Isolation of hepatoblasts based on the expression of Dlk/Pref-1

Mouse liver organogenesis is induced by a signal from the cardiac mesoderm at the 7-8 somite stage when the hepatic diverticulum emerges from the foregut endoderm (Gualdi et al., 1996; Houssaint, 1980; Jung et al., 1999; Zaret, 2000). The endodermal cells in the hepatic bud proliferate and invade the septum transversum mesenchyme. While the hepatic parenchyma at an early stage of liver organogenesis consists mostly of hepatoblasts, postnatal and adult liver parenchyma consists of mature hepatocytes and biliary epithelial cells (BECs)that form intrahepatic bile ducts. Histochemical analysis showed that ductal plates are formed around portal veins during mid to late gestation(Shiojiri, 1984; Shiojiri, 1997; Shiojiri et al., 2001). A number of metabolic enzymes expressed in the adult liver, such as tyrosine aminotransferase (TAT), glucose-6-phosphatase (G6Pase), and carbamoyl phosphate synthetase (CPS), become expressed at the late gestation or perinatal stage but not at the mid-gestation stage(Greengard, 1970; Haber et al., 1995). Based on these results, it is believed that hepatoblasts are common progenitors of mature hepatocytes and BECs and that the lineage is determined around the mid-gestation stage. However, the nature of hepatoblasts and the mechanism underlying the lineage determination are still largely unknown.

Hepatic progenitors are also known to be present in the adult liver(Fausto, 1994; Sell, 1994). Oval cells with the ability to differentiate into both hepatocytes and BECs appear in the periportal region following hepatic injury when hepatocyte proliferation is inhibited by a carcinogen, 2-acetylaminofluorene(Sell et al., 1981), or D-galactosamine (Lemire et al.,1991). Rat oval cells have been shown histochemically to express immature hematopoietic cell markers, c-Kit, CD34 and Thy1(Fujio et al., 1994; Omori et al., 1997; Petersen et al., 1998). In addition, small hepatocytes in adult rat liver are suggested to be hepatic progenitor cells (Mitaka et al.,1999; Tateno et al.,2000). However, it still remains unclear what roles oval cells and small hepatocytes play in normal liver development, homeostasis and regeneration. Besides adult liver, hepatic progenitor cells were shown to exist in pancreas (Rao et al.,1989; Reddy et al.,1991; Zulewski et al.,2001; Tosh et al.,2002;). Surprisingly, recent results showed that bone marrow cells could differentiate into hepatocytes(Lagasse et al., 2000; Oh et al., 2000; Petersen et al., 1999; Wang et al., 2002). Furthermore, hepatocytes were induced from ES cells as shown by expression ofα-fetoprotein and albumin, and they were engrafted into recipient liver as hepatocytes (Chinzei et al.,2002). However, the mechanisms of the transdifferentiation from bone marrow cells and the induction of hepatocytes from ES cells have not been investigated. In order to understand the molecular mechanism underlying these phenomena as well as the in vivo liver organogenesis, it is necessary to isolate hepatoblasts and to investigate their differentiation and proliferation.

The methodology using mAbs against cell surface antigens and a cell sorter has been extensively used to isolate HSCs and can be applicable for the isolation of the hepatic progenitor cells. In addition to the report that oval cells were isolated as Thy1⁺ cells from regenerating adult liver(Petersen et al., 1998),recently, attempts have been made to purify progenitors from fetal liver based on the expression of cell surface antigens by FACS. Suzuki et al. showed that the CD45^-TER119^-c-Kit^-CD29⁺CD49f⁺and CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/lofraction of E13.5 mouse liver contained hepatic progenitor cells(Suzuki et al., 2000; Suzuki et al., 2002). Kubota et al. showed that the RT1A1^-OX18^loICAM-1⁺fraction of E13 rat fetal liver contained hepatoblasts(Kubota and Reid, 2000). While these studies demonstrate the power of cell sorters to enrich hepatoblasts,there are not enough surface antigens to identify hepatoblasts. In this study,we used the signal sequence trap method to identify proteins with a signal sequence in E14.5 mouse fetal liver cells and found that Dlk was abundantly expressed in fetal liver.

Dlk is a type I membrane protein that has six EGF-like repeats in its extracellular domain and a short cytoplasmic domain(Laborda et al., 1993; Smas and Sul, 1993). The extracellular domain shows homology to Delta, one of the Drosophila melanogaster Notch ligands, but lacks the DSL domain that is important for binding to Notch. Dlk was found to be highly expressed in a small lung carcinoma cell line (Laborda et al.,1993) and was also identified as preadipocyte factor-1 (Pref-1) in 3T3-L1 preadipocytes (Smas and Sul,1993). Because Dlk orthologues were identified in human, rat and bovine as well as in mouse independently, many names were given to the same molecule; pG2 (Helman et al.,1990), fetal antigen-1 (FA-1)(Jensen et al., 1994), Pref-1(Fahrenkrug et al., 1999; Smas and Sul, 1993), stromal cell derived protein-1 (SCP-1) (GenBank/D16847), zona glomerulosa-specific factor (ZOG) (Halder et al.,1998), and Dlk (Laborda et al., 1993). Here we show that Dlk is strongly expressed in the fetal liver between E10.5 and E16.5 and that Dlk can be used as a marker to enrich hepatoblasts.

C57BL/6 mice (Nihon SLC, Japan) were used for all the experiments. Hamster anti-mouse Dlk mAb was prepared as described previously(Kanata et al., 2000). Rabbit anti-mouse cytokeratin 19 (CK19) polyclonal antibody was raised against the C-terminal peptide, HYNNLPTPKAI. Rabbit serum was used for immunohistochemistry. Dulbecco's modified Eagle's medium (DMEM) (Nissui,Tokyo, Japan), fetal bovine serum (FBS), liver perfusion medium, and liver digestion medium (Gibco BRL, Gaithersburg, MD) were used to prepare fetal hepatocytes and for primary culture. Dulbecco's modified Eagle's F12 medium(Sigma, St Louis, MO) was used for a low density culture.

Fetal hepatic cells of E14.5 liver were prepared and cultured according to the method of Kamiya et al. (Kamiya et al., 1999). Hepatic cells were suspended in DMEM containing 10%FBS (Gibco BRL), 2 mM L-glutamine (Gibco), 1× nonessential amino acid solution (Gibco), 1× insuline/transferrin/selenium (ITS) (Gibco), 50μg/ml of gentamycin, 10^-7 M dexamethasone (Dex) (Sigma), and 10 ng/ml of mouse oncostatin M (OSM) and plated on gelatin-coated dishes.

The signal sequence trap method using a retroviral vector, pMX-SST,developed by Kojima and Kitamura (Kojima and Kitamura, 1999) was used to identify cDNA clones encoding secreted and membrane proteins. cDNA was synthesized from poly(A) RNA of E14.5 CD45^-TER119^- hepatic cells using the Timesaver cDNA synthesis kit (Pharmacia, Peapack, NJ) with random hexamer primers. After the addition of the BstXI adaptor (Invitrogen, Carlsbad, CA), cDNA was inserted into the BstXI site of the pMX-SST vector. The cDNA library used in this study contained 5.0×10⁶ independent clones.

A fragment of Dlk cDNA was amplified by the reverse-transcription polymerase chain reaction (RT-PCR) with two primers: 5′-ATG CTT CCT GCC TGT GC-3′ and 5′-GCA CGG GCC ACT GGC-3′. The PCR fragment was subcloned into the pCRII vector (Invitrogen). Sense and antisense single-stranded RNA probes were prepared by in vitro transcription using the dioxigenin (DIG) RNA labeling kit (Roche, Basel, Switzerland). E10.5 embryo was fixed in 4% PFA and bleached in 6% hydrogen peroxide. After treatment with proteinase K, the embryo was hybridized with 0.5 μg/ml sense or antisense probes at 65°C overnight. After washing and blocking procedures, the embryos were incubated with alkaline phosphatase (AP)-conjugated anti-DIG antibodies (Roche) at 4°C for overnight. The signal was developed in 4-nitroblue tetrazolium chloride/5-bromo-4-chloro-3-indolyl phosphate(NBT/BCIP) solution (Roche).

Total RNA was extracted from tissues or cultured cells with Trizol reagent(Gibco). After electrophoresis, RNA transferred to a nylon membrane was hybridized with DIG-labeled antisense probe at 48°C overnight and then incubated with AP-conjugated anti-DIG antibodies (Roche) at room temperature. The signal was developed with CDPstar (Roche).

Fetal and adult livers were embedded in OCT compound (Sakura Finechemical,Tokyo, Japan). Frozen sections were prepared using a Microtome cryostat HM 500(Microm, Walldorf, Germany) and mounted on glass slides coated with MAS(Matsunami glass, Japan). They were then fixed in 4% PFA and incubated with normal goat serum. Primary and secondary antibodies were diluted to 1 μg/ml and 5 μg/ml, respectively, in 3% normal goat serum. The samples were added with anti-Dlk mAb followed by biotinylated goat anti-hamster IgG (Vector,Burlingame, CA). The samples were incubated in each antibody solution at 4°C in a moist chamber. Expression of Dlk was visualized with the Vectastain ABC kit (Vector) and 3,3-diaminobenzidine tetrahydrochloride (DAB)(Roche). To detect CK19, samples were incubated with 1000-fold dilution of anti-CK19 serum followed by AP-conjugated goat anti-rabbit IgG (Vector). The signal was developed with NBT/BCIP (Roche).

For double immunofluorescence staining, fetal hepatic cells mounted on glass slides were fixed in 4% PFA and incubated with anti-Dlk mAb and rabbit anti-albumin polyclonal antibody (Nordic, Sweden). Dlk was detected with biotinylated goat anti-hamster IgG (Vector) and FITC-conjugated streptavidin(PharMingen) and albumin was detected with rhodamine-conjugated goat anti-rabbit IgG (Chemicon, Temecula, CA). The samples were examined under a fluorescence microscope, Nikon Eclipse E800 (Nikon, Tokyo, Japan).

Dlk⁺ cells isolated by AutoMACS were mounted on glass slides and fixed in 4% PFA. They were incubated with 1 μg/ml anti-albumin or 20μg/ml rabbit anti-human α-fetoprotein (AFP) antibodies (ICN Biomedicals, Costa Mesa, CA). Both signals were detected by rhodamin-conjugated anti-rabbit IgG (Chemicon).

E14.5 fetal hepatic cells were incubated with anti-Dlk mAb and rat mAbs against CD45, TER119 and PECAM-1 (PharMingen, San Jose, CA). After washing with PBS, cells were incubated with FITC-conjugated goat anti-hamster IgG(Vector) and PE-conjugated goat anti-rat IgG (Cedarlane, Ontario, Canada). All these antibodies were diluted 100-fold and used for staining. The samples were then washed with PBS and mixed with 1 μg/ml propidium iodide (PI) before flow cytometric analysis with a FACScallibur (Becton Dickinson, San Jose,CA).

Dlk⁺ cells were isolated by an automatic magnetic cell sorter(AutoMACS) (Miltenyi Biotec, Bergisch Gladbach, Germany). E14.5 hepatic cells were incubated with anti-Dlk mAb, biotinylated goat anti-hamster IgG. After a wash with PBS, cells were resuspended in AutoMACS running buffer(1×10⁸ cells/ml of PBS containing 0.5% BSA) and 100 μl/ml of streptavidin-labeled microbeads (Miltenyi Biotec) were added. After a wash with the running buffer, the cells were loaded onto a magnetic column and Dlk⁺ cells were eluted from the column after the depletion of Dlk^- cells. Alternatively, E14.5 CD45-TER119^- cells were incubated with anti-Dlk mAb and FITC-conjugated goat anti-hamster IgG and sorted into Dlk^- and Dlk⁺ fractions by using a FACSvantage (Becton Dickinson).

Total RNA (1 μg) was used to synthesize cDNA using the First-strand cDNA synthesis kit (Amersham Pharmacia Biotech, Piscataway, NJ) and random hexamer primers. The samples were denatured at 94°C for 2 minutes, followed by the thermal cycles; denaturation at 94°C for 30 seconds, annealing at the temperature set for each pair of primers for 30 seconds, extension at 72°C for 2 minutes. The thermal cycle was repeated 20 times for AFP and albumin, 25 times for Dlk and GAPDH, and 30 times for other genes. The primers used for RT-PCR are shown in Table 1. The primers used for Dlk were the same as those used for in situ hybridization.

Quantitative PCR analysis was also performed to measure the mRNA levels for AFP and albumin by using LightCycler (Roche).

A low density culture was performed to examine the growth potential of Dlk⁺ cells. E14.5 Dlk^- and Dlk⁺ cells isolated by AutoMACS were cultured in DMEM F-12 (Sigma) at a density of 1000 and 50 cells/cm², respectively, on 6-well plates coated with type IV collagen (Nitta Gelatin, Osaka, Japan). The medium was supplemented with 10% FBS, 1×ITS, 10 mM nicotineamide (Wako, Tokyo, Japan), 0.1 μM Dex,and 5 mM L-glutamine. Various combinations of 20 ng/ml epidermal growth factor(EGF) (PeproTech, London, UK), hepatocyte growth factor (HGF) (R&D,Minneapolis, MN), and OSM were added 18 hours after the initiation of the culture. After 5 days of culture, cell nuclei were stained with hematoxylin(Muto Pure Chemicals, Japan) and the cells in each colony were counted. Colony numbers in three wells were determined in each set of culture and the experiment was repeated four times. These data were analyzed statistically using JMP program to obtain standard deviations and P-values.

The expression of albumin and CK19 in colonies was analyzed by immunocytochemistry. The cells cultured on chamber slides (NUNC, Roskilde,Denmark) coated with type IV collagen were fixed in methanol at -20°C for 10 minutes. Alternatively, each large colony formed on 6-well plates was placed with a cloning ring and treated with trypsin. The cells removed from the dish were mounted on glass slides and fixed in methanol. After washing with PBS and blocking with 3% donkey serum (Chemicon) for 30 minutes, the cells were incubated with 2 μg/ml goat anti-mouse albumin antibody (Bethyl laboratories, Inc., Mongomery, TX) and rabbit anti-CK19 serum (1000-fold dilution) at 4°C overnight. After they were washed with PBS containing 0.05% Tween 20 (PBST), the samples were incubated with Cy3-conjugated donkey anti-goat IgG antibody (Rockland, Gilbertsville, PA) and FITC-conjugated donkey anti-rabbit IgG antibody (Rockland) for 2 hours at 4°C.

FACSvantage was used for the single cell sorting. Each Dlk⁺ cell sorted from E14.5 hepatic cells was individually plated in one well of a 96-well plate coated with type IV collagen. After 5 days of culture, cells were fixed in methanol and the expression of albumin and CK19 was examined as described above.

Ultrastructures of E14.5 Dlk⁺ cells were examined with a transmission-electron microscope before and after the low density culture. Purified Dlk⁺ cells from fetal livers were fixed in 2.5%phosphate-buffered glutaraldehyde for 30 minutes at room temperature and in 1%phosphate-buffered OsO₄ for 15 minutes at room temperature. Dlk⁺ cells cultured for 5 days on 6-well plates coated with type IV collagen were detached from dishes by trypsin treatment and similarly fixed. Then, both cells were dehydrated and embeded in epoxy resin. Thin sections for electron microscopy were counterstained with uranil acetate and lead citrate,and examined with a JEM-1220 electron microscope (JEOL, Tokyo, Japan).

Dlk⁺ cells were isolated from E14.5 fetal livers of GFP transgenic mouse by using AutoMACS. Acute liver injury was induced in recipient mice by intraperitoneal administration of anti-Fas antibody Jo2 (200μg/kg; PharMinagen) before transplantation. After 24 hours of the injection of anti-Fas antibody, 2×10⁵ Dlk⁺ cells were transplanted intrasplenically into the anesthesized recipient mice as described in the previous work (Ponder et al., 1991). After 8 or 36 weeks, the recipient mice were sacrificed and frozen sections of their livers were prepared. Donor-derived GFP⁺ cells were detected under a fluorescence microscope. The sections including donor-derived GFP⁺ cells were used for immunostaining with anti-albumin antibody.

In order to identify surface markers for fetal hepatic cells other than blood cells, we applied the signal sequence trap using the cDNA library of E14.5 hepatic cells deprived of CD45⁺TER119⁺hematopoietic cells. Among various molecules identified by this approach,serum proteins such as AFP and albumin were most frequently found; 40% of the clones obtained comprised cDNAs for these two serum proteins(Table 2). In addition to these clones, we repeatedly isolated cDNA fragments encoding Dlk also known as Pref-1, suggesting that Dlk is one of the most abundant membrane proteins expressed in E14.5 fetal liver. Additional known membrane and secreted proteins identified were carrier proteins such as vitamin D binding protein and retinol binding protein, cytokines such as M-CSF and IGF-II, cytokine receptors such as interferon and lymphotoxin β receptors, proteases such as HGF-activator and plasminogen, and extracellular matrix proteins such as collagen, vitronectin and nidogen.

Dlk/Pref-1 was previously shown to be expressed in E8.5 fetus and in liver,pituitary, lung, vertebra and tongue at E13.5(Smas and Sul, 1993). In order to examine Dlk expression at the onset of liver organogenesis, whole-mount in situ hybridization was performed using E10.5 embryo. In this experiment, Dlk was detected in the liver bud as well as vertebra(Fig. 1). The Dlk expression in the E10.5 liver bud was also confirmed by RT-PCR(Fig. 2A). To examine the expression of Dlk during later liver development, Northern blot analysis was performed using total RNA extracted from developing livers. Dlk mRNA was strongly expressed in fetal liver between E12.5 and E16.5(Fig. 2B). Its expression was downregulated later in gestation and disappeared in the neonatal and adult livers (Fig. 2B). The expression of Dlk during liver development was also examined by using primary cultures of fetal hepatocytes. We previously showed that E14.5 fetal hepatocytes are induced to differentiate morphologically and to express various metabolic enzymes such as TAT, G6Pase and CPS by OSM in the presence of Dex in vitro (Kamiya et al.,1999). Northern blot analysis showed that Dlk expression gradually disappeared without OSM, and was more rapidly disappeared in the presence of OSM, a condition that induced expression of CPS and TAT(Fig 2C). Expression of TAT and CPS was induced after birth (Greengard,1970; Haber et al.,1995), while expression of Dlk was downregulated after E16.5(Fig. 2B). It is thus likely that E14.5 hepatocytes in the primary culture spontaneously differentiated to a later fetal stage and OSM was required for further differentiation to the postnatal stage in vitro.

Consistent with the results of in situ hybridization, immunohistochemical examination using anti-Dlk mAb on frozen sections indicated that Dlk was expressed in endodermal cells in the liver bud at the E10.5 stage, while it was not expressed in the foregut, from which the hepatic diverticulum was generated (Fig. 3A-D). Dlk was also detected in the E14.5 liver (Fig. 3E), but not in the adult liver(Fig. 3F). These results collectively indicate that Dlk is expressed in immature liver cells from the onset of the liver organogenesis.

The fetal liver parenchyma consists of immature hepatocytes, BECs, and their common progenitors, hepatoblasts. In order to know which endodermal cells express Dlk, we performed double immunostaining analysis of Dlk with either albumin or CK19. First, E14.5 fetal hepatic cells mounted on glass slides were stained with anti-Dlk mAb and anti-albumin antibody. Since the expression of albumin is induced at the beginning of liver organogenesis(Jung et al., 1999), albumin was expected to be detected not only in hepatocytes but also in hepatoblasts. We found that large hepatic cells expressed both Dlk and albumin in fetal liver between E12.5 and E18.5 (Fig. 3G-J and data not shown). We then stained fetal liver sections with anti-Dlk mAb, and anti-CK19 antibody that stains BECs. Although the formation of ductal plates has already started at the E14.5 or E15.5 stage(Clotman et al., 2002; Coffinier et al., 2002; Shiojiri, 1997; Shiojiri et al., 2001), the cells comprising the ductal structure were not uniformly stained with anti-CK19 antibody (data not shown). Therefore, in order to investigate whether Dlk is expressed in BECs, we used E17.5 liver sections in which CK19⁺ ductal plate structures became apparent. Double immunostaining showed that CK19⁺ BECs were Dlk^- at E17.5(Fig. 3K-M). These results suggest that Dlk is expressed in both immature hepatocytes and hepatoblasts,and the expression is downregulated when hepatoblasts differentiate into CK19⁺ BECs.

As Dlk is a membrane protein, it might be useful as a surface antigen to separate immature hepatocytes from other types of fetal hepatic cells such as hematopoietic and endothelial cells. Flow cytometric analysis using anti-Dlk mAb showed that about 10% of total hepatic cells were Dlk⁺ and that they were clearly separable from Dlk^- cells, which mainly consisted of CD45⁺, TER119⁺ and PECAM-1⁺ cells(panspecific hematopoietic, erythroid and endothelial cells, respectively)(Fig 4A). We then separated Dlk^- and Dlk⁺ cells from E14.5 fetal liver by AutoMACS. Flow cytometric analysis showed that both Dlk^- and Dlk⁺fractions were over 95% pure (data not shown), and immunostaining revealed that about 95% of cells in the Dlk⁺ fraction were AFP⁺and albumin⁺ (Fig. 4B). Furthermore, we employed quantitative RT-PCR analysis to examine the expression of AFP and albumin in CD45^-TER119^-Dlk^- and CD45^-TER119^-Dlk⁺ cells, which were separated by FACSvantage. Dlk⁺ cells expressed AFP and albumin about 60 and 40 times, respectively, higher than Dlk^- cells(Fig. 4C). These results indicate that immature hepatocytes, possibly including hepatoblasts, are enriched in the Dlk⁺ fraction.

To further characterize the E14.5 Dlk⁺ cells isolated by AutoMACS, expression of several genes including hepatocyte and BEC markers was examined by RT-PCR (Table 3). Dlk⁺ cells expressed strongly AFP and albumin, and significantly CK8, CK18, Cx26, Cx43, GS and GGT. Although GGT is known to be expressed in BECs of adult liver, it was previously reported that rat GGT was expressed also in fetal hepatocytes (Shiojiri et al., 1991; Holic et al.,2000). In contrast to early hepatocyte markers, expression of CPS,TAT, G6Pase, TO and Cx32, which are known to be expressed in mature hepatocytes, was undetected or barely detectable. Consistent with the result that Dlk was downregulated in BECs (Fig. 3K-M), Dlk⁺ cells did not express CK19. In addition to early hepatocyte marker genes, Dlk⁺ cells expressed liver enriched transcription factors, HNF1β, HNF3β, HNF4 and HNF6, whereas none of them was expressed in Dlk^- cells. On the contrary, expression of c-Kit was detected in Dlk^- but not Dlk⁺ cells.

To examine the proliferative potential of each sorted cell, we cultured Dlk^- and Dlk⁺ cells at a low density to evaluate clonal growth. Dlk^- and Dlk⁺ cells were isolated from E14.5 fetal liver using AutoMACS and were plated at a density of 1000 and 50 cells/cm², respectively, on 6-well plates coated with type IV collagen. After 5 days of culture, we counted the number of large colonies containing over 100 cells, which were considered to be formed from highly proliferative cells. First, we examined various combinations of cytokines,EGF, HGF, and OSM, for growth in vitro(Table 4). As Dlk⁺cells efficiently proliferated with HGF or HGF plus EGF, HGF was the most effective growth factor for Dlk⁺ cells among three cytokines,consistent with the fact that Dlk⁺ cells expressed c-Met, the receptor for HGF (data not shown). Since the number of large colonies formed in the presence of HGF plus EGF slightly exceeded that with HGF in this low density culture condition, we used the combination of HGF plus EGF in the following experiments.

We then evaluated the growth potential of Dlk⁺ cells in a low density culture with EGF and HGF. The growth of a single Dlk⁺ cell was followed for 5 days. Dlk⁺ cells proliferated exponentially between culture day 2 and 4, and some colonies still actively proliferated beyond day 4 and reached 100 cells at the end of the culture. After 5 days of culture, 24±3% of input Dlk⁺ cells proliferated and formed colonies with various sizes (Fig. 5). Half of the colony-forming Dlk⁺ cells formed small colonies containing less than 40 cells, while 10% of the colony-forming Dlk⁺ cells were highly proliferative. Since the number of colonies formed from Dlk^- cells was less than 5% of that from Dlk⁺ cells, colony-forming cells were enriched in the Dlk⁺ fraction.

Dlk is expressed in the E10.5 liver bud in which hepatic parenchyma is believed to consist of hepatoblasts. In order to know whether Dlk⁺cells are bipotential at the mid-gestation, we tested expression of albumin and CK19 in colonies formed from E14.5 Dlk⁺ cells. The results showed that over 60% of colonies derived from E14.5 Dlk⁺ cells contained both albumin⁺ and CK19⁺ cells(Fig. 6A-C). As colonies with various sizes contained both types of cells, there seemed no direct correlation between differentiation potential and proliferation potential. In this low density culture condition, the colonies contained cells which expressed either albumin or CK19 and also those that expressed both. To confirm the existence of albumin⁺CK19⁺ cells, a large colony was picked from the culture dish and the cells mounted on glass slides were subjected to immunostaining. We did find cells that expressed both albumin and CK19 (Fig. 6D). These results strongly suggested that single Dlk⁺ cells were able to differentiate into albumin⁺, CK19⁺, and albumin⁺CK19⁺ cells. However, there still remained the possibility that some colonies were derived from multiple cells even in the low density culture. In order to exclude such possibility we performed single cell sorting. By using FACSvantage each single Dlk⁺ cell was sorted and plated in one well of a 96-well plate. After 5 days of culture, colonies were formed in about 20% of wells inoculated with a single Dlk⁺cell. About 10% of those colonies contained over 100 cells that consisted of both albumin⁺ and CK19⁺ cells(Fig. 6E), indicating that a single Dlk⁺ cell was able to differentiate into both hepatocyte and BEC lineages. Taken together, it is concluded that the majority of colony-forming Dlk⁺ cells at E14.5 are hepatoblasts that display bilineage gene expression and some of Dlk⁺ hepatoblasts are highly proliferative.

We also analyzed ultrastructure of Dlk⁺ cells before and after 5 days of culture by electron microscope. The majority of E14.5 Dlk⁺cells had a round-shaped nucleus, a small nucleolus, and elongated mitochondria (Fig. 7A). The former two characteristics suggest that E14.5 Dlk⁺ cells are rather primitive cells that are not actively synthesizing mRNAs. After 5 days of culture, most of the cultured cells exhibited a large number of microvilli on their cell surfaces and cleaved nucleus, larger nucleolus and many round-shaped mitochondria in the cytoplasm(Fig. 7B), suggesting that Dlk⁺ cells differentiated during the culture. However, because this culture condition was for proliferation rather than maturation, as described above, the Golgi apparatus, lysosomes and accumulation of glycogen were not apparent in the cytoplasm.

In order to investigate differentiation potential of E14.5 Dlk⁺cells in vivo, we transplanted Dlk⁺ cells into the spleen of recipient mice that were pretreated with anti-Fas antibody to induce acute liver damage. GFP⁺ cells were found in recipient liver at 8 weeks after transplantation (data not shown) and GFP⁺ cells in the recipient liver were more frequently found after 36 weeks(Fig. 8A,B). These GFP⁺ cells expressed albumin(Fig. 8C), indicating that Dlk⁺ cells can differentiate to hepatocytes in vivo under these experimental conditions. By contrast, no GFP⁺ cells were found to express CK19, suggesting that this protocol did not produce a condition to replace recipient BECs.

Cell surface antigens are useful not only for identification but also for isolation of a particular type of cells and numerous such antigens have been identified in hematopoietic cells. On the contrary, very few such antigens have been known in hepatic cells and hepatoblasts have not been well defined by the expression of cell surface antigens, making identification, isolation and characterization of hepatoblasts difficult. We applied the signal sequence trap to identify cell surface antigens of fetal hepatic cells and found that Dlk is highly expressed in fetal immature hepatocytes. Dlk was previously shown to be expressed in fetal tissues including liver(Smas and Sul, 1993); however,its expression profile during development had been unknown. We found that Dlk was expressed in immature hepatocytes as early as E10.5 and its strong expression continued until E16.5. In contrast, Dlk expression was undetectable in neonatal and adult liver.

We demonstrated that Dlk⁺ cells isolated from E14.5 liver were able to proliferate to form colonies in 5 days of culture in the presence of HGF and EGF. More than 60% of such colonies contained both albumin⁺and CK19⁺ cells. Thus, E14.5 Dlk⁺ cells are mostly, if not entirely, hepatoblasts. Furthermore, since 10% of colony-forming Dlk⁺ cells formed large colonies that contained over 100 cells,highly proliferative hepatoblasts are enriched in the Dlk⁺population. Since the vast majority of Dlk⁺ cells expressed albumin, and almost all the adherent cells were albumin⁺ after 1 day of culture, colonies were mostly formed from Dlk⁺albumin⁺ cells. We also examined the ability of Dlk⁺ cells at different stages of gestation to proliferate and to differentiate. Dlk⁺ cells at E12.5 and E16.5 stages contained highly proliferative and bipotential hepatoblasts, while E18.5 Dlk⁺cells lost highly proliferative potential. As the expression of Dlk in each cell was also downregulated at E18.5 (data not shown), the Dlk level may be correlated with the growth potential of Dlk⁺ cells. We also investigated the differentiation of Dlk⁺ cells in vivo by transplantation of Dlk⁺ cells into the recipient mice treated with anti-Fas antibody, Jo2, which induces apoptosis in hepatocytes. In this model system, we detected donor-derived hepatocytes but not BECs. The inability to demonstrate differentiation of hepatic progenitors to BECs seems to be a common problem for in vivo transplantation assays because transplanted progenitor cells differentiated into only hepatocytes in many cases using mice as recipient animals (Suzuki et al.,2000; Lagasse et al.,2000; Chinzei et al.,2002; Forbes et al.,2002; Malhi et al.,2002). One exception is that transplanted CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells formed bile-duct like structures; however, such structures were found only in the spleen but not in liver(Suzuki et al., 2002). Thus,it is possible that the protocol we used did not create a condition that allowed the transplanted cell to replace the recipient BECs. However, the possibility remains that Dlk⁺ cells at E14.5 have lost the ability to become BECs in vivo.

Recent studies indicated that hepatoblasts could be enriched by using FACS:mouse hepatoblasts were enriched in the CD45^-TER119^-c-Kit^-CD29⁺CD49f⁺fraction (Suzuki et al., 2000)and were further enriched in the CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/lofraction (Suzuki et al.,2002), and rat hepatoblasts were present in RT1A1^-OX18^loICAM-1⁺ cells(Kubota and Reid, 2000). In the present study, we demonstrated that hepatoblasts are enriched in the Dlk⁺ population of mouse fetal hepatic cells. Flow cytometric analysis indicated that Dlk⁺ cells were CD45^-TER119^- and cKit^-. Although both Dlk⁺ and CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells contained highly proliferative and bipotential cells, there is a substantial difference in abundance between Dlk⁺ cells and CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells. Dlk⁺ cells constituted about 10% of E14.5 total fetal hepatic cells, while CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/loconstituted only 0.3% of them. The potential for differentiation also highlights a significant difference between Dlk⁺ cells and CD45^-TER119^-c-Kit^-CD29⁺CD49f⁺or CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells. For example, a longer incubation time was required for CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells to differentiate into hepatocytes and BECs: 70% of large colonies derived from Dlk⁺ cells contained albumin⁺ and CK19⁺ cells only after 5 days of culture(Fig. 6), while 80% of large colonies derived from CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells contained both albumin⁺ and CK19⁺ cells only after 21 days of culture. Thus, it is likely that CD45^-TER119^-c-Kit^-c-Met⁺CD49f^+/locells represent more immature cells than Dlk⁺ cells. Rat RT1A1^-OX18^loICAM-1⁺ hepatoblasts formed colonies containing albumin⁺ and CK19⁺ cells after 5 days of culture. Although they were derived from different animals, mouse Dlk⁺ cells and rat RT1A1^-OX18^loICAM-1⁺ cells appear to exhibit similar characteristics as a progenitor of hepatocytes and BECs.

The result that Dlk is expressed specifically in fetal liver suggests that Dlk might be implicated in proliferation and/or differentiation of hepatocytes. Consistent with this idea, Dlk, also known as Pref-1, was previously shown to be involved in differentiation of pre-adipotcytes as overexpression of Dlk resulted in inhibition of adipogenesis(Smas et al., 1997; Smas and Sul, 1993). It was also reported that Dlk expression increased when proliferation of pancreaticβ cells reached a maximum (Carlsson et al., 1997). To test whether Dlk might have a role for hepatic differentiation in a manner similar to adipocyte differentiation, we expressed Dlk in the fetal hepatocyte primary culture by using a retrovirus vector. However, expression of hepatic differentiation marker genes was not altered(data not shown). Another possibility is that Dlk is involved in hematopoiesis. Moore et al. reported that Dlk was expressed in fetal liver stroma cells that were able to support hematopoiesis(Moore et al., 1997). In addition, there are reports that Dlk modulated proliferation of thymocytes(Kanata et al., 2000), fetal liver hematopoietic cells (Ohno et al.,2001) and pre-B cells (Bauer et al., 1998). Interestingly, fetal liver hematopoiesis is most active in mid-gestation when Dlk is strongly expressed, which suggests that Dlk is involved in hematopoiesis. However, recent studies on Dlk deficient mouse show that the mutant mice are viable without apparent defects in liver formation and hematopoiesis, while they show growth abnormality and altered lipid metabolism (Moon et al.,2002). Thus, roles of Dlk in hematopoiesis as well as liver development still remain elusive and await further investigation.

In this study, we successfully isolated Dlk⁺ cells and demonstrated that hepatoblasts are abundant in E14.5 Dlk⁺ cells. Using the present method, 95% pure Dlk⁺ cells can be easily isolated from total fetal hepatic cells using AutoMACS. In addition, since Dlk expression in fetal liver is similar between mouse and human, the present method may be applicable for the isolation of human hepatoblasts.

We are grateful to Minoru Tanaka, Akihide Kamiya, Koji Nakamura, Hiroyuki Yanai and Hiroko Anzai for technical assistance and helpful discussion. We thank Takashi Sekiguchi for cell sorting by using FACSvantage and Minoru Kakeda for preparation of the monoclonal antibody against Dlk. This work was supported in part by Grants-in-Aid for Scientific Research and Special Coordination Fund for Promoting Science and Technology from the Ministry of Education, Culture, Sports, Science and Technology, a research grant from the Ministry of Health, Labour and Welfare, the Japan Government and CREST of Japan Science and Technology.