Heterocellular coupling was assessed in glioma-astrocytes co-cultures, using the dual-label technique, in which glioma cells were pre-labeled donors cells and astrocytes potential recipient cells labeled through gap-junctional diffusion of the dye. Heterocellular coupling was more extensive for GL15 cells (heterocellular coupling astrocytes/glioma ratio: 4.05 ± 0.6 recipient astrocytes per one GL15 donor cell) than for 8-MG cells (0.6 ± 0.2; p < 0.001) (Fig. 1A).

Analysis of tumor cells invasion was assessed in brain slice cultures, in which both homocellular coupling (between glioma cells) and heterocellular coupling with astrocytes can occur. GL15 cells displayed a significantly higher invasive potential than 8-MG cells [(S - S₀) / S₀ = 7.82 ± 0.68 vs. 2.67 ± 0.67, p < 0.001] (Fig. 1B).

Addition of CBX induced a significant decrease in the heterocellular coupling index of GL15 cells with astrocytes (down to 1.89 ± 0.74; p < 0.001) in co-cultures (Fig. 2B) whereas coupling was similar in control cultures either untreated, or treated with the functionally inactive analogue of CBX, GZA. Inhibition of GJC by CBX following implantation of GL15 into brain slices, resulted in a significant decrease in GL15 cells invasion into the brain parenchyma. The surface ratio (S-S₀) / S₀ of migration was reduced by inhibition of the gap junction function (5.59 ± 0.49 for CBX-treated vs. 7.82 ± 0.70 in untreated controls; p < 0.01) (Fig. 2C), as well as the number of GL15 cells that left the tumor mass (105 ± 6.3 cells/mm of perimeter vs. 132 ± 7.2; p < 0.01) (Fig. 2D). There was no difference between results obtained in the GZA and control groups. Those results were confirmed when migration of glioma cells out of spheroids was analyzed over an astrocyte monolayer culture in a 42 hours-long time lapse experiment (Fig. 2E).

In contrast, CBX treatment had no significant effect on 8-MG migration, either in glioma-astrocytes co-cultures or following implantation into brain slices (data not shown).

The composite cellular content of glioma fragments and the technical difficulty in maintaining cells alive after seeding them in culture as cell suspensions, precluded analysis of the functional coupling of tumors, either maintained in nude mice or freshly biopsied. We, therefore, relied upon the effects of CBX to assess whether, similarly to results obtained in established glioma cell lines, inhibition of GJC may affect the migration of tumor cells out of a fragment and into the neural parenchyma of a brain slice.

Invasive capacities of the seven tumors maintained in nude mice were quite heterogeneous. Three were minimally invasive according to our criteria [total number of cells that detached from the tumor mass was less than 15 cells and the distance of migration less than 50 μm (T3, T6 and T8)]. These three tumors were precisely the ones in the series of 7 tumors to not exhibit any response to CBX treatment (Table 1). Among tumors characterized as "invasive" according to our criteria, CBX treatment significantly inhibited invasion in three (T2, T4 and T7), and had no statistically significant effect in one (T1). Inhibition concerned both the surface ratio of migration (35 to 50% decrease) and the number of migratory cells (25% to 40% decrease) (Table 1). Treatment with the inactive analogue GZA had no effect on migration as compared to controls.

In order to test whether the results obtained with experimental gliomas could be extrapolated to native tumors, the migration potentials of biopsies from 4 human high grade gliomas (HG) were studied in brain slice culture, with or without addition of CBX or GZA. Overall, CBX treatment reduced migration indices of these four tumors by 30% (range 20 to 50%; p < 0.05 unpaired t test and Mann & Whitney – Table 2); this was the most pronounced for HG-23 (p < 0.02, Fig. 3A).

In order to investigate whether Cx43 was instrumental in the establishment of functional coupling between glioma cells and astrocytes, Cx43 expression was analyzed in glioma cells from GBM cell lines and fresh human biopsies used in previous experiments. We chose to focus on Cx43 because it is the major Cx in astrocytes and is expressed by glioma cells 10111213.

In agreement with previous studies 16, three isoforms of Cx43 could be separated on SDS-PAGE and visualized by immunoblotting as three distinct bands in control astroglial primary cultures: Cx43-NP, a non-phosphorylated isoform, Cx43-P1 and Cx43-P2, the two phosphorylated isoforms of the protein (Fig. 4). The Cx43-P1 isoform was prominent and Cx43-P2 and Cx43-NP isoforms were equally present (Fig. 4). As for invasive potential, Cx43 expression pattern was different in the two GBM cell lines with the GL15 cells exhibiting a three-fold higher concentration of the Cx43-P1 as compared to 8-MG cells (p < 0.001; Fig; 4A).

Histological analysis of Cx43 immunoreactivity in brain slices implanted with GL15 cells, showed that the protein was overexpressed in the tumor margin where glioma cells and astrocytes processes were intimately apposed and intermingled (Fig. 5). Confocal microscopy of brain slices allowed us to identify regions of enhanced immunostaining for Cx43 at the interface between GL15 cells and astrocytes, suggesting the existence of heterocellular gap junctions (Fig. 5). Interestingly, these astrocytes displayed a bipolar morphology that extended in arrays a long process that paralleled the outward migrating cells. This was not observed when 8-MG glioma cells were implanted.

A general expression of the P1 isoform was observed in the seven tumors maintained in nude mice, and a variable expression of the Cx43-NP isoform. The four invasive tumors (T1, T2, T4 and T7) expressed abundantly both Cx43 isoforms Cx43-NP and Cx43-P1, whereas the three others (T3, T6 and T8) which were not invasive and did not exhibit any response to CBX treatment, expressed lower levels of Cx43-P1 isoform and did not present detectable levels of Cx43-NP. (Fig. 4B). No expression of Cx43-P2 was recorded.

The amount of tissue available for experimental analysis did not allow us to carry out western blotting on freshly biopsied tumor fragments. Thus Cx43 expression in the human tumor biopsies was evaluated by immunohistochemistry on paraffin embedded tissue section.

Cx43 immunostaining was seen in all 4 specimens. Staining was patchy, and both the intensity of staining and the number of immunostained cells in patches differed from one sample to another. In the typical case, a positive microscopic field contained several strongly stained cells, and several unequivocally labeled cells of lower intensity (Fig. 3B). Homogenous fields of particularly strong immunolabeling intensity was observed in the HG-23 specimen.

We have next evaluated homocellular coupling between glioma cells by the scrape loading method, in which homocellular coupling is demonstrated by the diffusion of the dye from mechanically lesioned cells to their neighbors in a monolayer culture. Assay revealed extensive coupling of GL15 (homocellular coupling index: 63.75 ± 4.4 %), comparable to that displayed by astrocytes in parallel control experiments (60 ± 5.2%), whereas 8-MG cells were only weakly coupled (12.7 ± 1.87 %; p < 0.001 vs. GL15), a result in keeping with their low level of Cx43 expression. Homocellular coupling of all cell types was strongly affected by addition of the gap junctional blocker CBX: down to 7.4 ± 1.3 % for astrocytes (p < 0.001), to 1.97 ± 0.47 for GL15 (p < 0.001) and to 2.2 ± 0.47 for 8-MG cells (p < 0.001) (Fig. 6), whereas GZA had no effect.

To ensure that CBX-induced decreased migration of GL15 cells was mediated by the effect of the compound on gap junction communication, rather than more generally on the phenotype of tumor cells, analysis of tumor cell migration was performed in experimental conditions in which only homocellular coupling could exist, i.e. after seeding of glioma spheroids on a collagen IV matrix. Surprisingly, when homocellular coupling between tumor cells was inhibited by CBX, both tumor cell lines displayed a significant increase in migration out of the spheroids [(S-S₀) / S₀ = 55 ± 4.1 and 25.4 ± 5.25 for GL15 and 8 MG, respectively, p < 0.001 vs. untreated in both cases] (Fig 7).

In order to validate further the hypothesis of a role of GJC formed by Cx43 in tumor cell migration, we have used a technique adapted from that used by Lin et al. (2002) 8. C6 tumor cells have been genetically engineered by transfection, using a plasmid containing a fusion gene encoding GFP attached to the carboxyl terminus of Cx43. The gap junctions formed by this chimeric protein were dye-permeable 17. C6 cells are known to proliferate without migrating out of the tumor mass, following implantation into the rat brain, and do not express a significant amount of Cx43. By transplanting mixed population of native and GFP-Cx43-transfected cells, we explored the capacity for the latter to display specifically a migratory behavior. Indeed, GFP-Cx43 expressing C6 cells were preferentially located at the periphery of the tumor mass and some were even observed scattered in the surrounding brain parenchyma, strongly suggesting a migratory behavior (Fig. 8). This was not observed with C6-GFP cells which were, as classically described, only present in the tumor mass.

The two human glioblastoma (GB) cell lines used in the study, GL15 42 and 8-MG 43 were grown as cell culture in glioma-cell medium: 50% minimum essential medium (MEM) and 50% Dubelco Modified Eagle's Medium (DMEM) complemented with 10% fetal calf serum (ATGC Biotechnological, France), glutamine 2 mM, D-glucose 3.3 mM, Penicillin 100 UI/mL, Streptomycin 100 μg/mL (all from Eurobio, France) at 37°C in a 5% CO2 humidified incubator. On confluence, cells were trypsinized, centrifuged at 300 G for 10 minutes and resuspended in 1 mL of glioma-cell medium. 5 × 10⁵ cells were seeded in 1.5% agar-coated flasks (25 cm³, Falcon, France) for 7–10 days to obtain GL15 and 8-MG spheroids. Spheroids with a diameter ranging from 50–100 μm were selected for migration assays.

Seven grade III-IV human glioma biopsies xenografted and maintained subcutaneously in Swiss Nu/Nu mice (Charles River, France), kindly provided to us by Dr M-F. Poupon and P. Leuraud 44, were used in this study. For the sake of simplicity these tumors are referred to in the text as T1, T2, T3, T4, T6, T7 and T8 (see table 1). Mice were sacrificed when the tumor bulk reached a diameter larger than 1 cm. The tumors were retrieved and immediately transferred in the glioma-cell medium. Areas of necrosis and hemorrhage were identified under a microscope and discarded. The remaining tissue was cut into three fragments that were used as follows: -one was immediately implanted into a healthy nude mouse, in order to maintain the model; -a second one was immediately frozen for Western Blot analyses; -the third one was cut into smaller pieces for implantation into rodent brain slices maintained in organotypic cultures (see below).

Finally, four fresh human biopsy specimens were obtained from stereotactic or resection biopsies in the operating room. All samples were retrieved by the neurosurgeon from areas of enhanced contrast on magnetic resonance imaging, which correspond to a hypercellular zone on histological examination of the tumor. After retrieval, samples were immediately transferred in glioma-cell medium and prepared for implantation into brain slices. Pathological analysis of the same tumors was carried out in parallel on adjacent specimens; it systematically included immunohistochemistry for the connexin protein Cx43 in addition to usual stains. According to the WHO 2000 classification, these tumors were glioblastomas in three cases, an anaplastic grade III astrocytoma in one. For the sake of simplicity the generic term of glioblastoma will be used in the rest of the text.

Functional assessment of gap junctional communication (GJC) was performed on glioblastoma cell lines. Homocellular GJC was determined by the scrape-loading/dye transfer technique 4546. Briefly, either GL15, 8-MG or astrocytic cell cultures were incubated for 10 minutes in HEPES buffer solution (140 mM NaCl, 5.5 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂ 10 mM glucose and 10 mM HEPES pH 7.3). Cells were then washed in a Ca²⁺-free HEPES solution to prevent uncoupling. After scraping with the razor blade cells were incubated in the same solution containing 1 mg/ml of Lucifer Yellow (LY) (Sigma, France) and 1 mg/mL of Rhodamin Dextran 10,000 MV (Molecular Probes) for 1 minute. Dye transfer through GJC was assessed 10 minutes after scraping, using an Axioplan-2 microscope and appropriate filters (Carl Zeiss, Inc; Oberkochen, Germany). Fluorescent and bright light images digitized with a CoolSNAP camera (Ropper Scientific; Tuscon, AZ) from every microscopic field were merged, and the homocellular coupling index was calculated as the ratio of the number of LY-labeled cells to the total number of cells.

Heterocellular GJC was evaluated in glioma/astroglial cells co-cultures by a method adapted from Goldberg et al. (1995) 27. Tumor cells were loaded for 20 min with a dual-label dye solution containing isotonic PBS-glucose (0.3M), 5 μM Calcein-AM and 10 μM DiI (both from Molecular Probes, France). They were then rinsed three times with isotonic PBS-glucose 0.3M, trypsinized and centrifuged. Five hundred Calcein/DiI-labeled cells (donor cells) were seeded on an unlabeled confluent astrocytic monolayer (recipient cells), and incubated at 37°C in a 5% CO2 humidified incubator for 20 h. Heterocellular coupling was calculated as the ratio of recipient cells (calcein-labeled astrocytes) to donor cells (Calcein/DiI-labeled glioblastoma cells). It is important to underline that this technique is different from the scrape loading method and uses another dye tracer to assess the gap junctional coupling. Hence, their results cannot be readily compared in quantitative terms.

Effects of inhibition of GJC were evaluated by treating the cultures with carbenoxolone (CBX); controls received its inactive analog, glycyrrhyzic acid (GZA) 23 (both from Sigma, France). CBX is a widely used specific inhibitor of GJC likely exerting its action by changing the conformation of gap-junction channels 22, while GZA has no effect on GJC. CBX and GZA treatments were carried out at a concentration of 30 μM. Cytotoxicity of these molecules on the two glioblastoma cell lines and on astrocytes had been tested in preliminary experiments. There was no significant increase in the lactate dehydrogenase medium/cell ratio for up to 7 days at the concentrations used (Cytotoxicity Detection Kit, Roche Molecular Biochemicals, France).

Migration of glioblastoma cell lines was analyzed in three different experimental in vitro set ups: on collagen IV-coated cell culture dishes, after seeding on an astrocytic monolayer culture, and following implantation into organotypic brain slices (ISIS, Intra-Slice Implantation System). Migration out of tumor fragments issued from xenografted gliomas or from freshly biopsied specimens were only studied using the ISIS method.

Migration of GL15 and 8 MG cells out of spheroids deposited onto collagen-coated (0.5 μg/mL; Sigma, France) cell culture dishes was analyzed over 4 days.

For astrocytic co-cultures, primary astrocyte cultures were obtained from cerebral hemispheres of neonatal C57/bl6 mice (Charles River, France) as previously described 47. Cells were grown in astrocyte medium: MEM (GIBCO, France) complemented with 10% FCS (ATGC Biotechnological, France), glutamine 2 mM, D-glucose 33 mM, Penicillin 100 UI/mL, Streptomycin 100 μg/mL and amino acids 0.5X (all from Eurobio, France). Medium was changed every 3 days. On confluence, GL15 and 8-MG spheroids were seeded on top of the astroglial monolayer, and cultures were continued for 4 days.

The ISIS migration assay was carried out as described previously 14, using brain slices obtained from postnatal day 6 C57/bl6 mice. Briefly, after decapitation the two hemispheres were separated and 400 μm-thick coronal slices were cut using a McIllwain tissue chopper (Mickle Laboratory, England). Slices were transferred into brain slices medium: MEM (GIBCO, France), 1 g/l D-glucose, 10% heat-inactivated fetal calf serum, 0.1 g/l transferrin, 16 μg/l putrescin, 40 μg/l N-selenium, 30 μg/l tri-iodothyronin, 5 mg/l insulin, and 60 μg/l progesterone (all from Sigma, France). Slices were separated and transferred onto Millicell-CM membranes (Millipore, France). The Millicell-CM membranes were kept in six-well plates, above 1 ml of brain slice medium. Slices were incubated at 37°C in a humid atmosphere with 5% CO₂. The medium was changed three times a week. 24 hours after brain slice preparation, GL15 and 8-MG spheroids or small fragments of xenografted tumors and freshly biopsied specimens were deposited onto the slice surface and gently pushed with the tip of a needle, until the tumor tissue was well enveloped by the parenchyma of the slice. Cultures were maintained for 5 or 7 days for cell lines and tumor fragments, respectively.

In all experimental conditions, effects of the blockade of GJC were assessed using CBX (30 μM for collagen-coated and astrocytic co-cultures; 60 μM for brain slice cultures, a stronger dose because of the thinness of the slice), with two negative controls: untreated and GZA supplemented to the medium instead of CBX. These compounds were added to the culture medium 24 h after glioma spheroid or tumor implantation.

For a control experiment, we transfected the C6 glioma cell line with a plasmid vector encoding either a green fluorescent protein (GFP) fused to the carboxyl terminus of Cx43 17 kindly provided by Dr P. Martin, or else only GFP. C6-Cx43-GFP cells and C6-GFP control cells were implanted into the striatum of nude mice (5. 10⁴ cells in 1 μL) (n = 5). Mice were sacrificed at day 8 and perfused with paraformaldehyde (4%) 8 days after implantation. Brains were sectioned serially on cryostat at 30 μm thickness as previously described 48.

Immunoblot analysis for Cx43 was performed on cells and tumor samples that were suspended in lysis buffer containing Tris base 50 mM, pH 7.4, NaCl 150 mM, Triton X-100 0.5%, EDTA 1 mM, protease inhibitor cocktail 0.5% (Sigma, St. Quentin, France) and triturated to homogeneity. Homogenates were centrifuged at 13,000 G for 10 or 30 minutes (for cell lines and tumor samples, respectively), and supernatants were aliquoted and stored at -80°C. Total protein content was determined by the Lowry method with bovine serum albumin as a standard. Total protein (30 μg) was boiled for 5 min after addition of 10% glycerol/5% mercaptoethanol and separated on SDS-PAGE 12.5% gels, transferred on PVDF membrane (Bio-RAD, France) under classical conditions 49 and immunoblotted overnight at 4°C, using a polyclonal anti-Cx43 antibody (1:500, Zymed, France). Bound primary antibody was detected with a horseradish peroxydase-coupled anti-rabbit antibody (1:5,000; Amersham Pharmacia Biotech, France). In preliminary positive control experiments, Cx43 was revealed as three bands of about 41, 43 and 45 kDa, representing three different isoforms in heart sample. The two higher molecular mass bands have been shown to correspond to phosphorylated forms of the protein, and the 41 kDa migrating species to the unphosphorylated form of the protein 16.

Immunohistochemistry was carried out on cell and slice cultures and brain slices from in vivo experiments following fixation using 4% paraformaldehyde in PBS at 4°C (30 min for mono- and co-cultures, 4 h for ISIS). For fluorescent immunostaining, cultures were rinsed several times in PBS, then incubated for 1 hour in PBS containing 10% normal horse or goat serum and 0.6% Triton X-100 (Sigma). They were incubated overnight at 4°C with monoclonal antibodies against vimentin (1/400, Neo Markers, France), or polyclonal antibodies against glial fibrillary acidic protein-GFAP (1/100, Dako) or Green Fluorescent Protein-GFP (1/200,. Molecular Probes). After 3 washes in PBS, they were then incubated with a PBS solution containing anti-rabbit or anti-mouse immunoglobulin conjugated either to fluorescein isothiocyanate-FITC, AMCA (1/400; Vector, France), or to cyamidine-Cy3 (1/1000; Jackson Immuno Research, France). For human biopsies, immunocytochemical detection of Cx43 was carried out on formalin-fixed, paraffin-embedded 5 μm sections. Antigen-retrieval consisted in heating slides immersed in citrate buffer (pH 6) in a microwave oven (3 × 5 min at 750 W). To block non-specific binding sites slides were incubated in PBS containing 10% BSA. A primary polyclonal antibody (Zymed) directed against the intracellular portion of the Cx43 molecule was applied (1:200) to sections overnight. Bound primary antibodies were detected by a Vector Elite ABC kit (Vector Laboratories, Burlingame, California) following the manufacturer's instructions and using DAB as the chromogen. Slides were lightly counterstained in Harris's hematoxylin. Negative controls had the primary antibodies omitted. A specimen of normal human cortex from epilepsy surgery showed diffuse staining of astrocytic cells accentuated at perivascular and superficial locations, corresponding to astrocytic endfeet and the glia limitans, respectively.

Fluorescent microscopy on a Zeiss Axioplan2 microscope and image analysis using the KS.400 (3.0 version) software were performed to quantify invasion of glioma cells, according to techniques that have been detailed elsewhere 14. Two invasion parameters were assessed, (i) the maximum area of invasion, given as the ratio (S-S₀) / S₀, where S is the maximum area containing migrating glioma cells and S₀ is the initial tumoral mass area, this ratio compensating for the heterogeneous sizes of implanted spheroids and tumor fragments, and (ii) the number of invading cells (cells/mm of S₀ perimeter). For collagen-coated migration assay and astrocytic co-cultures, only the maximum area of invasion was assessed. The statistical significance was evaluated using Mann & Whitney test and Student unpaired t test.