween-20 and 5% donkey serum, and embryos were washed rocking for at least 5 hr at 4° in PBS with 1% Tween-20.

After staining, whole-mount embryos were transferred to 1:1 glycerol:PBS with 1% Tween and placed between two coverslips. Confocal imaging was performed using an LSM 510Meta confocal microscope (Carl Zeiss Inc., Germany). Pictures of whole embryos were taken using a fluorescent dissecting microscope (Leica Microsystems, Germany).

Quantitative Image Analysis

Total nuclear transcription factor expression was measured using Imaris 6.0.1 software (Bitplane, St. Paul, MN). The program reconstructed individual samples in three dimensions from confocal optical sections. The Surpass “add new spots” automated function was used to approximate nuclei as solid spheres of diameter 5 μm in user-defined regions of interest. Spots were filtered and manually edited to obtain one spot centered in each endoderm nucleus. When necessary, nuclei were divided into subgroups using the “Duplicate selection to new spots object” function. Intensity mean was taken to reflect the gene expression and exported to Microsoft Excel 2007 (Microsoft, Redmond, WA), and statistical analysis was performed in Excel as required.


The authors thank Brian Tilton, Anastasie Kweudjeu, George Kenty, and the Molecular Genetics Core Facility at Children's Hospital Boston for technical assistance, Christopher Wright for reagents, and William Anderson for review of the manuscript. The G8.8, F6A11, and 74.5A5 antibodies were obtained from the Developmental Studies Hybridoma Bank. D.A.M. is an investigator of the Howard Hughes Medical Institute. R.I.S. was supported by a National Science Foundation Graduate Research Fellowship and the Sternlicht Fellowship.