The formation and maintenance of the definitive endoderm lineage in the mouse: involvement of HNF3/forkhead proteins

The molecular mechanisms underlying the development of the definitive endoderm in vertebrates have remained obscure; this important lineage gives rise to the lining of the gut and its many outpockets during organogenesis. Even in Xenopus, where there is considerable information on the sig nalling mechanisms involved in the formation of the mesoderm lineage, there is a paucity of studies on the spec ification and determination of the endoderm lineage.

Cell fate studies in chick and mouse show that the defin itive endoderm does not arise from the primary hypoblast or primitive endoderm (Sanders et al., 1978; Gardner and Rossant, 1979), but arises de novo from ectoderm or epiblast at gastrulation (Rosenquist, 1971, 1972; Schoenwolf et al., 1992; Poelmann, 1981; Lawson and Pederson, 1987; Lawson et al., 1991). Single cell marking studies in the mouse have shown that, at the early primitive streak stage, there is a subpopulation of cells in the axial endoderm overlying the anterior of the streak that contributes to the gut endoderm (Lawson and Pederson, 1987). The rest of the endoderm layer at this stage will be displaced from the embryonic region and become extraembryonic endoderm, whereas definitive endoderm cells are most likely derived from underlying epiblast cells (Lawson et al., 1991). By the mid-streak stage, the anterior end of the streak reaches the distal tip of the embryo and forms the node, by analogy to Hensen’s node in the chick. The head process arises as a cranial extension of the node and appears to contribute to definitive endoderm by displacing primary endoderm cells in the midline (Poelmann, 1981). Cell fate studies in chick have shown that cells in early Hensen’s node can contribute to gut endoderm, as well as to notochord and the floorplate of the neural tube (Selleck and Stern, 1991; Schoenwolf et al., 1992). The limited information available in mouse is consistent with may be two early phases of definitive endoderm formation, one from anterior of the streak before formation of the node and one from the later head process.

With these cell fate studies in mind, one can predict where genes involved in the formation of the lineage should be expressed and thus focus attention on appropriate candidate genes. One source of candidate genes for endoderm formation are the tissue-specific regulatory factors involved in the later differentiation of endoderm-derived tissues. Most information is available on the liver, where a number of factors have been identified by their ability to bind to reg ulatory elements of liver-specific genes. These factors include HNFl, HNF3, HNF4, C/EBP, and DBP, all of which are highly enriched in expression in the adult liver (reviewed by Zaret, 1993). However, examination of their embryonic expression eliminates many of them from having a role in early definitive endoderm development. For example, the POU-homeodomain protein HNFla, which is necessary for high-level expression of many liver-specific genes (Courtois et al., 1987), is first expressed in the early mouse liver at day 10.5 p.c., well after initiation of the endoderm lineage (Blumenfeld et al., 1991; DeSimone et al., 1991). Similarly, mRNA for the liver-enriched leucine zipper protein, C/EBP, is first detected in the mouse liver at day 12.5 (Kuo et al., 1990). HNF4 is not detected in definitive endoderm until about 8.5 to 9.5 days p.c. (Sladek et al., 1990; Cereghini et al., 1992), and DBP is expressed late in ontogeny, after birth (Mueller et al., 1990).

By contrast, HNF3 transcription factors display charac teristics expected for proteins important for the endoderm lineage. HNF3α, HNF3β, and HNF3γ are products of three distinct genes, all of which are expressed exclusively in endoderm-derived tissues in the adult, such as liver, lung, and intestine (Lai et al., 1991), and play an important role in gene activation (Costa et al., 1989; Liu et al., 1991). The HNF3 family is characterized by a highly conserved DNA binding domain and two short carboxy-terminal domains (Lai et al., 1991). The DNA binding domain is 90% identical to a segment of the Drosophila forkhead gene (Weigel et al., 1989; Weigel and Jackie, 1990); this conserved sequence does not conform to any previously defined DNA binding domain, and therefore defines a new class of transcription factors, designated HFH (Clevidence et al., 1993). Numerous HFH genes expressed in various cell types are being discovered in organisms as diverse as yeast, nematodes, zebrafish, and humans (see Clevidence et al., 1993; Miller et al., 1993; Strahle et al., 1993); the HFH family may be as extensive as the homeobox genes. Several different HFH genes have now been isolated from Drosophila and shown to have highly specific expression patterns, suggestive of roles in different developmental processes (Grossinklaus et al., 1992; Hacker et al., 1992). However, the original forkhead gene, which shows the strongest homology to the HNF3 genes, is required for the formation of the anterior and posterior gut in Drosophila (Jurgens and Weigel, 1988). This suggested to us that HNF3 or related genes might be involved in the establishment of the endoderm lineage as well as its terminal differentiation.

Accordingly, we searched for HFH-related sequences in a cDNA library derived from late gastrulating mouse embryos, around the time of definitive endoderm formation. We found the mouse homolog of HNF3 as well as a novel HFH gene product, termed HFH-E5.l, and examined their expression in early embryogenesis. We further investigated the relative expression of HNF3 and other hepatocyte enriched transcription factors in a conditionally differentiat ing hepatic cell line. We conclude that HNF3 expression satisfies conditions expected for a factor that helps not only to establish the endoderm lineage, but also to maintain it. However, this may not be its only role, as it is expressed in other embryonic tissues. The expression of the novel HFH gene is also very restricted around the time of gastrulation, but this gene is not expressed in the endoderm, suggesting that HFH genes play multiple roles in development.

Low-stringency hybridization conditions for isolating HFH genes were established by examining the Southern blot hybridization profiles of mouse genomic DNA probed with different segments of the rat HNF3a cDNA (Lai et al., 1990). When the nitrocellu lose blots (Sartorius) were hybridized in 5x SSC, 0.5% SDS, 1 mM EDTA at 32°C for 18 hours, and washed in 0.lx SSC, 0.1% SDS at 23°C four times for 5 minutes, then at 39°C twice for 1 hour, coding segment pr.obes from both upstream and downstream of the DNA binding domain hybridized to single EcoRI and HindIII fragments, whereas a 473 bp PpuMl fragment encompassing the DNA binding domain hybridized to about 10 different DNA fragments in each digest (data not shown). The PpuMl fragment was used with the same hybridization conditions with BA85 filters (Schleicher & Scheull) to screen a total of l.5x10⁵ phage from a 11,gtl0 cDNA library prepared from mRNA of 8.5 day mouse embryos (kindly provided by Brigid Hogan; Farhner et al., 1987); E. coli C600 was the host. Restriction fragments encompassing the HFH homology were identified by hybridization and sequenced with the dideoxy procedure. cDNA sequences encoding the carboxy terminus of HNF3P were isolated by PCR from a 11,gtl 1 cDNA library prepared from mouse liver (kindly provided by R. Costa, E. Paulson, and J. Darnell). DNA sequences were analyzed with software from_the Genetics Computer Group, Inc.

Mouse embryos at various stages of gestation were obtained by mating random-bred CDl animals (Charles River Canada, Montreal). Noon on the day of vaginal plug appearance was con sidered to be embryonic day 0.5; gastrulation occurred between 6.5 and 8.5 days. Very early streak embryos showed asymmetry and were classified morphologically by the thickening of the epiblast in the posterior end, these lacked a distinguishable mesoderm layer. Mid-streak embryos contained mesoderm only in the posterior region. In late-streak embryos, mesoderm has reached its anterior most extent, the head process has begun to extend, and the amnion is closed. Headfold stage embryos showed neural folds at the anterior end. Late headfold stage/foregut stage embryos contain a foregut pocket and more prominent neural folds.

Paraffin sections of different stages of mouse embryos were hybridized to ³⁵S-labelled anti-sense RNA probes, according to Yamaguchi et al. (1992). The HNF3P probe was synthesized from a 566 bp cDNA clone consisting mostly of unique sequences outside the DNA binding domain, using T7 polymerase. RNA probe for HNF3α was transcribed from a 480 bp cDNA clone, which does not include the HFH DNA binding domain. HFH-E5.l probe was made from a 580 bp cDNA fragment (Fig. 1B) subcloned into pGEM3 using T7 polymerase. Emulsion coated slides were exposed for 2-4 weeks. After developing slides, sections were stained with toluidine blue.

The same HNF3α, HNF3β and HFH-E5.l cDNA fragments, as described above, were used as templates to make digoxygenin labelled RNA probes for whole-mount in situ hybridizations, as described (Conlon and Rossant, 1992). Sites of hybridization were detected with an anti-digoxigenin antibody coupled to alkaline phosphatase. Embryos were subsequently cleared in 50% glycerol prior to photography. To obtain sections of embryos stained by the whole-mount procedure, the embryos were first rehydrated into phosphate-buffered saline (PBS) containing 0.1% Tween 20, then postfixed in 4% paraformaldehyde in PBS at 4°C for 1 hour. Following this, embryos were dehydrated through 70%, 80%, and 90% ethanol for 30 minutes each. This was followed by four 15 minute incubations in 100% ethanol. Embryos were then soaked in xylene (2x 15 minutes) and paraffin wax (3x 1 hour) at 60°C. After orientation and embedding in paraffin wax, 10 μm sections were cut from the paraffin blocks. Sections were dewaxed through two changes of xylene (10 minutes each) and then mounted in Permount. The sections were photographed using Nomarski optics.

Mouse embryos were dissected free of decidua, and either mid sections (days 9.5 and 10.5 p.c.) or newly formed livers (day 12.5) from multiple embryos were pooled and kept at 4°C. Tissues were homogenized in 0.5 ml (9.5 day) or 2 ml (all other samples) volumes of an ice cold solution of 25 mM Tris pH 7.5, 138 mM NaCl, and 2.7 mM KC! (TBS) in a 1.5 ml microfuge tube or 7 ml glass tube, respectively, with a teflon pestle. Cells were collected by centrifugation, resuspended into 0.5 ml TBS, transferred to a fresh 1.5 ml tube, and pelleted. Nuclei were isolated and protein was extracted as described by Schreiber et al. (1989), except that nuclei were incubated on ice in buffer C with vortexing every 2 or 3 minutes. Adult liver nuclear protein was isolated, and HNF3a protein was translated in vitro, as described (Jackson et al., 1993). Electromobility shift assays employed either 5 μg of nuclear protein or 1 μI of in vitro translation product in a 20 μI volume, as described (Jackson et al., 1993). Binding reactions included either 500 ng (9.5 and 10.5 day extracts), 750 ng (12.5 day extract), 1500 ng (adult liver extract), or 250 ng (HNF3 in vitro translation product) poly (dl-dC). Oligonucleotide probes, end-labelled with ³²P, consisted of the eG binding site for HNF3 from the mouse serum albumin enhancer (Liu et al., 1991), the APF-1 binding site for HNF4 from the apolipoprotein CIII promoter (Costa et al., 1990), and the pC binding site for NF-Y from the mouse albumin promoter (Raymondjean et al., 1988). Nonlabelled competitor DNAs were used at a 100-fold molar excess, and included the eG and pC sites as well as a strong HNF3 binding site from the mouse transthyretin gene promoter (Costa et al., 1989). Western blot analysis was performed with nuclear extracts from liver and H2.35 cells, as described (Liu et al., 1991). The HNFla-specific antibody was kindly provided by Dr G. Crabtree.

HNF3 binding site probes suitable for UV-crosslinking were prepared by annealing an 8 bp primer to an oligonucleotide of the bottom strand sequence of the eG site, and extending the primer with Kienow polymerase in the presence of dGTP, [³²P]dCTP, ^[32P]dATP, and bromodeoxyuridine-substituted dUTP (Br-dUTP) (Chodosh et al., 1987). Binding reactions were performed in lids of 0.5 ml rnicrofuge tubes and were subsequently exposed to ultra violet light from a Spectroline EB-280C lamp with a 302 nm filter, yielding 860 mW/cm² at 15 cm, at a distance of 7 mm for 1 minute. Reaction products were resolved by the electromobility shift assay, identified by autoradiography, excised as gel fragments, and equi librated in stacking gel loading buffer (Laemmli, 1970) for several hours. The fragments were then placed into the wells of 10% poly acrylarnide gels containing SDS (Laemmli 1970); after elec trophoresis, gels were dried and autoradiographed. In some exper iments, the gel fragments were pre-equilibrated in 0.125 M Tris-HCI pH 6.8, 0.1% SDS, and 1 mM EDTA, then digested with 50-500 ng VS protease in pre-equilibration buffer containing 10% glycerol (Gooderham, 1984), prior to electrophoresis in a 15% polyacrylarnide-SDS gel.

To isolate HFH genes expressed during endoderm differen tiation, we used low-stringency hybridization conditions to screen a AgtlO cDNA library prepared from mouse embryos at 8.5 days gestation (Fahrner et al., 1987), using a rat HNF3α DNA binding domain probe (Lai et al., 1990). DNA sequence analysis of the most strongly hybridizing cDNA showed that it contained 570 bp that were 96% identical to rat HNF3 over the carboxy-terminal sequences unique to HNF3 (sequence submitted to Genbank). By comparison, both rat HNF3 and the mouse clone only show extensive similarity to rat HNF3a or HNF3y in the DNA binding domain (Lai et al., 1991). We conclude that this cDNA encodes mouse HNF3β. Additional segments of HNF3 cDNA were cloned by screening a mouse liver cDNA library; sequences of overlapping segments obtained from the liver and embryo libraries were identical. Northern blot analysis showed that mouse HNF3 is encoded by a 2.2 kb mRNA (data not shown), similar to HNF3 in the rat (Lai et al., 1991), indicating that the cDNA isolated is not full length. Comparison with the rat gene indicates that the cDNA sequence begins in the first third of the DNA binding domain and lacks amino-terminal sequences.

The second cDNA isolated from the mouse embryo library contained an open reading frame which encodes a novel HFH-related sequence, HFH-E5. l (Fig. 1A). The open reading frame begins with an ATG at nucleotide 355, and has an A residue 3 bp upstream, as expected for an efficient translation start site (Kozak, 1986). Another ATG exists at position 270 in a different reading frame, but it contains a C residue at position -3, which is inhibitory to translation (Morie et al., 1985). There is a stop codon in frame 183 bp upstream of the ATG at 355, making it likely that this ATG is the initiation codon. The cDNA is incom plete, and ends in the middle of the putative HFH domain. The putative DNA binding domain of HFH-E5.1 is 66% identical to HNF3, compared to the 90% identity between HNF3 proteins andforkhead proteins (Fig. 1B). The HFH E5.l cDNA is similar in structure to that for HTLF (Li et al., 1992) and lin-31 (Miller et al., 1993), in that all appear to have relatively long 5′ noncoding regions and only several amino acids prior to their HFH domains. However, of the HFH genes that have been cloned to date, HFH-E5.l exhibits the most similarity, 85% over the HFH domain, with the FD4 cDNA segment recently isolated from Drosophila (Häcker et al., 1992), and with the XFD5 genomic sequence from Xenopus (Knöchel et al., 1992).

Alignment of published HFH domain amino acid sequences with the HFH-E5.l clone shows that they can be arranged in groups that contain common amino acid substi tutions; the distinctions are emphasized when amino acids upstream of the HFH domain are considered (Fig. 1B). HFH-E5.l, FD4 and FD5 (Häcker et al., 1992), and XFD5 (Knöchel et al., 1992) contain a highly similar sequence at the amino terminus of the HFH domain, whereas HNF3 proteins,forkhead protein, and XFD3 (Knöchel et al., 1992), XFKHl (Dirksen and Jamrich, 1992) and pintallavis proteins (Ruiz i Altaba and Jessel, 1992) from Xenopus comprise a different, highly related group. We conclude that HFH-E5.1 is distantly related to HNF3/forkhead genes and that it represents another class of HFH genes expressed in mouse embryos during gastrulation.

Both whole-mount and sectioned in situ hybridization tech niques were used to determine the temporal and spatial expression of HNF3β during gastrulation and neurulation. The earliest expression of HNF3β detected by whole-mount in situ hybridization was at the very early primitive streak stage, in a small patch of epiblast cells at the anterior of the streak (Fig. 2A). This initial staining pattern expands to include more cells so that by the mid-streak stage, a large group of cells within the anterior third of the streak express HNF3β (Fig. 2B). Sections of such embryos showed that expression at these early stages is confined to epiblast cells and delaminating mesoderm cells in the anterior streak (Fig. 2F). At the late streak stage, as the definitive node becomes apparent at the distal tip of the embryo, expression of HNF3β was found in both layers of the definitive node (Fig. 2C and data not shown). Anteriorly migrating mesoderm and endoderm cells of the head process, which begins around this stage, also express HNF3β (Fig. 2C). Slightly later, at the headfold stage, whole-mount stained embryos revealed that the neural plate, notochord and the underlying midline endoderm all expressed the gene (Figs. 2D). By late headfold/early foregut pocket stage, endoderm expression became broader at the anterior end and spread into the devel oping foregut pocket (Fig. 2E). ³⁵S-labelled in situ hybrid ization to tissue sections of late streak and headfold stage embryos basically confirmed these results, but showed broader HNF3 expression in the endoderm lineage (Fig. 3B,E). At the late streak stage, expression was seen in patches of cells in the endoderm layer overlying the entire extent of the streak, and not just at the anterior end. At headfold stages, anterior endoderm expression was not restricted to the midline but extended laterally on either side (Fig. 3E). The broader domains of endoderm expression in late-streak and headfold stages detected on sections, compared to whole embryos, presumably reflects a higher sensitivity of the former method.

At 8.5 days (6–10 somites), expression persisted in the ventral part of the developing neural tube, from the forebrain to the posterior end, in a region encompassing the floorplate and cells lateral to it (Fig. 4A). Within the midbrain region, sections revealed that the staining spread even more dorsally on either side of the floorplate (Fig. 3H). These sections also showed expression in notochord and gut. In the gut, stronger expression was observed in foregut and hindgut at early somite stage (data not shown), but by the 10 somite stage the entire gut showed similar levels of expression (Fig. 4A). Interestingly, HNF3P expression was elevated in a portion of the gut beneath the cardiac region (Fig. 4A,C). The locally elevated levels of HNF3P appear to coincide with or precede morphological differentiation of the hepatic endoderm.

At 9.0–9.5 days (15–25 somites) of development, the expression domains of HNF3P established earlier basically persisted. Expression was seen throughout the notochord and the ventral part of the neural tube, as well as the entire gut (data not shown). Strong expression of HNF3P was also observed by 9.0–9.5 days in the liver primordium, which begins to be morphologically distinguishable at this stage (Figs 3K, 4C).

Later in development, at 16.5 days, HNF3β expression continued to mark the developing endoderm-derived tissues, such as lung, liver, pancreas and gut, with expression in the lung restricted to the developing respiratory epithelium; no expression was observed in mesoderm-derived organs such as the heart and kidney (Fig. 3N). Thus, early expression of HNF3P in endoderm tissue persists throughout development and remains expressed in these tissues in the adult (Lai et al., 1991). However, even by 16.5 days, expression was not entirely endoderm-specific, because expression was still observed in the posterior hypothalamus (data not shown) and cartilage of the vertebrae (Fig. 3N).

We examined the expression of the related gene, HNF3a, to see if early expression in the endoderm lineage was a common feature of HNF3 genes. Localized expression of HNF3a was not detectable by whole-mount in situ hybrid ization in the early primitive streak stage. Expression was first detected in the endoderm, by in situ hybridization to tissue sections at the late-streak stage of development (Fig. 3C). In contrast to HNF3P expression, no staining was detected in axial mesoderm and neural plate at the late streak and headfold stages respectively (Fig. 3C,F). From the 4-10 somite stage onwards, expression of HNF3a in the ventral floorplate, notochord, and gut appeared very similar to HNF3P, but weaker (Fig. 4B). Some differences in expression patterns, however, were also observed. At the 10 somite stage, expression of HNF3a was only observed in midbrain, but not in forebrain or hindbrain (Fig. 4D,E). Ventral floorplate expression of HNF3a was also not detected at this stage in the spinal cord (Figs 3I, 4D). Expression in the gut at 8.5 days and 9.5 days extended further anteriorly for HNF3 than HNF3a (Fig. 3K,L).

The presumptive liver primordium between 8.5 and 9.5 days showed similarly strong expression of HNF3a and HNF3 (Figs 3K,L, 4C,F). Expression of HNF3a in the ventral neural tube spread caudally to hindbrain and spinal cord regions by 9.0–9.5 days, but expression was restricted to the floorplate region (Fig. 3L). By 16.5 days, expression patterns of both genes were similar, except for the lack of expression of HNF3a and HNF3 in the vertebrae and the lining of the bladder, respectively (Fig. 3N and data not shown). Thus, although the patterns of expression were similar for the two genes, HNF3 was generally expressed earlier and more broadly than HNF3a. While the expression pattern in endoderm lineages appears to be very similar for both genes from the late-streak stage to the 16.5 day embryo, the earlier expression of HNF3 in the primitive streak suggests that it could play an earlier role in specifying the endoderm lineage. The difference in expression patterns detected by the HNF3 and HNF3a probes also confirms their hybridization specificity. In addition, a shorter HNF3 probe, which does not include the HFH DNA binding domain, gave similar results in whole-mount in situ hybrid ization analysis of 7.5 and 8.5 day embryos (data not shown).

High levels of mRNA for liver-enriched transcription factors are sometimes found in various non-liver tissues where the corresponding proteins are und tectable (e.g., Birkenmeier et al., 1989); these precedents demonstrate that conclusions about active transcription factors cannot be drawn from mRNA studies alone: Furthermore, it is important to establish that HNF3 proteins are not only present in the developing endoderm lineage, but also competent to bind target DNA response elements.

To address this issue, nuclear proteins from midsections of mouse embryos at 9.5 and 10.5 days gestation, and from nascent livers at 12.5 days, were subjected to an electromo bility shift assay with the eG binding site for HNF3 from the mouse serum albumin enhancer (Liu et al., 1991). As controls for the mobility of eG-HNF3 complexes, we used an adult liver nuclear extract and in vitro translated HNF3α.. As expected, the liver extract produced an abundant protein DNA complex that migrated slightly faster than the synthetic HNF3α.-DNA complex, and a less abundant, faster migrating complex (Fig. 5A); the upper complex contains both HNF3α and.HNF3β, while the lower complex contains HNF3γ (Lai et al., 1991). Extracts from the 9.5 and 10.5 day embryo midsections produced several protein-DNA complexes; one of these comigrated with the HNF3a/HNF3j3 complexes from liver (left arrow, Fig. 5A), with an additional faster-migrating complex in the 10.5 day midsections. All embryonic complexes were specifically competed by two different HNF3 sites, but not by a site for the ubiquitous factor NF-Y (Fig. 5A). The integrity of the embryonic extracts was demonstrated by the presence of NF-Y binding activity similar to that observed in adult liver (Fig. 5B). We conclude that the 9.5 day embryo midsection contain proteins with the DNA binding properties of HNF3α and HNF3β; no evidence was obtained for the presence of HNFγ. The assay also shows that 9.5–10.5 day embryos contain other proteins with a similar DNA binding speci ficity.

By 12.5 days of development, the DNA binding complexes present in the newly formed liver contained HNF3γ in addition to HNF3a and 13, and were identical in pattern and binding specificity to those seen with adult liver extracts, albeit lower in abundance (Fig. 5A). The lower abundance could be due to about half of the cells of the 12.5 day liver being hematopoietic (Paul et al., 1969), reducing the concentration of hepatocyte transcription factors. Control binding reactions with the NF-Y probe also showed several-fold less binding activity (Fig. 5B).

To demonstrate conclusively that HNF3 proteins were present in the embryonic extracts, we analyzed the migration and protease sensitivity of UV-crosslinked protein-DNA complexes. Specific binding complexes were cross-linked, excised from the gel, denatured in situ, and subjected to elec trophoresis in a second dimension in an SDS-polyacry lamide gel. As seen in Fig. 5B, the most prominent band from each nuclear extract comigrates with in vitro translated HNF3a. In addition, when the excised bands were incubated with increasing concentrations of V8 protease, the main terminal digestion products were identical in migration to that of in vitro translated HNF3 (Fig. 5B). These data show that HNF3 proteins are present and competent to bind their target sequences during endoderm differentiation, and that the adult pattern of HNF3 DNA binding activities is estab lished by 12.5 days of development.

Detailed studies of differentiated and de-differentiated somatic cell lines have provided insight into transcription factors important for the endoderm lineage. For example, de-differentiated cell lines derived from liver tumors typically lack the expression of HNFla (Cereghini et al., 1988; Baurnhueter et al., 1988), and a recent study has shown this could be due to a lack of HNF4 (Kuo et al., 1992). By contrast, such studies have shown that members of the HNF3 family often continue to be expressed by the de-differentiated cells (Kuo et al., 1992; Nitsch et al., 1993). Possibly, HNF3 proteins confer the ability of rare, re-dif ferentiated variants to be recovered from the cell population during selection or screening (Cereghini et al., 1988; Kuo et al., 1992). We sought to investigate this issue in greater detail.

Previously, we demonstrated that the H2.35 cell line, cloned from SV40-infected mouse hepatocytes, expresses extremely low levels of serum albumin mRNA under standard culture conditions, but when cultured on a collagen gel substratum and in a serum-free medium, albumin mRNA is induced over a hundred fold (Zaret et al., 1988). This system provides the opportunity to examine factors expressed in a cell population capable of homogenous re differentiation within the endodermal lineage.

In earlier studies we showed that de-differentiated H2.35 cells lack C/EBP and related DNA binding activities, but that they contain low levels of HNF3 that are increased upon re-differentiation (Liu et al., 1991). Importantly, the low amounts of HNF3 in de-differentiated H2.35 cells are suffi cient to activate transcription of a transfected reporter gene bearing HNF3 binding sites (DiPersio et al., 1991). We therefore wanted to address whether or not HNF4 and HNFla are present in the de-differentiated state. Electro mobility shift assays readily detected HNF3 in de-differen tiated H2.35 cell nuclear extracts, but no HNF4 DNA binding activity was found, using a high affinity HNF4 binding site from the apolipoprotein CIII gene (Fig. 6A). Northern blot analysis of H2.35 mRNA also failed to detect HNF4 mRNA (data not shown). As expected, abundant HNF4 DNA binding activity was observed in nuclear extracts from mo.use liver and a partially differentiated hepatoma cell line, HepG2 (Fig. 6A). No HNFla protein was detected in de-differentiated H2.35 cells in a western blot assay, while abundant levels of the protein were readily detected in liver extracts (Fig. 6B). Subsequent detection of NFl, a ubiquitous transcription factor (Santoro et al., 1988), demonstrated the integrity of the H2.35 cell extracts. We conclude that under de-differentiating culture conditions, H2.35 cells lack HNFl, HNF4, and C/EBP, and that HNF3 is the only known liver transcription factor family expressed and active in the cells. These findings suggest that all liver transcription factors other than HNF3 can be lost from cells in culture without affecting the ability of the cells to re-dif ferentiate along the hepatic endodermal lineage.

Expression of HFH-E5.l was first detectable at the midto late-streak stage of development, in a specific domain in the posterior-distal region of the embryo (Fig. 7A). Optical sec tioning through whole-mount stained embryos and in situ hybridization to tissue sections showed that expression at this stage was found in both ectoderm and mesoderm, but not in the endoderm (data not shown). By the early headfold stage, there was expansion of this expression to define a new anterior boundary. Immediately posterior to this boundary, a band of strong staining in mesoderm and overlying neural plate was observed (Fig. 7B,C). From this stage onwards, expression diminished in the mesoderm, but increased in the developing neural tube so that by the early somite stage (12 somites), expression was strong in neurectoderm in the anterior half of the embryo (Fig. 8B). The only staining in mesoderm anterior to the node at this stage was found in the region surrounding the midbrain, which probably corre sponds to the initial band of strong staining anteriorly (data not shown). Posteriorly, weaker staining was observed in presomitic mesoderm and in the midline ectoderm (Fig. 8B). In the neural tube, stronger expression was observed in midbrain and parts of the hindbrain. By the 6–10 somite stage, the entire neural tube from midbrain to the posterior end expressed HFH-E5.l (Fig. 7D). Mesoderm staining was then found in medial stripes of somites adjacent to the neural tube, and stronger expression was also found in the pre somitic mesoderm. Mesoderm at the midbrain level also continued to show weak expression (data not shown). This pattern of HFH-E5.l expression continued into 9.5 days.

Examination of sectioned in situs at 13.5 days and 15.5 days revealed that HFH-E5.1 continued to be expressed in the central nervous system (CNS). Specific groups of cells in the hypothalamus, midbrain, and medulla oblongata expressed HFH-E5.1 (Fig. 8F,H). In addition, staining was also observed in scattered cells of the ventral spinal cord (data not shown); no expression was detected outside the CNS at these later stages.

Knowledge of the forkhead family of regulatory proteins is rapidly expanding and reveals a group of genes that appear important in several developmental processes. The require ment of the Drosophilaforkhead gene for proper gut devel opment (Jurgens and Weigel, 1988) prompted us to ask whether the expression of forkhead homologs in the mouse could fulfill criteria expected for regulatory factors important for the formation of mammalian gut endoderm. While this work was in progress, Sasaki and Hogan (1993) independently reported the expression of HNF3 and HNF3a in mouse embryos between days 6.5 and 9.5 of gestation. Our focus on the expression of these genes in the endoderm lineage has revealed differences with respect to the early expression of HNF3 reported previously (Sasaki and Hogan, 1993). We show that HNF3 is expressed at the earliest stage of gastrulation in a group of cells that includes the progenitors of the definitive endoderm. Slightly later, at the late-streak stage, HNF3a begins to be expressed in the endoderm, together with HNF3 . Our studies of later stages showed that both HNF3 and HNF3a are expressed similarly throughout the subsequent differentiation of gut endoderm, during organogenesis, and that they appear to maintain the endoderm lineage after development.

To determine whether all progenitors of the definitive endoderm express HNF3, gene expression patterns must be compared with fate mapping studies. Fate mapping of the developing endoderm in mice has shown that there are already definitive endoderm cells on the surface of the anterior primitive steak by the mid-streak stage of develop ment (Lawson and Pederson, 1987), although much of the definitive endoderm seems to arise later from the head process (Lawson et al., 1991). HNF3P was clearly expressed in the group of cells at the anterior end of the developing primitive streak; this region of the epiblast is fated to give rise to the definitive endoderm (Lawson et al., 1991). Thus, HNF3 expression marks at least a subset of the earliest endodermal precursors, suggesting that this gene plays a fundamental role in establishing the lineage. Given that HNF3 proteins represent the vertebrate HFH products with the greatest homology to the Drosophila forkhead gene (Weigel et al., 1989), there appears to be evolutionary con servation of mechanisms for specifying the developing gut.

As the head process forms, both HNF3 and HNF3a are strongly expressed in the midline endoderm cells, which are known to subsequently form part of the definitive gut (Poelmann, 1981), as well as in cells more anterior in the midline, which may represent the progeny of the early defin itive endoderm precursors. Cells expressing lower levels of HNF3 were also seen lateral to the midline; the exact rela tionship between these cells and the later definitive endoderm is not yet clear. From this stage on, the expression of the two genes seems to be virtually identical in endoderm derivatives. Both genes are clearly expressed in the devel oping foregut and hindgut, although with different anterior boundaries, and expression is elevated in the early liver pri mordium. Later in development, expression persists in the liver and gut and then appears in the developing lung, a pattern consistent with the expression of these genes in the adult (Lai et al., 1991). These results suggest that both genes could play a role in the development and maintenance of the endoderm lineage.

The elevated level of expression of HNF3a and HNF3 mRNA in liver primordia is interesting in light of the known mechanisms of organogenesis. Tissue explant studies in the chick have shown that cardiac mesenchyme induces adjacent endodermal cells to assume a hepatic fate (Le Douarin, 1975); limited studies with mouse embryo tissues are consistent with this model (Houssaint, 1980; Cascio and Zaret, 1991). We therefore suggest that one of the first responses of endoderm to hepatic induction is an increase in the level of expression of HNF3 genes. Indeed, this response seems to parallel or precede morphological differentiation of the endoderm. Our microscale assays showed that the region containing mouse liver primordium at 9.5 days gestation contains HNF3 proteins competent to bind DNA. This indicates that HNF3 is capable of functioning very early in the definitive endoderm lineage. We further showed that HNF3γDNA binding activity was not detectable at 9.5 days, but that it was detectable in the early liver at day 12.5; thus, HNF3ymay act later in development than HNF3β or HNF3α.

A role in early endoderm specification is supported by the finding that only HNF3a continues to be expressed and active when a hepatocyte-derived cell line is maintained in the de-differentiated state; all other known liver-specific transcription factors, which are readily detectable in mature hepatocytes, are undetectable in the de-differentiated H2.35 cells. HNF3 may serve, therefore, to impart an endoderm identity to cells both in vivo and in vitro, perhaps by main taining liver genes in a state that is permissive for induction during differentiation. The ability of HNF3P to bind its own promoter (Pani et al., 1992) fits nicely with this model; once the HNF3β gene is activated, it would be sufficient to maintain its own synthesis as well as potentiate the expression of genes that are downstream in the regulatory cascade (Fig. 9).

HNF3α and HNF3β are both expressed in tissues other than t_he developing gut, making it possible that they play roles in other developmental processes. HNF3P, in particu lar, shows strong expression throughout the developing node at the anterior end of the streak, and in the developing notochord and floorplate, which are structures known to derive in part from the node. The node in mammals is thought to be equivalent to Heusen′s node in chick or the Spemann organizer in amphibians, both of which ca induce axial structures when grafted to ectopic sites. In Xenopus, there are two HFH genes whose expression domains include the organizer; pintallavis (Ruiz i Altaba and Jessell, 1992) and Xjkd (Dirksen and Jamrich, 1992), the latter of which is highly similar to HNF3p. The axial gene in zebrafish is also highly similar to HNF3P, and is required for the normal development of structures derived from the fish ′organizer′ (Strahle et al., 1993). Thus, there may be a conserved role in the organizer for this subgroup of HFH genes. Other classes of transcription factors are also expressed in the organizer (Blum et al., 1992; Taira et al., 1992; von Dassow et al., 1993), suggesting that it possesses considerable infor mation content.

HNF3α and HNF3β are also expressed in the developing ventral neural tube, from the midbrain to the posterior end. Expression in the midbrain clearly extends beyond the limits of the floorplate into neural tissue. Expression of ma alian HFH genes in neural tissue has been reported prev10usly: rat BF-1 expression is restricted to the telen cephalon in late stage embryos and remains specific to the brain in the adult (Tao and Lai, 1992). However, the expression of BF-1 prior to day 11.5 has not been reported. The findmg of early neural domains of expression of the t.INF3 en s opens the possibility that there may be sequen tial activation, and possibly interaction, between HFH genes in the development of the nervous system.

During our search for HFH genes potentially involved in endoderm development, we isolated a novel partial cDNA, HFH-E5.1, whose sequence helps define a different subgroup of HFH genes. Expression of HFH-E5. l is first detectable at the late primitive steak stage and is restricted to the ectoderm and mesoderm, with a sharp anterior boundary and a broad posterior region of expression. As development proceeds to the head fold stages, expression in he anterior mesoderm is matched by overlying expression m the neural plate. At the early somite stage, expression is found throughout the neural tube, being strongest in the future mid-brain. By days 13.5 and 16.5, the gene is expressed in specific groups of cells in the central nervous system. HFH-E5. l shares a high degree of sequence simi larity with the Drosophila HFH gene FD4 (Hacker et al., 1992), which is expressed in embryogenesis in neuroblasts along th longitudinal axis, and subsequently in sensory neurons m the head. Considering that both FD4 and HFH E5.l have such highly conserved structures and neural patterns of expression, this subgroup of HFH proteins, like the HNF3/forkhead subgroup, may have a developmental function that has been conserved through evolution.

The developmental pattern of expression of HFH-E5.1 also implicates the gene in regionalization along the neural plate. Anterior mesoderm at the late streak to early head fold stage can induce markers of mid-hindbrain structures, like the Engrailed genes, in naive ectoderm (Ang and Rossant, 1993), but posterior mesoderm cannot. The localized expression of HFH-E5.1 in the anterior mesoderm, followed by i s neur exp ession, suggests that this gene may help restnct the mducmg capacity of mesoderm. The posterior mesoderm expression of HFH-E5.1 persists and eventually becomes restricted to the presomitic mesoderm,.suggesting th t the gene may also play a role in somite formation, along with other genes expressed strongly in this area, such as FGFRl (Yamaguchi et al., 1992) and mouse Notch (Reaume et al., 1992).

W have shown that HFH3α and HNF3β show a strong association with the development of the definitive endoderm, but are also expressed in restricted areas in other germ layer derivatives, and we have identified HFH-E5 1 which shows no expression in the endoderm, but has int r esting domains of expression in the developing mesoderm and neurectoderm. Clearly the HFH gene family in mammals is large and probably not yet fully defined, and the varied and highly specific expression patterns suggests that these genes will play roles in a number of developmental processes, including early lineage specification as well as later cell type differentiation. Mutational analysis of these genes should elucidate their individual roles and possible interactions between them.

We thank Brigid Hogan for the mouse embryo cDNA library, Rob Costa, Eseng Lai, and James Darnell for the rat HNF3a. cDNA, Gerald Crabtree for the HNFla. antibody, Fran9ois Gmllemot for help with in situ hybridization on sections, Kendrapasad Harpal for providing tissue sections, Kirstie Lawson for dis cussion on the endoderm lineage, David Jackson for advice on protein-DNA crosslinking, and John Burch for advice on cloning by PCR. The research was supported by grants from the MRC and NCI of Canada to J. R. and from the NIH (GM36477) to K. S. Z. J. R. is a Terry Fox Cancer Research Scientist of the NCIC and an International Research Scholar of the HHMI.