E-cadherin inhibits nuclear accumulation of Nrf2: implications for chemoresistance of cancer cells

An excess of reactive oxygen species (ROS) or reactive nitrogen species (RNS) causes injuries on molecules and components of the cell. As an adaptive response, NF-E2-related factor-2 (Nrf2), which is a member of the cap’n’collar (CNC) family of basic leucine zipper proteins, regulates antioxidant gene expression. Therefore, Nrf2 might serve as a transcription factor responsible for the induction of detoxifying enzymes and consequently for the prevention of chemical carcinogenesis (Jeong et al., 2006; Satoh et al., 2010). However, there is controversy on the role of Nrf2 in the progression of cancer; a series of studies have shown that mutational activation of Nrf2 might cause malignancy and increase chemoresistance (Ikeda et al., 2004; Lau et al., 2008; Wang et al., 2008). Thus, Nrf2 is somewhat of a double-edged sword in cancer biology. Because the turnover and activation processes of Nrf2 are linked to cellular molecule(s) and component(s) that have differential binding affinities, the upstream regulators of Nrf2 expressed in a cell-type-specific manner need to be identified, particularly in association with cancer resistance.

Cadherins are the proteins responsible for cell–cell adhesion at the adherens junctions, and their expression affects cell morphogenesis, rearrangements, migration and epithelial–mesenchymal transition (EMT) (Gumbiner, 2005). Cadherin switching from E-cadherin to N-cadherin is a key feature of the EMT process: the loss of E-cadherin and/or transcriptional repression of bone morphogenetic protein 7 (BMP-7) and inhibitor of DNA binding 2 (Id2) represent EMT. Loss of E-cadherin might also be associated with the resistance of cancer cells to chemotherapy because it promotes cancer progression and metastasis (Arumugam et al., 2009; Huber et al., 2005; Wheelock et al., 2008). Oxidative stress disrupts the E-cadherin and catenin cell-adhesion complex (Parrish et al., 1999; Rao et al., 2002). Thus, it is plausible that E-cadherin might affect Nrf2 activity in response to oxidative stress. This concept is particularly important because a mutational increase in Nrf2 activity might stimulate tumor growth and metastasis. However, the link between E-cadherin and modulation of Nrf2 activity has never been explored.

Catenins link E-cadherin to the actin cytoskeleton in cell adhesion (Gumbiner, 2005). In addition, E-cadherin regulates actin assembly and organization at cadherin-mediated cell–cell adhesion sites through catenins (Drees et al., 2005). In particular, the N-terminal region of β-catenin binds with α-catenin, which interacts both indirectly and directly with the actin cytoskeleton (Wheelock and Johnson, 2003). β-Catenin interacts with the distal region of the cytoplasmic domain of E-cadherin. Previously, we reported that depolymerization of actin filaments in response to oxidative stress causes a complex of Nrf2 and actin to translocate into the nucleus to effect transactivation of target genes (Kang et al., 2002). Moreover, as a negative regulator of Nrf2, kelch-like ECH-associated protein 1 (Keap1) has the ability to bind actin filaments as a scaffold protein in the cytoplasm under unstressed conditions (Kang et al., 2004). Because E-cadherin regulates actin dynamics, it might affect the activity of Nrf2 through its binding to actin and/or Keap1.

In view of the importance of E-cadherin for the signaling of cell growth and transdifferentiation, and the lack of clear understanding of the link between E-cadherin and Nrf2, this study investigated whether E-cadherin affects Nrf2 activity and if so, what the molecular basis is. Our findings revealed a novel inhibitory effect of E-cadherin on Nrf2, also showing that E-cadherin might enhance the interaction between Nrf2 and Keap1. Moreover, the results of this study identified β-catenin as a molecule necessary for complex formation of E-cadherin with Nrf2 and Keap1. Our data presented here reveal the ability of E-cadherin to recruit Nrf2, which creates an inactive pool of Nrf2 in association with β-catenin binding.

First, we sought to determine the effect of enforced E-cadherin expression on the activity of Nrf2 in HEK293 cells. Immunoblot analysis showed that overexpression of E-cadherin prevented nuclear accumulation of Nrf2 (Fig. 1A). Overexpressed Nrf2 had a molecular mass of ∼110 kDa (supplementary material Fig. S1A); endogenous Nrf2 was detected at ∼66 kDa and ∼110 kDa, although its predicted molecular mass is 66 kDa, presumably as a result of post-translational modification (Taguchi et al., 2011). To functionally assess the effect of E-cadherin on Nrf2 activity, we examined the expression of NAD(P)H quinone oxidoreductase 1 (NQO1) and hemeoxygenase-1 (HO-1) (Fig. 1B). As expected, overexpression of E-cadherin reduced the mRNA and protein levels of Nrf2 target genes. In addition, E-cadherin expression decreased the basal or tert-butylhydroquinone (t-BHQ)-inducible luciferase expression from the pGL-1651 reporter gene construct, which contains an antioxidant response element (ARE) in the promoter region (Fig. 1C, left). E-cadherin also prevented the Nrf2-induced pGL-1651 luciferase activity (Fig. 1C, right). The lack of inhibition of luciferase expression by E-cadherin from pGL-1651-ΔARE, a reporter construct with a deletion mutation in ARE, verified the role of Nrf2 in reporter gene expression. In HepG2 cells, a cell line that has a higher level of endogenous E-cadherin than HEK293 cells (supplementary material Fig. S1B), E-cadherin knockdown using siRNA significantly enhanced the basal or sulforaphane (5 μM, for 6 hours)-inducible nuclear accumulation of Nrf2 (Fig. 1D). Consistently, luciferase activity from the NQO1-ARE reporter construct was increased by a deficiency in E-cadherin, confirming the impact of endogenous E-cadherin on the activity of Nrf2. All of these results indicate that E-cadherin represses the activity of Nrf2 by inhibiting nuclear accumulation of Nrf2.

To explore the possibility of an interaction between E-cadherin and Nrf2, the ability of E-cadherin to bind Nrf2 was determined in HEK293 cells in which E-cadherin was overexpressed. Ectopically expressed E-cadherin bound with Nrf2 (Fig. 2A); as expected, Nrf2 interacted with Keap1. Of note, E-cadherin was also capable of interacting with Keap1, a protein that sequesters Nrf2 for its ubiquitylation and facilitates proteasomal degradation of ubiquitylated Nrf2 (Kobayashi et al., 2004). In HepG2 cells, the endogenous E-cadherin also interacted with Nrf2 and Keap1 (Fig. 2B). Moreover, in vitro GST pull-down assays confirmed the ability of in vitro translated [³⁵S]E-cadherin to bind Nrf2, as did [³⁵S]Keap1 (Fig. 2C, middle). In addition, we verified the ability of E-cadherin to directly bind with Keap1 (Fig. 2C, right). These results indicate that E-cadherin has the ability to bind Nrf2 and Keap1, presumably facilitating the interaction between Nrf2 and Keap1.

Immunocytochemical analysis showed that ectopically expressed Nrf2 was predominantly located in the nuclei but that simultaneously expressed E-cadherin prevented nuclear accumulation of Nrf2 (Fig. 2D), confirming the functional role of E-cadherin in inhibiting nuclear localization of Nrf2. As expected, endogenously expressed E-cadherin in the plasma membrane of HepG2 cells also interacted with Nrf2 (Fig. 2E). Collectively, our findings show that E-cadherin prevents nuclear accumulation of Nrf2 as a consequence of their interaction close to the plasma membrane.

The C-terminus of Nrf2 comprises Neh1 and Neh3 domains that are responsible for heterodimerization of Nrf2 and small Maf and DNA binding (Itoh et al., 2010). To locate the region of Nrf2 responsible for its interaction with E-cadherin, immunoprecipitation and immunoblot assays were performed in cells transfected with truncated mutants of Nrf2 in combination with E-cadherin. Our findings indicated that the C-terminus, but not the N-terminus, of Nrf2 was capable of binding E-cadherin (Fig. 3A). PKC∂ phosphorylates Nrf2 at Ser40 residue, which is located in the Neh2 domain of the N-terminus, and this activation process is reciprocally coupled with ubiquitylation and degradation of Nrf2 by Keap1 (Niture et al., 2009). We wondered whether the phosphorylation of Nrf2 is affected by E-cadherin. As expected, t-BHQ treatment promoted Nrf2 activation, as indicated by a decrease in the binding of Keap1 and Nrf2 (Fig. 3B). However, the intensity of E-cadherin and Nrf2 binding was not diminished. Moreover, Ser40 to alanine (S40A) mutation of Nrf2 failed to affect this effect, suggesting that E-cadherin and Nrf2 binding might not be coupled with the phosphorylation of Nrf2. This hypothesis was further supported by our result that E-cadherin inhibited luciferase expression from pGL-1651 elicited by an S40A mutant of Nrf2 (Fig. 3C). In addition, we observed that endogenous E-cadherin had the ability to bind Nrf2 in HepG2 cells treated with t-BHQ or sulforaphane (Fig. 3D), supporting our proposal that E-cadherin binds Nrf2 independently of Nrf2 phosphorylation.

E-cadherin contains domains that interact with p120-catenin and β-catenin. To further understand the molecular basis of E-cadherin and Nrf2 binding, we measured the abilities of several mutant constructs of E-cadherin to bind Nrf2 and inhibit luciferase expression from the NQO1-ARE reporter containing three copies of ARE. Either a C-terminal fragment of E-cadherin (ECAD-CT) or a mutant of E-cadherin in the p120-catenin-binding domain (ECAD-CT-Δp120-catenin) interacted with Nrf2 (Fig. 4A). However, E-cadherin deficient in the β-catenin-binding domain (ECAD-CT-Δβ-catenin) showed a decreased binding with Nrf2. Likewise, ECAD-CT or ECAD-CT-Δp120-catenin decreased the Nrf2 induction of luciferase from NQO1-ARE to a similar extent as did full-length E-cadherin (i.e. luciferase activity from pGL-1651) (Fig. 4B). The induction of luciferase by Nrf2 was significantly higher in the cells transfected with ECAD-CT-Δβ-catenin than those transfected with either ECAD-CT or ECAD-CT-Δp120-catenin. Similarly, the percentage of cells showing Nrf2 localization in nuclei or nuclei+cytoplasm was 16.8%, 18.0% and 40.6% after transfection of the cells with ECAD-CT, ECAD-CT-Δp120-catenin and ECAD-CT-Δβ-catenin, respectively (Fig. 4C). All of these results indicate that the β-catenin-binding domain is associated with binding between E-cadherin and Nrf2.

We next examined whether β-catenin affects the binding of E-cadherin with Nrf2. siRNA knockdown of β-catenin decreased the intensity of E-cadherin and Nrf2 binding (Fig. 5A). Consistently, a deficiency in β-catenin relieved Nrf2 from repression by E-cadherin, as shown by the nuclear accumulation of Nrf2 as well as the induction of its target gene (Fig. 5B,C). Nrf2 is degraded by the 26S proteasome system after multiple ubiquitylation (Nguyen et al., 2003). In this process, Keap1 is required for formation of the proteasome complex that degrades Nrf2. In HepG2 cells, β-catenin bound endogenous E-cadherin, Nrf2 and Keap1 (Fig. 5D), supporting the notion that these molecules form a quaternary complex. To precisely understand the molecular interactions, we tested whether β-catenin mediates the interaction of E-cadherin and Nrf2 in conjunction with Keap1 ubiquitylation of Nrf2. Interestingly, knockdown of β-catenin markedly decreased the ubiquitylation of Nrf2 (Fig. 5E, left). In the absence of E-cadherin, knockdown of β-catenin had no effect on Nrf2 ubiquitylation (Fig. 5E, middle). As expected, knockdown of Keap1 decreased Nrf2 ubiquitylation (Fig. 5E, right). In HepG2 cells, a deficiency in E-cadherin caused a decrease in Nrf2 ubiquitylation (Fig. 5F, top), and diminished β-catenin binding to either Nrf2 or Keap1 (Fig. 5F, bottom). Our results indicate that E-cadherin interacts with Nrf2 through its β-catenin binding, which might facilitate the ubiquitylation of Nrf2 by Keap1.

Treatment with transforming growth factor β1 (TGFβ1) inhibits E-cadherin expression through a process of cadherin switch (Zavadil and Böttinger, 2005). To functionally assess the effect of E-cadherin loss on Nrf2 activity, we determined whether a decrease in E-cadherin has an impact on the transcriptional activity of Nrf2. In HepG2 cells, TGFβ1 treatment (5 μg/μl, for 48 hours), which caused a loss of E-cadherin, increased the content of Nrf2 in the nuclear fraction, and significantly enhanced NQO1-ARE reporter gene activity (Fig. 6A). Moreover, sulforaphane treatment (5 μM, for 6 hours) further increased the activity of Nrf2 in cells treated with TGFβ1, compared with untreated cells.

Finally, we compared expression of E-cadherin and the nuclear accumulation of Nrf2 in a set of hepatocarcinoma cell lines. In these cells, E-cadherin levels were reciprocal to the levels of nuclear Nrf2 (Fig. 6B, left). It is now accepted that Nrf2 activation in cancer cells contributes to their resistance to chemotherapeutic agents (e.g. doxorubicin and cisplatin) (Lau et al., 2008; Shim et al., 2009; Wang et al., 2008). As expected, the SNU449 cell line, which had the lowest level of E-cadherin was the most resistant to doxorubicin treatment (0.3–3 μM, for 12 hours) (Fig. 6B, right). Levels of nuclear Nrf2 were the highest in SNU449 cells. Overall, resistance of the cell lines to doxorubicin paralleled not only E-cadherin repression, but also the level of Nrf2 expression. Functionally, E-cadherin knockdown decreased the sensitivity of HepG2 cells to doxorubicin and induced Nrf2 target genes (Fig. 6C).

EMT is a fundamental biological process that involves not only physiological conditions such as embryonic development, but also pathological changes including cancer progression and organ fibrosis (Arumugam et al., 2009; Huber et al., 2005; Wheelock et al., 2008). Cells that have passed through EMT exhibit a higher level of ROS production (Felton et al., 2009). In this process, persistent oxidative stress might induce adaptive responses within the tumor cell, which might increase resistance to apoptosis. Nrf2 is a key regulator determining cell survival or death in stressful situations, and might help cancer cells to survive in a deleterious environment. In cancer biology, the upstream regulators of Nrf2 remain to be established, particularly in association with tumor malignancy because the degree of Nrf2 activation might vary depending on the transdifferentiation status of the cell (Ikeda et al., 2004; Lau et al., 2008; Wang et al., 2008). Here, we demonstrate for the first time that E-cadherin has an inhibitory effect on Nrf2 by making it inactive, which might result from the interaction of E-cadherin with Nrf2.

The cadherins consist of five superfamily members: classical cadherins, desmosomal cadherins, atypical cadherins, proto-cadherins and cadherin-related signaling proteins (Gumbiner, 2005). Classical cadherins comprising E-(epithelial), N-(neuronal), P-(placental) and VE-(vascular-endothelial) types have an extracellular domain for cell–cell adhesion, a transmembrane domain, and a cytoplasmic tail that binds intracellular catenins (Angst et al., 2001). In the EMT process, cadherin switch from E-cadherin to N-cadherin is a key feature. Our results from GST pull-down and pGL-1651 reporter assays showed that E-cadherin interacts with Nrf2, repressing the activity of Nrf2 by inhibiting its nuclear translocation. Moreover, E-cadherin bound the C-terminus of Nrf2, which contains cap’n’collar, DNA-binding domain, leucine zipper and Neh3 domain (Itoh et al., 2004). Although the functional effect of E-cadherin on the nuclear localization signal of Nrf2 remains to be elucidated, the ability of E-cadherin to interact with the C-terminus is likely to prevent nuclear accumulation of Nrf2 because Nrf2 has a putative nuclear localization signal and a nuclear export signal at its C-terminus (Jain et al., 2005). In the present study, N-cadherin had only a minor effect. N-cadherin transfection decreased the expression of luciferase from pGL-1651, but the degree of Nrf2 inhibition by N-cadherin was much lower than that by E-cadherin (supplementary material Fig. S2A). It also diminished luciferase expression from pGL-1651-ΔARE (supplementary material Fig. S2A), suggesting that the minor inhibitory effect of N-cadherin is due to another molecule and/or component rather than Nrf2.

In the present study, the ability of t-BHQ to activate Nrf2 was confirmed by measuring the release of Myc–Nrf2 from Keap1. The lack of change in overexpressed Nrf2 by t-BHQ (Fig. 3B) might be due to an experimental condition using exogenous Myc–Nrf2 and MG132, a proteasomal inhibitor that causes accumulation of the Keap1–Nrf2 complex and Keap1-dependent Nrf2 ubiquitylation. The functional impact of endogenous E-cadherin on Nrf2 ubiquitylation was strengthened by decreased Nrf2 ubiquitylation in HepG2 cells transfected with siRNA against E-cadherin; the lack of a detectable change in Keap1-dependent Nrf2 ubiquitylation by E-cadherin overexpression might result from saturation of ubiquitylated Nrf2 caused by MG132. Overall, our data showing endogenous protein–protein interactions support the functional role of E-cadherin in Nrf2 activity and physiological relevance.

β-Catenin is a protein whose core armadillo repeat structure allows it to bind many partners and conveys structural and functional information (Daugherty and Gottardi, 2007; Hoogeboom and Burgering, 2009). It functions as a latent signaling molecule responsible for transcription factor activation and gene induction through recruitment of co-activator of the Tcf/Lef DNA-binding protein family (Daugherty and Gottardi, 2007).

Under normal conditions, β-catenin enhances cell proliferation and differentiation through Wnt signaling (Daugherty and Gottardi, 2007; Hoogeboom and Burgering, 2009; Thévenod and Chakraborty, 2010). It also serves as a mediator that promotes cell survival under conditions of ROS challenge, which would increase stress resistance and ROS clearance. β-Catenin interacts with some transcription factors activated by ROS challenge, such as HIF-1, c-Jun, TCF and Foxo (Thévenod and Chakraborty, 2010). Interactions of β-catenin with these molecules enhance the transcriptional activity of target genes (Thévenod and Chakraborty, 2010). Our results identified the role of β-catenin in the binding of E-cadherin with Nrf2 or Keap1. They also support the notion that the complex of E-cadherin and β-catenin is required for E-cadherin-mediated repression of Nrf2 activity. Thus, β-catenin might have the ability to reprogram transcriptional activity in response to stimuli, which is consistent with a previous report (Hoogeboom and Burgering, 2009). In our study, E-cadherin interacted with β-catenin for Nrf2 binding, supporting the idea that E-cadherin forms a complex with β-catenin in association with its binding to Nrf2 and Keap1. Nonetheless, the finding that a deficiency in E-cadherin diminished the interaction of β-catenin and Nrf2 does not exclude the possibility of direct binding of β-catenin with Nrf2. Collectively, E-cadherin apparently provides a docking site for the molecules. Our observations showed that E-cadherin deficient in the β-catenin-binding domain (ECAD-CT-Δβ-catenin) still weakly binds Nrf2, suggesting that another region of E-cadherin (e.g. the intervening region) also provides a direct binding site for Nrf2. Our finding showing a decrease in Keap1 activity by β-catenin knockdown corroborated the concept that β-catenin enhances the interaction between E-cadherin and Nrf2 and might contribute to Keap1-mediated ubiquitylation.

Nrf2 binds depolymerized actin after a challenge of oxidative stress, and then the complex of Nrf2–actin can translocate into the nucleus for ARE activation (Kang et al., 2002). Thus, treatment with phalloidin, an actin-stabilizing agent, prevented the activation of Nrf2 (Kang et al., 2002). In our experiment, phalloidin treatment did not alter the interaction between E-cadherin and Nrf2 (supplementary material Fig. S2B). However, this interaction was weakened by a deficiency in β-catenin. Keap1 can also bind with actin filament and this Keap1–actin complex functions as a scaffold to retain Nrf2 in the cytoplasm under unstressed conditions (Kang et al., 2004). Our result raises the possibility that E-cadherin provides a docking site for Nrf2 and Keap1 in association with actin and β-catenin interaction. Therefore, it is likely that β-catenin contributes to the regulatory effect of E-cadherin for the activity of Nrf2, presumably facilitating Keap1-mediated ubiquitylation.

It is well recognized that TGF-β1 represses E-cadherin. Moreover, E-cadherin deficiency facilitates TGF-β1 induction because E-cadherin inhibits RhoA-dependent phosphorylation of Smad3 (Cho et al., 2010). Our findings corroborated the notion that an increase in Nrf2 activity plays a role in the EMT process of HepG2 cells treated with TGF-β1. Consistently, TGF-β1 enhanced Nrf2-mediated ARE activity and HO-1 induction in smooth muscle cells (Churchman et al., 2009). Presumably, this phenomenon is associated with enhanced superoxide production and the expression of NAD(P)H oxidase subunit p22(phox). Evidence is accumulating that the activating mutations of Nrf2 frequently occurred in human cancers and such mutations might help to increase the constitutive expression of pro-survival factors (Hayes and McMahon, 2009). It should also be pointed out that the expression levels of Nrf2 and its target genes were elevated in several aggressive cancer cells (i.e. LK2, EBC1 and HO1-U1) or in lung and colon cancer tissues (Kim et al., 2011; Shibata et al., 2008). All of these findings are in agreement with our observation that cancer cells deficient in E-cadherin exhibited higher levels of Nrf2 and were more resistant to the cytotoxicity caused by doxorubicin.

In summary, the results of this study showed that (1) E-cadherin inhibits nuclear accumulation of Nrf2 and Nrf2-mediated gene transcription by its interaction with Nrf2; (2) E-cadherin binds the C-terminus of Nrf2, thus it does not affect phosphorylation of Ser40; (3) E-cadherin interacts with Keap1; (4) E-cadherin forms a quaternary complex with Nrf2, Keap1 and β-catenin, which might contribute to Keap1-dependent ubiquitylation of Nrf2. Collectively, E-cadherin binding with Nrf2 might restrain nuclear localization of Nrf2 and its transcriptional activity, implying that the loss of E-cadherin promotes tumor proliferation and metastasis by increasing the activity of Nrf2.

L-[³⁵S]methionine (>1000 Ci/mmol) was purchased from PerkinElmer (Waltham, MA). Anti-E-cadherin antibody was obtained from BD Bioscience (San Jose, CA). Antibodies directed against Nrf2, lamin A/C, heat shock protein 70, NQO-1, Keap1, Gal4 DNA-binding domain (GBD) and β-catenin were supplied from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-HO-1 antibody was purchased from Stressgen (San Diego, CA). Sulforaphane, anti-ubiquitin antibody, and anti-β-actin antibody were purchased from Sigma (St Louis, MO). Horseradish-peroxidase-conjugated goat anti-rabbit and goat anti-mouse IgG were obtained from Zymed Laboratories (San Francisco, CA). MG132 was provided by Calbiochem (Darmstadt, Germany).

pGL-1651 reporter construct containing the promoter region of the GSTA2 gene was generated according to method described previously (Park et al., 2004). Briefly, the region containing –1651 to +66 bp of the GSTA2 gene was amplified by PCR using pGTB-1.65 as a template. The DNA product was cloned into pGEM-T EasyVector (Promega, Madison, WI) and subcloned into the pGL3 luciferase reporter. The deletion mutants of GSTA2 promoter luciferase plasmid, pGL-1651-ΔC/EBP and pGL-1651-ΔARE, in which the C/EBP-binding site and ARE were deleted, respectively, were constructed previously (Park et al., 2004). Mammalian expression plasmids for GBD-Nrf2 functional domain fusion proteins were provided by Masayuki Yamamoto (Tohoku University, Sendai, Japan); GBD-NT contains the N-terminal 1–317 amino acid region of Nrf2, whereas GBD-CT contains the C-terminal 317–597 amino acid region of Nrf2 (Katoh et al., 2001). A Ser40 to alanine mutant of Nrf2 was generated by oligonucleotide-mediated mutagenesis (Cho et al., 2007). The constructs encoding deletion mutants of catenin-binding domains in E-cadherin were gifts from Richard A. Anderson (University of Wisconsin-Madison, Madison, WI).

HEK293 and HepG2 cells were supplied by the American Type Culture Collection (Manassas, VA), and SNU886 and SNU449 cells were obtained from the Korean Cell Line Bank (Seoul, Korea). The cells were maintained in Dulbecco’s modified Eagle’s medium containing 10% fetal bovine serum (FBS), 50 U/ml penicillin, and 50 μg/ml streptomycin at 37°C in a humidified atmosphere with 5% CO₂. Cells that had undergone fewer than 20 passages were used. They were incubated in the presence or absence of t-BHQ (30 μM), sulforaphane (5 μM) or recombinant human TGF-β1 (5 μg/μl) for the indicated time periods.

Transient transfection was done with Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA). Briefly, cells were re-plated 24 hours before transfection at a density of 7×10⁵ cells in six-well plates. The cells were incubated with the plasmids of interest and Lipofectamine reagent for 3 hours for transfection, and continuously incubated for additional 24 hours in minimal essential medium containing 1% FBS. Control cells were transfected with an equal amount of the respective empty plasmid (i.e. mock transfection). In luciferase reporter assays, the cells were transiently transfected with pGL-1651 or pGL-NQO1-ARE plasmid that contains three copies of ARE derived from the promoter of the NQO1. The activity of luciferase was measured by adding Luciferase Assay Reagent (Promega, Madison, WI).

Cell lysates and subcellular fractions were prepared according to previously published methods (Kang et al., 2002). Membrane fractions were prepared using ProteoExtract subcellular proteome extraction kit (Calbiochem, Darmstadt, Germany). SDS-polyacrylamide gel electrophoresis and immunoblot analyses were performed as previously described (Ki et al., 2005). Immunoreactive proteins were visualized using an ECL chemiluminescence detection kit (Amersham Biosciences, Amersham, UK). At least three separate experiments were performed with different samples to measure changes in protein levels.

Total RNA was extracted using Trizol (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. Total RNA was reverse-transcribed using an oligo(dT)₁₆ primer to obtain cDNA. The cDNA was amplified by polymerase chain reaction (PCR). PCR was performed with a Light Cycler 1.5 apparatus (Roche, Mannheim, Germany) using a Light Cycler DNA master SYBR green-I kit according to the manufacturer’s instructions. The following primer sequences were used: human NQO1, 5′-AGGCTGGTTTGAGCGAGT-3′ (sense) and 5′-TTGAATTCGGGCGTCTGCTG-3′ (antisense); and human HO-1, 5′-CAGGCAGAGAATGCTGAGTTC-3′ (sense) and 5′-CATCACCAGCTTAAAGCCTT-3′ (antisense).

To determine protein–protein interaction, a fraction of cell lysates (500 μg protein in 1 ml) was incubated with antibody directed against the protein of interest overnight at 4°C. The antigen–antibody complex was immunoprecipitated following incubation for 2 hours at 4°C with Protein-G–agarose. Immune complexes were solubilized in 2× Laemmli buffer and boiled for 5 minutes. The samples were immunoblotted with the antibody of interest.

E-cadherin and Nrf2 binding was analyzed using agarose-immobilized GST fusion proteins according to methods described previously (Ki et al., 2005). Briefly, Escherichia coli BL21(DE3) cells harboring the plasmid encoding GST-Nrf2 plasmid was cultured at 37°C with shaking. GST fusion protein was produced by incubating the cells with 1 mM isopropyl-β-D-thiogalactopyranoside for 6 hours. Bacteria expressing GST–Nrf2 or GST were precipitated and resuspended with one-tenth volume of PBS containing 0.1% Tween-20, 2 mM EDTA, 0.1% β-mercaptoethanol and 0.2 mM PMSF. Bacterial cell lysates was mixed with GSH agarose beads (100 μl, 50% slurry) pre-swollen with PBS containing 0.1% Tween-20. GST–Keap1 was purchased from Abnova (Taipei City, Taiwan). GST fusion protein was immobilized to GSH agarose by incubation at 4°C for 20 minutes and washed three times with PBS containing 0.1% Tween-20 and 0.2 mM PMSF and once with 20 mM HEPES buffer (pH 7.8, buffer A) containing 50 mM KCl, 5 mM MgCl₂, 5 mM EDTA, 6% glycerol, 0.05% Triton X-100, 0.1 mM PMSF, 1 mM DTT and 0.05 mg/ml BSA.

Radiolabeled protein was produced with the TNT system (Promega, Madison, WI) using pcDNA-E-cadherin or pcDNA-Keap1 or pcDNA-Nrf2 (1 μg). For binding reactions, GSH agarose immobilized to GST–Nrf2 (or GST as control) (25 μl) was mixed with [³⁵S]E-cadherin, [³⁵S]Keap1 or [³⁵S]Nrf2 (10 μl) in 500 μl buffer A. The protein-bound agarose was washed three times with 20 mM HEPES buffer (pH 7.8) containing 50 mM KCl, 5 mM MgCl₂, 5 mM EDTA, 6% glycerol, 0.05% Triton X-100 and 0.1 mM PMSF. Bound proteins were eluted with 20 μl of 50 mM Tris-HCl (pH 8.0) containing 10 mM reduced GSH and 10% glycerol. The eluate was resolved by SDS-PAGE. Labeled proteins were visualized by autoradiography.

HEK293 cells were grown on Lab-TEK chamber slides (Nalge Nunc International, Rochester, NY). Standard immunocytochemical methods were used for immunostaining of Nrf2 or E-cadherin, as previously described (Kang et al., 2002). For immunostaining, the cells were fixed in 100% methanol for 30 minutes. After blocking in 5% bovine serum albumin in PBS for 1 hour, the cells were incubated for 1 hour with anti-Nrf2 antibody and monoclonal mouse anti-E-cadherin antibody. The cells were incubated with Alexa-Fluor-488-conjugated goat anti-rabbit IgG antibody and Alexa-Fluor-555-conjugated rabbit anti-mouse IgG antibody (Invitrogen, Carlsbad, CA). Stained cells were examined using a laser-scanning confocal microscope (Leica TCS NT, Leica Microsystems, Wetzlar, Germany).

To knockdown the protein of interest, cells were transfected with siRNA directed against β-catenin, E-cadherin or Keap1, or a nontargeting control siRNA (100 pmol/ml) using Dharmafect (Dharmacon, Lafayette, CO). All siRNAs were purchased from Dharmacon. The knockdown effect was confirmed by immunoblotting for target protein.

HepG2, SNU886 or SNU449 cells were plated in a 48-well dish (1×10⁵ cells/well), and incubated in Eagle’s minimum essential medium without 10% FBS for 12 hours. The cells were then incubated with doxorubicin dissolved in dimethylsulfoxide at indicated concentrations for 12 hours. 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-tetrazolium bromide (MTT) assays were performed to assess cell viability as previously described (Kim et al., 2008).

One-way analysis of variance was used to assess significant differences among treatment groups. For each significant effect of treatment, the Newman–Keuls test was used for comparisons of multiple group means. The criterion for statistical significance was set at P<0.05 or P<0.01.