CXADR-Mediated Formation of an AKT Inhibitory Signalosome at Tight Junctions Controls Epithelial–Mesenchymal Plasticity in Breast Cancer

Tight junctions (TJ) are multiprotein complexes located at apical–lateral borders in epithelial cells. They control paracellular permeability and regulate epithelial proliferation, polarization, and differentiation (1). TJs are composed of transmembrane proteins including MARVEL domain proteins (i.e., occludin), claudins, and Ig-type proteins (i.e., JAM-A, CXADR), which form TJ-based networks with adaptor proteins like members of the zonula occludens (ZO) and the membrane-associated guanylate kinase inverted (MAGI) protein families. These adaptor proteins contain PDZ domains, which mediate protein-protein interactions and the formation of TJ-based networks.

Deregulation of TJs is a hallmark of EMT, whereby cancer cells lose epithelial characteristics and gain an increased capacity to disseminate and metastasize (2, 3). EMT is also linked to chemo-resistance (4). Recent studies indicate that EMT is a plastic process and that cancer cells may transit in and out of various degrees of epithelial and mesenchymal states. Partial EMT is characterized by loss of epithelial properties like cell–cell junctions, whereas the acquisition of mesenchymal traits is not compulsory (5). TGFβ1 is a potent inducer of EMT and is overexpressed in breast cancer (6). Remodeling of TJs is a key early state of TGFβ–induced EMT, which precedes deregulation and loss of E-cadherin. This suggests that TJs can function as so called “gate-keepers” protecting cells from undergoing EMT. As such, TJs might be useful targets to strengthen epithelial characteristics and thereby prevent, or even reverse EMT, as a new type of therapeutic strategy to inhibit cancer invasion, migration, and metastasis. However, knowledge about how TJ-based proteins and signaling networks control EMT is lacking.

TGFβ–induced EMT involves recruitment of the TGFβ type I receptor to TJs, which is mediated by occludin (7). Thus, the signaling cascade that eventually induces an EMT program downstream of TGFβ1 originates at TJs. Upon receptor stimulation, Smad transcription factors translocate to the nucleus and regulate various target genes (6, 8). However, for an EMT program to be induced, Smads need to cooperate with EMT factors including members of the Snail, Twist, and ZEB families (9, 10). These factors are regulated by non-Smad pathways, of which, AKT signaling has been attributed an important role (9, 11). A downstream target of AKT is glycogen synthase kinase 3β (GSK-3β), which in its active state targets Snail and Twist for nuclear export and degradation (9, 12, 13).

In a search for TJ-based signaling networks regulating EMT we studied the possible role of CXADR, a transmembrane component of TJs, which molecular function is not understood. CXADR was identified as a high-affinity receptor for coxsackie B and adenoviruses type C (14). CXADR engages in homophilic interactions, and has been implicated in regulating cell–cell adhesion and permeability (15). CXADR contains a PDZ-binding motif in its C-terminus and can interact with adaptor proteins, including MAGI-1, ZO-1, MUPP-1, and LNX (15–18). CXADR is distinct from other TJ proteins as it is essential for proper organ development and tissue homeostasis. Mouse embryos carrying targeted deletion of the Cxadr gene die early through impaired development of several organ systems, including the heart and the lymphatic system (19–21). Moreover, conditional deletion of Cxadr in adult mice results in dilatation of the intestinal tract and atrophy of the exocrine pancreas (22). Transmission electron microscopy examination of the intestinal epithelium of such mice revealed structural changes but no obvious changes in TJs.

CXADR is repressed during TGFβ–induced EMT (10, 23), and loss of CXADR has been linked to signs of carcinoma progression, including dedifferentiation, invasion, and poor outcome in various types of human cancer (24–26). In breast cancer, contradictory results exist with reports showing either low, elevated, or not significantly changed CXADR levels (27–30). Manipulation of CXADR in murine tumor models has revealed that CXADR act as a suppressor of tumor progression and metastasis (31–33).

Here, we used a combination of EMT models, conditional Cxadr^−/− mice and human breast cancers to gain mechanistic insight into the molecular function of CXADR, and its capacity to regulate EMT. Our studies led to the identification of a previously unknown role for CXADR in forming an AKT inhibitory signalosome at TJs. We found that in addition to CXADR, the signalosome is composed of MAGI-1, PTEN, and the pleckstrin homology domain leucine-rich repeat protein phosphatase-2 (PHLPP2). CXADR was found to play a critical role for the stability and function of the signalosome and for proper control of the AKT signaling pathway. Loss of CXADR promoted hyperactivation of AKT and TGFβ1–induced EMT. The results identify CXADR as a critical gate-keeper, which controls epithelial–mesenchymal plasticity in breast cancer.

EpH4, EpRas, and EpXT (kindly provided by H. Beug, Vienna Medical University) cells were cultured in DMEM/F-12 medium (#11330057, Gibco Cell Culture/Thermo Fisher Scientific). EpH4 cells, which are the parental cells to EpRas and EpXT cells, display epithelial features, including a cobblestone-like morphology and positive expression for E-cadherin, occludin, and CXADR (Supplementary Fig. S1A and S1B). EpRas cells were generated through stable overexpression of the H-Ras oncogene in EpH4 cells (34). EpRas cells maintain epithelial characteristics but in contrast with EpH4 cells, they are tumorigenic and sensitive to TGFβ1–induced EMT. EpXT cells were derived by ex vivo culturing of subcutaneous EpRas tumors and are stably in EMT through an autocrine TGFβ1 loop. They display an elongated morphology, express low levels of CXADR and other junction proteins, and are positive for vimentin (35). MCF-7, A549, T47D, Cos-7 and HEK-293 cells were from the ATCC and were cultured in DMEM (#Gibco, 22320030) or RPMI (#Gibco 21875034). All cells were cultured at 37°C in a humidified chamber with 5% CO₂, used for 10–15 passages and regularly stained with DAPI (Vector Labs) to test for Mycoplasma contamination. All media was supplemented with 10% heat inactivated FBS (Invitrogen) and 1% penicillin and streptomycin (PEST). EMT was induced by the addition of TGFβ1 (Peprotech) at doses of 2–10 ng/mL for designated time. For Pip3 inhibition, cells were treated with 10 nmol/L of PITenin-7 (Calbiochem/MERCK, Solna, Sweden) for 24 hours. For AKT inhibition, cells were treated with 1–2 μmol/L of MK2206 (#S1078 Selleckchem) 20 minutes before adding of TGFβ1 for indicated time courses.

Cells were lysed in RIPA buffer (Thermo Fisher #89900) supplemented with protease and phosphatase inhibitors (Invitrogen/Thermo Fisher #87785) and total protein extracts were boiled in Laemmli sample buffer (Life Technologies/Thermo Fisher #NP0007), separated by SDS–PAGE under reducing conditions and transferred to nitrocellulose using iBlot2 (Invitrogen). The membrane was blocked using Blocking reagent (Roche, #11520709001) for 1 hour in room temperature and then incubated with primary antibodies over night at 4°C. Then, the membrane was incubated with appropriate anti-IgG secondary antibodies conjugated with horseradish peroxidase antibodies for 1 hour at room temperature. All antibodies were diluted in blocking reagent (Roche). Immunoreactive bands were visualized by chemiluminescence (Roche # 11520709001 or Millipore #WBLUF0100) and developed using a LAS1000 system (Fuji Photo Film Co.) or ChemiDoc XRS+, Image Lab Software (Bio-Rad). Blots were checked for equal loading by probing with an anti-Calnexin antibody. Calnexin was used as a loading control in all Western blot analyses.

See Supplementary Table S1.

Pools of scrambled control or CXADR siRNA (Dharmacon, Thermo Fisher Scientific) were transfected at an end concentration of 50 nmol/L into cells using Dharmafect reagent according to standard protocol. Plasmids (see Supplementary Table S2) were transfected into HEK-293, EpXT and COS-7 cells using lipofectamine 2000 (Invitrogen, #11668027), and into MCF-7 and A549 cells using Lipofectamin LTX (Invitrogen/Thermo Fisher Scientific #15338100) according to the manufacturer's protocol. An expression plasmid for PHLPP2 was generated through the cloning of mouse PHLPP2 cDNA (Dharmacon) into an empty pcDNA3.1 vector.

The same number of EpH4 cells (10⁵) were seeded in triplicate in 6-cm dishes. Low (10%–20%) and high (>100%) confluency cells were collected after 48 hours and 6 days, respectively, lysed in RIPA buffer supplemented with protease and phosphatase inhibitors, and analyzed by SDS-PAGE.

Cells were trypsinized, re-suspended in growth factor-reduced (GFR) Matrigel (2 mg/mL; BD Biosciences) diluted 1:5 in DMEM with 1% FBS. 50,000 cells were seeded into the cell culture inserts (8-μm pore; Millipore). Medium with TGFβ1 (10 ng/mL) was added in the lower chamber. After 12 hours, incubation in 37°C/ 5% CO₂ incubator, non-migrated cells were removed with cotton swabs and inserts fixed in methanol. The membrane was removed and mounted with Vectashield containing DAPI (Vector Labs). Images were captured by using a Nikon Eclipes E800 microscope and the number of invaded cells was counted.

EpRas-shControl and EpRas-shCAR cells were seeded into 96-well plates. After 24 hours, cells were infected with 10 μL lentivirus encoding for shControl or ShCAR (Santa Cruz Biotechnology) and 8 μg/μL polybrene. Medium was changed after 24 hours incubation in 37°C and selection medium with 10 μg/μL puromycin was added. Cells were under selection for 1 week during that time the cells were expanded.

Cells were grown on cover slips, and fixed shortly in ice-cold absolute Acetone or 2% PFA for 15 minutes. In case of PFA fixation, cells were further incubated with 10 nmol/L glycine in PBS for 15 minutes and 0.1% Triton-X in PBS for 5 minutes. The cover slips were incubated for 1 hour in blocking solution (5% goat or donkey serum, and 0.2% BSA in PBS), and later primary antibodies were added and incubated at +4°C overnight. After incubation with primary antibodies the cells were washed adequately in PBS containing 0.2% BSA. Cells were then incubated in secondary antibody for 1 hour. The cover slips were mounted with Vectashield containing DAPI (Vector Labs). Images were captured by a Nikon Eclipes E800 microscope and Zeiss LSM700 confocal microscope.

Staining of frozen sections (OCT embedded-5 μmol/L thickness) from mouse tissues and human tumors was performed after fixation with absolute Acetone for 2 minutes, blocking in 10% negative donkey serum, and 4% BSA in PBS, followed by staining with primary antibody incubate +4°C overnight. Sections were incubated with appropriate secondary antibodies for 2 hours at room temperature, washed and mounted with Vectashield containing DAPI. For image analysis and quantification of the intensity of immunofluorescence staining in the 14 different human invasive breast carcinomas, three to five images per tumor were captured (×200 magnification), depending on tumor size. Images were imported into the ImageJ software (http://rsbweb.nih.gov/ij/). Three to five randomly selected tumor areas in each image were analyzed separately for mean fluorescence intensity (CTCF) staining of CAR and correspondently matches the exact same aria for PHLPP2, PTEN, Magi-1 and E-cadherin staining. CTCF = Integrated Density – (Area of selected cell × Mean fluorescence of background readings).

Junction localization of the CAR was analyzed by analyzing the junction staining of CAR divided by the total CTCF in each image. Image analysis was performed in a setting blinded to the patients' clinical characteristics.

Confluent cells were lysed in NP40 buffer (50 mmol/L TRIS pH 8.0, 150 mmol/L NaCl, 1% NP40 and phosphatase inhibitor supplemented with phenylmethylsulfonyl fluoride (PMSF). The antibody coupling of magnetic beads (Dynabeads; Life Technologies) was done according to the manufacturer's protocol for EpH4 and EpRas cells. Samples were mixed with coupled beads and incubated overnight in +4°C on rotor. Beads were washed in NP40 buffer and prepared by boiling beads in SDS containing loading buffer (Invitrogen). In case of HEK293 cells, in the domain identification IP, cells were transfected with correspondent plasmids. Forty-eight hours posttransfection, confluent cells were lysed in the IPB7 buffer (TEA 20 mmol/L, NaCl 700 mmol/L, Igepal CA-630 0.5%, DOC 0.2%) supplemented with halt phosphatase and protease inhibitor and PMSF. 400 μg of whole-cell lysate was incubated with primary antibody for 5 hours at +4°C then Protein A/G plus agarose (Santa Cruz #sc-2003) was added and incubate overnight at +4°C. Samples were washed in in the lysis buffer and PBS. Samples were analyzed on Western blot.

Cells (EpXT) were seeded and transfected with a cDNA-control or cDNA-CXADR using Lipofectamin 2000. Twenty-four hours post transfection cells were split and reseeded to a new 12-well plate and treated with cycloheximide (Abcam#ab120093; 12 μg/mL) or DMSO (as control). Cells were collected in RIPA lysing buffer at the indicted time points. In case of EpRas cells, shcontrol or shCXAR cells were seeded and grew overnight. The day after cells were treated with cycloheximide (50 μg/mL) or DMSO for indicated time points. Obtained total proteins lysate were subjected to Western blot analysis and blotted with proper antibodies (mentioned in the Ab-list). Western signals were quantified using ImageJ. software. Statistical analysis and T test was based on at least three independent experiments.

Cells were fixed shortly in ice-cold-Ethanol. After fixation the cells were stained with antibodies for anti-CAR, PHLPP2, PTEN or E-cad. After incubation with primary antibodies at 4°C overnight, Proximity ligation assay (PLA) was performed according to the manufacturer's protocol using the Duolink detection kit with PLA PLUS (#DUO92001or #DUO92003) and MINUS (#DUO92005 or #DUO92004) probes for mouse and rabbit (Sigma-Aldrich). Samples were mounted in DAPI (Mounting medium with DAPI for PLA assay) from Sigma-Aldrich and analyzed by fluorescence microscopy equipped with a ×40 immersion lens.

Mice with a loxP-flanked Cxadr allele (F/F mice) were generated at the MCI/ICS (Mouse Clinical Institute—Institute Clinique de la Souris, Illkirch, France) as previously described (22). F/F mice were crossed with the transgenic mouse line B6.Cg-Tg (CreEsr1)5AmC/J (The Jackson Laboratory) expressing a tamoxifen-inducible Cre–ERTM fusion protein under the control of a chicken actin/cytomegalovirus (CMV) promoter. Further breeding was performed to generate the mouse line F/F;Cre, which was back-crossed for three generations onto C57Bl/6J and then used for experiments. Deletion of the Cxadr gene was performed by injecting adult animals with tamoxifen (0.1 mg/g body weight) once a day for five consecutive days. Mice were sacrificed on day 19 after injecting the initial dose of tamoxifen. Animal experiments were performed in accordance with standards of animal care and were approved by Stockholm North Animal Ethical Board (N179/08).

Clinical samples of human ductal carcinoma in situ and invasive ductal breast cancer were obtained after breast cancer surgery and were freshly frozen and sectioned according to standard procedures. A pathologist at Umeå University Hospital, Sweden independently classified all tumors according to the status of estrogen, progesterone and HER2 receptor as well as the grade of the tumors and Ki67 expression level (see Supplementary Table S3). Written informed consent was obtained from all patients. All studies using human material were conducted in accordance with the ethical guidelines of the Helsinki Declaration of 1975, and approved by the Research Ethics Review Board of Northern Sweden (EPN).

The Gene expression-based Outcome for Breast cancer Online (GOBO) tool was used for prognostic validation of sets of genes in a pooled breast cancer data set comprising 1881-samples (http://co.bmc.lu.se/gobo/). To study how gene expression correlated with overall survival, tumors were divided into two quantiles, in which gene expression was either above, or below median. Tumors with expression levels in the lower quantile were defined as having loss of gene expression. Each of the two quantiles were compared with overall survival.

All values are presented as means ± s.e.m of three independent experiments. Significance between two groups was determined by t test and significance as assessed by ANOVA, followed by the Bonferroni's test for multiple comparisons. Linear regression analysis of mean fluorescence intensity values for CAR with PHLPP2, PTEN, MAGI-1 or E-cadherin expression in human tumors was analyzed using PrismGraphPad. A P value of <0.05 was considered significant for all statistical analysis.

To study the role of CXADR in EMT, we manipulated CXADR expression in an established model system of TGFβ1-induced EMT in mouse mammary tumor cells, either transiently induced in EpRas cells, or constitutive in EpXT cells (34). Knocking down CXADR in EpRas cells using siRNA resulted in an enhanced EMT response compared with control cells after exposure to TGFβ1 (10 ng/mL, 24 hours), as evident by more prominent repression of E-cadherin and occludin, and induction of vimentin (Fig. 1A and B). In line with this, we found that EpRas cells with stable knockdown of CXADR were more migratory than control cells in invasion assays toward a source of TGFβ1 (5 ng/mL, 12 hours; Fig. 1C). Conversely, overexpression of CXADR (48 hours) in EpXT cells promoted mesenchymal–epithelial transition (MET; Fig. 1D), and inhibited their invasive capacity (Fig. 1E) compared with control cells. The results imply that CXADR, directly or indirectly regulates epithelial–mesenchymal plasticity in mammary tumor cells through yet unknown mechanisms.

To explore the mechanisms by which CXADR controlled EMT, we analyzed whether CXADR regulates any of the signaling arms promoting EMT downstream of TGFβ1. Stable knockdown of CXADR in EpRas cells did not affect phosphorylation of Smad3 (p-Smad3), either under baseline conditions, or after TGFβ1 treatment (Fig. 2A). Neither did CXADR knockdown have any effects on phosphorylation of ERK1/2 (p-ERK1/2), or p38 (p-p38) MAP kinases. In contrast, the levels of phosphorylated AKT (p-AKT) were increased in CXADR knockdown compared with control cells, even in the absence of TGFβ1 (Fig. 2A). After TGFβ1 exposure (10 ng/mL, 1 hour), p-AKT levels were further enhanced. These results indicated that CXADR is negative regulator of the AKT signaling pathway, and that loss of CXADR may promote hyperactivation of AKT and facilitate TGFβ-induced EMT. Additional studies using MK2206 (1 μmol/L, 72 hours), an allosteric AKT inhibitor, confirmed that TGFβ1-induced EMT in EpRas cells was AKT dependent (Supplementary Fig. S2A).

To further explore the capacity of CXADR to regulate the AKT pathway, we studied how loss- and gain-of-CXADR expression affected phosphorylation of GSK-3β (p-GSK-3β), the downstream target of AKT. Increased levels of p-GSK-3β were detected in EpRas cells compared with control cells both after stable (Fig. 2B), and transient knockdown of CXADR (Supplementary Fig. S2B). Reciprocally, decreased levels of p-GSK-3β were detected in EpXT cells after overexpression of CXADR (Fig. 2C). As GSK-3β regulates the stability of Snail and Twist factors, we studied whether knockdown of CXADR affected these factors. We observed enhanced nuclear localization of Snail1 and Twist in CXADR knockdown compared with control EpRas cells upon TGFβ1 treatment (Fig. 2D–F). Additional analysis revealed that Snail1 and Twist protein levels were increased in CXADR knockdown cells, and further enhanced after TGFβ1 exposure (Supplementary Fig. S2C). Together, these results indicated that CXADR controls epithelial–mesenchymal plasticity by inhibiting the AKT/GSK-3β-Snail/Twist signaling axis.

Next, we aimed to elucidate how CXADR could be molecularly linked to AKT. As a first step, we used PITenin-7, an inhibitor of phosphatidylinositol (3,4,5)-triphosphate (PIP3), to study whether PIP3 was involved in CXADR-mediated inhibition of AKT. The results showed that PITenin-7 could only partially inhibit the hyperactivation of p-AKT, which was induced in EpRas cells by knocking down CXADR (Fig. 2G). This indicated that CXADR-mediated inhibition of AKT could involve both PIP3-dependent, and PIP-3-independent mechanisms. A mediator of PIP3-dependent inhibition of AKT is the phosphatase PTEN, which has been linked both to TJs and EMT (36, 37). We therefore speculated that PTEN might be regulated by CXADR. Indeed, compared with control cells, the levels of PTEN were decreased in CXADR knockdown EpRas cells, and increased in CXADR overexpressing EpXT cells (Fig. 2H and I). On the basis of a recent report showing that loss of another phosphatase-PHLPP2, a direct and PIP-3–independent inhibitor of AKT, is linked to hyperactivation of AKT and EMT (38), we hypothesized that also PHLPP2 might be linked to CXADR. In support of this, we detected significantly decreased levels of PHLPP2 in CXADR knockdown compared with control EpRas cells (Fig. 2H). Opposite to this, PHLPP2 levels were increased in EpXT cells after overexpression of CXADR (Fig. 2I). Further results showed that the levels of PHLPP2 and PTEN were lower in CXADR-low EpXT cells compared with EpH4 cells and EpRas cells (Supplementary Fig. S2D). Together, these results suggested that PHLPP2 and PTEN could function as mediators of AKT inhibition downstream of CXADR.

As a next step, we performed immunofluorescence staining to explore whether PHLPP2 and PTEN localized to TJs, similar to CXADR. The results showed that PHLPP2, and PTEN colocalized with CXADR at TJs in EpRas cells, and in EpH4 cells (Fig. 3A; Supplementary Fig. S3A). Staining for the adaptor protein MAGI-1, a protein known to interact with CXADR at TJs, was used as a positive control. In comparison, PHLPP2 and PTEN localized mostly to the cytoplasm in EpXT cells, in which TJ staining for CXADR and MAGI-1 was diffuse and discontinuous (Fig. 3B). These results showed that PHLPP2 and PTEN colocalize with CXADR at TJs in mammary epithelial cells.

To study whether PTEN and PHLPP2 could physically interact with CXADR at TJs, we performed proximity ligation assays. Positive signals—mostly concentrated at cell junctions, were detected for CXADR + PHLPP2, as well as CXADR + PTEN in EpH4 cells (Fig. 4A). Combinations of CXADR + MAGI-1, or CXADR + E-cadherin antibodies were used as positive and negative controls, respectively. On the basis of reports showing that PTEN can interact with MAGI proteins (39) and that PHLPP2, similar to CXADR and PTEN, contains a C-terminally located PDZ-binding motif (40), we hypothesized that MAGI-1, which contains 6 PDZ domains (Fig. 4B), could serve as an adaptor for a protein complex containing CXADR, PHLPP2, and PTEN. In support of this, we found that coimmunoprecipitation with an antibody against MAGI-1 was sufficient to pull down endogenous CXADR, PHLPP2 and PTEN in cell lysates from EpRas cells (Fig. 4C).

Previous, independent studies have shown that CXADR interacts with PDZ domain 3, and PTEN with PDZ domain 2 of MAGI-1 (18, 37). These results were confirmed after overexpression in HEK-293 cells (Supplementary Fig. S4A and S4B). To study the capacity of PHLPP2 to interact with MAGI-1 we co-expressed wild-type (WT) PHLPP2 together with HA-tagged, either WT or mutants of MAGI-1, each lacking one of the PDZ domains, in HEK-293 cells. Coimmunoprecipitation revealed that PHLPP2 was capable of interacting with WT MAGI-1, and all MAGI-1 mutants except for the one lacking PDZ domain 3 (Fig. 4D).

Together, the results identified CXADR as a membrane anchor for a TJ-based network, which contains MAGI-1 and the two phosphatases and AKT inhibitors PHLPP2 and PTEN (Fig. 4E). PHLPP2, which directly inactivates AKT through dephosphorylation, was found to interact with the same PDZ domain (#3) of MAGI-1 as CXADR. In contrast, PTEN, which acts upstream of AKT, preventing its activation by dephosphorylating PIP3, binds to PDZ domain #2 of MAGI-1. This suggests that CXADR forms a complex, which both prevents and reverses AKT activation, and thus act on dual levels.

Considering the data, we hypothesized that CXADR might stabilize the other components of the network at TJs. Similar to PHLPP2 and PTEN, MAGI-1 levels were decreased in CXADR knockdown EpRas cells (Supplementary Fig. S4C). Results from cycloheximide chase assays showed that CXADR knockdown in EpRas cells resulted in a significant loss of PHLPP2 protein stability over a time period of 33 hours (Fig. 5A and B). Similar to this, PTEN and MAGI-1 stability was reduced in CXADR knockdown cells (Supplementary Fig. S4D). Overexpression of CXADR in EpXT cells resulted in a significantly increased PHLPP2 protein stability (Fig. 5C and D) whereas PTEN and MAGI-1 were less affected (Fig. 5C; Supplementary Fig. S4D). On the basis of the fact that stabilization of TJs is linked to high cell confluency and contact inhibition, we reasoned that the CXADR network might be regulated by cell confluency. Indeed, dramatically increased levels of CXADR, PHLPP2, PTEN and MAGI-1 were detected in confluent versus sub-confluent EpH4 cells, paralleling inhibition of AKT (Supplementary Fig. S4E).

Furthermore, we analyzed whether CXADR manipulation affected the levels of phosphorylated PTEN (p-PTEN, Ser380/Thr382/383), which is an unstable and closed (inactive) form of PTEN (41). We found that although the total PTEN levels were reduced in EpRas cells after CXADR knockdown and increased in EpXT cells after CXADR overexpression, the levels of p-PTEN showed the opposite pattern with increased levels in EpRas cells and decreased levels in EpXT cells after loss- and gain-of-CXADR, respectively (Fig. 5E). To study the importance of CXADR for function of PHLPP2 and PTEN, we performed experiments with COS-7 cells, which are CXADR negative. Overexpression of PHLPP2, individually or together with GFP-MAGI-1, had minor effects on AKT phosphorylation in COS-7 cells (Fig. 5F). However, when CXADR was co-expressed with PHLPP2 and GFP-MAGI-1, p-AKT was potently inhibited. A similar effect was detected when CXADR was co-expressed with PTEN (HA-tagged) and GFP-MAGI-1 (Fig. 5G). Immunostaining indicated that HA-PTEN was recruited to cell junctions in COS-7 cells when overexpressed together with CXADR (Supplementary Fig. S4F). Together, these results demonstrated that CXADR plays a critical role in regulating the stability and function of an AKT inhibitory signalosome at TJs.

It has been reported that conditional deletion of Cxadr in adult mice results in intestinal dilatation and subtle structural changes of the epithelium in the small intestine (22). However, molecular insight into how the intestinal phenotype is linked to CXADR deficiency is lacking. On the basis of our results, we hypothesized that these phenotypic changes could be linked to destabilization of the AKT inhibitory signalosome. Immunostaining of sections of the small intestine from WT mice showed colocalization of CXADR with PHLPP2 and MAGI-1 in intestinal epithelial cells (Supplementary Fig. S5A). In contrast, PTEN was mostly concentrated in a peri-junctional area, and only showed partial co-localization with CXADR. As expected, CXADR staining was not detected in the small intestine from conditional Cxadr^−/− mice (Supplementary Fig. S5B). PHLPP2 and MAGI-1 were detected but were not concentrated at TJs as in WT mice. Even PTEN was less concentrated at cell junctions Cxadr^−/− mice. The results indicated that CXADR plays a similar architectural role in intestinal epithelial cells as in mammary epithelial cells in controlling the stability of a TJ-based complex composed of MAGI-1, PHLPP2, and PTEN.

Finally, we were interested to study whether CXADR was linked to deregulation of the AKT inhibitory signalosome and epithelial–mesenchymal plasticity in human breast cancer. We analyzed the expression of CXADR mRNA in the GOBO database of 1881 samples of human breast cancer (42). CXADR expression did not correlate with overall survival when all tumors were included (Supplementary Fig. S6A). However, when tumors were separated into luminal (A/B) and basal breast cancers, a significant and exclusive correlation (P = 0.00244) was found between low CXADR expression and poor survival in luminal A tumors, but not in basal tumors (Fig. 6A; Supplementary Fig. S6B). In comparison, E-cadherin (CDH1) was not associated with survival in any of the breast cancer subtypes (Supplementary Fig. S6C–S6E). The expression of other transmembrane components of TJs including Claudin 3 (CDLN3), JAM-A (F11R), and occludin (OCLN), was not associated with survival in luminal A tumors (Supplementary Fig. S6F–S6H). The expression of MAGI1 mRNA did not correlate with survival in any of the tumors (Supplementary Fig. S7A and S7B). Similar to CXADR, low expression of PTEN mRNA correlated with poor survival in luminal A (P = 0.03621), but not in basal tumors (Supplementary Fig. S7C and S7D). On the contrary, high expression of PHLPP2 correlated with poor survival in luminal A (P = 0.00269), but not in basal tumors (Supplementary Fig. S7E and S7F).

Furthermore, we analyzed the expression of CXADR in 51 human breast cancer cell lines previously classified as luminal or basal (A or B) tumor cells (Supplementary Fig. S8A; ref. 43). Overall, the expression of CXADR was lowest in basal B cells, which display EMT properties. Among luminal breast cancer cells, substantial heterogeneity in CXADR expression was observed. Loss- and gain-of-function experiments were performed to determine whether CXADR levels affected AKT signaling and epithelial characteristics in MCF-7 and T47D cells, luminal breast cancer cells expressing low and high levels of CXADR, respectively. Overexpression of CXADR in MCF-7 cells resulted in potent inhibition of AKT, increased expression of PHLPP2 and MAGI-1, and promoted an N-to-E-cadherin switch (Fig. 6B). PTEN levels were less affected. Immunofluorescence staining demonstrated that overexpressed CXADR localized to junctions in MCF-7 cells (Supplementary Fig. S8B). PHLPP2, which localized to the cytoplasm in mock-transfected MCF-7 cells, and MAGI-1, colocalized with CXADR at junctions in CXADR-overexpressing MCF-7 cells, whereas junction localization of PTEN was less evident. Knocking down CXADR in T47D cells resulted in increased activation of AKT and decreased levels of PHLPP2, PTEN, and E-cadherin (Fig. 6C). Together, these results showed that CXADR levels affect AKT signaling and epithelial vs. mesenchymal characteristics in human luminal breast cancer cells. In line with this, we found that AKT inhibition through MK2206 (2 μmol/L, 48 hours) resulted in a N- to E-cadherin switch in MCF-7 cells, and thus mimicked the effect of CXADR overexpression (Supplementary Fig. S8C).

To study whether this concept was applicable to other types of human cancer, we overexpressed CXADR in lung A549 adenocarcinoma cells, another frequently used model of TGFβ1-induced EMT. Similar to the effects observed in MCF-7 cells, overexpression of CXADR promoted increased expression of PHLPP2 and MAGI-1, inhibited AKT, and promoted a more epithelial phenotype (Supplementary Fig. S8D and S8E). The expression of PTEN was also enhanced in A549 cells after overexpression of CXADR. In addition, overexpression of CXADR partially rescued the repression of E-cadherin, and the induction vimentin and Twist in A549 cells exposed to TGFβ1 (Supplementary Fig. S8E). Collectively, these results showed that overexpression of CXADR can stabilize the signalosome, inhibit AKT signaling, strengthen epithelial characteristics and protect against EMT in both breast and lung cancer cells.

To get more insight into the expression and localization of CXADR in human breast cancer, we performed coimmunostaining for CXADR and PHLPP2, PTEN, or MAGI-1 in 14 human breast tumors classified as luminal A or basal breast cancer. Staining intensity for CXADR showed a high degree of both inter- and intra-tumoral variation, and no significant differences were detected between luminal A and basal breast cancer (Fig. 6D). Similar patterns were observed for PHLPP2, PTEN and MAGI-1 (Supplementary Fig, S9A–S9C), and E-cadherin (Supplementary Fig. S9D). Staining of CXADR correlated significantly with PHLPP2 (P < 0.0001), MAGI-1 (P < 0.0001), PTEN (P < 0.001), and E-cadherin (P < 0.0001) in different regions of luminal A tumors (Fig. 6F and G; Supplementary Fig. S9E). Furthermore, the variability in CXADR staining was not only related to intensity but also to localization. CXADR staining was diffuse and less concentrated at cell junctions in basal versus luminal A breast cancer (Fig. 6E and F; Supplementary Fig. S9F and S9G).

In this study, we identified CXADR as a critical regulator of epithelial–mesenchymal plasticity in breast cancer. This was linked to a previously unknown function for CXADR in controlling the AKT signaling pathway. AKT is known to promote EMT by inhibiting GSK-3β, which results in stabilization and nuclear retention of Snail and Twist. Thus, the activity of AKT is a key determining factor for the capacity of cells to undergo EMT. Our findings demonstrate that AKT activity is controlled from TJs by CXADR-mediated formation of a signalosome composed of the adaptor protein MAGI-1, and the phosphatases PHLPP2 and PTEN. The stability and function of PHLPP2 and PTEN was critically dependent on CXADR as loss of CXADR resulted in loss of PHLPP2 and PTEN, hyperactivation of AKT, and sensitized cells to TGFβ-induced EMT. Conversely, restoration of CXADR stabilized PHLPP2 and in some cells also PTEN, inhibited AKT signaling, promoted epithelial characteristics and inhibited EMT in breast and lung cancer cells. Together, these data demonstrate that CXADR mediates preventive and reversing mechanisms of AKT inhibition.

From a cellular point of view, it may be strategic to provide CXADR with a leading role in controlling AKT. First, this links AKT inhibition to the formation of cell–cell interactions and contact inhibition of cell proliferation and migration. CXADR-deficiency during development results in subcutaneous edema, which is associated with deformed and dilated lymphatic vessels (21). Moreover, conditional deletion of CXADR in adult mice causes dilatation of the intestinal tract due to unknown mechanisms (22). The results presented here provide a clue that these defects are linked to deregulation of the AKT signaling pathway.

Second, by giving CXADR a leading role, AKT signaling is negatively balanced at an intracellular localization, where the EMT response is initiated by activated TGFβ receptors (7). By operating at the top of the signaling chain the signalosome can negatively balance to what extent AKT is activated at a given time point. In normal cells, CXADR-mediated formation of the signalosome mediates a break on AKT, and importantly, prevents AKT from being hyperactivated even in the presence of stimulatory signals, such as TGFβ1. In line with this, it is known that the capacity of TGFβ1 to induce EMT is significantly inhibited by the formation of cell–cell contacts (44).

Third, by establishing a CXADR/MAGI-1 scaffold at TJs, PHLPP2 and PTEN can be rapidly recruited to TJs when AKT inhibition is needed. It will be of interest to further elucidate how the recruitment of PHLPP2 and PTEN to CXADR/MAGI-1 is regulated. Posttranslational mechanisms likely control how intracellular pools of PTEN and PHLPP2 are shuttled to TJs. Throughout our studies, PHLPP2 was more dependent on CXADR than PTEN. Considering that PHLPP2 interacts with the same PDZ domain of MAGI-1 as CXADR, it is tempting to speculate that there is a structural prerequisite for this. However, another plausible explanation is that the recruitment of PHLPP2 is more crucially dependent on the interaction with the CXADR/MAGI-1 complex, whereas PTEN is more promiscuous and interacts also with other junction-based complexes. On this notion, it is known that the PDZ-binding motif of PTEN can interact with various adaptor proteins (45), and also with the E-cadherin/β-catenin (46). Cell type–specific signalosomes may also exist. For example, both PTEN and PHLPP have been shown to interact with the adaptor protein Na(+)/H(+) exchanger regulatory factor 1 (NHERF1) in glioma cells, and are downregulated in high-grade gliomas (47). Intriguingly, our earlier studies show that glioma progression also is associated with loss of CXADR expression (48). It will be of interest to study if CXADR acts as a membrane anchor for PHLPP2 and PTEN, and controls AKT signaling also in glioma cells.

We found it interesting that loss of CXADR mRNA correlated with poor survival in luminal A breast cancer, but not in basal breast cancer. At a first glance, these results appeared counterintuitive since basal tumors are less differentiated, and express more EMT markers than luminal tumors. However, considering the results from our mechanistic studies, the implications could be that loss of CXADR in luminal tumors may promote dedifferentiation and EMT, features already present in basal tumors. Similar to CXADR, low expression of PTEN correlated with poor survival in luminal A breast cancer, but not in basal breast cancer. This is in line with previous studies showing that PTEN levels affect the fate of luminal breast cancers (49). On the contrary, high PHLPP2 mRNA levels were associated with poor survival in luminal A breast cancer. This may be interpreted as a discrepancy compared with the other results. However, in contrast with CXADR, which is an epithelial cell marker (50), PHLPP2 is expressed in various cell types and its expression levels in tumors may therefore be more difficult to interpret. For example, PHLPP2 is highly expressed in regulatory T cells (Tregs; ref. 51), and infiltration of such cells is linked to poor prognosis in ER⁺ breast cancer (52).

Immunostaining revealed important aspects of CXADR protein expression and localization in human breast cancer. Even in tumors classified as luminal A breast cancer by classical markers, regions were found in which CXADR expression was low and associated with low levels of PHLPP2, PTEN, and EMT features. This emphasizes the importance of heterogeneity in tumor characterization. Another aspect was that CXADR was not only deregulated at the expression level, but also at the level of localization. This was more commonly observed in basal tumors. Protein localization may provide important information about the properties of tumor cells, and to what extent they have the capacity to invade the surrounding tissue and metastasize, and will be an interesting aspect to study more in the future.

Our results provide a novel mechanistic link between deregulation of TJs, hyperactivation of AKT and EMT. Although mutations in genes regulating the AKT pathway are frequently observed in patients with metastatic tumors, genetic loss of PTEN is only found in 5% of breast cancers (53). Instead, loss of PTEN immunoreactivity is significantly more common and detected in nearly 40% of breast tumors. Thus, protein destabilization due to posttranslational mechanisms including loss of CXADR, play important roles for the loss of PTEN and PHLPP2 during tumor progression. Studies on larger cohorts may provide information on whether loss of CXADR could be used as a marker to predict malignant progression in luminal breast cancer. Finally, our results may open up possibilities to target CXADR to modulate AKT signaling and EMT in cancer cells.

No potential conflicts of interest were disclosed.

Conception and design: A. Nilchian, J. Johansson, A. Ghalali, S. Travica, O. Rosencrantz, C.T. Vincent, U. Stenius, J. Fuxe

Development of methodology: A. Nilchian, J. Johansson, A. Ghalali, S. Travica, O. Rosencrantz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Nilchian, J. Johansson, S. Travica, M. Sund

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Nilchian, J. Johansson, S. Travica, A. Santiago, O. Rosencrantz, C.T. Vincent, M. Sund, J. Fuxe

Writing, review, and/or revision of the manuscript: A. Nilchian, J. Johansson, A. Ghalali, C.T. Vincent, M. Sund, J. Fuxe

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Nilchian, J. Johansson, A. Ghalali, K. Sollerbrant

Study supervision: U. Stenius, J. Fuxe

We thank Dr. Yuichi Ono at the KAN Research Institute at Kobe MI R&D Center in Japan for providing the MAGI-1 constructs and Dr. Joseph Zabner from the University of Iowa for providing the GFP-MAGI-1 plasmid. J. Fuxe is supported by the Swedish Research Council (2018-3114), the Swedish Cancer Society (CAN 2015/773), and the Karolinska Institutet. C.T. Vincent is supported by the Swedish Research Council (2017-03056).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.