Retroviral Delivery of Connexin Genes to Human Breast Tumor Cells Inhibits in Vivo Tumor Growth by a Mechanism That Is Independent of Significant Gap Junctional Intercellular Communication*

The mechanism by which gap junction proteins, connexins, act as potent tumor suppressors remains poorly understood. In this study human breast tumor cells were found to exhibit diverse gap junction phenotypes including (a) undetectable Cx43 and no intercellular communication (HBL100); (b) low levels of Cx43 and sparse intercellular communication (MDA-MB-231); and (c) significant levels of Cx43 and moderate intercellular communication (Hs578T). Although retroviral delivery of Cx43 and Cx26 cDNAs to MDA-MB-231 cells did not achieve an expected substantial rescue of intercellular communication, overexpression of connexin genes did result in a dramatic suppression of tumor growth when connexin-expressing MDA-MB-231 cells were implanted into the mammary fat pad of nude mice. Subsequent immunolocalization studies on xenograph sections revealed only cytoplasmic stores of Cx43 and no detectable gap junctions. Moreover, DNA array and Western blot analysis demonstrated that overexpression of Cx43 or Cx26 in MDA-MB-231 cells down-regulated fibroblast growth factor receptor-3. Surprisingly, these results suggest that Cx43 and Cx26 induce their tumor-suppressing properties by a mechanism that is independent of significant gap junctional intercellular communication and possibly through the down-regulation of key genes involved in tumor growth. Moreover, our studies show that retroviruses are effective vehicles for delivering connexins to human breast tumor cells, facilitating potential gene therapy applications.

Gap junctional intercellular communication (GJIC) 1 plays a crucial role in direct cell-cell signaling via the exchange of small molecules such as calcium ions, IP 3 , and cAMP between the linked cells (1,2). A family of homologous proteins called connexins comprise the basic building blocks of gap junctions (1,2). Connexins are co-translationally inserted into endoplasmic reticulum membrane and oligomerized into hemichannels (connexons), which are transported to the plasma membrane (3). Connexons from two contacting cells dock at the plasma membrane to form intercellular gap junction channels, which cluster to form gap junction plaques (3).
Presently, in excess of fifteen connexin genes have been identified and cloned (1,2). However, only three connexins, connexin26 (Cx26), connexin43 (Cx43), and connexin32 (Cx32), have been found in rodent mammary epithelium in different temporal and spatial patterns (4 -8). Cx43 is localized mainly to myoepithelial cells throughout all of the stages of mammary gland development, whereas Cx26 is localized primarily to the luminal cells (6). Expression of Cx26 increases in pregnancy and peaks during lactation. Cx32 is a minor connexin expressed in luminal cells, and it can be detected only in lactating gland (6,8). Monaghan et al. (5) reported Cx26 and Cx43 to be the only connexins expressed in the resting human mammary gland.
Connexin expression, gap junction assembly, and GJIC are down-regulated in a variety of neoplastic cells or primary tumors both in vivo and in vitro (9 -14). In many cases the restoration of connexin expression and GJIC has reduced tumor growth and promoted cell differentiation (12). Therefore, connexin genes have been deemed tumor suppressors. Further evidence supporting the role of connexins as tumor suppressors came from Cx32 knockout mice, which were found to be more susceptible to liver tumor formation in comparison to wild-type littermates (15). In the human breast, Wilgenbus et al. (16) reported the lack of Cx43 expression in malignant tumors. Lee et al. (17) also found that Cx43 mRNA was not detected in several mammary tumor cell lines, and subtractive hybridization was used to identify Cx26 as a potential tumor-suppressing molecule (18). Consistent with these studies, we found a deficiency of Cx43 gap junctions in several human breast carcinomas (7). Interestingly, Jamieson et al. (19) compared the expression of Cx43 and Cx26 in normal human breast tissues and human breast tumor tissues in situ and found that many invasive carcinomas expressed one of these two connexins but that most often these connexins were not assembled into gap junctions. These latter results suggest that a population of transformed mammary epithelial cells continue to express connexins but have lost the ability to assemble connexins into functional gap junctions.
If connexin transgene approaches ever are to be used as part of a potential human breast cancer therapy regimen, it is necessary to develop a delivery system for introducing connexins in vivo and to determine whether human breast tumor cells are generally capable of expressing exogenous connexins and forming functional gap junctions. Although a clear relationship between connexin expression, GJIC, and carcinogenesis has been elucidated, the understanding of gap junctions in the development of human breast cancer has yet to be established. Moreover, it remains unclear whether the tumor-suppressing properties of connexins is restricted to the ability of connexins to form gap junction channels.
In the present study we identified human breast cancer cells that mechanistically exhibit two types of GJIC defects. First, in cells that lack connexin expression (i.e. HBL100 cells) GJIC could be restored effectively using a retroviral system to deliver connexins. However, in cells that were defective in connexin trafficking and gap junction assembly (i.e. MDA-MB-231) exogenous connexins were delivered preferentially to lysosomes. Nevertheless, in vivo data revealed that, even in the absence of significant GJIC, overexpression of connexin genes in MDA-MB-231 human breast tumor cells suppressed tumor growth as the possible result of down-regulation of regulatory genes involved in tumor growth.

MATERIALS AND METHODS
Cell Culture-All of the cells were grown at 37°C with 5% CO 2 in the recommended medium including RPMI 1640 (Invitrogen) for MDA-MB-231/Hs578T cells and Dulbecco's modified Eagle's medium (Invitrogen) for normal rat kidney (NRK), HeLa, and HBL100 cells. The medium was supplemented with 10% fetal bovine serum (or 10% calf serum for HBL100 cells), 2 mM glutamine, and 100 units/ml penicillin, 100 g/ml streptomycin. Human mammalian epithelial cells (HMEC) was purchased from Clonetic Human Cell Systems (BioWhittaker, Walkersville, MD) and cultured in the mammary epithelial cell growth medium (BioWhittaker) according to the manufacturer's instructions.
Immunocytochemistry and Confocal Imaging-Cells were immunolabeled as described previously by Laird et al. (20) with anti-Cx43 polyclonal antibodies directed against amino acid residues 360 -382 of Cx43 (21) or antibodies to Cx26 (Zymed Laboratories Inc., San Francisco, CA). The images were captured using a Zeiss LSM 410 inverted confocal microscope equipped with a ϫ63 oil lens as described previously (20).
Single Cell Microinjection-Single cells were pressure microinjected with 5% Lucifer Yellow (Molecular Probes, Leiden, Netherlands) using an Eppendorf microinjection system and a Zeiss inverted epifluorescent microscope. Digital images were acquired using a CCD Sensicam. The percentage of microinjected cells that exhibited dye coupling and the order of dye transfer (first order, cells contacting the microinjected cell only; second order, cells distant from the microinjected cells by one cell, etc.) were determined. Because many of these tumor cells are prone to overlapping and piling up, dye that spread only to a single neighboring cell was quantified separately to allow for artifacts that might occur from the inadvertent penetration of two cells during the microinjection procedure.
cDNA Constructs, Transfection, and Infection-Cx43 and Cx26 cDNAs were provided by Dr. C. C. G. Naus. To engineer the retroviral vectors that encoded Cx43 or Cx26, we removed the enhanced green fluorescent protein reporter gene from an AP2 viral vector (22) by NotI digestion. Cx26 cDNA was engineered into the AP2 vector between BglII restriction sites while Cx43 cDNA was inserted between NotI and CalI. Recombinant retroviral vectors were transfected into 293GPG package cell lines to produce virus-containing supernatant that was further used for infecting the tumor cells as described previously (23). More than 90% of the mammary tumor cells expressed connexins after three rounds of viral infection.
Cell Growth and Tumor Formation in Nude Mice-To examine cell growth in vitro, 5 ϫ 10 4 cells were plated in 35-mm tissue culture dishes and allowed to grow. At days 2, 3, 4, 5, and 6 cells from replicated plates were collected and counted using a Coulter Z TM 1 particle counter (Beckman-Coulter, Miami, FL). These data were plotted as the average cell number Ϯ S.E.
For in vivo tumor growth 6 -8-week-old female Nu-Nu mice (18 -22 g) were obtained from Charles River Laboratories (St. Zotique, Quebec). The experimental protocol was approved by the Animal Care Commit-tee of McGill University, Montreal, Quebec. Animals (5-15 per treatment) were housed 5 to a cage and fed rodent chow and water ad libitum. After 1 week of acclimatization, a suspension of 1 ϫ 10 6 MDA-MB-231 viral control tumor cells or MDA-MB-231 tumor cells expressing Cx43 or Cx26 were transplanted into the mammary fat pad in a total volume of 0.1 ml of medium. Tumor length and width were measured every 3-8 days using a caliper accurate to 0.5 mm. Tumor volume was calculated using the formula V ϭ L ϫ W ϫ 0.5W. The experiment was terminated when the control tumors reached the maximal size allowed. In this mouse study, MDA-MB-231 cells infected with the control retrovirus that did not encode a connexin, one mouse died prematurely and was eliminated from the data set. This data set was plotted as the average tumor size Ϯ S.E. In the case of the mice transplanted with MDA-MB-231 expressing either Cx43 or Cx26, four mice failed to develop a tumor, three mice developed tumors that were too small to accurately measure, and three mice developed small tumors.
Immunohistochemistry on Xenograph Sections-The xenografts in nude mice were excised, fixed in 4% formalin, embedded in paraffin, and cut into 4-m thick sections for immunohistochemistry. Tissue sections were deparaffinized in xylene, graded alcohol, and water, boiled in citric acid buffer for antigen retrieval, and blocked by 10% horse serum. The tissue sections were stained with anti-Cx43 polyclonal antibodies (Sigma) followed by incubation with biotinylated secondary anti-rabbit IgG (Vector Laboratories, Burlington, ON). Positive immunolocalization of Cx43 was determined using the Elite ABC kit (Vector Laboratories). For contrast, tissue sections were counterstained with hematoxylin. Finally, all images were acquired using the Zeiss Axioscope light microscope.
DNA Array and Western Blotting-Total RNA was extracted from cells using Trizol Reagent (Invitrogen) for the preparation of biotinlabeled cDNA probes. Nonrad-GEArray kits were purchased from SuperArray, Inc. (Bethesda, MD) and supplemented with positively charged nylon membrane on which cDNA fragments from genes had been immobilized in duplicate. The membrane used in our study contained 23 genes closely related to tumor growth and metastasis plus two positive control genes (GAPDH and ␤-actin) and one negative control (pUC18). Probe synthesis, hybridization, and chemiluminescent detection were performed according to the manufacturer's protocol. In brief, 10 g of total RNA was reverse-transcribed into biotin-labeled cDNA probes at 42°C for 120 min. The hybridization was performed with denatured cDNA probe at 68°C overnight after a 1-h incubation of the membrane with sheared salmon sperm at 68°C. For detection, the membrane was incubated with AP-streptavidin and then subjected to CDP-Star chemiluminescence substrate. The results were analyzed using the Multi-Analyst 1.0.1 software from Bio-Rad. Because no signal amplification was performed, the membranes were quantified first by subtracting the average background intensity for the pUC18-negative control and secondly by examining the signal ratio in comparison to the internal GAPDH-positive control (see Fig. 2, bottom right corner of blots). Differential hybridization levels of over 2-fold were considered physiologically relevant. Only fibroblast growth factor receptor-3 (FGFR3) and CXCR4 were significantly regulated by both Cx43-and Cx26-expressing cells and pursued further.
For Western blots, cells were lysed in solubilization buffer as described previously (20,24,25), and protein quantification was determined using BCA protein assay reagent (Pierce). Equal amounts of proteins were separated by 10% SDS-PAGE and transferred to nitrocellulose membranes that were subsequently immunoblotted with the indicated antibodies (anti-Cx43 and anti-FGFR3, Sigma). Specific antibody binding was determined using the West-Pico chemiluminescence detection system (Pierce). To assure equal gel loading, nitrocellulose membranes were stripped in 62.5 mM Tris-HCl, 2% SDS, and 100 mM ␤-mercaptoethanol, and blots were reprobed for the housekeeping enzyme GAPDH (Cedarlane Laboratories, Hornby, ON, Canada).

Characterization of Connexin Expression in Human Breast
Tumor Cells-We initially characterized the expression of the only known human breast connexins, Cx26 and Cx43, in three different human breast tumor cells by immunofluorescent staining and Western blotting. Endogenous Cx43-positive NRK cells were used together with HMEC to demonstrate the distribution and assembly of Cx43 into gap junction plaques (Fig.  1, A and B, arrows). In a subpopulation of Hs578T cells, Cx43 was capable of assembling into gap junction plaque-like struc-tures on the cell surface (Fig. 1C, arrows), whereas in MDA-MB-231 cells intracellular Cx43 was visible, but no gap junction plaques were observed (Fig. 1D). Immunofluorescence failed to detect Cx43 in HBL100 cells (Fig. 1E). By immunoblotting, Cx43 was detected in cell lysates from NRK, HMEC, Hs578T, and MDA-MB-231 cells but not in HBL100 cells (Fig.  1F). Phosphorylated species of Cx43 (slower migrating forms) were apparent in NRK cells, suggesting significant levels of assembled gap junctions, but the same Cx43 species typically were less abundant or undetectable in tumor cells. The expression of Cx26 protein was not detected in these three tumor cell lines by immunofluorescence (data not show).
Defective GJIC in Human Breast Tumor Cells-To examine GJIC in human breast tumor cells we microinjected live cells with 5% Lucifer Yellow and examined intercellular dye coupling (Table I). Cx43-and Cx26-deficient HBL100 cells exhibited only trace amounts of dye coupling to the first order cells (Table I). Moreover, although MDA-MB-231 cells expressed Cx43 this connexin was not assembled into gap junctions as only 6% of the microinjected cells exhibited dye transfer, and this was further restricted to first order cells (Table I). Interestingly, even though a significant number of Hs578T cells expressed Cx43 that was assembled into gap junction plaques these cells were only 47% coupled, and dye spreading was restricted to first order cells (Table I). These results suggest that this tumor cell line is quite heterogeneous with respect to GJIC. NRK cells were found to be extensively dye coupled (Table I). Together these findings suggest that each of these tumor cell lines is significantly or extensively defective in GJIC.
Retroviral Delivery of Connexin Genes Differentially Rescues GJIC in Human Breast Tumor Cells in Vitro-In an attempt to rescue the defects in GJIC we used AP2 retroviruses encoding Cx26 or Cx43 to deliver these genes into HBL100 and MDA-MB-231 cells. The retroviral system enabled an efficient targeting of exogenous connexins to both tumor cell lines with Ͼ90% of cell populations expressing the transgene after three rounds of infection. In HBL100 cells, exogenous Cx43 (Fig. 2B) or Cx26 (Fig. 2D) were assembled into gap junction plaques (arrows). Immunoblots revealed that Cx43 was detected in the cells infected by the Cx43 gene-containing AP2 virus but not in the control cells (Fig. 2F). Compared with their wild-type cell counterparts, introduction of either Cx43 or Cx26 genes efficiently rescued GJIC in HBL100 cells, resulting in over 80% of the microinjected cells exhibiting dye transfer to second or third order cells (Table I). In contrast, when Cx43 or Cx26 were expressed in MDA-MB-231 cells, these cells remained poorly coupled with dye spreading to only 8 -15% of first order cells (Table I). In many cases, Lucifer Yellow spread to only one neighboring cell when MDA-MB-231 overexpressed Cx43 or Cx26 (Fig. 3, D and E). The majority of overexpressed Cx43 in MDA-MB-231 cells remained as a single band likely representing the unphosphorylated species of Cx43 (Fig. 2E), although slower migrating phosphorylated species of Cx43 also were evident. Immunolabeling studies revealed that the exogenous Cx43 ( Fig. 2A) or Cx26 (Fig. 2C) were localized aberrantly to perinuclear intracellular compartments. Double immunolabeling revealed that Cx26 (Fig. 3, A and C) colocalized with the lysosomal-associated membrane protein LAMP-1 (Fig. 3, B and C), demonstrating that Cx26 had been delivered to lysosomes. Likewise, we showed previously that Cx43 was delivered to lysosomes in MDA-MB-231 cells (26). Consequently, the defect in MDA-MB-231 cells is not in connexin synthesis but rather in their inability to assemble functional gap junctions efficiently in vitro.
Connexin Expression in MDA-MB-231 Cells Dramatically Inhibits Tumor Cell Growth in Vivo Independently of the Formation of Appreciable Gap Junctions-In the absence of functional gap junctions, we examined the role of connexins in tumor growth control in vitro by comparing the growth kinetics of connexin-overexpressing cells and their control counterparts, including wild-type cells or cells infected with the retroviral vector that did not encode a connexin. When MDA-MB-231 cells were engineered to overexpress Cx43 or Cx26 in vitro these cells exhibited growth characteristics similar to the control (Fig. 4A). However, a dramatic inhibition of tumor growth was observed when Cx43-or Cx26-overexpressing MDA-MB-231 cells were implanted into mammary fat pad of the nude mice (Fig. 4B). In addition to the smaller tumor volume, in comparison to controls, it took more time for connexin-expressing cells to develop tumors, and in over 50% of the mice tested no measurable tumor developed. Xenograph removal and immunostaining for Cx43 revealed that Cx43-expressing MDA-MB-231 cells retained the connexin within the cell and did not assemble gap junction plaques at cell-cell interfaces (Fig. 5, C  and D), whereas only low levels of detectable Cx43 immunostaining were observed on sections of control MDA-MB-231 xenoplant tumors (Fig. 5B). As a positive control, punctate Cx43-positive gap junctions were readily observed between cardiac myocytes (Fig. 5A, arrows). Together these results suggest that connexins exert tumor suppression properties in MDA-MB-231 cells by a mechanism that is independent of appreciable GJIC.
Fibroblast Growth Factor Receptor Was Down-regulated in MDA-MB-231 Cells Overexpressing Cx43 or Cx26 -To examine the molecular mechanism of a possible GJIC-independent tumor suppression by connexins we performed DNA array analysis to screen several well characterized genes related to either tumor growth or metastasis. FGFR3 mRNA was found to be down-regulated by 2.6-fold in Cx43-overexpressing MDA-MB-231 cells in comparison to the control (Fig. 6, A and B, boxed). Similarly, Cx26-overexpressing MDA-MB-231 cells down-regulated FGFR3 by 2.3-fold in comparison to the control (data not shown). In accordance with DNA array data, Western blots revealed that FGFR3 protein expression was reduced significantly by 30 -50% in MDA-MB-231 cells engineered to overexpress Cx43 or Cx26 in comparison to both wild-type and viral control cells (Fig. 6, C and D). The DNA array analysis also revealed that the chemokine receptor, CXCR4, was down-regulated in Cx43-expressing MDA-MB-231 cells (Fig. 6, A and B   The overlay (C) revealed that Cx26 is localized primarily to lysosomes. A single MDA-MB-231 cell expressing Cx26 was pressure-microinjected with 5% Lucifer Yellow, and dye was observed to spread to one contacting cell only (D and E, asterisk). Bar, 10 m human breast may exhibit GJIC and a susceptibility to growth control or a GJIC deficiency and may be more sensitive to growth stimulation (28), suggesting a correlation between growth control and GJIC. It has been reported that there is a down-regulation of connexins, gap junctions, and/or impaired GJIC in human breast carcinomas and in other neoplastic cells (12, 29 -31), but the mechanism responsible for this downregulation remains largely unknown. We observed in this study not only that breast tumor cells can have reduced connexin expression but also that they may have impaired ability to transport, localize, and assemble connexins into gap junctions. We have characterized diverse gap junction phenotypes represented by three distinct human breast tumor cell lines. HBL100 cells displayed an acute deficiency in connexin expression, resulting in an absence of functional gap junction chan-nels. In contrast to HBL100 cells, a moderate level of Cx43 protein was detected in Hs578T cells, and these cells exhibited a significant level of GJIC. MDA-MB-231 cells represent a third phenotype in which GJIC was lacking, even though the cells expressed significant amounts of Cx43. Although no Cx26 protein was detected within these cells, it has been reported that MDA-MB-231 have low levels of Cx26 mRNA, which can be up-regulated if the cells are treated with the tumor promoter and phorbol 12-myristate 13-acetate, a protein kinase C activator (17,32). MDA-MB-231 cells are malignant and may mimic breast carcinomas much like what has been described in situ by Jamieson et al. (19), where the cells continue to produce connexins but appear to be defective in the ability to assemble gap junctions.
Together our results suggest that GJIC defects in human breast tumor cells can occur at two distinct locations. First, as a consequence of cell transformation the connexin gene may be turned off as evident by the lack of Cx43 and Cx26 in HBL100 cells. Successful and efficient rescue of GJIC in HBL100 cells upon infection with retroviruses that encode connexins would suggest that they do not have further downstream defects in the secretory pathway or in connexin assembly. These breast tumor cells are not aggressive and will not grow in nude mice. Our previous survey of human breast tumors in situ (7) suggests that several human breast tumors lacking connexins may be suitable candidates for GJIC rescue. It is likely that many breast tumors and carcinomas of other types may exhibit a shutdown of the connexin promoter (12,33). In these cases and in breast tumor cells of similar phenotype, exogenous expression of connexin or chemical up-regulation of connexins may represent a useful approach to restore GJIC with a possible therapeutic advantage of inhibiting tumor cell growth and metastatic potential.
The second type of defect in GJIC represented by human MDA-MB-231 breast tumor cells occurs when cells continue to express connexins but lack the appropriate protein folding, transport, or targeting machinery to assemble functional gap junctions efficiently. Our studies revealed that the overexpression of connexins alone in MDA-MB-231 cells did not efficiently rescue GJIC. Therefore, the rescue of GJIC in human breast tumor cells relies not only on the expression of connexins but also on more downstream machinery to properly fold, assemble, or stabilize connexins in functional gap junction channels.
Because MDA-MB-231 cells have defects in assembling both endogenous and exogenous connexins, it was not a surprise to see that connexin-expressing MDA-MB-231 cells did not exhibit significant differences in vitro growth. However, overexpression of Cx43 or Cx26 dramatically inhibited tumor formation in nude mice, resulting in a 6-fold decrease in tumor growth. Interestingly, Bond et al. (10) found that Cx32-overexpressing C6 glioma cells did not have a reduced growth rate in vitro but that growth was retarded in vivo; however, in their study GJIC was known to be highly up-regulated. Our finding here is the first report in which Cx43 and Cx26 not only exhibit differential growth effects in vitro and in vivo but also, more intriguingly, this was mediated via a GJIC-independent mech-anism or via minimal increases in GJIC. The fact that excised xenograph sections revealed only intracellular Cx43 and no detectable gap junctions ruled out the possibility that connexin-expressing MDA-MB-231 cells assembled functional gap junctions more efficiently when implanted into the milieu of the mammary fat pad. We cannot completely rule out the possibility that a second connexin was up-regulated in the xenoplant tumors, but this appears unlikely given the fact that it also should have occurred in mice transplanted with MDA-MB-231 cells infected with the empty retroviral vector.
It is becoming increasingly more evident that connexins may exert tumor growth inhibition via both GJIC-dependent and -independent mechanisms. When Cx43 was expressed in human glioblastomas, cell proliferation was reduced dramatically in vitro and in vivo in the absence of the establishment of GJIC (34). Moreover, Omori and Yamasaki (35) showed that the expression of a mutant form of Cx43 reverses the growth control properties of wild-type Cx43 in rat C6 glioma cells without affecting GJIC, suggesting a mechanism in which Cx43-mediated growth suppression is unrelated to GJIC. In the case of Neuro2a cells, expression of the carboxyl-terminal end of Cx43 alone suppressed cell growth, suggesting a possible second mechanism by which connexins may suppress tumor cell growth (36). Together with our findings, this suggests that connexins may activate signaling pathways that affect in vivo cell growth.
The question remains as to how connexin expression in the absence of substantial increases in GJIC induces a reduction in tumor cell growth. Given that our studies revealed little increase in GJIC upon high expression of Cx43 or Cx26 in MDA-MB-231 cells we propose one of three possible mechanisms. Because connexins are known to oligomerize into hemichannels early within the secretory pathway (37,38), organelle-localized or plasma membrane-distributed hemichannels possibly may be gated open to allow the passage of small molecules (i.e. calcium) that signal to the nucleus. It is well understood that connexin channels can allow the passage of secondary messengers (39 -41), and calcium has been demonstrated to regulate gene transcription (42,43). However, it remains to be demonstrated that intracellular hemichannels can be gated open. In a second model, it is conceivable that connexins, or more likely FIG. 6. FGFR3 was down-regulated in MDA-MB-231 cells engineered to overexpress Cx43 or Cx26. FGFR3 mRNA was less abundant in Cx43-overexpressing MDA-MB-231 cells in comparison to wild-type tumor cells (A and B, square outline). In accordance with the DNA array analysis, Western blot analysis revealed that Cx43 was down-regulated in cells overexpressing Cx43 or Cx26 in comparison with wild-type or viral controls (C). HeLa cells were used as an internal control for anti-FGFR3 antibody labeling. Quantification of three independent experiments revealed a 30 and 50% reduction in FGFR3 expression in comparison to the viral control when MDA-MB-231 cells overexpressed Cx43 and Cx26, respectively (p Ͻ 0.05). No statistical difference was found between wild-type and viral controls (D). connexin fragments, regulate gene transcription via possible interactions with transcription factors. The fact that the carboxyl terminus of Cx43 can regulate tumor cell growth in Neuro2a cells is consistent with this model (36). Targeting of Cx43 to the nucleus is plausible given the putative nucleartargeting sequence encoded within the carboxyl terminus, although there is no such putative nuclear-targeting sequence encoded in Cx26. Finally, this study does not exclude the possibility that minimal increases in GJIC have a dramatic effect on the inhibition of tumor cell growth in vivo. In our study, GJIC was assessed via dye coupling methodologies that may be insensitive to minor changes in the intercellular coupling status of the tumor cells. Future expression of loss-of-function Cx26 mutants in GJIC-incompetent human breast tumor cells should resolve this issue.
To further study the molecular mechanism of tumor suppression by connexin expression we applied DNA array technology to screen a series of genes closely related to tumor growth and metastasis. Western blot analysis confirmed that FGFR3 was down-regulated in both Cx43-and Cx26-expressing communication-deficient MDA-MB-231 cells, suggesting that connexins do indeed have dual functions not only in mediating cell-cell communication but also in regulating gene expression. Because our DNA array study revealed a second candidate molecule (CXCR4) down-regulated in cells expressing Cx43 or Cx26, it is quite possible that larger arrays will reveal a number of cell growth-or metastasis-related molecules that are regulated by connexin expression. Although our array study included only a few selected genes, it supports the conclusion that connexin expression can impact several factors that play key roles in regulating tumor proliferation. For instance, FGF/FGFR is a potent angiogenic factor that has been reported to promote tumor growth as well as tumor invasion, a phenotype that is blocked when this angiogenic marker is inhibited (44). The fact that connexins reduce the expression of FGFR3 raises the possibility of an important role of connexins in inhibiting paracrine regulatory loops involved in angiogenesis. Certainly additional studies are required to address this important mechanism in appropriate in vivo systems. The complementation of growth factors and the complexity of receptor/growth factor binding within the milieu of the mammary fat pad of nude mice in comparison to the simplicity of tumor cells grown in culture may explain why connexin expression inhibited tumor growth in vivo and not in vitro. These results strongly suggest that both Cx43 and Cx26 are potent tumor suppressors in vivo and emphasize the importance of examining the tumor-suppressing properties of connexins in the three-dimensional milieu of an organism. Finally, these studies highlight the possibility that it may not be a prerequisite for connexins to be assembled into functional gap junctions in order to be of potential therapeutic value in the treatment of human breast cancer. If connexins can be introduced or chemically up-regulated in breast tumors they may act as a tumor suppressor irrespective of their ability to assemble into functional gap junctions efficiently.
In summary, we have shown that human breast tumor cells in vitro have impaired or abolished GJIC. Moreover, we have identified that the defect in GJIC may occur at the level of the gene or during connexin folding, transport, and gap junction assembly. Retroviral-based delivery of connexin genes into human MDA-MB-231 breast tumor cells suppressed tumor growth in vivo, possibly caused by the down-regulation of tumor growth-facilitating factors. These results suggest that the retroviral delivery system in combination with connexin gene therapy may have future value as part of a treatment regimen for human breast cancer.