The main result of this study is the demonstration that functional GJC between tumor cells and host astrocytes is instrumental in the invasion of the brain parenchyma by human glioblastoma cells. Establishment of functional coupling between those two cell populations may be mediated by the gap junction protein Cx43 as ability of glioblastoma cells of various origins to migrate out of the tumor bulk into the brain parenchyma, appeared related to the level of Cx43 expression. In support of this hypothesis, overexpression of a fusion Cx43-GFP gene construct triggered a migratory behavior in otherwise non-infiltrative C6 glioma cells implanted into the brain. The similarities between those results and mechanisms of neural precursor cell migration during embryonic development suggest that glioma cells may subvert and take advantage of a physiological mechanism of cell migration into the brain. In contrast, homocellular interaction between glioma cells through gap junctions hampers their dissemination out of the tumor mass, pointing to an additional role for gap junctions, in cell-to-cell adhesion.
Over the past decade, studies on GJC in gliomas have mainly addressed its role in uncontrolled cell proliferation [18–20], by analogy with epithelial cancers in which a low level of connexin expression is a sign of profoundly perturbed tissue homeostasis and malignant evolution . The notion that, in gliomas, GJC may also favor migration draws scientific argumentation from embryology and from observations that other neurectodermal tumors such as melanoma need to establish GJC with host cells in order to migrate out . This concept has been strongly supported recently by the demonstration that a C6 rodent glioma cell line that had been genetically engineered to express Cx43 displayed a migratory behavior following transplantation into the rat brain . This phenomenon was the more striking that the value of the rodent C6 cell line as an experimental model for glioblastoma has always been questioned because it did not reproduce the infiltrative phenotype which is a major characteristic of the human brain tumor. The induction of C6 migration was related to the establishment of functional GJC between glioma cells and host astrocytes through homotypic Cx43-Cx43 gap junctions.
Several lines of evidence drawn from our results converge to indicate that this concept is clinically relevant, as GJC with astrocytes also plays a role in the migration of brain tumor cells for human glioblastomas. Experimental results obtained here with native human glioblastoma cells of various origins, can indeed be interpreted as a demonstration of a similar role for GJC. First, pharmacological blockade of GJC by the specific inhibitor CBX  had negative effects on tumor cell migration, whereas its inactive analogue GZA  did not. Second, the gap junction protein Cx43, which is overexpressed by reactive astrocytes in the margin of glioblastomas , was also expressed by all studied human tumors, including biopsy specimens, providing a molecular basis for the potential formation of homotypic heterocellular gap junctions. Last, close membrane appositions at regions of reinforcement of Cx43 immunostaining, suggesting intercellular contacts between astrocytes and glioma cells migrating out of a tumor spheroid or the body mass, were observed in co-cultures of glioma cells on an astroglial monolayer, as well as following implantation of tumors into brain slices maintained in organotypic cultures.
The pharmacological approach used here had both advantages and drawbacks, which should be indicated as they may limit the conclusions to be drawn. The main advantage of that technique is the fact that it allowed us to study GJC inhibition in fresh tumor samples, in which biological properties and diversity of glioblastoma cells are well preserved, in contrast to the clonal selection and genetic variation that are commonly observed with cells lines. Conversely, we probably introduced some biological variability in the experimental system. This may be relevant to the fact that GJC inhibition with CBX did not abolish but rather reduced the rate of human glioma cell migration in our study, whereas Lin et al. (2002) obtained an "all-or-nothing" effect following genetic engineering of the C6 line. The partial effect on migration observed here may be due to the partial pharmacological inhibition consistent with the results of other studies [24, 25] that demonstrated that CBX only reduced GJC by 50%. Alternatively, glioblastoma cell migration into the brain parenchyma does not relate exclusively to cell-to-cell interaction with astrocytes. Interactions between tumor cell and matrix components (reviewed by Giese et al., 2003)  are also major actors in glioma cell migration, and they are most likely not effected by inhibition of GJC, at least not directly. The use of a pharmacological compound also leads to questions about specificity. Various compounds used to block GJC have additional non-specific and toxic effects, among which halotane, octanol or acidifying agents. CBX has been shown to be nontoxic and has no effect on cell proliferation, total protein synthesis or cell viability [23, 26, 27]. This is ascribed, by some authors, to its mode of action, through direct binding on connexon particles . We have checked the specificity of the GJC effects via different experimental ways, including a systematic negative control with glycyrrhyzic acid (GZA), the CBX precursor which only differs from it by its inability to disrupt GJC. CBX and its inactive analog GZA are known to both interact with mineralocorticoïd and glucocorticoïd receptors (see discussion in ), but these receptors do not mediate CBX-decreased GJC . We also tested CBX effect on cell apoptosis, cell proliferation or cell viability in our conditions. No significant effect was observed. Although it has been reported that Cx43 may induce cell growth [18, 29], we did not observe any effect of CBX on cell proliferation. It is interesting to mention that CBX-decreased GJC is mainly due to alteration of gap junction channel permeability, whereas Cx overexpression is sufficient to regulate cell growth [20, 30]. As an interesting positive control of the specificity of the effects recorded in co-cultures and following implantation of tumor cells into brain slices, we have observed an opposite effect of CBX on cell migration when only homocellular interaction between glioblastoma cells was concerned (on collagen IV). Altogether, these results support the conclusion that non-specific pharmacological effects, if any, did not alter results of our study in a consequential manner.
Levels of Cx43 expression were positively related to glioma invasiveness. Analysis of Cx43 expression by western blot allowed detection of two distinct isoforms of the Cx43 protein, Cx43-P and Cx43-NP. Both isoforms were detected in the four highly invasive tumors (T1, T2, T4, T7), whereas the three minimally invasive tumors (T3, T6, T8) displayed no detectable levels of Cx43-P isoform. The presence of both Cx43 isoforms strongly correlated with the invasive capacities of the tumor. However, whether the migratory capacity may depend on the overall levels of expression, and/or upon the concomitant presence of both Cx43 isoforms is, at this stage, a matter of speculation. Published data on the association of either isoform with the functional state of gap junctions are conflicting. Phosphorylation of Cx43 either on serine or tyrosine residues has been reported to disrupt GJC (for review see [31, 32]). This could in part explain why T3, T6 and T8 tumors, which only displayed the phosphorylated isoform -the inactive isoform of Cx43-, were non-invasive tumors. However, the reverse association between Cx43 dephosphorylation and GJC inactivation has been suggested in astrocytes . Further biochemical analysis of Cx43 isoforms function in these cells may reveal whether the pattern of Cx43 isoform expression is involved in the invasive process and how this pattern can be altered by specific phosphorylation. On the other hand, it cannot be excluded that the abundance of both -P and -NP isoforms in migrating tumors simply relates to a dynamic turnover of connexons between the cytoplasm (Cx43-NP) and the cell membrane of a migrating cell, where Cx43-P may participate in short-lived contacts with host astrocytes . Recent data have shown that CBX itself did not have any effect on the phosphorylation pattern of Cx43 .
In the biopsy specimens, levels of Cx43 expression were different from one tumor to another, and from one zone to another within a specimen, in agreement with results reported by other authors [10–13]. Interestingly, the highest expression of Cx43 was seen in HG-23, the tumor in which migration was the most substantially inhibited by CBX. In addition, transfection of C6 glioma cells with a Cx43 protein fused to a green fluorescent protein allowed us to trace C6-Cx43 cells after intrastriatal stereotactic implantation. Consistent with results of Lin et al (2002) and , only C6-Cx43 cells switched to an invasive phenotype providing a direct evidence of the positive correlation between glioma cells Cx43 expression and invasive potential. We hypothesize that Cx43 may favor glioma cells migration by facilitating glioma cells to form gap junction interactions with host astrocytes that will allow them, through currently unknown molecular mechanisms, to drag out from the tumor mass.
A peculiar morphological finding in this study was the phenotypic alteration of astroglial processes around tumors. Changes in the astroglial phenotype induced by glioma cells through GJC have been observed by other authors . The ISIS protocol used here allowed us to gain a comprehensive three-dimensional view of the astroglial meshwork, and to identify the radial orientation of astrocytic processes toward the body of the tumors. This organization is reminiscent of the scaffolding of radial glial cells observed in the developing neocortex in vivo and in vitro . Radial glial cells extending long processes from the ventricular zone to the cortical plate communicate with migrating neurons through Cx43 gap junctions [38–40]. Although tentatively, we may therefore speculate that astroglial Cx43 provide similarly support and guidance to the migrating tumor cells. In the same way as immature neuronal cells do [38, 39, 41], glioma cells may trigger the phenotypic change of surrounding astrocytes into radial glia-like cells, in order to make them suitable as a substrate for migration.
The three different migration assays allowed us to explore the importance of intercellular interactions between migrating cells and cellular environment occurring during invasion. We have shown that inhibition of homocellular GJC in GL15 spheroids seeded on collagen IV (i.e. in the absence of astrocytes), boosted glioma cell motility, resulting in a higher centrifugal dispersion of cells. This result is in keeping with that obtained by Lin et al (2002) in an experiment in which Cx43 was shown to mediate glioma cell adherence and aggregation. However, accelerated motility of GL15 glioma cells when CBX blocked homocellular glioma/glioma GJC was not observed when glioma cells additionally formed heterocellular GJC with astrocytes, i.e. on astrocyte monolayer or in brain slices. Quite the contrary, heterocellular glioma-astrocytes GJC seemed to be necessary for the invasive process to occur normally, as CBX significantly blocked it in those conditions.