Multiple mechanisms are responsible for altered expression of gap junction genes during oncogenesis in rat liver

Gap junctions (GJs) represent aggregates of transmembrane channels that provide cells with a mechanism to regulate the sharing of second messengers and other hydrophilic molecules less than about 1200 Da in mass. Recent studies have identi-fied a family of GJ proteins, termed connexins, that form inter-cellular pores exhibiting different unitary conductance values and gating characteristics. Although the precise functional roles of the different connexins are unknown, normal biologi-cal processes including development, cell proliferation and modifications in organ physiology can be modulated by GJ-mediated intercellular communication (GJIC) (Loewenstein, 1981; Larsen, 1989; Beyer et al., 1990; Musil and Goode-nough, 1990; Bennett et al., 1991).

Since both positive and negative growth regulatory molecules may pass through GJs, a role of alterations in GJIC has been postulated for neoplastic development (Loewenstein, 1979). In accord with this hypothesis, many tumor promoters, oncogenic proteins and growth factors have been demonstrated to modulate GJIC (Loewenstein, 1990; Trosko et al., 1990; Musil and Goodenough, 1990; Yamasaki, 1991). Furthermore, subtractive hybridization of normal and tumor mRNAs has identified connexins as potential tumor suppressors (Lee et al., 1991). Tumorigenic cells transfected with either connexin32 (C×32) or connexin43 (C×43) exhibit growth retardation in vitro and in nude mice (Eghbali et al., 1991; Mehta et al., 1991; Naus et al., 1992).

Models of multistage hepatocarcinogenesis in the rat represent a useful system in which to study the biochemical and molecular mechanisms of neoplasia. Discrete populations of cells can be identified to study the stages of initiation, promotion and progression (Saeter and Seglen, 1990; Pitot et al., 1991). Approximately 90% of the total rat liver GJ protein has a predicted mass of 32 kDa (connexin32; C×32), while a minor form has been termed connexin 26 (C×26) (Paul, 1986; Zhang and Nicholson, 1989). Whereas C×32 is expressed by hepatocytes throughout the rat acinus, C×26 is restricted to periportal hepatocytes (Traub et al., 1989; Berthoud et al., 1992a). Connexin 43 (C×43), first identified in cardiac myocytes, is expressed by several liver-derived cell lines (Asamoto et al., 1991; Spray et al., 1991) and in Glisson’s capsule and perisinusoidal cells in fetal rat liver (Berthoud et al., 1992a). Furthermore, immortalized embryonic mouse hepatocytes express C×32 and C×26, but shift to C×43 when serum is added to the medium (Stutenkemper et al., 1992).

During rat hepatocarcinogenesis C×32 immunoreactivity and GJIC are reversibly downregulated in preneoplastic foci (Neveu et al., 1990). Unlike other markers of hepatic foci, these alterations were observed with several chemical carcino-genesis regimens (Neveu et al., 1990; Klaunig et al., 1990; Klaunig, 1991; Krutovskikh et al., 1991). In addition to alter-ations in C×32, some hyperplastic nodules display increased C×26 immunostaining (Sakamoto et al., 1992). While all hepatic neoplasms examined to date (from rats, mice and humans) display decreased abundance of morphologically identifiable GJs (Weinstein et al., 1976; Swift et al., 1983), the molecular mechanisms regulating these changes remain unclear. Janssen-Timmen and colleagues (1986) showed that rat hepatic neoplasms display reduced expression of C×32 by immunofluorescence and western blotting. However, subse-quent studies by our laboratory and others showed near normal levels of C×32 mRNA in hepatic neoplasms from rats, mice and humans (Beer et al., 1988; Fitzgerald et al., 1989; Beer and Neveu, 1990; Oyamada et al., 1990; Sakamoto et al., 1992).

To better understand the regulation of connexins during rat hepatocarcinogenesis, we examined C×32, C×26 and C×43 expression in preneoplastic and neoplastic lesions using cDNA, cRNA and antibody probes. The questions addressed were: (1) do modifications in C×32 mRNA and protein expression commonly occur in rat hepatic neoplasms? (2) are changes in C×32 expression reversible in promoter-indepen-dent lesions? and (3) do modifications in C×26 and C×43 mRNA or protein expression occur in preneoplastic or neo-plastic rat hepatocytes?

Inbred Fisher F344 and outbred Sprague-Dawley rats (Harlan Sprague Dawley Co., Madison, WI) were maintained at 27°C on a 14 hour light/10 hour dark cycle and fed defined AIN-76 or crude NIH-07 diet (Teklad Test Diets, Madison, WI) and water ad libitum. Preneoplas-tic hepatic foci were examined in newborn female F344 rats initiated with a single non-necrogenic dose of diethylnitrosamine (DEN) (10 mg/kg, p.o.) and promoted with either 0.05% phenobarbital (PB) for 8 months or 0.1 μg/kg 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) for 14 months (Dragan et al., 1991). Promoter-independent lesions (Pitot et al., 1991), defined as those that continue to proliferate following withdrawal of PB promotion, as well as promoter-dependent foci, were induced in neonatal female F344 rats initiated with DEN (10 mg/kg, p.o.) followed by 8 months of 0.1% PB. At this time, osmotic minipumps (model 2001 Alzet, Alza Corp., Palo Alto, CA) containing 600 μCi of [3H]thymidine (Amersham, Arlington Heights, IL; specific activity, 84.0 Ci/mmol) were implanted i.p. (Neveu et al., 1990). Eight days following the implant, the animals were withdrawn from the PB-containing diet and placed on purified AIN-76 diet for 20 days before being killed. Hepatic neoplasms were generated by a single non-necrogenic dose of DEN (10-30 mg/kg, p.o.) followed by chronic promotion with either 0.05% PB or 0.1 μg/kg per day TCDD for 14 months (Dragan et al., 1991; Beer et al., 1988). Following their respective regimens, rats were killed, and rep-resentative sections of the liver were fixed for histopathology by rapid freezing on solid CO2 for staining studies or in liquid nitrogen for mRNA and protein analyses.

The neoplastic nature of 22 lesions was evaluated by histopathologi-cal examination of H&E-stained sections as previously described (Beer et al., 1988). Tumors were dissected to remove surrounding liver tissue prior to isolation of protein and RNA samples. The histology of DEN/PB-treated neoplasms is listed in Fig. 4, below. Several hepatomas were also examined from rats treated with DEN/TCDD: TCDD1-hepatocellular carcinoma with mixed cell types with focal areas of lymphoid infiltration; TCDD2-hepatocellular carcinoma; TCDD3-hepatocellular carcinoma with fibrosis; TCDD4-hepatocellular carcinoma with bile duct proliferation, fatty metamor-phosis and inflammation; TCDD5-neoplastic nodule with focal fibrosis, bile duct proliferation and inflammation; TCDD6-cholangio-cellular carcinoma.

Because of the similarity of the amino acid sequences between the members of the connexin family (reviewed by Bennett et al., 1991), as well as the possibility of epitope masking, several antibodies generated against different cytoplasmic epitopes were employed to verify immunoreactivity (Table 1). A rabbit polyclonal antibody generated against the placental form of rat glutathione S-transferase (GST) identified altered hepatic foci (AHF; Hendrich et al., 1987). Immunocytochemistry and immunoblotting procedures were optimized so that no reactivity was observed when preimmune serum or nonspecific ascites fluid was substituted for the antibody prepara-tions.

Crude homogenates were prepared by disrupting tissue for 30 seconds with a Brinkmann homogenizer (Brinkmann, Westbury, NY); 50 mg (wet weight) of tissue samples was disrupted in 1 ml of isolation buffer A containing 4 mM NaHCO3, 2 mM phenylmethylsulfonyl fluoride, 2 μg/ml aprotinin, 2 mM EDTA, 5 mM diisopropylfluo-rophosphate, 100 mM sodium fluoride, 10 mM sodium pyrophos-phate, and 2 mM sodium orthovanadate. In addition, 50 mg of tissue samples was homogenized in 20 mM NaOH followed by 30 seconds of sonication at 50% power with a Branson sonifier 250 microtip (VWR, Chicago, IL). By centrifugation at 12,000 g for 15 minutes, NaOH-insoluble protein pellets were recovered, resuspended in 2 ml 20 mM NaOH, sonicated for 30 seconds at 50% power, repelleted at 12,000 g for 15 minutes, and then suspended in 1 ml of isolation buffer A (Hertzberg and Skibbens, 1984).

Samples were sonicated for 30 seconds at 50% power with a Branson sonifier 250 microtip, and protein concentrations were deter-mined (Peterson, 1977). Proteins were solubilized in 2% SDS (Gallard-Schlesinger, Carle Place, NY), 62.5 mM Tris-HCl, 10% glycerol, and 50 mM dithiothreitol (Sigma), pH 6.8, for 30 minutes at room temperature (RT) and resolved by SDS-PAGE (Laemmli, 1970) with 3% to 5% stacking and 12.5% separating gels cast in a minigel apparatus (BioRad, Richmond, CA). Transfer of the protein to positively charged Immobilon-P membranes (Millipore, Bedford, MA) was carried out in modified Towbin transfer buffer (10% methanol) at 300 mA for 90 minutes at 4°C. The SDS-polyacrylamide gels were then stained with Coomassie Blue R-250 (Bio-Rad) to evaluate equal loading and transfer of proteins.

Nonspecific protein binding of the membranes was blocked with filtered Blotto (5% Carnation non-fat dry milk powder in 40 mM Tris-HCl, pH 7.4, 0.1% Tween-20, 0.05% sodium azide) for 1 hour at RT (Johnson et al., 1984). Primary antibodies were incubated with the blots for 2 hours at RT, washed several times in TBS (50 mM Tris-HCl, pH 7.4, 0.9% NaCl, 0.05% sodium azide), followed by incuba-tion with the appropriate affinity-purified rabbit secondary antibody (Chappel, Malvern, PA) for 1 hour at RT. Antibody-connexin complexes were identified with 40 nCi/ml 125I-Protein A (ICN, specific activity > 70 μCi/μg). Rainbow 14C-methylated protein markers were used for molecular mass determinations (Amersham). Autoradiograms with XAR-5 film (Kodak, Rochester, NY) were exposed at –70°C with an intensifying screen.

Immunocytochemical staining of cryosections (6-8 μm) was performed as previously described (Neveu et al., 1990). Briefly, acetone-fixed sections were blocked for endogenous biotin and per-oxidase activity, and incubated with primary antibody overnight at 4 °C. Protein-antibody complexes were visualized by the biotin/strep-tavidin/peroxidase method with aminoethyl carbazole as the chromogen (Zymed Laboratories Inc., San Francisco, CA). Double staining of the placental form of glutathione S-transferase (GST) and connexins was performed with biotin-conjugated goat anti-rabbit antibody, followed by avidin conjugated to α-galactosidase and X-gal as the chromogen (Neveu et al., 1990). For quantitative stereological analysis, sections were also stained for gamma-glutamyltranspepti-dase (GGT) activity with enzyme histochemistry (Hendrich et al., 1987). Nuclei were counterstained with Mayer’s hematoxylin, and coverslips were applied with Crystal/Mount (Biomedia Corp., Foster City, CA). All slides were viewed with a Zeiss Axiophot microscope (Carl Zeiss, Germany) with either light-or dark-field microscopy.

Double immunofluorescence of C×32 and C×26 was performed with acetone-fixed cryosections. Sections were blocked for endoge-nous biotin (Neveu et al., 1990), incubated with an affinity-purified rabbit polyclonal antibody to C×26 (α19, see Table 1), followed by incubation with goat anti-rabbit IgG conjugated with biotin (Sigma) for 1 hour at RT. A mouse monoclonal antibody to C×32 (M12.13) was then applied for 2 hours at RT, incubated with Texas Red-con-jugated to streptavidin (Vector), and then with goat anti-mouse IgG conjugated with FITC (Sigma). The sections were washed three times in phosphate-buffered saline between steps. To identify AHF in these sections, we incubated the slides with a third primary antibody to GST for 1 hour at RT, with goat anti-rabbit IgG conjugated to biotin for 1 hour at RT, and then with 7-amino-4-methyl-coumarin-3-acetic acid conjugated to streptavidin for 30 minutes at RT. Coverslips were placed on the sections with Fluoromount-G (Fisher, Chicago, IL) sup-plemented with 2.5% 1,4-diazabicyclo[2.2.2]octane to reduce quenching. Yellow (C×32), red (C×26) and blue (GST) fluorescences were visualized with epifluorescence illumination.

Total RNA was isolated from pulverized, quick-frozen tissues by the acid-guanidinium method (Chomczynski and Sacchi, 1987), and northern blots were performed with 10 μg total RNA per lane (Beer et al., 1988). Ethidium bromide at a concentration of 0.2 μg/ml was added to the gels before electrophoresis to verify the integrity of the RNA as well as to monitor the equivalence of loading before and after transfer to GeneScreen Plus (Dupont, Boston, MA). EcoRI fragments of the cDNAs corresponding to nearly full-length coding mRNA for C×32 (1.5 kb) (Paul, 1986), C×26 (2.1 kb) (Zhang and Nicholson, 1989), and C×43 (3.1 kb) (Beyer et al., 1989) were isolated from plasmid sequences by agarose electrophoresis and purified with GeneClean (Bio101 Inc., La Jolla, CA). Radiolabeled probes ([32P]dCTP, 3000 Ci/mmol, Amersham, Arlington Heights, IL) were generated with the multiprime DNA labeling system (Amersham) followed by G-50 Quickspin purification (Boehringer Mannheim, Indianapolis, IN). Prehybridization and hybridization were performed as previously described (Beer et al., 1988). Blots were washed in 1× SSC with 1% SDS at 65°C for 60 minutes, 0.1× SSC with 1% SDS at 65°C for 60 minutes, and 0.1× SSC for 30 minutes at RT followed by exposure to XAR-5 X-ray film (Kodak) at –70°C with an intensi-fying screen (Dupont).

Sense and anti-sense radiolabeled ([35S]thiophosphate-UTP, 1282 Ci/mmol, Amersham) riboprobes were transcribed in vitro from the C×32 cDNA as previously described (Beer et al., 1988). The probes were hydrolyzed at pH 10.2 to lengths of about 200 nucleotides. The specificity of the probes was examined by northern blotting of normal liver and heart total RNA. In situ hybridization was performed as pre-viously reported, with slight modifications (Beer et al., 1988; Nakat-sukasa et al., 1990). Briefly, cryosections were thaw-mounted onto 3-aminopropyltriethoxysilane-coated slides (Rentrop et al., 1986) and fixed in fresh 4% paraformaldehyde in PBS for 20 minutes at RT. Subsequently, the sections were incubated in PBS supplemented with 5 mM MgCl2 for 10 minutes at RT, and then in 0.1 M glycine/0.2 M Tris-HCl, pH 7.4, for 10 minutes at RT. After being washed in phosphate-buffered saline, the sections were post-fixed in 4% paraformaldehyde for 20 minutes at RT. Washed sections were then dehydrated in 50, 75, 90 and 100% ethanol, air dried, and prehy-bridized at 45°C for 2 hours in 50% deionized formamide (Clonetech, Palo Alto, CA) containing 0.3 M NaCl, 100 mM dithiothreitol, 10 mM Tris-HCl, pH 7.5, 20 mM vanadyl ribonucleosides (Bethesda Research Labs, Bethesda, MD), 1× Denhardt’s solution, 0.05% yeast total RNA, 0.005% yeast tRNA, 0.05% sonicated salmon sperm DNA, 1 mM EDTA, and 10 mM DL-methionine. The sections were washed briefly in 2× SSC, air dried, and hybridized in the prehy-bridization buffer supplemented with 20% dextran sulfate (Mr 100,000; Sigma) and 1.5×108 cpm/ml of heat-denatured 35S-labeled cRNA probe for 5 hours at 48°C. Washing of the sections, RNase treatment, dehydration, dipping in photoemulsion and development were performed as previously described (Beer et al., 1988; Nakat-sukasa et al., 1990). After development, sections were coverslipped with Crystal/Mount (Biomedia Corp.) and viewed by both light-and dark-field microscopy.

As previously described (Traub et al., 1989), the antibodies used in this study (Table 1) show that C×32 is homogeneously distributed as discrete spots on hepatic membranes throughout the acinus in untreated rat liver (Fig. 1A), whereas punctate C×26 staining is restricted to periportal hepatocytes (Fig. 1B). Modifications in C×32 and C×26 immunostaining were observed in foci generated by either PB or TCDD promotion. In both protocols, many foci that were positive for the placental form of glutathione S-transferase (GST) displayed diminished C×32 immunoreactivity (Fig. 1C). A distinct population of GST-positive foci, exhibiting near-normal punctate C×32 staining (Fig. 1D), was identified by increased C×26 staining (Fig. 1E). Immunofluorescence microscopy of sections doubly stained for C×32 (FITC) and C×26 (Texas Red) showed that both proteins co-localize in C×26-positive foci (Fig. 1F,G). Although C×43 immunoreactivity was never observed in foci, staining was apparent between cells of the Glisson’s capsule (Fig. 1H) and areas of biliary hyperplasia (Fig. 1I). Bile ducts present in portal triads from normal untreated rats did not express Cx43 immunoreactivity (data not shown).

The number of foci and the percentage of the liver occupied by foci exhibiting diminished C×32 or enhanced C×26 staining were examined by quantitative stereology as previously described (Hendrich et al., 1987; Neveu et al., 1990). In addition, we compared these parameters with those obtained using GST and γ-glutamyltranspeptidase (GGT) to mark foci in serial sections. Table 2 compares the number and volume percentage of the liver of foci scored by any one marker, with the cumulative number and volume of foci scored by any of the four markers (i.e. C×32, C×26, GST, GGT). Differential upregulation of C×26 scored 43% of the total volume of foci promoted by TCDD, but only 6% of the total volume of lesions promoted by PB. In both protocols, approximately 75% of the volume of foci was identified by changes in either C×32 or C×26 staining.

Previously, we observed that about 94% of GST-positive foci exhibit a normal distribution of C×32 staining if PB is withdrawn from the diet 20 days before killing the animal (Neveu et al., 1990). Utilizing osmotic minipumps containing [³H]thymidine, we simultaneously examined the relationship

Fig. 1. Immunostaining of C×32 (A,C,D,F,J,K), C×26 (B,E,) and Cx43 (H,I) protein in normal and preneoplastic rat livers. Cryosections were immunolabeled with specific antisera: M12.13 for C×32, α19 for C×26, and 18A for C×43 (see Table 1). Antigen-antibody complexes were visualized with dark-field (A-E,J,K), fluorescence (F,G), or bright-field (H,I) microscopy (see Materials and Methods). (A) C×32 in normal liver. Homogeneous acinar distribution between the portal tract (p) and central vein (v). of C×32 staining and cell proliferation. Promoter-independent lesions were generated in rats initiated with DEN and promoted with PB for 8 months. PB was then withdrawn from the diet for 20 days prior to killing. [³H]thymidine incorporation was detected with photoemulsion and dark-field microscopy as pre-viously described (Neveu et al., 1990). Fig. 1J shows normal C×32 immunostaining observed in most GST-positive foci after cessation of PB promotion. Although no significant dif-ference in the lobular distribution of hepatocyte [³H]thymidine incorporation was observed after removal of PB, two popula-tions of GST-positive foci were identified by differences in [³H]thymidine incorporation (Table 3). Only promoter-inde-pendent foci, identified by their ability to proliferate in the absence of PB promotion, continued to exhibit decreased C×32 staining (Fig. 1K). These findings indicate that irreversible downregulation of C×32 staining correlates with foci that are promoter-independent for growth.

Fig. 2A demonstrates that Cx32 transcripts are homogeneously expressed in hepatocytes throughout the liver lobule with no detectable cRNA-mRNA hybrids in nonparenchymal cells.

The specificity of the hybridization was apparent, since only background hybridization was observed when a sense probe was employed (Fig. 2B) or when sections were treated with RNase prior to hybridization. In contrast to decreased C×32 immunoreactivity in foci, mRNA in situ hybridization of liver sections from rats initiated with DEN and promoted with PB showed that, of the 50 C×32-deficient AHF (Fig. 2C), only four displayed reduced C×32 message throughout the lesion (Fig. 2D). The remaining foci displayed levels of cRNA-mRNA hybrids comparable to surrounding non-neoplastic hepatocytes (Fig. 2E). A fraction of cells within these foci usually exhibited reduced C×32 staining. Fig. 2F shows that C×32 is absent in Glisson’s capsule; Fig. 2G demonstrates Cx43 transcripts in these cells. These observations indicate that punctate C×32 staining is downregulated in most foci in a manner indepen-dent of mRNA abundance.

All 24 hepatic neoplasms, generated by DEN initiation and promotion with either PB or TCDD and examined for C×32, C×26 and C×43 immunoreactivity, exhibited alterations in C×32 and/or C×26 staining. Although 15 of the tumors did not exhibit detectable C×32 staining (Fig. 3A), most neoplastic hepatocytes within the other nine lesions exhibited either diffuse immunoreactivity (PB11; PB12; PB15) (Fig. 3B) or random punctate staining (PB5; PB9; PB10; PB13; TCDD2; TCDD4) (Fig. 3C). The disorganized C×32 staining did not appear to be localized as discrete punctate areas on the plasma membrane between juxtaposed cells, but was randomly dis-tributed (Fig. 3D) or localized in sinusoidal regions of the cell that did not contact neighboring tumor cells (Fig. 3E). Most neoplasms with altered C×32 staining also displayed altered C×26 staining. In contrast to the co-localization of C×32 and C×26 observed in foci, double immunofluorescence staining demonstrated that C×32 (FITC) and C×26 (Texas Red) immunoreactivities do not always co-localize in tumors (Fig. 3F, G).

All areas of bile duct proliferation, cholangiomas and cholangiocarcinomas were completely unreactive towards several antibodies to C×32 and C×26. Although C×43 staining was present in 18 of 24 neoplasms, it was always restricted to non-parenchymal cell types. Both punctate and intracellular C×43 staining was observed in areas of fibrosis (Fig. 3H) and bile duct proliferation (Fig. 3I). Whereas benign proliferations of bile ducts and cholangiomas always displayed Cx43 staining, all five cholangiocarcinomas were deficient in C×43 immunoreactivity (Fig. 3J).

Northern blotting determined that some of the observed alter-ations in C×32, C×26 and C×43 immunoreactivities (see above) could be attributed to changes in steady-state mRNA abundance (Fig. 4). Ethidium bromide staining of the gel demonstrated that near-equivalent amounts of total RNA were loaded in each lane. Detectable qualitative alterations in the mobility of connexin mRNAs were not apparent in neoplasms relative to normal rat liver controls for C×32 (1.6 kb) and C×26 (2.1 kb) (Paul, 1986; Zhang and Nicholson, 1989), and relative to rat heart for C×43 (3.1 kb) (Beyer et al., 1989). C×32 mRNA abundance was greater than or equivalent to that of untreated adult liver in 11 of 16 neoplasms; the remaining samples displayed reduced or undetectable levels. Fig. 4 shows that the abundances of C×26 and C×43 mRNAs were low in normal adult rat liver, but abundant steady-state transcripts are apparent in many neoplasms (7/16-C×26; 8/16-C×43). Whereas increased C×43 mRNA expression was identified in the neoplasms containing bile duct proliferation or fibrosis, specific alterations in C×26 and C×32 mRNA expression were not correlated with routine histopathology. Similar variations in the patterns of C×32, C×26 and C×43 mRNA expression were seen in neoplasms from rats initiated with DEN and promoted with TCDD, mestranol or chlorendic acid (data not shown).

A representative immunoblot showing the spectrum of observed alterations in the abundance and/or electrophoretic mobility of C×32, C×26 and C×43 in hepatic neoplasms from rats promoted with PB is depicted in Fig. 5. As previously described (Hertzberg and Skibbens, 1984), monomer and dimer forms of C×32 (27 and 47 kDa) were present in normal liver tissue homogenates. Although C×32 immunoreactivity was observed in several neoplasms, the electrophoretic mobility of C×32 was altered compared with that of normal liver. Tumors PB13, PB12 and PB15 displayed only the dimer form of C×32; PB5 exhibited altered mobility at 31, 47 and 88 kDa; PB1 exhibited altered mobility at 29 kDa and PB9 exhibited only monomeric C×32. The abnormal C×32 immunoreactive bands in tumors were found to be localized to post-nuclear crude membranes (data not shown). Although anti-C×26 antibodies failed to detect specific immunoreactiv-ity in crude liver homogenates, a single immunoreactive band of 24 kDa was detected in NaOH-insoluble pellets of total tissue homogenates. Whereas the level of C×26 was reduced in most neoplasms, immunoreactivity was apparent in PB1, PB5 and PB12. Increased C×43 immunoreactivity at 43 kDa was observed in PB5, PB13 and PB15 (Fig. 5), which did not co-migrate with C×43 (46 kDa) from rat heart (data not shown).

Comparison of northern and western blot experiments indicated that alterations in C×32 mRNA in neoplasms were not always paralleled by changes in protein expression. For example, mRNA isolated from tumor PB4 exhibited normal levels of C×32 mRNA, but failed to show immunoreactivity by staining (Fig. 3A) or western blotting (Fig. 5). A similar disparity between C×32 mRNA and protein expression was observed in preneoplastic foci (Fig. 2C,D). Despite use of several different antibodies to C×32 (Table 1), antigenic sites may have been masked in some tumors and AHF. Since C×26 immunoreactivity was apparent in NaOH-insoluble pellets from several neoplasms that did not display staining, antigenic sites of this protein may also have been masked in tissue sections. Furthermore, some neoplasms were also observed to display increased C×26 mRNA, but C×26 protein was unde-tectable in NaOH-insoluble pellets. In either case, the masking of epitopes or the loss of protein expression demonstrates alter-ations in C×26 protein expression.

Our results confirm and extend previous observations that C×32 staining is diminished in preneoplastic rat hepatocytes (Beer et al., 1988; Neveu et al., 1990; Krutovskikh et al., 1991). Although decreased C×32 staining is reversible in rat hepatic foci (Neveu et al., 1990), we found that promoter-independent foci and neoplastic rat hepatocytes lack normal punctate dis-tribution of C×32 whether PB is present or absent (see Table 3). In agreement with these observations, mouse foci and adenomas exhibit a similar pattern of C×32 staining after with-drawal of PB promotion (Klaunig, 1991). Krutovskikh and col-leagues (1991) found that most GST-positive foci that cease to proliferate (i.e. remodel) following the Solt-Farber model of hepatocarcinogenesis display normal C×32 expression and transfer the GJ-permeable dye Lucifer Yellow. Since the expression of several markers of rat hepatocarcinogenesis have been shown to be promoter dependent (Pitot et al., 1991), irre-versible downregulation of C×32 may represent a necessary change for promoter-independent growth or tumor progres-sion.

Northern blotting (Fig. 4) and mRNA in situ hybridization (Fig. 2) showed that C×32 mRNA is present in most preneo-plastic and neoplastic hepatocytes and is comparable in abundance to untreated adult liver. While these observations concur with previous studies of rat, mouse and human hepatomas (Beer et al., 1988; Beer and Neveu, 1990; Oyamada et al., 1990), decreased C×32 mRNA was seen in rat hepatic neoplasms induced by chronic administration of N-ethyl-N-hydroxyl ethylnitrosourea, or after a necrogenic dose of DEN followed by several cycles of acetylaminofluorene and carbon tetrachloride (Fitzgerald et al., 1989; Sakamoto et al., 1992). Since we found no obvious correlation between the level of C×32 mRNA and histopathology, the reported differences in the abundance of C×32 mRNA may result from the different agents utilized to induce neoplasia. For example, a single administration of hepatotoxic levels of DEN or carbon tetra-chloride transiently resulted in a greater than 90% reduction of C×32 immunoreactivity in rat liver (Miyashita et al., 1992).

Several changes in the pattern of C×32 immunostaining were evident in rat hepatic neoplasms (Fig. 3A-D). The diversity of both punctate and diffuse staining suggests that C×32 can be sequestered in several different subcellular compartments. As described in nonhepatic neoplasms, intracellular staining may represent internalized GJs (Larsen, 1983). Alternatively, immunoreactivity may be sequestered in the smooth endoplas-mic reticulum, lysosomes and other membrane components that are altered in neoplasms from rats initiated with DEN and promoted with PB (Feldman et al., 1981). Although immuno-electron microscopy is needed to define the intracellular local-ization of C×32 in neoplasms, our results indicate that several post-translational mechanisms may alter the transport, assembly and/or turnover of C×32 into GJs during hepatocar-cinogenesis.

Alterations in connexin staining in some hepatomas may represent a secondary change because transfection of commu-nication-deficient cell lines with various cell adhesion molecules can restore GJIC (Musil et al., 1990; Jongen et al., 1991). Interestingly, studies by Faris and colleagues (1991) show that the expression of the major liver cell adhesion molecule (cell CAM 105) is modified in many hepatic neoplasms.

Immunoblot analysis of crude extracts from neoplasms with modified C×32 staining revealed several alterations in the migration of immunoreactive bands on SDS-PAGE (Fig. 5). The observed changes in mobility may result from alterations in the amino acid sequence of C×32, folding of C×32, or post-translational processing (Jaenicke, 1987; Amara et al., 1992).

Several reports have described shifts in the electrophoretic mobility of C×43 after protein phosphorylation, although these have not been observed for phosphorylated forms of C×32 (Bennett et al., 1991). Additional studies will be necessary to determine what mechanisms post-translationally downregulate C×32 expression during hepatocarcinogenesis.

Immunofluorescence studies of human hepatic neoplasms also found C×32 immunoreactivity, but in a disorganized pattern (Oyamada et al., 1990; Wilgenbus et al., 1992). Since morphological studies demonstrate that GJs are diminished in cholestatic and neoplastic human livers (Robenek et al., 1981; Swift et al., 1983), the abnormal C×32 staining could represent alterations similar to those we identified in rat hepatic neoplasms.

Although morphological studies show that some neoplasms display normal GJs (Larsen, 1983), differential expression of connexins could potentially alter GJIC without changing the morphological appearance or abundance of GJs. Consistent with this hypothesis, a subset of foci (44%) and neoplastic nodules (16%) generated by the Solt-Farber protocol exhibit increased C×26 staining (Sakamoto et al., 1992). In this report, we show enhanced C×26 protein in preneoplastic foci (Fig. 1D). Upregulation of C×26 staining identified different numbers and volumes of foci in livers from rats promoted with either PB or TCDD. This observation suggests that modifica-tions in C×26 staining may be dependent on the nature of the promoting agent (Table 2). Furthermore, increased C×26 staining was observed in foci generated by a choline/methion-ine-deficient diet, but not by the peroxisome proliferator WY-14,643 (data not shown).

As previously described in normal mouse and rat liver (Zhang and Nicholson, 1989; Traub et al., 1989), both C×32 and C×26 immunoreactivity co-localized in periportal hepato-cytes and in C×26-positive foci. In contrast, C×32 and C×26 immunoreactivity did not always co-localize in neoplasms (Fig. 3F, G). These results suggest that C×32 and C×26 have unique requirements for assembly/turnover or, less likely, that one connexin may be mutated, resulting in differential cellular localization. Unlike C×32, diffuse C×26 immunoreactivity was not observed in hepatomas. The disparity between increased C×26 mRNA (Fig. 4) and decreased C×26 immunostaining could be owing to antigen masking or to downregulation inde-pendent of mRNA abundance. Similar results were seen in hepatic neoplasms generated by the Solt-Farber protocol (Sakamoto et al., 1992). Immunofluorescence analysis of two human hepatocellular carcinomas also failed to identify C×26 immunoreactivity (Wilgenbus et al., 1992). Furthermore, post-transcriptional regulation has been partly implicated in the acinar heterogeneity of C×26 in normal rat liver (Rosenberg et al., 1992).

The physiological consequence(s) of increased C×26 staining in AHF may be important for the formation of selective GJIC and/or changing the gating properties of the intercellular channel. Heterotypic junctions formed with C×32 and C×26 in Xenopus oocytes exhibit gating properties different from corresponding homotypic junctions (Barrio et al., 1991). Therefore, GJIC may be restricted at the interface of C×26-positive AHF and surrounding hepatocytes such that the passage of molecules of certain size, shape and/or charge may be limited.

In this study, we show that C×43 mRNA (Fig. 4) and protein levels (Fig. 5) are enhanced in rat hepatic neoplasms. However, histological examination identified that the C×43 staining was restricted to bile ducts, Glisson’s capsule and cholangiomas (Figs 2, 3). Cholangiocarcinomas, thought to be derived from non-parenchymal liver cells (Saeter and Selgen, 1990), failed to exhibit C×32, C×26 or C×43 immunoreactivity (Fig. 3). Tumor PB3 exhibited abundant C×43 mRNA levels, but failed to display C×43 staining. Increased C×43 transcripts in human hepatic neoplasms (Oyamada et al., 1990) and C×43 staining were restricted to non-parenchymal cell types (Wilgenbus et al., 1992). Although rat liver epithelial cells have the ability to switch between C×43 and C×32/C×26 expression depending on culture conditions (Stutenkemper et al., 1992), we did not observe C×43 expression in preneoplastic or neoplastic hepa-tocytes with several polyclonal antibodies generated against different epitopes of rat C×43. Furthermore, foci from rats initiated with DEN and promoted with PB, TCDD, CI Solvent Yellow, WY-14,643, chlorendic acid, tamoxifen or ciprofibrate failed to exhibit C×43 staining.

Evidence for a role of altered GJIC during oncogenesis has been strengthened by recent studies showing restoration of growth control following transfection of connexin cDNAs (Eghbali et al., 1991; Mehta et al., 1991; Naus et al., 1992). While transfection studies show that the mechanism(s) respon-sible for downregulation of GJIC in some neoplasms may be defeated by overexpressing connexin mRNAs, similar gene replacement strategies were not successful in several rat hepatoma cell lines (Mehta et al., 1991; M. J. Neveu, unpub-lished observation). In this study, we found that rat hepatic neoplasms utilize several different modifications to alter the expression of C×32, C×26 and C×43. Fig. 6 summarizes how the observed changes may effect homologous (A) and heterol-ogous (B) communication between initiated cells and sur-rounding hepatocytes. The wide spectrum of alterations in connexin mRNA and protein expression observed in this study is consistent with cell fusion analysis of GJIC-deficient tumori-genic cell lines that identified at least two genes that modulate GJIC (Loewenstein, 1979; MacDonald, 1982). Taken together, the diverse array of mechanisms to downregulate connexin expression indicates that restoration of GJIC in some tumors could require the correction of several genetic loci/pathways. Since connexin expression was reversibly downregulated early during oncogenesis, upregulation of GJIC during tumor promotion may represent a more amenable target for cancer chemoprevention.

The studies described in this investigation were supported by grants from the United States Public Health Service National Cancer Institute (CA-07175, CA-45700, and GM30667). M.J.N. was a pre-doctoral trainee in Environmental Toxicology of the National Institute of Environmental Health Sciences (ESO-7015). E.L.H. is a recipient of a Career Scientist Award from the Hirschl Trust. The authors express their gratitude to Jennifer Vaughn for expert animal experi-mentation. Critical editing of the manuscript by Dr Ilse Riegel is gratefully acknowledged by the authors.