Loss of hepatocyte ERBB3 but not EGFR impairs hepatocarcinogenesis

Abstract

Epidermal growth factor receptor (EGFR) and ERBB3 have been implicated in hepatocellular carcinogenesis (HCC). However, it is not known whether altering the activity of either EGFR or ERBB3 affects HCC development. We now show that Egfr^Dsk5 mutant mice, which have a gain-of-function allele that increases basal EGFR kinase activity, develop spontaneous HCC by 10 mo of age. Their tumors show increased activation of EGFR, ERBB2, and ERBB3 as well as AKT and ERK1,2. Hepatocyte-specific models of EGFR and ERBB3 gene ablation were generated to evaluate how the loss of these genes affected tumor progression. Loss of either receptor tyrosine kinase did not alter liver development or regenerative liver growth following carbon tetrachloride injection. However, using a well-characterized model of HCC in which N-nitrosodiethylamine is injected into 14-day-old mice, we discovered that loss of hepatocellular ERBB3 but not EGFR, which occurred after tumor initiation, retarded liver tumor formation and cell proliferation. We found no evidence that this was due to increased apoptosis or diminished phosphatidylinositol-3-kinase activity in the ERBB3-null cells. However, the relative amount of phospho-STAT3 was diminished in tumors derived from these mice, suggesting that ERBB3 may promote HCC through STAT3 activation.

hepatocellular carcinoma (HCC) is the second-leading cause of cancer deaths worldwide, resulting in over 600,000 deaths per year (22). Its incidence is rising, particularly in the United States, where it has increased by 70% over the last 25 years and is expected to double by 2030. Men are two to four times more often afflicted with HCC, which may be due to HCC-promoter effects of androgens and the inhibitor effects of estrogens (48). Hepatomas are usually the result of chronic liver disease, including hepatitis B/C or alcohol-induced cirrhosis. No effective secondary prevention or systemic treatments are available, and it is one of the most lethal cancers, partly because of the delay in diagnosis (19). Novel chemopreventive and therapeutic strategies to target key molecules associated with HCC are needed to improve patient well-being and survival.

Recent clinical evidence suggests that epidermal growth factor receptor (EGFR or ERBB1) tyrosine kinase inhibitors such as erlotinib show promise in the treatment of human HCC (45). Consistent with human studies, a study in rats demonstrated that gefitinib administration reduced HCC in an N-nitrosodiethylamine (DEN)-induced liver injury model in which cirrhosis progresses to HCC (41). The effect was ascribed to a direct inhibitory effect of EGFR inhibition on the tumors themselves. More recently, erlotinib was also shown to reduce the development of HCC in a rat model in which repeated low-dose injections of DEN were administered. However, the inhibitory effect was attributed to decreased initiation of HCC by DEN in hepatocytes as opposed to a direct effect on HCC progression itself (15). Conversely, others have reported that a genetically modified mouse deficient for EGFR in hepatocytes and bile duct cells showed increased HCC following a two-stage model of HCC using DEN and phenobarbital (23). The increased number and size of tumors was attributed to increased proliferation associated with the loss of hepatocellular EGFR, which was unexpected given the general belief that EGF is a mitogenic factor and tumor promoter. However, the recent demonstration that phenobarbital is itself a ligand for EGFR introduces a confounding factor in this study (30, 32).

EGFR belongs to the ERBB protein family, which consists of four family members (27). Although frequently studied as individual proteins, they are highly interactive and form signaling homodimers and heterodimers upon binding to EGF-like ligands or heregulins (HRG). Our laboratory has shown that three members of this family are expressed in hepatocytes: EGFR (ERBB1), ERBB2 and ERBB3 (7, 38). By forming heterodimers, different pairs of ERBB proteins can initiate unique intracellular signaling cascades. In human HCC, it has been reported that ERBB3 is one of 218 proteins (of 9,000 examined) that is consistently upregulated and that either EGFR or ERBB2 (but not both) tend to be upregulated along with ERBB3 (33). ERBB3 was previously thought to lack intrinsic kinase activity; however, it is now known to be a weak kinase, activated by the interaction with other ERBB molecules (43). In contrast to other ERBB kinases, which are activated by ligand binding alone within a homo- or heterodimeric kinase signaling unit, ERBB3 kinase activation requires a transient physical interaction with its dimeric ERBB binding partner. ERBB3 monomers, once activated, can dissociate from the initial heterodimeric pairings, forming HRG-activated ERBB3 homodimers. In addition to interacting with other ERBB kinases, ERBB3 can directly interact with other signaling receptor tyrosine kinases, such as MET (11) and IGF1R (20), providing a strategy through which a cancer cell can elude pharmacological ERBB inhibition.

To investigate the physiological role of EGFR in hepatocarcinogenesis of adult mice, we analyzed HCC in the Egfr^Dsk5 mutant mouse model (9). The EGFR in this mouse bears a mutation that results in a leucine to glycine at position 863 (Leu863Gln) within a region of the kinase domain that stabilizes the receptor activation loop, producing a gain-of-function allele that increases basal EGFR kinase activity. We observed spontaneous HCC in older male adults heterozygous for the Egfr^Dsk5 allele. We then used a genetic approach to ablate EGFR or ERBB3 in hepatocytes [hepatocyte-specific (HS)-EGFR-knockout (KO) or HS-ERBB3-KO] of male mice to assess their importance in the progression of chemically induced HCC. Using the well-characterized one-stage DEN-induced model of hepatocarcinogenesis, we evaluated cancer progression in these mice, administering the chemotoxin to 14-day-old mice before significant hepatocellular EGFR or ERBB3 ablation occurred. In this model, we find no essential role for EGFR in HCC progression. However, we do show that loss of ERBB3 retards HCC growth and results in fewer tumors, demonstrating that ERBB3 is required for optimal HCC growth, possibly involving STAT3 phosphorylation.

EXPERIMENTAL PROCEDURES

Mice and genetic crosses.

The Egfr^Dsk5 allele was initially generated by random mutagenesis with N-ethyl-N-nitrosourea (ENU) and maintained isogenic on the C3H background (13). 129S1/SvImJ-Egfr^Dsk5 congenic mice were generated by backcrossing C3H-Egfr^Dsk5 heterozygous stocks to 129 wild-type mice for more than 10 generations (9). We used heterozygotes primarily because of the difficulty in breeding and maintaining sufficient number of homozygotes. Genetic mouse models Egfr^tm1Dwt (Egfr^f) (26), Erbb3^tm1Dwt (Erbb3^f) (24), and B6.Cg-Tg 21 Mgn/J (Alb-Cre) (The Jackson Laboratory) mouse strains were generated as previously described. Egfr^f has loxP sites flanking exon 3 of the Egfr. Cre recombinase-mediated deletion of exon 3 causes a frameshift with two stop codons in exon 4 and early termination of translation. To generate hepatocyte-specific EGFR-KO, Alb-Cre mice were crossed with Egfr^f/f to generate Alb-Cre, Egfr^f/f or their littermate controls Egfr^f/f (40). A similar approach was used to establish Alb-Cre, Erbb3^f/f mice. Mice were fed Purina Mills Lab Diet and water ad libitum under specific pathogen-free conditions in an American Association for the Accreditation of Lab Animal Care-approved facility. Mice were raised under conditions of regulated lighting (lights on 0600–1800), temperature, and humidity. The Vanderbilt Institutional Animal Care and Use Committee approved all experiments.

Genotyping.

DNA was extracted from ear punches or tail biopsies for genotyping by incubating at 95°C in 100 μl of 25 mM NaOH/0.2 mM EDTA for 20 min and then neutralizing with 100 μl of 40 mM Tris·HCl, pH 5.0. For the subsequent genotyping reactions, 1 μl of lysed tissue sample was used per reaction. PCR conditions were 35 cycles at 94°C for 45 s, 62°C for 60 s, and 72°C for 120 s. The Egfr alleles were amplified by PCR with the following primers: EGFR, 5′-CTTTGGAGAACCTGCAGATC-3′ and EGFR, 5′-CTGCTACTGGCTCAAGTTTC-3′. PCR products were run on a 3% agarose gel to separate a 320-bp product corresponding to wild-type Egfr and a 370-bp product corresponding to the KO Egfr allele. The Erbb3 alleles were amplified by PCR with the following primers: ERBB3, 5′-TCCAGCGTGGAAAAGTTCAC-3′ and ERBB3, 5′-AAGCCTTCTCTATGGAAAGTG-3′. PCR products were run on a 3% agarose gel to separate a 354-bp product corresponding to wild-type Erbb3 and a 488-bp product corresponding to the KO Erbb3 allele. +/Egfr^Dsk5 mice were genotyped as previously described (40).

N-nitrosodiethylamine hepatocarcinogenesis.

Mice received a single intraperitoneal injection of 10 mg/kg body wt of N-nitrosodiethylamine (Sigma-Aldrich, St. Louis, MO) at day 14 of age (36). Subgroups of male and female mice were euthanized at 3, 4, 5, 6, and 8 mo for EGFR and 6 and 8 mo for ERBB3. Solitary tumors manifested in some of the male mice as earlier as 3 and 4 mo of age, but only in about half of the mice. The subgroups from 3 to 6 mo for both EGFR and ERBB had 5 mice per subgroup whereas the four 8-mo subgroups each had 15–20 mice and were the main focus of this study.

Collection of livers and other organ samples.

Mice were anesthetized with 3% isoflurane before being subjected to a thoracotomy and cardiac puncture to obtain blood. Organs were rapidly dissected from each animal and the wet weights were recorded. Portions of the liver were either frozen in liquid nitrogen for protein and RNA analyses or fixed in phosphate-buffered 4% paraformaldehyde for subsequent paraffin embedding and various histological analyses.

CCl₄ model of liver injury and regeneration.

Carbon tetrachloride (CCl₄) (0.6 μl/g body wt) in olive oil (200 μl) was injected into the peritoneum of adult male mice and they were then euthanized at 48, 96, or 120 h after injections. All injections were performed during the middle of the light phase to minimize circadian differences in toxicity. There were five or more mice per time point and no postinjection mortality. Since deletion of floxed DNA in hepatocytes is age dependent, we used mice that were at least 8 wk old.

Immunohistochemistry.

Mice that underwent CCl₄ injection were injected with 100 μg/g of 5-bromo-5′-deoxyuridine (BrdU; Boehringer Mannheim, Indianapolis, IN) 1 h before euthanasia. Livers were fixed in PBS buffered 4% paraformaldehyde. After overnight fixation, liver samples were transferred to 70% ethanol, dehydrated in a graded series of ethanols and xylenes, and embedded in paraffin. Sections 5 μm thick were cut using a Leica Biocut 2030 microtome. Sections were deparaffinized and rehydrated in a graded series of ethanols. They were immunostained with anti-BrdU (Ab-2, Oncogene Research Products, La Jolla, CA); anti-Ki67 (Clone SP6, Lab Vision, Fremont, CA); and anti-cyclin A (sc-596, Santa Cruz Biotechnology, Santa Cruz, CA). For BrdU immunostaining, tissue sections were pretreated with 1 N HCl for 8 min at 60°C to denature the DNA. Histological images were photographed on an Olympus Vanex AHBT3 microscope using a NIKON E5000 connected by a PTEM 257009 camera to an objective adaptor. In the CCl₄ experiments, sections were stained with a blue-colored total RNA (HealthGene, Toronto, ON, Canada) stain to highlight and facilitate quantification of the necrotic areas. These images were photographed, uploaded to a computer and the percent necrotic area was determined by use of the Image J program (42). In addition, we immunostained tumor sections from the various genotypes for Ki67 (Clone SP6, Lab Vision, Fremont, CA); TACE (Ab19027, Chemicon International, Billerica, MA); and ERRFI1 (R-2903, Sigma-Aldrich). For immunohistochemistry, sections were processed by the manufacturer's protocol (Vectastain Elite ABC kit; Vector Laboratories, Burlingame, CA). Endogenous peroxidase activity was quenched by placing the slides in 0.3% H₂O₂ in methanol for 30 min. After blocking the slides in a 3.0% goat serum solution for 1 h, the tissue sections were incubated with the antibodies for 90 min in a humidified chamber. The sections were then washed, incubated with biotinylated goat anti-rabbit secondary antibody, rinsed in PBS, and incubated with the Vectastain Elite ABC reagent. Diaminobenzidine was used as the peroxidase substrate.

Western blotting.

Pieces of liver (about 100 mg) were weighed and then homogenized on ice by using a 2-ml Wheaton glass tissue homogenizer in TGH buffer (20 mM HEPES, 1% Triton X-100, 10% glycerol 50 mM NaCl). This buffer included protease inhibitors (1 mM PMSF, 10 μg/ml aprotinin and 1 μg/ml leupeptin) as well as phosphatase inhibitors (10 mM sodium molybdate, 1 mM sodium orthovanadate, and 10 mM β-glycerol phosphate). Lysates were immunoblotted as previously described (37). We used the following affinity-purified antibodies from Santa Cruz Biotechnology (Santa Cruz, CA): sc-03 for EGFR; sc-285 for ERBB3; sc-283 for ERBB4; sc-15335 for α-actinin and sc-596 for cyclin A. Antibodies from Cell Signaling Technology (Beverly, MA) included pSer473-AKT (CSD9E); AKT (CS9272); Thr 202, Tyr 204, Erk 42, 44 antibody (CS4370); pY705STAT3 (CS9131); STAT3 (CS7907); Histone H3 (CSD1H2); pY1068 EGFR (CSD7A5); pY1197 ERBB3 (CS56E4). For pY1112ERBB2, we used antibody 06–562 from Upstate Biotechnology For ERRFI1 we used R-2903 from Sigma-Aldrich (St. Louis, MO). For ERBB2, we used Ab2428-1 from Abcam (Cambridge, UK). The lanes in each immunoblot contained equal amounts of protein, as determined by the Bio-Rad DC Protein Assay (Bio-Rad Laboratories; Hercules, CA). After each transfer, we confirmed equal protein loading and transfer by Ponceau S staining of immunoblots, scanning the image for future reference. We also used various antibodies that recognize relatively invariant proteins, such Histone H3 and α-actinin, as loading controls. Sample cohorts were analyzed on a single blot to ensure reliable comparison. Immunoreactive signal was detected by the ECL method with either the Supersignal West Pico Chemiluminescent Substrate or the Supersignal West Femto Maximum Sensitivity Substrate (Thermo Scientific, Rockford, IL). The decision to use one or the other depended on the quality of the antibody as well as the abundance of the target protein. We performed densitometry using an Epson scanner and the Image J program (42).

Statistical analysis.

Data are expressed as means ± SE. Statistical analysis was performed by an unpaired, two-tailed Student's t-test assuming equal variances between compared groups. A P value of <0.05 was determined to be statistically significant.

RESULTS

HCC arises spontaneously in Egfr^Dsk5 mutant mice.

We euthanized heterozygous Egfr^Dsk5 mutant male mice and their wild-type counterparts at various ages. Mutant but not wild-type mice developed spontaneous hepatocarcinomas after 8 mo of age. Although visible tumors were not present in mice younger than 8 mo of age, their hepatocytes had an elevated mitotic rate as well as large, dark and atypical nuclei, many of which were polyploid. By 10 mo, tumors began to appear in isolated mice. By 12 mo, all Egfr^Dsk5 mice had at least one large tumor. By 14 mo, the livers from each mouse had two to four large tumors. Figure 1A shows the onset of tumors in these mice as a function of age. Figure 1, B–D, show representative photomicrographs of various tumors immunostained for Ki67, a nuclear marker for proliferation-competent cells. The very black nuclei identify Ki67-positive cells. In Fig. 1B, the tumor borders nontumorous liver. The tumor situated to the right of the arrows has many proliferating cells (both cancer and endothelial cells) in contrast to the nontumorous liver to the left. The tumor cells resembled hepatocytes and have a sinusoidal stroma lined by a single layer of endothelial cells (Fig. 1B). They did not resemble the large polyploid cells seen in many fields in these mice (data not shown). Most tumor sections had this appearance, but some sections had cords (Fig. 1C), glands, or large cells with a clear cytoplasm (Fig. 1D). The clear cytoplasm was not glycogen (based on the absence of periodic acid-Schiff staining) and presumably is lipid.

Fig. 1.

Spontaneous hepatocarcinogenesis in +/Dsk5 mice. A: percentage of Egfr^Dsk5 mice at different ages that developed visible tumors. Over 20 tumors were identified in 10 mice at 13–14 mo of age. Usually, there were two to five 3- to 8-mm tumor nodules in each liver by 14 mo of age. No tumors were seen in the livers of wild-type mice. Each subgroup had 5 to 10 mice. B: tumor from the liver of an Egfr^Dsk5 mouse immunostained for MKI67, a cell proliferation marker. Many cell nuclei stained positively for MKI67 in the tumor to the right of the arrows whereas the adjacent liver to the left was negative, except for some nonparenchymal cells. C: in some MKI67-stained sections, the tumors had a more cordlike or even glandular appearance. D: other MIKI67 sections had clear cells that were periodic acid-Schiff negative and presumably represent lipid inclusions. Bar = 50 μM.

Evaluation of ERBB and signal transduction pathways in the liver and tumors of Egfr^Dsk5 mutant mice.

To begin to characterize the liver tumors in the Egfr^Dsk5 mutant mice, we evaluated the expression of the ERBB proteins and several signaling molecules in wild-type and +/Dsk5 livers as well as individual liver tumors from 14-mo-old mice. Figure 2 shows representative immunoblots. The blots were quantified by densitometry and the results that were significantly different are shown in Fig. 3. In general, the +/Dsk5 livers and tumors showed several-fold elevated levels of tyrosine phosphorylation for EGFR (pY-1068), ERBB2 (pY-1112), and ERBB3 (pY-1197) compared with the wild-type liver. The increased phosphorylation of ERBB3 was notable in the tumors, but there was variation between tumors. The “total” levels of EGFR were decreased in +/Dsk5 mouse liver, particularly in the tumors. This has been reported previously and has been ascribed to accelerated turnover of EGFR. Serine phosphorylation of AKT and ERK1,2 were also notably increased in the tumors relative to the nontumorous liver of the +/Dsk5 mouse. Consistent changes in ERRFI1 and pY-STAT3 were not observed. However, five of the eight tumors analyzed showed a prominent increase in expression of ERRFI1 whereas the remaining three showed diminished levels relative to the adjacent liver. The increased expression in five tumors was associated with a protein band having perceptibly greater mobility. Interestingly, there was a close correlation between the expression levels of ERRFI1 and pY-STAT3 levels in the tumors but not in the livers of these mice (this includes examination of data from other tumors not shown).

Fig. 2.

Immunoblot analysis of signaling molecules in wild-type (+/+) livers, +/Dsk5 livers, and +/Dsk5 tumors. A: immunoblots were carried out to analyze the phosphorylation of pY-EGFR (Y1068), pY-ERBB2 (pY1112), pY-ERBB3 (Y1189), pAKT (ser473), ppERK 42,44 (Thr 202, Tyr 204), and pY-STAT3 (Y705). The expression levels of ERRFI1 were also evaluated. B: immunoblots were carried out to determine the expression levels of EGFR, ERBB2, ERBB3, AKT, α-actinin, STAT3, and Histone H3. All data in this figure are from identical replicate blots. Each lane is from the same liver or tumor of the same mouse. All of the tumors were from the +/Dsk5 mice since the wild-type mice were tumor free. Note the enhanced phosphorylation of pAKT and ppERK1,2 in the +/Dsk5 tumors compared with the +/Dsk5 tumor-free livers.

Fig. 3.

Densitometric analysis of Fig. 2. Scanning densitometry was carried out for the blots represented in Fig. 2. There were 4 +/+ livers, 4 +/Dsk5 livers, and 8 tumors for analysis. Only the blots that showed statistically significant differences are displayed in this figure. *P < 0.001.

EGFR and ERBB3 expression and functionality are effectively reduced in the HS-EGFR-KO and HS-ERBB-KO mice.

We have previously reported successful disruption of EGFR expression in hepatocytes to create an HS-EGFR-KO mouse line. This line, which relied on Alb-cre recombinase to excise a floxed Egfr gene, showed reduced expression of EGFR after weaning (40). In this paper, we used a similar strategy to ablate the expression of ERBB3 in hepatocytes. To validate that we created an HS-ERBB3-KO mouse line, liver homogenates were prepared from the livers of male and female adult mice and immunoblotted for ERBB3. In the HS-ERBB3-KO mice, ERBB3 was markedly reduced by 2–3 mo of age, but not before weaning (Fig. 4A).

Fig. 4.

Generation of a HS-ERBB3-KO mouse. A: to generate hepatocyte-specific ERBB3-KO, Alb-Cre mice were crossed with ERBB3^f/f to generate Alb-Cre ERBB3^f/f or their siblings ERBB3^f/f (40). As shown in this immunoblot, this mouse showed ablation of ERBB3 in hepatocytes in adult but not preweaned mice. B: we injected the HS-ERBB3-KO or wild-type mice with CCl₄, a centrilobular toxin or the control vehicle (olive oil). This figure shows the necrotic liver surrounding 2 central veins (CV) that have been highlighted with a black line. The image represents tissue stained with a blue total RNA stain (HealthGene). The contrast has been increased in this image to permit grayscale viewing. The black lines within the necrotic area represent RNA in viable nonparenchymal cells. C: the area of necrosis in liver sections was monitored at 48 h after injection. Images such as the one in B were analyzed for the amount of necrotic area (excluding CV area). The extent of centrilobular necrosis in the livers of the CCl₄-treated HS-ERBB3-KO mice and wild-type littermates was similar. D: to assess cell proliferation status, we analyzed nuclear 5-bromo-5′-deoxyuridine (BrdU) labeling in both male and female mice. Left image shows representative nuclear BrdU labeling in a knockout mouse surrounding a necrotic area. A CV in the center of this image has been highlighted with a black line. No differences were observed in cell proliferation status in knockout (KO) or wild-type mice. E: we confirmed the BrdU results by immunoblotting for CCNA (cyclin A) expression in livers obtained from mice that had been injected 48 h earlier. No CCNA expression was detected in the olive oil-treated controls (data not shown). No differences were observed in CCNA expression for KO or wild-type mice. Bar = 100 μM.

Normal cell proliferation in ERBB3-deficient liver epithelium during development or in liver regeneration following exposure to CCl₄.

We identified no abnormalities in liver development or size in the HS-ERBB3-KO mouse. The liver displayed a normal acinus and grew at a normal rate (data not shown). We evaluated by immunoblot the effect of the loss of ERBB3 on various markers, such as ALB (albumin), phospho-AKT, the dually phosphorylated ERK1/2, other ERBB tyrosine kinases, MET, HMOX (heme oxygenase), and PLCγ (phospholipase gamma), but no differences were found in their expression or activation (data not shown). We hypothesized that loss of ERBB3 would augment chemotoxic injury or delay the regenerative response to it. CCl₄ is toxic to hepatocytes in the centrilobular zone of the liver, causing a necrotic area around the central vein within hours after injection. These hepatocytes are subsequently replaced by proliferating hepatocytes that migrate into the wounded area. This requires not only cell proliferation but also cell migration and inflammatory removal of the necrotic and apoptotic cellular debris. We injected CCl₄ into wild-type and HS-ERBB3-KO mice to determine whether ERBB3 ablation affected liver injury or cell proliferation. We had previously demonstrated that loss of hepatocyte EGFR did not affect the regenerative response to CCl₄ (40); however, there are non-EGFR-mediated mechanisms that could depend on ERBB3 to diminish injury or promote regeneration, including potential interactions with receptor tyrosine kinases, such as ERBB2 or MET. We harvested livers from mice euthanized at 48, 96, and 120 h after CCl₄ administration. We found that ERBB3 gene disruption had no effect on the initial injury (Fig. 4B), on cell proliferation as judged the levels of BrdU nuclear labeling (Fig. 4C), or on CCNA (cyclin A) expression (Fig. 4D) at 48 h. By 120 h after injection, the injured area was repaired in all female KO and wild-type mice and all male wild-type mice (data not shown). However, three of the five male HS-ERBB3-KO mice showed focal injury, suggesting that the regenerative response in male HS-ERBB3-KO mice may be mildly impaired.

Loss of ERBB3 but not EGFR results in diminished DEN-induced HCC.

We injected both male and female HS-EGFR-KO and HS-ERBB3-KO mice with DEN at day 14 after birth. Mice were euthanized at 3, 5, 6, and 8 mo after the initial injection for the EGFR genotypes and at 6 and 8 mo for the ERBB3 genotypes. Mice were euthanized and the number of tumor nodules and relative size were determined by gross inspection and microscopy. In the male mice, solitary tumors appeared by 3 mo. By 5 mo all mice had between 1 and 10 tumors. By 6 mo, all mice had between 14 and 20 tumors although most were small in size. It was not until 8 mo that a substantial number of large tumors manifested in all mice. There were only a few tumors in the female mice at any time point, consistent with previous reports of a sexual dimorphism in DEN tumorigenesis (21). We found no significant difference in the size of tumors in the HS-EGFR-KO mice, HS-ERBB3-KO mice or their respective wild-type counterparts. On average, there were slightly more tumors in HS-EGFR-KO mice, but this did not reach statistical significance (Fig. 5A). The liver weights (as a percent of body weight) of the tumor-bearing HS-EGFR-KO mice and their wild-type littermates were identical. In contrast, we did find a 50% reduction in tumor number in the HS-ERBB3-KO mice compared with the control mice (Fig. 5B). The liver weight in the HS-ERBB3-KO was also significantly reduced, correlating with a reduced tumor load (Fig. 5C).

Fig. 5.

Hepatocyte-specific (HS)-ERBB3-KO but not HS-EGFR-KO mice have decreased DEN tumorigenesis. A: the number of gross tumor nodules in the liver was counted for each HS-EGFR-KO (n = 15) and wild-type mouse (n = 15). The average number of tumors was slightly greater in the EGFR-KO mice, but the difference was not statistically significant. B: the gross number of tumor nodules in the liver was counted for each HS-ERBB3-KO (n = 15) and wild-type mouse (n = 15). The average number of tumors was statistically significantly greater in the wild-type than in the HS-ERBB3-KO mice. C: the weight of each liver was calculated as a percent of body weight for HS-EGFR-KO and HS-ERBB3-KO mice. The weight increased as a function of tumor load. No differences were seen for HS-EGFR-KO and wild-type mice, but the livers of the HS-ERBB3-KO mice were smaller than those of the wild-type mice, correlating with the difference in the number of tumors. *P < 0.0005, **P < 0.05.

Loss of ERBB3 but not EGFR decreases the rate of cell proliferation in HCC.

We evaluated the rate of cell proliferation in the various lines of DEN-injected mice by immunostaining for nuclear MKI67 (Ki67) and CCNA in normal and tumorous tissue. MKI67 is a marker for proliferation competent cells whereas CCNA is a marker of DNA synthesis. Tumors showed substantially increased MKI67 and CCNA immunolabeling relative to nontumorous tissue (shown later in Fig. 9). We found no differences in MKI67 or CCNA labeling in the HS-EGFR-KO mice and their wild-type counterparts (Figs. 6, A and B, left). However, we did see a statistically significant 30% reduction in MKI67 and CCNA immunostaining in the HS-ERBB3-KO mice compared with the wild-type mice, suggesting that ERBB3 ablation reduced cell proliferation (Fig. 6, A and B, right).

Fig. 9.

Histochemical and immunohistochemical analysis of ERBB3 wild-type and knockout tumors. A: hematoxylin and eosin stain of 2 representative tumors of the wild-type and knockout mice. The tumors tended to be basophilic compared with the adjacent nontumorous liver above the arrows. This is typical for the tumors initiated by DEN injection. Some tumors in both genotypes had clear cells. Bar = 100 μM. B: immunohistochemical localization of Ki67. TACE, and ERRFl1. Top: immunostaining for MKi67, a cell proliferation marker. Tumor (T) is situated to the left of each figure. One can see increased Ki67 labeling in the tumor compared with the nontumorous liver (L) in the right half of the figures. The images show decreased labeling in the knockout compared with wild-type tumors. Middle: increased diaminobenzidine (brown) immunostaining of TACE in the tumors compared with the livers, regardless of genotype. Bottom: markedly diminished ERRFI1 in the tumors of both genotypes relative to nontumorous liver. Bar = 100 μM.

Fig. 6.

Tumors in HS-ERBB3-KO mice, but not HS-EGFR-KO mice, have a lower rate of cell proliferation. A: we quantified the number of MKI67-positive nuclei in tumors from the KO or wild-type mice and found that the tumors in the HS-ERBB3-KO mice but not the HS-EGFR-KO mice had reduced MKI67 immunostaining, indicative of a reduced cell proliferation rate in the former. B: we quantified the number of CCNA-positive nuclei in tumors from the KO or wild-type mice and found that the tumors in the HS-ERBB3-KO mice but not the HS-EGFR-KO mice had reduced CCNA immunostaining, indicative of a reduced rate of DNA synthesis. Representative results are shown above each bar graph. *P < 0.0001, **P < 0.03, ***P < 0.0004.

Loss of ERBB3 in tumor results diminished phospho-STAT3 but has no consistent effect on the phosphorylation of ERK1/2 or AKT.

We examined the expression of the ERBB receptor tyrosine kinases in the HS-ERBB3-KO liver and tumor tissue. We were able to show the effective ablation of ERBB3 expression in liver and tumor tissue (Fig. 7A, top). A longer chemiluminescent film exposure did show ERBB3 immunoreactivity, consistent with the presence of nonparenchymal ERBB3 expression. We were also able to detect pY-1068 EGFR and pY-1112 ERBB2 (Fig. 7A, middle) in these tissues, which persisted in the absence of ERBB3. No ERBB4 was detected by immunoblot (data not shown). In contrast to the tumors observed in the +/Dsk5 mouse liver, we did not see statistically significant changes in the levels of the activated forms of ERK1,2 or Akt. Phospho-ERK1,2 tended to be higher in the tumors relative to the livers (Fig. 7A, bottom; Fig. 8) whereas phospho-Akt tended to be lower in the tumors. When we examined pY-Stat3, we found that it was significantly increased in tumors of wild-type compared to ERBB3 knockout mice (Fig. 7B). In both wild-type and knockout tumors, there was a striking reduction in ERRFI1 relative to nontumorous liver, confirmed by immunohistochemistry in Fig. 9. This protein was markedly decreased, but a longer chemiluminescent exposure did reveal a band. Immunohistochemistry also revealed varying levels of expression, even within a single tumor.

Fig. 7.

Immunoblot analysis of signaling molecules in wild-type and ERBB3 knockout livers and tumors. A: we monitored the levels of pY-ERBB3, ERBB3, pY-EGFR, EGFR, pY-ERBB2, pp-ERK1,2, and p-AKT in the livers of HS-ERBB3-KO or wild-type mice by immunoblot of liver or tumor lysates. Note that ERBB3 is knocked out in the livers of cre f/f mice. B: we also monitored the expression pY-Stat3, STAT3, and ERRFI1.

Fig. 8.

Densitometric analysis of Fig. 7. Scanning densitometry was carried out for several of the blots depicted in Fig. 2. There was an n = 6 for each genotype for this particular analysis. All samples for each protein were from the same blot; however, replica blots were used for different samples, and some of the blots were reprobed with different antibodies. Note that the tumors expressed less ERRFI1. The ERBB3 tumors expressed less pY-STAT3 than their wild-type counterparts. The ppERK1,2 tended to be greater in the tumors whereas the reverse was true for pAKT, but none of these reached statistical significance even when the genotypes were analyzed together. *P < 0.001.

Immunohistochemical characterization of tumors in the livers of ERBB3 wild-type and knockout.

The tumors from 15 wild-type and 15 hepatocyte-specific ErbB3 knockout were evaluated histologically and also by immunohistochemistry. The hematoxylin and eosin stain shown in Fig. 9A revealed the tumors to be basophilic and have a similar appearance to normal liver. These tumors had an increased number of Ki67-positive cells compared with adjacent nontumorous liver (Fig. 9B, top). The tumors were also positive for TACE, confirming prior reports (4, 10). (Fig. 9B, middle). As has been reported in non-small cell-lung tumors (29), the tumors showed decreased expression of ERRFI1, consistent with the immunoblot results (Fig. 9B, bottom). In normal liver, ERRFI1 was strongly expressed the centrilobular region as shown in the rightmost bottom panel of Fig. 9B. As in normal bronchial epithelia, the staining was both cytoplasmic and nuclear (29).

DISCUSSION

In this paper, we evaluated the expression and activation of the ERBB family of receptor tyrosine kinases in murine HCC. We previously showed that the normal hepatocyte expresses EGFR, ERBB2, and ERBB3, but not ERBB4 (7). ERBB2 is expressed mainly during neonatal liver growth, becoming markedly diminished in the liver of adult mice, but it does reappear in isolated hepatocytes after 48 h in primary culture (38) and has been detected in HCC (1, 2). We first examined the role of EGFR in HCC by studying spontaneous hepatocarcinogenesis in heterozygote Egfr^Dsk5 mice. These mice express an EGFR with a Leu863Gln change within a region of the kinase domain that stabilizes the receptor activation loop and produces a gain-of-function allele that increases basal EGFR kinase activity (14). The male but not the female mice have enlarged livers as well as increased hepatic and circulating cholesterol levels (39). We saw no evidence of HCC through 8 mo of age; however, by 10 mo, all Egfr^Dsk5 male mice had one or two hepatomas, indicative of the oncogenic potential of sustained EGFR activation in this cell (Fig. 1). Others have shown that transgenic mice that overexpress a secreted EGF fusion protein in the liver also develop hepatomas by 8 mo (5, 16). Indeed, an EGF gene polymorphism that results in increased EGF secretion in humans increases the risk of HCC in patients with cirrhosis (44). Amphiregulin, another ligand of EGFR, has also been implicated in HCC (3). Yet in human HCC somatic mutations have been described in the kinase domain for ERBB2 (2), but not as yet for EGFR (25).

The development of HCC in Egfr^Dsk5 mice is relevant to the recently demonstrated association of lipogenesis with human HCC (6). We previously analyzed the effect of this mutation on hepatic lipogenesis and made the unexpected finding that male but not female Egfr^Dsk5 mice have increased plasma and hepatic cholesterol levels. In Egfr^Dsk5 mice fed standard chow diets, the mutation caused hepatomegaly and doubled the plasma and liver cholesterol levels of adult male but not newborn or adult female mice (39). The Egfr^Dsk5 adult mouse liver expressed markedly reduced protein levels of low-density lipoprotein receptor, and a marked increase in fatty acid synthase and 3-hydroxy-3-methyl-glutaryl-CoA reductase. Increased expression of transcription factors associated with elevated cholesterol synthesis, such as STEBP1/2 and Early Growth Response 1, was also observed.

Given the strong evidence linking increased lipogenesis with HCC, the Egfr^Dsk5 mutant mice may be susceptible to HCC, partly because of the metabolic consequences of amplified lipogenesis. Work in transgenic mice has shown that sustained, increased phosphorylation of AKT but not ERK1/2 is sufficient to cause HCC and that this seems to be caused partly by AKT activation of CRTC1 (CREB-regulated transcription coactivator 1), which in turn leads to increased RPS6 (ribosomal protein S6) and eventually to lipogenesis and glycolysis (6). The tumors of the Egfr^Dsk5 did show extremely high levels of p-AKT (Ser 473), ppERK1,2 (Thr 202, Tyr 204), and tyrosine phosphorylated ERB proteins, notably p-ERBB3 (Tyr 1197) (Figs. 2 and 3). The increased p-AKT may be related to the increased phosphorylation of ERBB3, which has six YXXM motifs that bind the p85 subunit phosphatidylinositol-3-kinase (PI3K), leading to the activation of this enzyme. Because Egfr^Dsk5 mice exhibit increased pAKT, RPS6, and lipogenesis, the metabolic consequences of the Egfr^Dsk5 mutation may create a favorable environment for the survival of hepatocytes undergoing spontaneous mutagenesis, setting the stage for HCC. A recent review argued that oncogenes can sometimes initiate cancer not only through direct effects on tumor initiation or progression but also through secondary effects on cell survival, inflammatory cytokine production, or even metabolism (12). Since female Egfr^Dsk5 mutant mice do not show increased lipogenesis, it will be of interest to evaluate HCC in them, particularly because their livers, as determined by immunoblotting and radioligand binding analyses, have one-seventh as many EGFR as male mice and because female mice are less susceptible to HCC in general (40).

Because Egfr^Dsk5 mutant mice developed spontaneous HCC, we examined whether hepatocellular loss of EGFR or ERBB3 altered the progression of chemically induced HCC. Fourteen-day-old mice were injected with DEN, a mutagen, prior to ablation of EGFR or ERBB3. This enabled us to focus on the roles of these proteins in tumor expansion as opposed to initiation. We hypothesized that the loss of EGFR or ERBB3 would decrease tumorigenesis; however, only ERBB3 ablation had such an effect (Fig. 5).

EGFR, ERBB2, and ERBB3 have been implicated in HCC. The expression of these receptor tyrosine kinases is often upregulated in HCC. Somatic mutations have been described in the kinase domain of ERBB2 (2), and ERBB3 is consistently upregulated in hepatomas (17). It is one of 218 genes out of 9,000 examined that are upregulated in human HCC (33). Secreted ERBB3 isoforms have even been described as serum markers for early hepatomas in patients with chronic hepatitis and cirrhosis (18). An ERBB3 binding protein (EBP1) that negatively regulates ERBB3 signaling through diminished phosphorylation of AKT is decreased in HCC patients and is inversely related to HCC grade, size, and cell proliferation rate. Some have argued that it is a tumor suppressor gene for HCC (19).

We found that genetic disruption of EGFR did not decrease tumorigenesis in the DEN model. This work seems counter to the observation that the Egfr^Dsk5 mice developed spontaneous HCC. However, the phenotypical consequences for a gain-of-function model for a particular gene as opposed to the loss-of-function model will not always lead to opposite phenotypes. If the spontaneous tumorigenesis observed in Egfr^Dsk5 mice is secondary to the increased lipogenesis, then it is notable that we have not seen decreased lipogenesis in our HS-EGFR-KO mice on standard chow diets (data not shown).

Our results appear to conflict with recent work by another group that genetically ablated EGFR in parenchymal cells of mice before injecting them with DEN (23). This group reported that chronic DEN injection in EGFR-KO mice increased tumor formation and cell proliferation and decreased apoptosis. Although we did see an increase in the average number of tumors in our mice, it was not statistically significant, and the average liver weights of the mice in the HS-EGFR-KO and wild-type groups were identical. The experimental strategies of the two studies differed in several key ways. We injected DEN once in preweaned Alb-cre transgenic mice before EGFR or ERBB3 ablation occurred. In the other study, DEN was injected into postweaned Alfp-cre transgenic mice. This Alfp-cre recombinase is under the control of not only the liver-specific albumin promoter but also albumin and α-feto-protein enhancers, resulting in complete ablation of EGFR at the end of weaning. Thus our study focuses on how loss of hepatocellular EGF affects the growth phase of HCC after tumor initiation whereas the other study includes the initiation phase of DEN tumorigenesis as well. In addition, this group also employed a two-step model of HCC in which the initial DEN injection is followed by phenobarbital injections. EGFR has recently been shown to be a receptor for phenobarbital (30, 32). This might add an additional level of complexity to this study because phenobarbital has been shown to inhibit EGF-stimulated hepatocyte proliferation (31). If cells lack EGFR, then this inhibition would cease and cells would proliferate at a higher rate than EGFR expressing cells, perhaps accounting the appearance of elevated cell proliferation in EGFR-null HCC. Finally, our cre-recombinase model disrupts EGFR expression only in hepatocytes whereas the cre-recombinase model used in the other study targets bile duct cells as well as hepatocytes. It seems unlikely that EGFR loss in bile duct cells affect hepatocytes during the initiation or progression phases of DEN-mediated HCC, but communication between distinct cell types does occur.

Small molecule kinase inhibitors of the EGFR kinase have been developed that allow one to create a pharmacological knockout. This knockout is transient, requiring multiple injections, and indiscriminately targets EGFR-expressing cells in all organs. Although these inhibitors do not genetically ablate EGFR expression, they are effective in suppressing its kinase activity. They have been used in the clinic to treat EGFR-bearing tumors, but cancerous cells frequently recruit other tyrosine kinases to overcome the growth inhibition caused by EGFR suppression. When injected in rats treated chronically with DEN, pharmacological inhibitors of EGFR impaired the formation of HCC (15). Although this seems inconsistent with our findings, here again, the discrepancy is likely related to experimental differences, including coadministration of the DEN with the EGFR kinase inhibitor. Similar to the genetic EGFR knockout model described above, pharmacological EGFR inhibition interferes with the initiation phase of DEN-induced HCC. These authors did analyze cell proliferation in EGFR-KO tumors that formed in spite of EGFR inhibition. Consistent with our results, they found no effect of loss of EGFR signaling on cell proliferation in HCC. In aggregate, this suggests that EGFR plays a role in HCC initiation but not progression.

In contrast to EGFR, we did find that ERBB3 gene disruption retarded HCC. Other groups have shown that ERBB3 is important in mouse mammary tumorigenesis as well as in colorectal cancer in the Apc^Min mouse model. In the mammary tumorigenesis mode, ERBB3 ablation reduced PI3K-AKT-driven tumor formation driven by the polyomavirus middle T oncogene (8). ERBB3 appeared to be responsible for AKT phosphorylation, tumor growth, and lung metastases, and these results were ascribed to decreased PI3K signaling normally arising from ERBB2-ERBB2 heterodimerization. Genetic disruption of ERBB3 has also been shown to result in almost complete inhibition of intestinal tumor formation in the Apc^Min model of colon cancer (24). In this case, intestinal tumors also exhibited decreased PI3K/AKT signaling; however, the reduction in tumorigenesis was attributed not to decreased rate of cell proliferation but to enhanced apoptosis of nascent tumor cells.

We did not observe the expected decreases in AKT phosphorylation or increases in apoptosis in the liver or tumors of HS-ERBB3-KO mice. CASP3 (caspase 3) or TUNEL-positive cells in these tumors were rare (data not shown), and we cannot rule out the possibility that increased apoptosis may have occurred earlier during the long incubation period prior to overt tumorigenesis. Multiple tumors developed in the ERBB3-null breast and intestinal mouse models within 2–3 mo as opposed to 8–9 mo in the DEN-induced HCC model, and these more rapid models of tumorigenesis may be more conducive to identifying and analyzing signaling abnormalities between the knockout mice and their wild-type counterparts (7).

Nevertheless, we did observe that tumors in the HS-ERBB3-KO livers exhibited decreased STAT3 phosphorylation compared with wild-type livers (Figs. 7 and 8). STAT3 appears to be a key signaling molecule in human HCC (22) as well as liver regeneration (28), and it will be important to determine how hepatocellular ERBB3 diminishes maximal phosphorylation of STAT3. ERBB3 activation of STAT3 in a progesterone receptor-dependent pathway has been studied in breast cancer (34), but not in the liver. Since ERBB3 can signal in association with EGFR, ERBB2, FGFR, MET, and the PGR (progesterone receptor), it will be a challenge to identify a cellular system to analyze the specific mechanism(s) by which ERBB3 alone or in concert with these other molecules maximizes STAT3 activation. Along this line, we did see ERBB2 expression in the tumors (Fig. 7), and this is consistent with prior work showing that EGF-like ligands and HRG (heregulin) can induce ERBB2 in rat hepatocytes during primary culture (38). HRG, the ERBB3 ligand, has been shown to phosphorylate tyrosine 215 of SRC (47), which itself can activate STAT3. ERBB3 has also been shown to activate SRC in several cell systems (35, 47). Whether a SRC-mediated STAT3 phosphorylation is operative in hepatocytes and HCC requires further investigation.

There are several limitations of this study. First, preweaned mice differ from adult mice in their patterns of gene expression. For example, we have shown that ERBB2 is more highly expressed in neonatal than adult mice (7). Injection of DEN into 14-day-old mice vs. adult mice may cause different mutagenic profiles, leading to tumors that may be distinct. EGFR ablation may lead to different outcomes in other models of hepatocarcinogenesis. Secondly, it is difficult to compare the gain-of-function model with the loss-of-function models because we do not have a thorough understanding of the transformative genetic changes that lead to tumorigenesis. For example, in different tumors, we observed heterogeneous expression or activation of some genes expected to affect tumor growth, such as ERRFI1 and ERK1,2. Striking interstrain differences in the importance of EGFR in HCC may also exist (46). One study that would be relevant would be to determine whether ERBB3 ablation in Egfr^Dsk5 mutant mice would eradicate or delay tumorigenesis. Another line of future investigation would be to determine whether DEN-initiated tumorigenesis itself is accelerated or enhanced in the livers of the Egfr^Dsk5 mice.

Figure 10 summarizes the main findings of this paper. We showed that a gain-of-function mutation in EGFR can result in the spontaneous production of solitary tumors as early of 10 mo of life in male mice. By 14 mo, every +/Dsk5 mouse had two to four tumors. In contrast, the injection of DEN during the late neonatal phase prior to weaning generates a large number of tumors in male mice by 8 mo. We hypothesized that the loss of EGFR or ERBB3 in hepatocyte-specific knockout mice would show a diminished tumor number or cell proliferation rate based on our work with the gain-of-function Egfr^Dsk5 mice. This proved to be true for ERBB3 but not EGFR. Signaling in the different models showed several differences. The tumors of the Egfr^Dsk5 mice showed consistent elevations in the activated forms of AKT and ERK1,2 whereas the DEN-initiated tumors showed consistent increased pY-STAT3. Genetic ablation of ERBB3 led to the development of tumors with less pY-STAT3. Since STAT3 has been shown to be important in human carcinogenesis, the decrease in pY-STAT3 may be particularly relevant to the differences in hepatic tumorigenesis between wild-type and HS-specific ERBB3 knockout mice.

Fig. 10.

Summary figure of the +/Dsk5 and DEN models of carcinogenesis. This figure shows the major findings in this paper. A: we earlier reported that the livers of Dsk5 mice display hepatomegaly by 6 wk of age, including increased activation of Akt and ERK1,2. Hepatomegaly is caused primarily by increased cholesterol accumulation associated with increased 3-hydroxy-3-methyl-glutaryl-CoA reductase (HMG-CoA reductase) and decreased low-density lipoprotein receptor (39). We now report that all mice develop spontaneous carcinogenesis by 13 mo of age. B: we injected mice at day 14 of age with DEN, a well-studied hepatocarcinogen. This allowed us to evaluate tumor progression because ERBB3 gene ablation in our cre-alb recombinase has not occurred at the time tumors are initiated. We found that the loss of ERBB3 but not EGFR appreciably reduced the number of tumors 8 mo later.

GRANTS

This work was supported by R01DK53804 and R21CA149708 (to W. E. Russell) and R01CA092479 (to D. W. Threadgill).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

L.A.S., D.W.T., and W.E.R. conception and design of research; L.A.S., X.Z., M.C.S., M.A.W., A.A., and A.M.C. performed experiments; L.A.S., X.Z., M.C.S., M.A.W., A.A., and A.M.C. analyzed data; L.A.S., X.Z., M.C.S., M.A.W., A.A., and A.M.C. interpreted results of experiments; L.A.S. and M.C.S. prepared figures; L.A.S. drafted manuscript; L.A.S., D.W.T., and W.E.R. edited and revised manuscript; L.A.S. and W.E.R. approved final version of manuscript.