The metastasis suppressor NDRG1 modulates the phosphorylation and nuclear translocation of β-catenin through mechanisms involving FRAT1 and PAK4 Free

N-myc downstream-regulated gene 1 (NDRG1) is a ubiquitously expressed, predominantly cytosolic protein (Bae et al., 2013), which markedly reduces the metastasis of tumor cells in vivo, including those derived from prostate and colon cancers (Guan et al., 2000; Bandyopadhyay et al., 2003; Liu et al., 2012b). In fact, a variety of studies have revealed a negative correlation between NDRG1 expression and cancer grade and spread, indicating its role as a suppressor of metastasis (Fang et al., 2014).

A series of studies have demonstrated that NDRG1 modulates cellular signaling pathways and molecular motors (Chen et al., 2012; Dixon et al., 2013; Kovacevic et al., 2013; Sun et al., 2013b; Sun et al., 2013a) and interacts with key effectors of oncogenic signaling, including WNT (Liu et al., 2012b) and nuclear factor-κB (NF-κB) (Hosoi et al., 2009). Our recent investigations utilizing microarray analysis have also revealed that NDRG1 plays a vital role in regulating the transforming growth factor β (TGF-β) pathway (Chen et al., 2012; Kovacevic et al., 2013). In fact, NDRG1 inhibits the TGF-β-mediated epithelial-to-mesenchymal transition (EMT), which plays a crucial role in metastasis, by maintaining the membrane localization of E-cadherin, inhibiting the expression of the EMT marker vimentin and blocking cell migration and invasion (Chen et al., 2012; Liu et al., 2012b). These investigations have increased our understanding of NDRG1 function and have led to the current study, which examines the effects of NDRG1 on β-catenin, an important molecule integrally involved in cell adhesion and metastasis.

β-catenin is a key molecule participating in the WNT–β-catenin pathway, which is involved in regulating the EMT (Jamieson et al., 2012; Liu et al., 2012b; Mao et al., 2013a). The function of β-catenin is dependent on its cellular localization. When expressed at the cell membrane, β-catenin, together with E-cadherin, is part of the adherens junction complex, which mediates cell–cell adhesion (Brembeck et al., 2006; MacDonald et al., 2009). However, when localized in the nucleus, β-catenin acts as a transcriptional co-activator of transcription factors of the lymphoid-enhancing factor-1 (LEF-1)/T cell factor (TCF) family to activate oncogenic targets, such as cyclin D1 (Clevers and Nusse, 2012; Jamieson et al., 2012).

The mechanism through which β-catenin becomes localized in the nucleus begins with the activation of WNT signaling, through binding of secreted WNT glycoproteins to their receptors, which leads to inhibition of the Axin1–APC–CK1–GSK3β complex (Mao et al., 2001; Li et al., 2012a). Consequently, β-catenin becomes stabilized and translocates into the nucleus (Henderson and Fagotto, 2002; Jamieson et al., 2012). In the absence of the canonical WNT signal, cytoplasmic β-catenin turns over rapidly – it is first phosphorylated by the Axin1–APC–CK1–GSK3β complex and is subsequently degraded by the proteasome (MacDonald et al., 2009; Jamieson et al., 2012; Stamos and Weis, 2013). In fact, β-catenin is initially phosphorylated at Ser45 by CK1 (officially known as CSNK1A1), an event which then facilitates phosphorylation of Ser33, Ser37 and Thr41 by GSK3β, leading to β-catenin ubiquitylation and degradation (MacDonald et al., 2009; Verheyen and Gottardi, 2010).

There are several negative regulators of β-catenin phosphorylation, including protein phosphatase 1 and 2A (PP1 and PP2A) (Ikeda et al., 2000; MacDonald et al., 2009). In addition, the protein frequently rearranged in advanced T-cell lymphoma 1 (FRAT1), is known to inhibit GSK3β-mediated phosphorylation of β-catenin (Yost et al., 1998). FRAT1 prevents GSK3β from binding to the Axin1–APC–CK1 complex by competition with Axin1 for a common or closely overlapping binding site on GSK3β (Fraser et al., 2002). Furthermore, FRAT1 induces GSK3β dissociation from the Axin1–APC–CK1 complex, inhibiting β-catenin phosphorylation and promoting its stabilization (Thomas et al., 1999; van Amerongen and Berns, 2005; Hagen et al., 2006). Increased expression of FRAT1 is correlated with greater cancer cell invasion and metastasis, and poorer patient outcome (Zhang et al., 2011; Guo et al., 2013). Another protein affecting β-catenin stability is PAK4, which inhibits β-catenin degradation and acts as a shuttle for nuclear translocation (Li et al., 2012b; Dart and Wells, 2013). As such, PAK4 expression promotes the oncogenic and metastatic characteristics of tumor cells, including increased cell migration and invasion (Kimmelman et al., 2008; Wells et al., 2010).

We have previously demonstrated that NDRG1 inhibits the TGF-β-induced EMT by maintaining E-cadherin and β-catenin at the cell membrane (Chen et al., 2012). Hence, NDRG1 could play a role in the WNT–β-catenin pathway, acting to inhibit the EMT (Chen et al., 2012). However, the mechanism by which NDRG1 mediated β-catenin stabilization was unclear and required elucidation to provide an understanding of the potent role of NDRG1 in preventing metastasis.

In this investigation, we examined the phosphorylation and localization of β-catenin in response to the overexpression and silencing of NDRG1 in prostate and colon cancer cells. For the first time, our results demonstrate that NDRG1 inhibits the phosphorylation of β-catenin at Ser33/37 and Thr41 and maintains the localization of β-catenin at the cell membrane through the activity of FRAT1 and PAK4. Moreover, through its effects on PAK4, NDRG1 modulates the WNT–β-catenin pathway by inhibiting β-catenin nuclear translocation and, hence, its ability to upregulate cyclin D1.

To elucidate the molecular role of NDRG1 in terms of stabilizing membrane-associated β-catenin (Chen et al., 2012), we utilized two established models, including DU145 (Fig. 1A,B) and HT29 cells (Fig. 1C,D) that stably overexpress exogenous human NDRG1. We used these cell types as our previous studies demonstrated that, in both models, NDRG1 plays an important anti-metastatic role, decreasing cell migration and invasion (Chen et al., 2012; Dixon et al., 2013; Sun et al., 2013a). In these cell types, exogenous expression of FLAG-tagged NDRG1 was detected on western blots as a ∼45 kDa protein (Fig. 1A,C). In addition, endogenously expressed NDRG1 (e.g. in vector control cells) was observed at ∼43 and 44 kDa, indicating possible post-translational alterations e.g. phosphorylation and/or cleavage (Murray et al., 2004; Kovacevic et al., 2011; Ghalayini et al., 2013). Densitometric analysis of immunoblots was performed, and represents the total of all NDRG1 bands. The NDRG1-overexpressing DU145 (Fig. 1A) and HT29 cells (Fig. 1C) demonstrated a significant (P<0.001) increase in NDRG1 expression relative to their empty-vector-transfected control cells.

To further assess the effects of endogenous NDRG1, we generated NDRG1-silenced clones (sh-NDRG1) of these two cell lines (Chen et al., 2012). As shown in Fig. 1A,C, in comparison with control cells transfected with scrambled shRNA (sh-control), the sh-NDRG1 clones showed a significant (P<0.001) threefold to sixfold reduction in the expression of NDRG1 protein in DU145 (Fig. 1A) and HT29 cells (Fig. 1C). Considering the marked differences between tumors in terms of their genetic background, all studies were conducted with both cell lines, to gain insight into NDRG1 function.

Phosphorylation of β-catenin by GSK3β at Ser33/37 and Thr41 plays a crucial role in the degradation of this protein, as it leads to polyubiquitylation and proteasomal degradation (Kimelman and Xu, 2006; MacDonald et al., 2009; Wu and Pan, 2010; Stamos and Weis, 2013). Hence, we examined the effect of NDRG1 overexpression and silencing on the level of phosphorylation at these sites, to determine whether changes in phosphorylation stabilized β-catenin following NDRG1 expression (Chen et al., 2012). In both DU145 and HT29 cells, NDRG1 overexpression resulted in a significant (P<0.05 and P<0.01, respectively) reduction in β-catenin phosphorylation (Ser33/37 and Thr41; Fig. 1A,C) when it was expressed relative to total β-catenin. The expression of total β-catenin was not significantly altered relative to that observed in the vector control in DU145 (Fig. 1A) or HT29 cells (Fig. 1C). Concurrently, NDRG1 overexpression in both cell types led to a significant (P<0.05) increase in non-phosphorylated β-catenin (Ser33/37, Thr41) when normalized to total β-catenin (Fig. 1A,C). Studies using a third tumor cell type, namely HCT116 colon cancer cells, gave similar results to those observed with HT29 and DU145 cells – there was a significant (P<0.05) decrease in phosphorylated β-catenin (Ser33/37, Thr41) upon NDRG1 expression (data not shown).

Silencing NDRG1 in DU145 and HT29 cells resulted in a significant (P<0.01) increase in the amount of β-catenin that was phosphorylated at Ser33/37 and Thr41 when normalized to total β-catenin, which was significantly (P<0.01) decreased under these conditions relative to the sh-control (Fig. 1A,C). In addition, there was a marked and significant (P<0.05 in HT29 cells; P<0.01 in DU145 cells) decrease in the amount of β-catenin that was not phosphorylated at Ser33/37 and Thr41 in sh-NDRG1 cells relative to the sh-control cells, when normalized to total β-catenin (Fig. 1A,C). We also examined β-catenin mRNA levels using quantitative real-time PCR, which demonstrated that NDRG1 expression did not affect β-catenin mRNA levels (supplementary material Fig. S1A). This suggests that NDRG1 modulates β-catenin expression by post-transcriptional mechanisms.

Our data show that NDRG1 expression decreased the phosphorylation of β-catenin at Ser33/37 and Thr41 and increased the amount of non-phosphorylated β-catenin (Ser33/37, Thr41). However, although overexpression of NDRG1 did not significantly increase total β-catenin levels, NDRG1 silencing led to a decreased amount of total β-catenin. Further studies were conducted to examine whether NDRG1 could modulate the downstream effects of β-catenin. When expressed in the nucleus, β-catenin functions as part of a transcription factor complex to upregulate the transcription of proto-oncogenes, such as cyclin D1 (MacDonald et al., 2009; Clevers and Nusse, 2012; Fagotto, 2013). Therefore, the effect of NDRG1 overexpression and silencing on cyclin D1 expression was examined. Interestingly, in DU145 (Fig. 1A) and HT29 (Fig. 1C) cells, NDRG1 overexpression significantly (P<0.05) reduced cyclin D1 levels. By contrast, NDRG1 silencing (P<0.05) increased cyclin D1 expression. Importantly, these results suggest that although NDRG1 could prevent β-catenin degradation, it also negatively regulates cyclin D1.

Immunofluorescence studies demonstrated that NDRG1 overexpression resulted in a marked increase in the amount of non-phosphorylated (Ser33/37, Thr41) and total β-catenin at the plasma membrane relative to that observed in the vector control in DU145 cells (Fig. 1B) and HT29 cells (Fig. 1D). Concurrently, there was a decrease in cytosolic total β-catenin expression (Fig. 1B,D). In sh-NDRG1 cells relative to sh-control cells, there was a reduction in non-phosphorylated (Ser33, Ser37 and Thr41) and total β-catenin levels, which was observed in both cell types (Fig. 1B,D).

The studies above suggest that the regulation of β-catenin by NDRG1 involves multiple pathways that control the localization, phosphorylation and degradation of the former. To further investigate this novel finding, the above studies were complemented using fractionation (Fig. 2) and immunofluorescence (Fig. 1B,D) to assess the effect of NDRG1 on the cellular distribution of β-catenin.

Fractionation studies demonstrated that NDRG1 in vector control and sh-control DU145 (Fig. 2A,B) and HT29 cells (Fig. 2C,D) was localized predominantly in the cytoplasmic fraction. Overexpression of NDRG1 led to a marked and significant (P<0.01) increase in the levels of this protein in the cytosolic and membrane fractions (Fig. 2A,C). Silencing of NDRG1 (shNDRG1) decreased the amount of NDRG1 in the cytoplasmic and membrane fractions relative to the sh-control (Fig. 2B,D).

Both non-phosphorylated β-catenin (Ser33/Ser37 and Thr41) and total β-catenin in vector control and sh-control cells were evident in the cytoplasmic, membrane, cytoskeletal, nuclear-soluble and chromatin-bound fractions in DU145 and HT29 cells (Fig. 2A–D). For DU145 cells, NDRG1 overexpression led to a significant increase in non-phosphorylated β-catenin (Ser33/37, Thr41; P<0.05) and total β-catenin levels (P<0.05) in the cytoplasmic, membrane and cytoskeleton fractions relative to the vector control (Fig. 2A). A similar effect was also observed for HT29 cells (Fig. 2C). Silencing of NDRG1 resulted in a contrasting effect, whereby there was a significant decrease in both non-phosphorylated β-catenin (Ser33/37, Thr41; P<0.05) and total β-catenin (P<0.05) in the membrane and cytoskeletal fractions in DU145 and HT29 cells relative to the sh-control (Fig. 2B,D).

In comparison to the effects in cytoplasmic, membrane and cytoskeletal fractions, the overexpression or silencing of NDRG1 had the opposite effects on the amount of β-catenin in the nuclear-soluble and chromatin-bound fractions in both cell types (Fig. 2A–D). In fact, NDRG1 overexpression significantly (P<0.05) decreased the levels of both non-phosphorylated β-catenin (Ser33/37, Thr41) and total β-catenin in these latter fractions in DU145 and HT29 cells, relative to the respective vector controls (Fig. 2A,C). By contrast, sh-NDRG1 either slightly increased or significantly (P<0.05) increased both nuclear-soluble and chromatin-bound non-phosphorylated β-catenin (Ser33/37, Thr41) and total β-catenin relative to the sh-control (Fig. 2B,D).

In summary, immunoblotting (Fig. 1A,C), when taken together with the fractionation (Fig. 2A–D) and immunofluorescence studies (Fig. 1B,D) demonstrated that the increase in non-phosphorylated β-catenin (Ser33/37 and Thr41) and total β-catenin levels observed after NDRG1 expression is primarily related to β-catenin that is localized at the plasma membrane.

To further elucidate the mechanisms involved in altering β-catenin expression after changes in NDRG1 levels, we investigated the effect of NDRG1 on other key regulators of β-catenin expression and phosphorylation. This included an examination of FRAT1 expression (Hagen et al., 2006; MacDonald et al., 2009; Verheyen and Gottardi, 2010; Bae et al., 2013) and GSK3β phosphorylation (at Ser9 and Tyr216; Wu and Pan, 2010), and an assessment of the expression of CK1 (Stamos and Weis, 2013), PP1(Stamos and Weis, 2013) and PP2A (Ikeda et al., 2000).

We initially examined the effect of NDRG1 on the GSK3β-binding protein FRAT1, which inhibits the binding of GSK3β to the Axin1–APC–CK1 complex and prevents β-catenin phosphorylation (Thomas et al., 1999; van Amerongen and Berns, 2005; Hagen et al., 2006). For both DU145 and HT29 cells, NDRG1 overexpression significantly upregulated FRAT1 protein (P<0.01) and mRNA (P<0.05 for DU145 cells; P<0.01 for HT29 cells) levels relative to the vector control (Fig. 3A; supplementary material Fig. S1B). By contrast, sh-NDRG1 significantly reduced FRAT1 protein (P<0.05 for HT29 cells; P<0.01 for DU145 cells) and mRNA (P<0.05) levels relative to the sh-control (Fig. 3A; supplementary material Fig. S1B). Despite the effect of NDRG1 on FRAT1, there was no significant change in the amount of total GSK3β relative to the vector control or sh-control in either cell line (Fig. 3A). Previous studies have also shown FRAT1 did not affect total GSK3β (Hagen et al., 2006).

Immunofluorescence (supplementary material Fig. S1C) demonstrated that FRAT1 and GSK3β were localized mainly to the cytoplasm and, to a much lesser extent, the membrane in HT29 and DU145 cells. The overexpression and silencing of NDRG1 led to an increase and decrease, respectively, in FRAT1 expression in these compartments relative to the relevant controls. No marked expression of FRAT1 was observed in the nucleus and there was no relative change in its distribution between cellular compartments after alteration of NDRG1 levels (supplementary material Fig. S1C). Furthermore, alteration of NDRG1 expression led to no appreciable change in the cellular distribution of total GSK3β (supplementary material Fig. S1C).

Given that FRAT1 inhibits the binding of GSK3β to Axin1 and subsequent β-catenin phosphorylation, immunoprecipitation experiments were then performed to assess how this interaction was affected by NDRG1 overexpression and silencing in DU145 and HT29 cells (supplementary material Fig. S1D). Importantly, NDRG1 overexpression significantly (P<0.01) reduced the binding of GSK3β to Axin1 relative to the vector control, whereas sh-NDRG1 significantly (P<0.05 in DU145 cells; P<0.01 in HT29 cells) increased the interaction of GSK3β with Axin1 relative to the sh-control in both cell types (supplementary material Fig. S1D). These studies were consistent with the hypothesis that the increase in FRAT1 observed after NDRG1 expression (Fig. 3A) inhibits the binding of GSK3β to the Axin1–APC–CK1 complex and prevents β-catenin phosphorylation at Ser33/37 and Thr41 (Figs 1, 2).

To further establish whether the increased expression of FRAT1 in response to NDRG1 overexpression was responsible for the reduction in β-catenin phosphorylation at Ser33/37 and Thr41, we performed additional studies to silence FRAT1 in both HT29 and DU145 vector control and NDRG1-overexpressing cells. FRAT1 was significantly (P<0.01) reduced in both HT29 and DU145 cells upon treatment with its specific siRNA (Fig. 3B,C). Moreover, the silencing of FRAT1 led to a significant (P<0.01) increase in β-catenin phosphorylation at Ser33/37 and Thr41 and a significant (P<0.05) decrease in total β-catenin levels (Fig. 3B,C) in both vector control and NDRG1-overexpressing cells. This further confirms that NDRG1 stabilizes β-catenin through FRAT1.

We also examined the effect of NDRG1 on GSK3β phosphorylation at two different sites (Tyr216 and Ser9) that control the activity of the latter protein (Wu and Pan, 2010). However, there was no consistent effect observed between the cell types (supplementary material Fig. S2A,B) and, again, this might be related to the diverse molecular backgrounds of the cells used. Furthermore, we assessed two serine/threonine phosphatases, PP1 and PP2A, both of which associate with Axin1 and/or APC and counteract the action of GSK3β and/or CK1 in the Axin1 complex (MacDonald et al., 2009). However, NDRG1 had no significant effect on the expression of PP1α (officially known as PPP1CA) or PP2A (subunits A, B and C; supplementary material Fig. S2C,D).

Collectively, it can be concluded that NDRG1 consistently increased the expression of the GSK3β-binding protein FRAT1, which inhibits β-catenin phosphorylation at Ser33/37 and Thr41. Moreover, FRAT1 was found to play a vital role in the ability of NDRG1 to inhibit the phosphorylation and degradation of β-catenin.

The studies above demonstrate that NDRG1 increases the levels of non-phosphorylated β-catenin (Ser33/37 and Thr41) and total β-catenin at the plasma membrane and prevents the localization of β-catenin to the nucleus. PAK4 is a nucleo-cytoplasmic shuttling protein that is imported into the nucleus, promoting TCF/LEF transcriptional activity and leading to β-catenin stabilization through inhibition of its degradation (Li et al., 2012b). Furthermore, nuclear import of PAK4 accompanies the import of β-catenin and increased TCF/LEF transcriptional activity (Li et al., 2012b; Dart and Wells, 2013). Considering this, we investigated the effect of NDRG1 expression on PAK4 levels (Fig. 4A) and its distribution (Fig. 4B).

Overexpression of NDRG1 in DU145 and HT29 cells did not modulate PAK4 protein expression levels (Fig. 4A). By contrast, NDRG1 silencing led to a slight but non-significant decrease in the amount of PAK4 protein (Fig. 4A). Examining PAK4 distribution using immunofluorescence demonstrated that NDRG1 overexpression in DU145 and HT29 cells led to decreased nuclear levels of PAK4 and increased membrane localization of β-catenin relative to the vector control cells (Fig. 4B). By contrast, NDRG1 silencing increased nuclear PAK4 levels and decreased the amount of β-catenin at the cell membrane relative to sh-control DU145 and HT29 cells (Fig. 4B). Taken together, these data show that NDRG1 regulates PAK4 distribution.

Considering the role of NDRG1 in regulating PAK4, and its potential role in modulating β-catenin translocation, studies were performed to assess the effects of two different PAK4 siRNAs on β-catenin expression and distribution (Fig. 5). Furthermore, we examined the effect of silencing PAK4 on the β-catenin target, cyclin D1 (MacDonald et al., 2009; Jamieson et al., 2012). Using immunoblotting, we found that PAK4 silencing significantly (P<0.05) decreased the amount of β-catenin protein in DU145 cells, whereas in HT29 cells, a slight but non-significant reduction in β-catenin was observed (Fig. 5A). However, for both cell types, PAK4 silencing significantly (P<0.01 in HT29 cells; P<0.001 in DU145 cells) decreased cyclin D1 expression (Fig. 5A). Immunofluorescence demonstrated that PAK4 silencing led to a decrease in the amount of nuclear, cytosolic and membrane-bound β-catenin (Fig. 5B). This observation could be explained by the fact that PAK4 inhibits the phosphorylation and degradation of β-catenin (Li et al., 2012b; Dart and Wells, 2013), and hence, PAK4 silencing leads to a decrease in total β-catenin levels. We also assessed the effect of PAK4 siRNA on β-catenin distribution in NDRG1-silenced HT29 and DU145 cells. Immunofluorescence demonstrated that PAK4 silencing markedly increased membrane-associated β-catenin and E-cadherin in NDRG1-silenced cells (supplementary material Fig. S3A,B). These results indicate that NDRG1 increases the levels of membrane-associated β-catenin and E-cadherin through its inhibitory effect on nuclear PAK4.

Furthermore, TOP/FOP-flash reporter assays were also performed to examine whether the inhibitory effect of NDRG1 on β-catenin-mediated TCF/LEF transcriptional activity was modulated through PAK4. The results showed that PAK4 silencing significantly reduced TCF/LEF transcriptional activity in all cells, regardless of changes in NDRG1 expression (Fig. 5C). This further highlights the vital role that PAK4 plays in mediating the nuclear localization and oncogenic activity of β-catenin. Taken together, these results suggest that NDRG1 plays a role in preventing PAK4 nuclear translocation, which is important for preventing upregulation of oncogenic targets.

WNT ligand plays a crucial role in the signaling cascade that results in decreased β-catenin phosphorylation and degradation and, hence, increased nuclear translocation, resulting in the transcription of WNT target genes, such as that encoding cyclin D1 (Smolich et al., 1993; Clevers and Nusse, 2012). Considering the effect of NDRG1 on decreasing β-catenin phosphorylation at Ser33/37 and Thr41 and preventing nuclear translocation, while promoting membrane localization (Fig. 1), it was important to examine the effect of this metastasis suppressor on WNT activity.

Incubation of vector control or sh-control DU145 cells with WNT3a ligand (100 ng/ml for 48 h) (Willert et al., 2003; Liu et al., 2011a) led to a significant (P<0.05) decrease in NDRG1 expression relative to that of untreated cells, whereas it had no significant effect on this metastasis suppressor in NDRG1-overexpressing cells and shNDRG1 cells (Fig. 6A). A similar response was also observed using HT29 cells (Fig. 6C). The effect of WNT3a on downregulating endogenous NDRG1 might occur through its ability to upregulate N-myc or c-myc (Kuwahara et al., 2010; Liu et al., 2012a), which might then decrease NDRG1 levels (Shimono et al., 1999; Li and Kretzner, 2003).

In the vector controls or sh-controls of DU145 and HT29 cells, treatment with WNT3a significantly (P<0.01) decreased the amount of phosphorylated β-catenin (Ser33/37, Thr41) and resulted in a significant increase in total β-catenin (P<0.05) and cyclin D1 (P<0.01) relative to controls that were not incubated with WNT3a (Fig. 6A,C). By contrast, in NDRG1-overexpressing DU145 and HT29 cells, WNT3a did not further decrease the amount of phosphorylated β-catenin (Ser33/37, Thr41) (Fig. 6A,C). However, NDRG1 overexpression either decreased (DU145) or prevented (HT29) the ability of WNT3a to upregulate total β-catenin and cyclin D1 (Fig. 6A,C). In sh-NDRG1 DU145 or HT29 cells, treatment with WNT3a decreased levels of phosphorylated β-catenin (Ser33/37, Thr41) when normalized to total β-catenin, whereas this treatment significantly (P<0.05 in HT29 cells; P<0.01 in DU145 cells) increased the level of cyclin D1 relative to that of cells not treated with WNT3a (Fig. 6A,C). Hence, NDRG1 overexpression inhibited the oncogenic effects of WNT3a, whereas silencing of NDRG1 enhanced WNT signaling.

The results above were confirmed using immunofluorescence in DU145 (Fig. 6B) and HT29 cells (Fig. 6D). Indeed, addition of WNT3a increased the nuclear localization of β-catenin in vector control and sh-control cells, whereas this was prevented in NDRG1-overexpressing cells, where β-catenin was highly localized at the plasma membrane. In sh-NDRG1 cells, incubation with WNT3a resulted in a more nuclear localization of β-catenin relative to the pattern observed when these cells where incubated with medium alone (Fig. 6B,D). Collectively, immunoblotting and immunofluorescence data demonstrate that NDRG1 inhibits the oncogenic effects of WNT by promoting β-catenin localization at the plasma membrane and inhibiting its nuclear translocation. This prevents the expression of the β-catenin target cyclin D1 (Clevers and Nusse, 2012; Jamieson et al., 2012).

To further assess the inhibitory effect of NDRG1 on WNT–β-catenin signaling, TOP/FOP-flash reporter assays were performed (Fig. 7A). In the vector controls or sh-controls of DU145 and HT29 cells, treatment with WNT3a significantly (P<0.01) increased TCF/LEF transcriptional activity. Moreover, in sh-NDRG1 DU145 or HT29 cells, TCF/LEF transcriptional activity was even further increased with WNT3a stimulation (P<0.01) when compared with the sh-control cells. By contrast, NDRG1 overexpression prevented the ability of WNT3a to activate TCF/LEF transcriptional activity (Fig. 7A). Hence, NDRG1 overexpression inhibited the oncogenic effects of WNT3a.

The studies above demonstrated that the increased amounts of non-phosphorylated (Ser33/37 and Thr41) and total β-catenin after NDRG1 expression are localized at the plasma membrane (Fig. 1B,D; Fig. 2) and might promote cell adhesion through the adherens complex (Brembeck et al., 2006). To examine this, a well-established cellular aggregation assay was implemented (Kovacevic et al., 2008). These studies demonstrated that NDRG1 overexpression significantly (P<0.05 in HT29 cells; P<0.01 in DU145 cells) increased cellular aggregation relative to the vector control, whereas sh-NDRG1 significantly (P<0.05 in HT29 cells; P<0.01 in DU145 cells) decreased aggregation in these cells (Fig. 7B).

As a second assay of adhesion, DU145 or HT29 cells were treated with trypsin-EDTA for 2–20 min and the percentage of cells that remained adhered to the plastic substratum was assessed. NDRG1 overexpression significantly (P<0.05) delayed the detachment of cells from the substratum relative to that of the vector control, whereas the sh-NDRG cells displayed significantly (P<0.05) increased detachment relative to that of the sh-control (Fig. 7C). This result suggests that NDRG1 promotes strong cell–cell and/or cell–matrix adhesion, restricting access of the trypsin-EDTA to the cell junctions and thereby delaying cell detachment. Both these assays confirmed that NDRG1 promotes aggregation and adhesion, which is consistent with its effect of upregulating membrane-associated β-catenin, which would promote cell adhesion.

A variety of studies have shown the important role of NDRG1 in inhibiting cancer cell metastasis (Kovacevic and Richardson, 2006; Liu et al., 2011b; Chen et al., 2012; Liu et al., 2012b; Bae et al., 2013; Mao et al., 2013b; Sun et al., 2013a). However, complete understanding of its mechanism of action remains an important research goal. This investigation demonstrates a unique role for NDRG1 in increasing the levels of non-phosphorylated β-catenin (Ser33/37 and Thr41) and total β-catenin at the plasma membrane and preventing its nuclear localization. This effect was achieved through the NDRG1-mediated upregulation of the GSK3β-binding protein FRAT1, which prevents GSK3β association with the Axin1–APC–CK1 destruction complex and, thus, β-catenin phosphorylation. Furthermore, NDRG1 modulated the WNT–β-catenin pathway by blocking the nuclear translocation of β-catenin and, hence, the ability of β-catenin to increase TCF/LEF transcriptional activity and upregulate cyclin D1 transcription. This was mediated through NDRG1 inhibiting nuclear PAK4 localization, which would compromise the activity of PAK4 in terms of shuttling β-catenin to the nucleus. This is the first time that a metastasis suppressor has been shown to inhibit oncogenic WNT–β-catenin signaling by affecting a novel mechanism influencing FRAT1 and PAK4 (Fig. 8).

The WNT–β-catenin pathway is known to be associated with tumorigenesis (MacDonald et al., 2009; Clevers and Nusse, 2012). Activation of this pathway can induce the EMT (Liu et al., 2012b), which allows tumor cells to metastasize (Son and Moon, 2010; Mao et al., 2013a). In addition to the effects of NDRG1 on inhibiting oncogenic WNT–β-catenin signaling through FRAT1 and PAK4, as shown herein, it was recently described that NDRG1 interacts with the WNT co-receptor LRP6, leading to inhibition of WNT signaling (Liu et al., 2012b). NDRG1 was also reported to promote GSK3β phosphorylation at its activating site (Tyr279/216) to reduce total β-catenin levels (Liu et al., 2012b). The present study showed that NDRG1 decreased GSK3β activity by reducing its ability to bind to Axin1 and phosphorylate β-catenin (supplementary material Fig. S1D). Also, instead of altering total β-catenin levels, NDRG1 affected the distribution of this protein by promoting membrane localization and inhibiting nuclear translocation. Previous reports also demonstrated that NDRG1 inhibits the TGF-β-induced EMT, by maintaining E-cadherin and β-catenin at the cell membrane, enabling establishment of the adherens complex (Chen et al., 2012). However, this latter work did not dissect the mechanisms involved in promoting β-catenin localization at the cell membrane. Collectively, these studies demonstrate that NDRG1 acts through a variety of mechanisms to prevent tumorigenic WNT–β-catenin signaling.

The essence of WNT signal transduction is the WNT-mediated inhibition of β-catenin phosphorylation and degradation, which leads to cytosolic β-catenin accumulation and promotes its nuclear translocation (MacDonald et al., 2009; Clevers and Nusse, 2012; Jamieson et al., 2012). Once in the nucleus, β-catenin forms a transcriptional complex with TCF/LEF that increases the expression of oncogenic target genes (Jamieson et al., 2012). Intriguingly, FRAT1 has been reported to act as an oncogene, because it could prevent β-catenin phosphorylation and degradation, leading to promotion of the WNT signaling cascade (Salic et al., 2000; van Amerongen and Berns, 2005). By contrast, under the influence of NDRG1, we observed upregulation of FRAT1 (Fig. 3A–C), which did not lead to an oncogenic event. In fact, under these latter conditions, there was decreased expression of the WNT target cyclin D1 (Fig. 1A,C), increased cell–cell adhesion (Fig. 7B) and decreased detachment from the substratum (Fig. 7C), which might counter metastasis. Moreover, the effect of NDRG1 on β-catenin was found to be directly mediated by FRAT1, with the silencing of this latter protein abolishing the NDRG1-mediated reduction in the amount of phosphorylated β-catenin. Furthermore, NDRG1 markedly prevented the association of GSK3β with Axin1, which is consistent with the NDRG1-mediated increase in FRAT1 preventing GSK3β from binding to the Axin1–APC–CK1 complex, allowing the subsequent accumulation of β-catenin (Fraser et al., 2002).

Although NDRG1 increased the levels of non-phosphorylated β-catenin, it did not result in its nuclear translocation. In fact, NDRG1 inhibited WNT signal transduction by preventing the PAK4-mediated nuclear translocation of β-catenin. Moreover, non-phosphorylated β-catenin became localized at the plasma membrane (Fig. 1B,D; Fig. 2), where it might function as part of the adherens junction that inhibits the EMT and metastasis (Jamieson et al., 2012). This latter effect involved PAK4, which modulates β-catenin intracellular translocation and signaling (Li et al., 2012b). In this study, NDRG1 overexpression reduced nuclear PAK4 localization and β-catenin targeting to the nucleus, which was accompanied by decreased expression of cyclin D1 (Figs 4, 5). Moreover, silencing PAK4 reduced the nuclear levels of β-catenin (Fig. 5B), as well as WNT signaling (Fig. 5C), which further confirmed that the NDRG1-mediated effects on PAK4 have important implications for β-catenin expression and localization, as well as for its roles in both cell adhesion and WNT signaling. Hence, NDRG1, through its effects on FRAT1 and PAK4, stabilizes β-catenin and prevents its nuclear localization, while also promoting the association of β-catenin with the membrane, which potentially suppresses metastasis.

Whereas the overexpression of NDRG1 stabilized β-catenin and promoted its association with the membrane, the silencing of NDRG1 in the DU145 and HT29 cells resulted in an overall reduction in total β-catenin levels (Fig. 1A,C). This was accompanied by an increase in β-catenin phosphorylation (Ser33/37 and Thr41), suggesting that the proteasomal degradation of this protein is increased. Considering that NDRG1 did not modulate β-catenin mRNA levels (supplementary material Fig. S1A), these results indicate that endogenous NDRG1 levels are required to maintain a steady state of β-catenin protein in the cells, whereas a loss of NDRG1 allows β-catenin to be rapidly degraded.

Of note, the closely related NDRG family member NDRG2, has been demonstrated to function as a tumor suppressor through its effects on β-catenin (Kim et al., 2009; Hwang et al., 2011). However, in marked contrast to NDRG1, NDRG2 interacts with β-catenin through its α6 helix residue L172 (Hwang et al., 2011). Interestingly, although NDRG1 possesses the α6 helix, it does not contain the crucial L172 residue, which is actually replaced with I172 (Hwang et al., 2011). Furthermore, using immunofluorescence, we show here that NDRG1 does not directly colocalize with β-catenin (data not shown). Hence, the mechanism of how NDRG1 affects β-catenin expression is markedly different to that observed with NDRG2.

It is important to note that cellular adhesion is a complex process and that, in addition to β-catenin, other molecules, such as E-cadherin, also play a vital role (Brembeck et al., 2006; Canel et al., 2013). In fact, our previous studies demonstrated that NDRG1 maintains membrane-associated E-cadherin in both DU145 and HT29 cells (Chen et al., 2012) and inhibits stress fiber formation (Sun et al., 2013a). Moreover, others have also demonstrated that NDRG1 is involved in the recycling of E-cadherin, leading to its stabilization (Kachhap et al., 2007). Hence, the molecular mechanisms underlying the effect of NDRG1 on cellular adhesion are multifaceted, with the current study focusing on one aspect, namely β-catenin, which plays an important role in this process.

Although E-cadheren modulates β-catenin (Kachhap et al., 2007), other factors can also regulate β-catenin localization and turnover in response to NDRG1. For example, our data clearly demonstrate that: (1) WNT ligand upregulates nuclear β-catenin (Fig. 6B,D); (2) PAK4 plays a crucial role in the nuclear translocation of β-catenin (supplementary material Fig. S3) and is essential for oncogenic TCF/LEF transcriptional activity mediated by nuclear β-catenin (Fig. 5C); and (3) FRAT1 silencing increases β-catenin phosphorylation (Ser33/37, Thr41) and reverses the effect of NDRG1 on this latter molecule (supplementary material Fig. S1C,D). Hence, it was important to characterize these regulators of β-catenin in order to enable a comprehensive understanding of NDRG1 activity.

It is also notable that although NDRG1 had a very marked effect on the phosphorylation and localization of β-catenin, its effects on the TCF/LEF transcriptional activity, although significant (P<0.05) and in agreement with the alterations in β-catenin, were milder. This might indicate that, in addition to β-catenin, other factors could also be involved in altering the transcriptional activity of TCF/LEF (MacDonald et al., 2009).

In conclusion, NDRG1 utilizes a unique mechanism to suppress the WNT–β-catenin pathway. Notably, NDRG1 inhibits β-catenin phosphorylation at Ser33/37 and Thr41 and increases the amount of non-phosphorylated β-catenin at the plasma membrane. This was mediated by the NDRG1-induced upregulation of the GSK3β-binding protein FRAT1, which inhibits the interaction of GSK3β with the Axin1–APC–CK1 destruction complex. Furthermore, NDRG1 modulates the WNT–β-catenin pathway by inhibiting the nuclear translocation of β-catenin through PAK4. In fact, NDRG1 expression reduced the nuclear localization of PAK4, preventing the nuclear targeting of β-catenin. Hence, NDRG1 inhibits the oncogenic roles of β-catenin, while promoting its metastasis-suppressing effects.

The human prostate and colon cancer cell lines DU145, HT29 and HCT116 (American Type Culture Collection; Manassas, VA) were grown under established conditions, as described previously (Chen et al., 2012; Sun et al., 2013a). NDRG1 overexpressing and silenced clones of the DU145, HT29 and HCT116 cells were generated as described previously (Chen et al., 2012; Sun et al., 2013a). Recombinant human WNT3a (5036-WN-010; R&D Systems) was used to activate the β-catenin response.

Preparation of cell lysates and immunoblot analysis was performed by established protocols, as described previously (Gao and Richardson, 2001). The Subcellular Protein Fractionation Kit for Cultured Cells from Thermo Scientific (78840; Waltham, MA) was utilized to prepare fractions according to the manufacturer's instructions. Primary antibodies were used at a 1∶1000–2000 dilution and the antibodies were against: NDRG1 (ab37897), FRAT1 (ab108405) and GSK3β (phospho Y216, ab75745) from Abcam (Cambridge, UK); β-catenin (9562, 2677), phosphorylated β-catenin (Ser33,37/Thr41; 9561), non-phosphorylated β-catenin (8814), Axin1 (2087), GSK3β (9832), phosphorylated GSK3β (Ser9; 9336), CK1 (2655), PP1α (2582), PP2A (antibody sampler kit; 9780), cyclin D1 (2926), PAK4 (3242), histone 3 (9717), EGF receptor (2926), E-cadherin (5296) and vimentin (5741) from Cell Signaling Technology (Boston, MA); and SP1 (ABE135) from Millipore (Darmstadt, Germany). The secondary antibodies (1∶10,000 dilution) included: anti-goat-IgG (A5420), anti-rabbit-IgG (A6154) and anti-mouse-IgG (A4416) from Sigma-Aldrich (St Louis, MO). Antibody against β-actin (1∶10,000; A1978, Sigma-Aldrich) was used as a loading control.

Immunoprecipitation was performed using Dynabeads®–Protein-G (Invitrogen). Briefly, cells were washed with ice-cold PBS and lysed using SDS-free radioimmunoprecipitation assay (RIPA) buffer containing protease inhibitors (Roche Diagnostics, Basel, Switzerland). Protein (300 µg) was incubated with monoclonal GSK3β antibody (1∶50; 12456, Cell Signaling Technology) overnight at 4°C. This mixture was added to 30 µl of Dynabeads®–Protein-G (Invitrogen) and incubated for 3 h at 4°C. The beads were then washed with ice-cold PBS, resuspended in LDS loading buffer, and incubated at 70°C for 10 min. The supernatant was separated on a 10% Bis-Tris gel. Axin1 and GSK3β were detected by immunoblot.

Immunofluorescence was performed by standard methods as described previously (Chen et al., 2012), using a Zeiss Axio Observer.Z1 fluorescence microscope (Zeiss, Oberkochen, Germany).

Total RNA was extracted from cells and tissues using Trizol (Invitrogen) according to the manufacturer's protocol. Total RNA (0.5 µg) from each sample was used for first-strand cDNA synthesis using a reverse transcription kit (Promega, WI). The quantitative real-time (qRT)-PCR was carried out using cDNA as a template and Universal PCR Master Mix (Applied Biosystems, Carlsbad, CA) on an 7900HT sequence detection system (Applied Biosystems). Primers used for qRT-PCR analysis were as follows: β-catenin, 5′-CCTATGCAGGGGTGGTCAAC-3′ (forward) and 5′-CGACCTGGAAAACGCCATCA-3′ (reverse); FRAT1, 5′-GCCCTGTCTAAAGTGTATTTTCAG-3′ (forward) and 5′-CGCTTGAGTAGGACTGCAGAG-3′ (reverse) β-actin, 5′-CCCGCCGCCAGCTCACCATGG-3′ (forward) and 5′-AAGGTCTCAAACATGATCTGGGTC-3′ (reverse). The relative amount of mRNA was normalized using β-actin as an endogenous control.

Silencing FRAT1 or PAK4 expression using FRAT1 or PAK4 siRNA was performed according to the manufacturer's instructions. Briefly, DU145 and HT29 cells were transfected with Hs_FRAT1_1 (SI00077826), Hs_ FRAT1_2 (SI00077833), Hs_PAK4_5 (SI02660315) or Hs_PAK4_6 (SI02665432) from Qiagen (Düsseldorf, Germany) or the scrambled control siRNA (10 nM for 48 h at 37°C), using Lipofectamine® RNAiMAX (Invitrogen, Carlsbad, CA).

DU145 and HT29 cells were transiently transfected with 200 ng of luciferase reporter plasmid TOP-flash (Millipore, 21-170), which contains two sets of three copies of the TCF-binding site (wild type), or its control FOP-flash (Millipore, 21-169), which contains mutated TCF binding sites, in combination with 2 ng of pRL-TK vector (Promega) containing Renilla luciferase. Transfection was performed with Lipofectamine 2000 (Invitrogen). Luciferase activity was measured using the Dual Luciferase Reporter Assay (Promega). Firefly luciferase activity was normalized to Renilla luciferase activity for each well. All transfection experiments were conducted in triplicate and repeated three times independently.

The cell–cell adhesion assay was used as described previously (Kovacevic et al., 2008). Briefly, cells were harvested using 1 mM EDTA in Ca²⁺/Mg²⁺-free PBS. After washing in Hank's balanced salt solution (HBSS) containing 1% bovine serum albumin (BSA), cells were seeded on 24-well plates (blocked with HBSS containing 2% BSA for 3 h at 37°C; 5000 cells/well). As a positive control, 1 mM CaCl₂ was added to promote cell–cell adhesion. The cells were incubated in a gyrating shaker (90 rpm for 1–2 h at 37°C) and the reaction was terminated by the addition of 0.5 ml of 25% formaldehyde/well.

For detachment assays, we used a standard protocol, as described previously (Haraguchi et al., 2008). Briefly, cells (5×10⁴) were seeded onto plates and incubated for 24 h at 37°C. Cells were then dissociated using 0.125% trypsin-EDTA (Gibco®, Carlsbad, CA) at 37°C for the indicated times. RPMI-1640 or McCoy's 5A medium (Gibco®) containing 10% FBS (Gibco®) was then added to inhibit trypsin. Following removal of detached cells, the remaining cells were fixed with 70% ethanol for 10 min and stained with Crystal Violet (0.1%) for 30 min. The plates were rinsed with PBS, and Crystal Violet in adherent cells was dissolved with 0.5% Triton X-100. Optical density was then measured at 595 nm.

Student's t-test and ANOVA were used for statistical analyses, with P<0.05 being considered significant.