A novel nk-2-related transcription factor associated with human fetal liver and hepatocellular carcinoma.

A novel cDNA was partially isolated from a HepG2 cell expression library by screening with the promoter-linked coupling element (PCE), a site from the alpha-fetoprotein (AFP) gene promoter. The remainder of the cDNA was cloned from fetal liver RNA using random amplification of cDNA ends. The cDNA encodes a 239-amino acid peptide with domains closely related to the Drosophila factor nk-2. The new factor is the eighth vertebrate factor related to nk-2, hence nkx-2.8. Northern blot and reverse transcriptase polymerase chain reaction analysis demonstrated mRNA in HepG2, two other AFP-expressing human cell lines, and human fetal liver. Transcripts were not detected in adult liver. Cell-free translation produced DNA binding activity that gel shifted a PCE oligonucleotide. Cotransfection of nkx-2.8 expression and PCE reporter plasmids into HeLa cells demonstrated transcriptional activation; NH2-terminal deletion eliminated this activity. Cotransfection into AFP-producing hepatocytic cells repressed AFP reporter expression, suggesting that endogenous activity was already present in these cells. In contrast, cotransfection into an AFP-negative hepatocytic line produced moderate activation of the AFP gene. The cardiac developmental factor nkx-2.5 could substitute for nkx-2.8 in all transfection assays, whereas another related factor, thyroid transcription factor 1, showed a more limited range of substitution. Although the studies have yet to establish definitively that nkx-2.8 is the AFP gene regulator PCF, the two factors share a common DNA binding site, gel shift behavior, migration on SDS-acrylamide gels, and cellular distribution. Moreover, the nk-2-related genes are developmental regulators, and nkx-2.8 is the first such factor associated with liver development.

Developmental processes are frequently associated with expression of specific homeobox transcription factors. Although all share a common form of DNA binding domain, the hundreds of known homeobox factors belong to many subfamilies with a wide variety of secondary domains (1). The Drosophila factor nk-2 is the prototype of a distinct family of homeobox factors. nk-2-related homeodomain factors have been characterized in Drosophila (nk-2, nk-3/bagpipe, and nk-4/tinman/msh2), planarians (Dth1 and Dth2), leeches (lox10), Caenorhabditis (Ceh22), and vertebrates (nkx-2.1 to 2.7) (2). The nk-2-related factors contain a characteristic secondary domain, the "conserved peptide," which has an unknown function and is unrelated to known protein domains.
The three Drosophila homologues have important developmental functions. nk-2 is involved in early neurogenesis, nk-3 is required for visceral muscle formation, and nk-4 is essential for the formation of precardiac mesoderm.
The vertebrate nk-2 factors also regulate development. nkx-2.1, or thyroid transcription factor 1 (TTF-1), 1 is a regulator of thyroid-specific gene expression, thyroid development, thyroid cell differentiation, and thyroid cell proliferation (3,4) and is first expressed several days before thyroid differentiation (5). TTF-1/nkx-2.1 also regulates pulmonary development and gene expression (6,7). nkx-2.5/Csx and nkx-2.3 are both expressed in early cardiac primordia and thus replicate the function of tinman (8). nkx-2.6 is also expressed in heart. In Drosophila, ablation of tinman blocks cardiac development, whereas knockout of mouse nkx-2.5 arrests heart development at the looping stage (8). This is a less severe phenotype than the Drosophila knockout and probably reflects functional redundance with nkx-2.3 and nkx-2.6 (9). nkx-2.7, another tinman homologue related to cardiac development, has recently been described in zebrafish (10).
nkx-2.2 is expressed in developing mouse brain, with an onset of expression at about 9 days gestation. The transcripts are found in localized regions that correspond to anatomic boundaries in the developing forebrain. The localization suggests that nkx-2.2 specifies differentiation of the developing diencephalon into its anatomically and functionally defined subregions (11).
In mice the cardiac mesenchyme forms from the ventral wall of the foregut at 8.5 days gestation, and the hepatic primordium buds from an adjacent area of the foregut at 8.5-9 days gestation (12). There is a strong association between this early phase of cardiac differentiation and nk-2-related factors, but no homeodomain factors of any sort have been associated with initial differentiation of the liver, later differentiation of bile ducts, or regulation of hepatic stem cells. Two homeobox factors, HNF-1 (13,14) and HNF-6 (15), are associated with the liver, but they regulate only the postdevelopmental phenotype. Another homeobox gene, hlx, has a limited effect on liver development. However, hlx is expressed in hematopoietic cells of fetal liver, not hepatocytes. Mice with targeted disruption of this gene undergo normal early hepatic differentiation, but the failure of hematopoietic cells to colonize the fetal liver results in a small but normally formed liver (16).
Our research has focused on the ␣-fetoprotein (AFP) gene, characteristically expressed from the earliest stages of liver development, but silent after birth. A site near the AFP promoter, the promoter-linked coupling element (PCE), appears to interact with the main developmental regulators of AFP expression. In HepG2 cells, the PCE binding activity has been characterized as a distinct transcription factor, PCF (17,18). To characterize PCF further, we used a PCE-containing oligonucleotide for expression cloning, leading to the isolation of nkx-2.8, which shares numerous properties with PCF. This is the first demonstration of an association between an nk-2 homeobox factor and liver gene expression.

EXPERIMENTAL PROCEDURES
Expression Cloning-A gt11 library of oligo(dT)-primed cDNA from HepG2 cells (CLONTECH) was screened for plaques that bound the PCE in the AFP gene promoter (Ϫ166 to Ϫ155, TGTTCAAGGACA; Ref. 18). This sequence and its complement were included in a doublestranded oligonucleotide with added TCGA sticky ends. Probe preparation and screening were modified from previously described methods (19,20). The oligonucleotide was labeled with polynucleotide kinase and [␥-32 P]ATP and concatenated with DNA ligase to an average size of 5-10 repeats with a specific activity of 5 ϫ 10 8 d/min/g. Bacteriophage were plated using Escherichia coli strain Y1090. Plaques were lifted onto Millipore HATF membranes, air dried, incubated in blocking buffer (5% nonfat dry milk, 25 mM HEPES, pH 7.9, 5 mM MgCl 2 , 40 mM KCl, 1 mM dithiothreitol) for 30 min and then probe mix (0.25% nonfat dry milk, 25 mM HEPES, pH 7.9, 5 mM MgCl 2 , 40 mM KCl, 0.1 mM dithiothreitol, 10 g/ml sonicated denatured DNA, and 10 ng/ml labeled concatenated probe) for 12 h. DNA from Micrococcus lysodeiktikus was chosen for this mixture for its high GϩC content (70%) and provided a very effective background competition that greatly enhanced the sensitivity of the screen. A screen of 1,000,000 phage plaques produced a single clone with appropriate binding specificity. A second screen of 2,000,000 plaques from another HepG2 library in ZAP (Stratagene, La Jolla, CA) provided no additional positives. Upon sequencing, the positive clone was found to be truncated at the 5Ј-end by an in-frame deletion that also removed part of the ␤-galactosidase gene. PCR assays (see below) were used to screen pooled DNA isolated from both libraries but detected only the single partially deleted clone. The cloned insert was subcloned in pBluescript KSϩ (Stratagene). Because clonal instability was observed, this and all further subclones were propagated in a DNA repair and recombination-deficient E. coli host (SURE, Stratagene).
PCR-based Cloning-From the sequence of the new clone, direct and reverse transcriptase PCR assays (below) were used to screen a variety of RNA sources. A positive detection was obtained from a commercial cDNA prepared from human fetal liver (Marathon-Ready cDNA, CLONTECH) which was synthesized using random amplification of cDNA ends cloning adaptor annealed to the 5Ј-end. Amplification was carried out with primers AP1 and R5A and Tth DNA polymerase (denaturation at 95°C for 2 min; 40 cycles of 95°C for 40 s, 68°C for 4 min). A weak PCR product was then reamplified with two combinations of nested primers, using the same temperature-cycling protocol. Primers AP2 and R13A gave products of 160 and 340 bp; primers AP2 and R7A gave products of 410 and 590 bp. The PCR products were subcloned in pCR-Script SKϩ (Stratagene), and multiple clones were sequenced. All represented the same region, with predicted size differences based on primer position. The shorter PCR products were terminated at a high GϩC region within the longer one. Thus all clones represented a single mRNA. Two allelic variants were recognized. For PCR studies, except where otherwise specified, enzymes were used with buffers and conditions supplied by the manufacturers. The primers are listed in Table I. DNA Sequencing, cDNA, and Genomic Analysis-Because the high GϩC content caused many compression artifacts, all DNA sequencing was verified by sequencing with a thermostable DNA polymerase system (SequiTherm EXCEL, Epicentre Technologies, Madison, WI). The original clone and four plasmid clones from the PCR amplification of the 5Ј-end were sequenced fully on both strands. A 241-bp region of overlap between the bacteriophage and PCR-generated clones was identical in all clones. All sequences were subjected to BLAST analysis (National Center for Biotechnology, http://www.ncbi.nih.gov/BLAST/). Related sequences were downloaded and compared by FASTA alignment routines (21).
For additional mapping and to rule out splicing variants the fetal liver cDNA and the gene were analyzed by direct PCR, using Tth DNA polymerase and a "touchdown" amplification protocol (denaturation at 95°C for 2 min; 7 cycles of 95°C for 40 s, 72°C for 4 min; 40 cycles of 95°C for 40 s, 68°C for 4 min).
mRNA Analysis-RNA isolation from cell lines and tissues, agaroseurea gel electrophoresis, and Northern blot hybridization were carried out as described previously (22,23). Poly(A) RNA was purified using a magnetic bead system (Poly(A)Tract, Promega, Madison, WI). For each gel lane, 100 g of total RNA was processed, 50 g of E. coli tRNA was added, and the entire sample was ethanol precipitated and redissolved for electrophoresis. For a hybridization probe, the F9:R5 PCR product was cloned in pCR-Script SKϩ. A riboprobe was generated with T7 RNA polymerase following plasmid linearization with PvuII. For PCRs, 1 g total RNA was incubated in a single tube with primers R5 and PBDG-B and Superscript II reverse transcriptase (Life Technologies) at 45°C for 1 h. Amplification of nkx-2.8 transcripts was then carried out on an aliquot using primers F7 and R10A and KlenTaq DNA polymerase (CLONTECH). The sample was denatured at 95°C for 2 min followed by 7 cycles of 95°C for 40 s, 72°C for 4 min; and 40 cycles of 95°C for 40 s, 68°C for 4 min. A control PCR for a housekeeping mRNA, porphobilinogen deaminase (PBDG; Ref. 24), was carried out with primers PBDG-A and PBDG-B and Taq DNA polymerase. The sample was denatured at 94°C for 2 min followed by 40 cycles of 94°C for 30 s, 55°C for 20 s, and 72°C for 30 s. Products were visualized on an ethidium bromide-stained agarose gel or transferred to nitrocellulose and detected by hybridization to a specific probe, the product of an F10:R7 PCR.
Cells, Tissues, and Transfection-Human cell lines HepG2, HuH7, and HuH1-clone 2 (Clone 2) are derived from hepatocellular carcinomas (25)(26)(27); and RPMI 7451 is from a cholangiocarcinoma (28). H4C3 is a rat hepatocellular carcinoma cell line that expresses high levels of albumin and low levels of AFP (23). Human lung carcinoma cell line H441 was used as a source of TTF-1 for gel shifts (29). All were propagated in Williams E medium containing 1-5% fetal calf serum, penicillin-streptomycin, and glutamine. Specimens of human fetal and adult liver were provided by Dr. Stephen Strom (University of Pittsburgh).
CaPO 4 transfection of cell lines HepG2, Clone 2, and HeLa was carried out as described previously (30). H4C3 cells were transfected with LipofectAMINE (Life Technologies, Inc.); 10-cm culture plates were inoculated with 1 ϫ 10 6 cells and transfected after 2 days, using 10 g of DNA and 60 l of LipofectAMINE/plate, according to the supplier's protocols. Individual transfection experiments always consisted of a  series of identical plates transfected simultaneously under identical conditions. Each determination was the average of two transfections. All described results were reproducible in at least two separate experiments.
Reporter and Expression Plasmids-A PCF reporter plasmid, pPCE4-HIV-CAT, was constructed from plasmid pAPF1-HIV-CAT (31) by cutting with SalI and PstI to remove the HNF4 binding sites and substituting an 80-bp oligonucleotide array consisting of four copies of the PCE from the AFP promoter. The oligonucleotide contained PCF binding sites in the same orientation spaced at 20-bp intervals. A control plasmid containing no binding sites, pHIV-CAT, was produced by blunt ligating an SalI-PstI-digested plasmid. AFP and albumin gene expression plasmids have been described previously (18,30,32).
nkx-2.8 expression plasmids were constructed in pCI (Promega), which contains a CMV early enhancer/promoter and SV40 splicing and polyadenylation signals. The vector also contains a multicloning site with an adjacent T7 RNA polymerase promoter. Full-length pCMV-Nkx2.8 and 5Ј-deleted pCMV-Nkx2.8⌬ were cloned as follows. A KpnI-DraI segment of nkx-2.8 (bp 531-1180) was cloned into KpnI and HpaI sites of linker plasmid pSL1180 (Pharmacia Biotech Inc.) and subsequently excised as a KpnI-MluI segment. To construct pCMV-Nkx2.8⌬, a double-stranded synthetic oligonucleotide containing an NheI site, a consensus ribosome binding site, an ATG start codon, and nkx-2.8 bp from 299 to 332 (a TfiI site) was cleaved with NheI and TfiI and annealed to TfiI-KpnI (bp 332-531) and KpnI-MluI segments. The three-segment combination was cloned into the NheI and MluI sites of pCI. pCMV-Nkx2.8⌬ expresses a peptide shortened at the NH 2 terminus by 31 amino acids, corresponding to the original deleted gt11 clone. To construct the full-length expression plasmid pCMV-Nkx2.8, a segment containing the region from the initiation codon to the KpnI site (bp 532) was amplified by PCR using a 5Ј-primer that added an NheI site and a consensus ribosome binding sequence. The amplimer was digested with NheI-KpnI and substituted into pCMV-Nkx2.8⌬ at the same restriction sites. Expression plasmids pCMV-TTF1 (4) and pCGN-Nkx2.5 (33) were provided by R. DiLauro and R. J. Schwartz, respectively.
Gel Shift Analysis-pCMV-Nkx2.8 and pCMV-Nkx2.8⌬ DNAs were linearized with BamHI, transcribed with T7 RNA polymerase, and translated using a TNT Coupled Wheat Germ Extract System (Promega) according to the supplier's protocols. Labeling with [ 14 C]leucine was also described in these protocols. Cell extracts and gel shift procedures were previously described (18).

RESULTS
cDNA Cloning and Sequence-Using a combination of approaches, overlapping products have been cloned which encode a novel transcription factor (Fig. 1). BLAST analysis of translated sequences showed two characteristic domains, a homeobox ( Fig. 2A) and a conserved peptide (Fig. 2B). Both domains established a relationship to the Drosophila factor nk-2. These comparisons also demonstrated that the encoded factor is new, the shortest member of the family described so far. The new factor is the eighth vertebrate nk-2-related factor, hence nkx-2.8.
Despite its shorter length, the nkx-2.8 open reading frame defines a full-length factor. The 239-amino acid peptide of 25,810 Da is a basic protein with a predicted pI of 9.57. The cDNA also has an unusually high GϩC content, 65.8% over its full length and 71.0% in the open reading frame.
Other regions of the nkx-2.8 did not show significant relationships on BLAST analysis. However, FASTA alignment demonstrated significant homology to other nk-2-related factors. nkx-2.8 has an NH 2 -terminal region comparable to nk-2related factors, which have an 11-amino acid "5Ј-peptide" near the NH 2 terminus (Fig. 2C). This peptide lies in a similar position in other family members except for human TTF-1, which has a unique 30-residue extension at the 5Ј-end. Although the 5Ј-peptide is not conserved in nkx-2.8, the nine residues beginning at amino acid 9 (TVRSLLGLP) show 44% identity to the 5Ј-peptide consensus. Four residues are identical, and the other five represent conservative substitutions. Other alignments (not illustrated) indicate that the NH 2 -terminal segment of nkx-2.8 is shortened in the middle and near the junction with the homeobox compared with other members of the family. The NH 2 -terminal segment of nkx-2.5 is a weak transcription activation domain, rich in proline, alanine, acid, and basic amino acid residues (33). The NH 2 -terminal segment of nkx-2.8 has a similar composition but with a lower proportion of basic residues. At the COOH terminus, sequences of several factors align directly with nkx-2.8. Thus the major difference from the other factors is shortening of the NH 2terminal segment. Because comparisons establish homology directly at the NH 2 and COOH termini, nkx-2.8 appears to be a full-length factor.
nkx-2.8, like most nk-2-related factors, has a conserved peptide domain between the homeobox and the COOH terminus. The domain has a hydrophobic center, VAVPVLV, with a central proline that prevents helix formation. This center is surrounded by charged, mostly basic, residues. This suggests a single loop reminiscent of the nonhelical loops in known DNA binding domains and might therefore represent an accessory DNA binding domain. The paired class homeodomain factors have a second DNA binding domain and thus provide a precedent for homeodomain factors with two distinctive DNA binding domains. Alternatively, the domain might function to associate proteins.
nkx-2.8 is most closely related to TTF-1 and nkx-2.2. TTF-1, however, has a longer, more complex structure, with redundant domains. Even the untranslated region shows a relationship to TTF-1 because the short open reading frames in the 5Ј-untranslated region of the nkx-2.8 cDNA region can be aligned with the NH 2 -terminal region of human TTF-1 with about 30% identity. In addition, both TTF-1 and nkx-2.8 have polyglycine regions, although in slightly different locations. TTF-1 has a stretch of eight glycines between the homeobox and conserved peptides, whereas nkx-2.8 has six consecutive glycines just downstream of the conserved peptide.
Gene Structure-The predicted cDNA structure was confirmed with a series of PCR studies of fetal liver cDNA, and parallel studies were carried out on genomic DNA. This analysis verified the unique cDNA and also localized an intron. The following primer combinations were studied: F7:R10A, F7:R7A, F7:R12A, F7:R13A, F12:R13A, and F5A:R16 (Table I). In the cDNA analysis, each combination gave a single strong product, suggesting that the mRNA has no splicing variants. Amplification with F7:R7A and F7:R10A gave genomic DNA products ϳ750 bp larger than cDNA products, whereas genomic DNA did not amplify with F7:R12A. All other products were identical for both cDNA and genomic DNA. These results suggested a ϳ750-bp intron at the position of primer R12A. A genomic PCR product was cloned and partially sequenced (Fig. 3), confirming an intron within the codon located 31 amino acids from the NH 2 -terminal side of the homeobox. This is similar to the structure of the human TTF-1 gene, which has a single 966-bp intron within a codon 36 residues proximal to the homeobox (29).
mRNA Analysis-Standard Northern blot analysis of total RNA was attempted with several probes, but because of the high GϩC content, specific signals were obscured by nonspecific hybridization to rRNA. To solve this problem, a riboprobe plasmid was constructed containing the F9:R5 amplimer, a region within the homeobox which had lower GϩC content (Fig.  4A). In blots of total RNA, this probe detected a transcript with an apparent molecular mass of 1.6 kb but still showed crosshybridization to rRNA even under stringent conditions (not illustrated). However, when poly(A) RNA was isolated for blots, the hybridization was predominately to this single 1.6-kb transcript, which was clearly detected in RNA from three hepatocarcinoma cell lines, HepG2, HuH7, and Clone 2 (Fig. 4B). The cholangiocarcinoma cell line, R7451, and adult tissues including liver, appeared to lack this transcript, although very weak signals near 1.6 kb and at higher molecular masses could represent either low levels of transcript or cross-hybridizing species. mRNA was also isolated from three fetal liver specimens and compared with three additional adult liver speci- FIG. 2. Comparison of nkx-2.8 with other nk-2-related factors. Panel A, homeodomains. Related homeodomains were aligned and listed above nkx-2.8 in order of decreasing similarity. A consensus sequence was derived for nk-2-related factors, and a second consensus was derived from 100 HOX genes closely related to the Drosophila Antennapedia gene. The nkx-2.8 sequence is in bold type; amino acids that deviate from the consensus are shown in light type. The column labeled ⌬ shows the number of differences from the nk-2 consensus for each sequence. Predicted DNA base (:) and backbone (.) contacts are also marked (34). Panel B, conserved peptides. This nkx-2.8 domain (bold type) is closely related to a consensus derived from other nk-2 factors. The domain consists of a nonhelical hydrophobic loop surrounded by highly charged, predominantly basic amino acids. The column labeled ⌬ shows the number of differences from the consensus. Panel C, 5Ј-peptides. Conserved peptides found in many nk-2-related factors near the 5Ј-end are aligned. Position denotes the amino residue where each 5Ј-peptide is located. Conservative amino acid substitutions are underlined; nonconservative substitutions are shown in lowercase. Panel D, comparison of nkx-2 proteins. nkx-2.8 is shorter than the related factors but aligns at the 5Ј-ends of nkx-2.2, nkx-2.5, and murine TTF-1, and with amino acid 30 of human TTF-1, the longest protein in the family. All align at the 3Ј-end. Studies of protein functional regions, reported for murine nkx-2.5 (33) and human TTF-1 (4), are also summarized. mens. A weak signal at 1.6 kb was suggested in at least two of the fetal liver mRNAs. The Northern blot studies can only be considered semiquantitative because poly(A) RNA preparations vary significantly.
Reverse transcriptase PCR analysis was also set up (Fig. 4C) because the Northern blot detection in fetal liver was inconclusive, and the possibility of even lower transcript levels in adult liver could not be ruled out. Reverse transcriptase PCR demonstrated a 273-bp product predicted for spliced transcripts in two cell lines and fetal liver. Hybridization confirmed the identity of the PCR product and indicated that larger products detected in some of the RNA preparations were nonspecific. On long film exposures, a very weak signal was also detected in lung RNA, but the analysis of adult liver was completely negative. Thus nkx-2.8 mRNA was detected unequivocally in fetal but not adult liver and in three cell lines with phenotypes that resemble fetal liver.
Analysis of Translated nkx-2.8 Protein-The full-length and deleted nkx-2.8 expression plasmids were translated in vitro and the products analyzed by oligonucleotide gel shift. Gel shifts (Fig. 5A) were carried out with a PCE oligonucleotide (tcgaTGTTCAAGGACA). Full-length nkx-2.8 protein produced a band shift that comigrated with the PCF band shift of HepG2 cells but was distinctly lower than the band shift of the related factor, TTF-1. As expected, the 5Ј-deleted protein produced a band shift lower than that of intact nkx-2.8. The analysis demonstrated that the expression plasmids containing the cDNAs are functional and encode peptides that bind the PCE.
The gel shift analysis showed a similarity between PCF and nkx-2.8. However, the latter had a predicted molecular mass of 26 kDa, whereas two types of experiments indicated a higher molecular mass for PCF as measured on SDS-acrylamide gels: UV-cross-linking indicated peptides of about 34 kDa, whereas partial purification enriched three bands of 32-34 kDa (18). Translated protein migrated as a single band in SDS-acrylamide electrophoresis, with an apparent molecular mass of 33.6 kDa (Fig. 5B). Basic proteins like nkx-2.8 frequently show anomalous slow migration on SDS-acrylamide gels. The observed migration is common to both nkx-2.8 and PCF.
nkx-2.8 Function in a Nonhepatocytic Cell Line-HeLa cells, expected to be free of endogenous expression of nkx-2.8 and other hepatocytic transcription factors, were used for the initial evaluation of transcriptional activity. nkx-2.8 was analyzed directly and compared with the related factors nkx-2.5 and TTF-1 (Fig. 6). For this analysis, a reporter plasmid, pPCE4-HIV-CAT, was constructed in which four PCE sites were placed in tandem near a basal promoter. In HeLa cells, the reporter showed clear activation over unstimulated levels. Maximum 9-fold stimulation occurred at 1 g. Higher levels had somewhat lower activity but still stimulated the reporter. nkx-2.5 also showed comparable (14-fold) activation. The deleted form of nkx-2.8 showed no transcriptional stimulation, nor did TTF-1, even though the latter showed clear binding to a PCF oligonucleotide (Fig. 5). 2 Thus nkx-2.8 is a functional transcriptional activator, can act through a PCE site, and requires the NH 2 -terminal 31 amino acids for activation. nkx-2.5 has comparable function, although another nk-2-related factor, TTF-1, cannot activate the same reporter plasmid in HeLa.

nkx-2.8 Function in Hepatocytic Cell
Lines-For evaluation of nkx-2.8 in hepatocytic cells, we attempted to study differentiated hepatocytic cell lines with both fetal (AFP ϩ , albumin ϩ ) and adult (AFP Ϫ , albumin ϩ ) phenotypes (Fig. 7A). In general, hepatocellular carcinoma cell lines either have a fetal phenotype like HepG2 or are undifferentiated. Cell lines reported to have an adult hepatocytic phenotype are infrequent and problematic, although we utilized two such lines, Clone 2 and H4C3. Morinaga et al. (27)  Right, a separate hybridization study compares three human fetal liver specimens (1, 18 weeks gestation; 2, 18 weeks; 3, 21 weeks) and three adult liver specimens (1, 63 years; 2, 57 years; 3, 16 years). Each lane in both panels contains the poly(A) RNA purified from 100 g total RNA. Panel C, reverse transcriptase PCR analysis. 1-g aliquots of total RNA were treated with reverse transcriptase and primers from both nkx-2.8 and a control housekeeping gene, PBDG. These were then subjected to separate amplifications. Fetal liver 1 and adult liver 1 were used for these amplifications. Top panel, an ethidium bromide-stained electrophoresis gel in which the predicted 273-bp nkx-2.8 PCR product was detected in products from HepG2, HuH7, and fetal liver mRNA; other higher molecular bands were also generated by PCR. Middle panel, the gel was blotted and hybridized with a specific probe that verified the detection of the 273-bp band on the stained gel. Bottom panel, the 578-bp PBDG product was detected in the products from all of the RNA samples.
2 was reported to have selectively lost AFP expression. However, our analysis of Clone 2 showed reduced but still significant AFP expression. Moreover, we also observed clear nkx-2.8 expression in Clone 2 (Fig. 4). Reuber hepatoma cell lines like H4C3, common models for the adult hepatocyte phenotype, also have problematic features. Because H4C3 is rat-derived, we did not analyze for nkx-2.8 transcripts. However, we previously detected low AFP mRNA expression (23) and a weak PCF gel shift from these cells (18). 2 A variety of reporter genes were evaluated in the three hepatocytic cell lines. To indicate the cell differentiation states in these lines, the relative activities of the reporters are compared in Fig. 7A. Fig. 7B illustrates two experiments in HepG2 cells. The data on the left show the responses of two reporters to increasing levels of transfected nkx-2.8. The data on the right compare several nk-2 family expression plasmids transfected at a single DNA concentration. In HepG2 cells, the reporter pPCE4-HIV-CAT was stimulated strongly (14-fold) compared with the same promoter without the PCE binding site, indicating a significant level of an endogenous activating factor binding at the PCE. The transfected expression plasmids all reduced expression of this reporter gene; they also repressed a very active reporter containing intact AFP gene transcription controls. The repression was relatively weak and observed with intact and deleted nkx-2.8, TTF-1, and nkx-2.5, although these may not all repress by the same mechanism. Comparison of the two illustrated nkx-2.8 analyses shows the general reproducibility of transfections in separate experiments; the levels were 76 Ϯ 19% and 51 Ϯ 7% of the control (reporter plasmid alone).
Most likely, repression in HepG2 resulted from squelching, i.e. overexpression of transcription factor and competition for trans-activators or binding sites for protein-protein interaction. Such squelching is probably apparent in the HeLa cell studies, where greater amounts of transfected plasmid produced less stimulation than 1 g. Thus the repression in HepG2 cells may have occurred because HepG2 already contained significant levels of a PCE-activating factor. An alternate possibility is competition for binding sites in the absence of transcriptional activation. This is most likely true for the deleted nkx-2.8, which lacks an activation domain, but nkx-2.8, nkx-2.5, and TTF-1 are clearly activators in other settings (e.g. HeLa and H4C3 cells). Nevertheless, there is a formal possibility that the factors cannot interact with the promoters of either the PCE reporter or the AFP reporter in the transcriptional environment of HepG2 cells. In this case, it would have to be presumed that another trans-factor activates in these cells through binding at the PCE and that all three factors, nkx-2.8, nkx-2.5, and TTF compete for binding of this factor but cannot activate.
Transfection of both intact and deleted nkx-2.8 also repressed the AFP gene reporter in Clone 2 cells (Fig. 7C), although in these cells, the activity of the PCE reporter was too low to show significant effects. The fact that both HepG2 and Clone 2 showed repression by a PCE-targeted factor suggests that endogenous AFP gene controls active at the PCE were intact in both cell lines. Clone 2 has much less AFP gene expression than its parental line, but this apparently reflects altered regulation at gene sites other than the PCE and probably does not represent a true adult hepatocytic phenotype.
In contrast to the other hepatocytic lines, H4C3 cells ( 7D) showed some stimulation of both AFP and PCE reporters. Intact nkx-2.8 showed only weak effects. Although there was no significant stimulation of the PCE reporter in two separate experiments, an AFP gene reporter was stimulated 22% in the illustrated experiment and 20% in a separate experiment (not illustrated). The deleted nkx-2.8 had no effect on the reporters, whereas nkx-2.5 and TTF-1 showed greater than 3-fold stimulation in some combinations. The transfections with the latter two factors demonstrate that nk-2-related factors, acting through the PCE, are potentially strong AFP activators in H4C3 cells.
Notably, nkx-2.8 did not induce AFP expression in H4C3 cells up to the levels of HepG2 cells. However, such a result would be expected only if lack of nkx-2.8 expression was the single property that prevented strong AFP gene expression in H4C3 cells. Interestingly, TTF-1 activated the reporters in H4C3 but not in HeLa, suggesting that a variety of cofactors with distinct cellular distributions interact with various nk-2related factors.

DISCUSSION
An Unusual Gene and Transcript-nkx-2.8 mRNA has an unusual DNA sequence. The overall GϩC content is 66%, whereas the coding region is 71%. Moreover, the exons contain 129 CG dinucleotides, potential DNA methylation sites. Such CG islands are found upstream of some genes but are rarely incorporated into coding regions. A high GϩC content is also a feature of the mammalian TTF-1 (71% GϩC), nkx-2.2 (65%), and nkx-2.5 (70%) genes (11,29,36). Interestingly, when examples of these genes have been sequenced from lower vertebrates, the high GϩC content is not conserved (10,37). The Xenopus nkx-2.5 gene, for example, is only 48% GϩC.
Computer analysis of RNA folding predicts an unusually high degree of secondary structure for the nkx-2.8 mRNA. The combination of high GϩC content and secondary structure has led to technical problems in the cloning, purification, and quantification of the mRNA. The abundance of nkx-2.8 mRNA may have been underestimated because of these technical problems. Transfections were as described above, except that these LipofectAMINE transfections used 6.7 g of reporter plasmid, 1.3 g of expression plasmid, and 1 g of carrier DNA. All values were normalized to the basal expression of pPCE4-HIV-CAT (100%). In this cell line, the basal expression of pAFP6000 is 1.6 times this level.
Nevertheless, a low level of mRNA is characteristic of the nkx-2 gene family except for TTF-1 and may typify genes that have limited regulatory function. The studies of this paper provide a technical paradigm for a difficult class of mRNAs.
Relationship of Binding Sites-It is not surprising that there is a close relationship between the PCE and TTF-1 binding sites (Fig. 8) because the nkx-2.8 homeobox is extremely similar to the TTF-1 homeobox and to those of other members of the family. The sites share a common motif, TCAAG, which is also the central motif of the Drosophila nk-2 binding site (38). nkx-2.5 also binds strongly to a TTF-1 site (33). The structure and sequence of these nk-2-family homeodomains differ considerably from the more common Antennapedia class homeodomains. The latter bind to a TAAT central motif that corresponds to CAAG in the nk-2 family sites (39).
nkx-2.8 and PCF-nkx-2.8 was obtained through efforts to clone the AFP gene developmental regulator activity we characterized as PCF (18). There are marked similarities between PCF and nkx-2.8. Both bind the PCE and activate from that site, and both are expressed in the same cells, along with AFP. However, the exact peptide composition of PCF is not fully resolved. Photoaffinity labeling demonstrated a broad PCF band migrating at 34 kDa, and partially purified PCF showed multiple bands in the 32-34-kDa range. nkx-2.8 is basic, and PCF also purifies as a basic protein(s). Although only 26 kDa, nkx-2.8 has anomalous migration on SDS-acrylamide gels which causes it to migrate near 34 kDa, the position of PCF components. The multiple PCF bands might represent isoforms, post-translational modifications, related factors, coactivators, or even unrelated peptides that copurify. It appears likely, however, that nkx-2.8 is a component of PCF.
There is a more significant problem in the relationship between PCF and nkx-2. 8. Surprisingly, transfection of nkx-2.8 into AFP-positive lines HepG2 and Clone 2 caused transcriptional repression of AFP gene reporters, whereas transfection in H4C3 cells led only to weak AFP gene activation. In contrast, PCF has been characterized as a strong activator in HepG2 cells. Although other explanations are possible, repression may indicate squelching of an activity already present in HepG2 and Clone 2. The weak activation in H4C3 cells (with stronger activation by other nk-2-related factors) does indicate that nkx-2.2 and related factors can act as AFP gene activators in a hepatocytic cell setting. The weak effects are disappointing, but this may reflect the limitations of available cell lines as experimental models. H4C3 cells may lack other components necessary for full activation of AFP or even, according to standard models, actively repress the gene (40,41). It will be necessary to examine other cell systems, potential coactivators, and possibly transgenic systems before the role of nkx-2.8 is fully resolved.
Galarneau et al. (42) have used transfection experiments to demonstrate that another factor, FTF, can activate AFP transcription on binding to the PCE. FTF is also known as LRH and PHR and is related to Drosophila FTZ-TF1. However, PCF from HepG2 cells is a distinctive activity unrelated to FTF, and FTF is undetectable in HepG2 cells (18,42). Moreover, the levels of FTF are highest in adult liver, where the AFP gene is silent. Recent studies in Drosophila have shown an important relationship that may explain interaction of PCF, nkx-2.8, and FTF. In Drosophila, FTZ-TF1 and the homeodomain protein ftz are mutually dependent cofactors. The two factors interact to activate important low affinity sites (43,44). Similar interaction between nkx-2.8 and FTF, or other members of each family, might account for the complex regulation that is observed at the PCE.
nkx-2.8 and Other nk-2-related Factors-The nk-2-related factors are defined by distinctive homeoboxes, and the studies in this paper demonstrate that nkx-2.8, nkx-2.5, and TTF-1 all bind the PCE. nkx-2.5 substituted for nkx-2.8 in all of the transfection assays, suggesting very similar function for these two factors, whereas TTF-1 showed more limited ability to substitute. Like many activation domains, the nkx-2.8 NH 2 terminus is rich in proline and acid residues, and the deleted 31-amino acid region shares this composition. Direct NH 2 -terminal deletion eliminated activation function. In contrast, nkx-2.5 and TTF-1 have terminal regulatory domains with more internal activation domains because short NH 2 -terminal deletions increased transcriptional activation before more extensive NH 2 -terminal deletion removed the apparent activation domain (4,33). nkx-2.8 has a shorter NH 2 -terminal region (86 amino acids) than either nkx2.5 (136 amino acids) or murine TTF-1 (159 amino acids). nkx-2.8 is thus more compact than other members of the family and may lack an NH 2 -terminal regulatory region, but it nevertheless retains activation function in its shortened NH 2 terminus. Moreover, the unique NH 2 -terminal domain of nkx-2.8 might be subject to specific regulation in hepatocytic cells which is evaded by nkx-2.5 and TTF-1.
The present studies have established that nkx-2.8 gene expression is associated with AFP expression in fetal but not adult liver and in hepatocellular carcinoma. AFP is expressed not only in fetal liver but also in primitive endoderm and yolk sac. A common target in the AFP gene could be regulated by different members of the nk-2 gene family in these settings, just as cardiac development involves the partially redundant expression of several different nk-2-related genes (2, 8 -10). Future areas of investigation include the role of nkx-2.8 overexpression in neoplasia and the identification of additional nkx-2 genes related to endodermal development.
Our findings that associate nkx-2.8 with liver development remain preliminary, limited by our initial choice of a human cell model, because the human is not a suitable organism for the detailed study of development. For characterization of animal development, we recently cloned segments of the rat and mouse nkx-2.8 genes and found that they are extremely similar to their human counterpart in both NH 2 -terminal and homeobox regions. 3 Such strong conservation suggests a gene with important function.