Negative-feedback regulation of the Wnt pathway by conductin/axin2 involves insensitivity to upstream signalling

Stabilisation of β-catenin is a key step in the Wnt/β-catenin signalling pathway, allowing β-catenin to stimulate transcription of Wnt target genes in conjunction with TCF transcription factors (Clevers and Nusse, 2012). In the absence of Wnt proteins, β-catenin is earmarked for ubiquitylation and proteasomal degradation by phosphorylation mediated by a destruction complex consisting of the tumour suppressor APC, the kinases GSK3 and CK1, and the scaffolding proteins axin (also known as axin1) or conductin (also known as axin2) (Stamos and Weis, 2013). Binding of Wnt ligands to their receptors frizzled and LRP5 or LRP6 (LRP5/6) leads to membrane recruitment of axin proteins by the frizzled-associated phosphoprotein dishevelled 1, 2 or 3 (Dvl) ultimately resulting in the inhibition of β-catenin phosphorylation and degradation (Bilic et al., 2007; Cliffe et al., 2003; Cselenyi et al., 2008; Smalley et al., 1999; Wu et al., 2009; Zeng et al., 2008).

Axin molecules form dynamic oligomers by head-to-tail interaction of their DIX domains (Fiedler et al., 2011; Kishida et al., 1999). These complexes, often seen as cytoplasmic puncta upon overexpression, are thought to provide high-avidity interaction sites for β-catenin and other destruction complex components. Specific mutations of the DIX domain that abolish the head-to-tail interaction prevent axin oligomerisation and render axin ineffective at mediating β-catenin degradation (Fiedler et al., 2011). Dvl also has a DIX domain through which it binds to axin. Dvl can interfere with axin function in different ways. Frizzled-associated Dvl recruits axin to frizzled–LRP5/6 receptor complexes, which causes inhibition of axin-bound GSK3 mediated by phosphorylated LRP5/6 receptors. (Cselenyi et al., 2008; Wu et al., 2009; Zeng et al., 2008). In addition, the DIX–DIX domain interactions of Dvl and axin disturb self-assembly of axin homopolymers and thereby interfere with β-catenin degradation (Fiedler et al., 2011; Kishida et al., 1999; Schwarz-Romond et al., 2007).

Axin and conductin are related proteins that share key sequence elements (Behrens et al., 1998; Fagotto et al., 1999). In contrast to axin, which is constitutively expressed, conductin is a direct Wnt target gene upregulated after activation of the pathway (Jho et al., 2002; Lustig et al., 2002). Therefore, it is assumed that conductin acts as a negative-feedback regulator of Wnt signalling. However, it is unclear how conductin escapes upstream inhibition upon activation of Wnt signalling to remain active in β-catenin degradation. In principle, conductin might accumulate to levels sufficiently high to saturate available binding sites at Wnt-receptor–Dvl-complexes and might thus be able to continue degrading β-catenin. Alternatively, or in addition, conductin might be less susceptible to upstream inhibition. We provide evidence that favours the latter mechanism by showing, (1) that Wnt stimulation leads to only a modest increase in conductin protein compared to axin levels, and (2) that conductin interacts less efficiently with Dvl2 than axin making it largely resistant to inhibition mediated by Dvl2. Thus, functional rather than expression differences determine the differing roles of axin and conductin as constitutive versus inducible inhibitors of Wnt signalling.

The relative amounts of axin and conductin during Wnt signalling were determined by western blotting of extracts from Wnt3a-treated MDA-MB-231 cells. In line with previous reports, the amount of conductin increased after Wnt3a treatment for 6 h, whereas that of axin decreased (Fig. 1A, lanes 1,2; Lustig et al., 2002; Yamamoto et al., 1999). Normalisation of the western blot signals by using serial dilutions of GFP–conductin and GFP–axin present on the same blots allowed comparison of axin and conductin levels (Fig. 1A, lanes 3–6 and 7–10). This showed that conductin levels were lower than axin levels under resting conditions and, surprisingly, remained substantially lower after Wnt3a-induced upregulation of conductin and downregulation of axin (Fig. 1A, bar chart). These results largely exclude strong accumulation of conductin as a decisive factor for negative-feedback regulation. We also compared axin and conductin levels in three colorectal carcinoma cell lines, which exhibit constitutively high conductin expression due to activation of Wnt signalling because of mutations of APC (DLD1, SW480) or β-catenin (HCT116) (Lustig et al., 2002). In all these cell lines, axin levels exceeded those of conductin (Fig. 1B), confirming that conductin is only moderately upregulated by Wnt signalling.

We next compared the potency of axin and conductin in degrading β-catenin. Transfection of axin in SW480 cells led to a stronger reduction of endogenous β-catenin than transfection of conductin (Fig. 1C,D) over a range of different expression levels (Fig. 1E). Notably, axin was present in puncta, whereas conductin was more diffusely distributed, possibly explaining its lower activity in degrading β-catenin (Fig. 1C). Similarly, transiently expressed β-catenin was strongly reduced by coexpression of axin but less by coexpression of conductin (Fig. 1F).

To compare the individual contribution of endogenous axin and conductin to the inhibition of Wnt signalling, we knocked down their expression in MDA-MB-231 cells using small interfering RNAs (siRNAs) and monitored nuclear β-catenin levels by immunofluorescence staining. Depletion of axin or conductin led to an increase in β-catenin staining intensity in the absence of exogenous Wnt ligands (Fig. 2A), suggesting that both factors are active in degrading β-catenin. Of note, MDA-MB-231 cells express endogenous Wnts (Bafico et al., 2004) that might stimulate the pathway to a certain extent, but they are apparently not sufficient to fully block axin proteins. Wnt3a treatment increased β-catenin intensity, similar to treatment with the GSK3 inhibitor BIO, whereas knockdown of β-catenin reduced staining. Importantly, knockdown of conductin further increased the Wnt3a-induced β-catenin staining intensity, whereas knockdown of axin did not (Fig. 2A). The Wnt3a-induced increase of dephosphorylated β-catenin (ABC) was augmented by knockdown of conductin, but not axin, in three different cell lines, at similar knockdown rates (Fig. 2B–D). Moreover, knockdown of conductin led to a stronger increase of Wnt3a-induced TCF/β-catenin reporter activity than knockdown of axin (Fig. 2D). Thus, despite its lower expression level and activity conductin appears to retain negative regulatory activity under sustained Wnt activation, whereas axin activity is more strongly inhibited.

Recruitment of axin and conductin proteins to frizzled receptors by Dvl is widely considered the initial step towards inhibition of the β-catenin destruction complex (MacDonald et al., 2009; Metcalfe and Bienz, 2011). We therefore tested whether axin and conductin might differ in their interaction with Dvl. For this, recruitment of axin and conductin by CAAX-tagged Dvl2, a membrane-associated form of Dvl2, was determined by cell fractionation (Fig. 3A; Smalley et al., 1999). Axin was equally distributed between membrane and cytoplasmic fractions, and coexpression of Dvl2–CAAX increased the proportion of axin in the membrane fraction (Fig. 3A, lanes 1–4). Membrane recruitment of axin was strongly reduced when its DIX domain was mutated (construct axinM3; Fiedler et al., 2011) demonstrating dependence on the specific DIX–DIX-mediated interaction with Dvl2 (Fig. 3A, lanes 5–8). In contrast to axin, conductin was mainly present in the cytoplasmic fraction and only poorly recruited to the membrane by Dvl2–CAAX (Fig. 3A, lanes 9–12). Importantly, replacement of the DIX domain of conductin with that of axin, generating Cdt^AxinDIX, resulted in increased membrane recruitment of this protein by Dvl2–CAAX (Fig. 3A, lanes 13–16). These results show that conductin exhibits a weaker interaction with Dvl2 than axin, and that this is determined by the DIX domains of axin and conductin. The punctate pattern of GFP–axin in the cytoplasm was less predominant upon coexpression of Dvl2–CAAX, resulting in colocalisation of both proteins at the plasma membrane (Fig. 3B, upper panels). In contrast, GFP–axin^CdtDIX (axin containing the DIX domain of conductin) did not become associated with the membrane upon coexpression of Dvl2–CAAX (Fig. 3B, lower panels).

Like axin, Dvl2 localises in cytoplasmic puncta, and axin colocalises with Dvl2 in such puncta (Fig. 3C). The DIX domain mutant axinM3 does not form puncta (Fiedler et al., 2011) but, surprisingly, abolished formation of Dvl2 puncta (Fig. 3C). The M3 mutation impairs the head interaction surface of the axin DIX domain preventing its homopolymerisation, but leaves the tail interaction surface intact for interaction with Dvl2 (Fiedler et al., 2011). We propose that this results in inhibition of Dvl2 polymerisation, leading to smaller complexes no longer visible as puncta (Fig. 3C, see scheme). Conductin alone had a diffuse staining pattern (Fig. 1C), and colocalised with Dvl2 in puncta, albeit with a lower efficiency than axin, as indicated by its remaining partially diffuse staining pattern (Fig. 3C). ConductinM3 remained diffuse in the presence of Dvl2 but, in contrast to axinM3, did not abolish the formation of Dvl2 puncta. We propose that conductinM3 integrates to a lower extent in Dvl2 polymers than axinM3 and therefore fails to disrupt these polymers. These data support differences in the strength of the axin and conductin interaction with Dvl2.

We next analysed whether β-catenin degradation mediated by axin and conductin is differently inhibited by Dvl2. As a readout, we used immunofluorescence staining of endogenous β-catenin in SW480 cells (Fig. 4A). Axin efficiently degraded β-catenin, whereas conductin was less efficient, resulting in residual β-catenin staining (Fig. 4A,a,f). Coexpression of Dvl2–CAAX, but not of the DIX domain mutant Dvl2M2–CAAX, reduced the level of axin-mediated β-catenin degradation (Fig. 4A,b,c). Importantly, Dvl2–CAAX only weakly affected degradation mediated by conductin (Fig. 4A,g). Axin containing the DIX domain of conductin (GFP–axin^CdtDIX) became largely resistant to inhibition by Dvl2–CAAX (Fig. 4A,d,e), whereas conductin containing the DIX domain of axin (GFP–Cdt^AxinDIX) became more sensitive to inhibition by Dvl2–CAAX (Fig. 4A,h,i). These results are quantified in Fig. 4B,C. Similarly, degradation of transiently expressed β-catenin upon coexpression of axin was more efficiently inhibited by Dvl2–CAAX than degradation by conductin. Conversely, β-catenin degradation mediated by Axin^CdtDIX was unaffected by Dvl2–CAAX, whereas degradation mediated by Cdt^AxinDIX was blocked by Dvl2–CAAX (Fig. 4D,E). Taken together, these results show that the different capability of Dvl2 to interfere with axin and conductin depends on the respective DIX–DIX domain interactions. A previous bioinformatical analysis has predicted that there are ten amino acids that are likely to be required for specificity of DIX domain interactions (Ehebauer and Arias, 2009). Four of these amino acids differ between axin and conductin and, therefore, might mediate differences in their interaction with Dvl (supplementary material Fig. S1). Like Dvl2, Dvl1 and Dvl3 inhibited β-catenin degradation mediated by axin but not degradation mediated by Axin^CdtDIX (Fig. 4F). In line with the differences in inhibition of axin versus conductin mediated by Dvl2, knockdown of conductin but not of axin further increased Dvl2-stimulated activity of the TCF/β-catenin dependent reporter (Fig. 4G).

Our results shed light on the molecular basis of negative-feedback regulation in Wnt signalling by showing that the amount of conductin in Wnt-stimulated cells remained lower than that of axin and that conductin is relatively insensitive to Dvl2. This rules out mere overproduction of conductin as the basis of negative pathway regulation and favours the idea that the capacity of conductin to act as a negative-feedback regulator is based mainly on its reduced responsiveness towards upstream signalling. It remains to be determined whether the qualitative rather than quantitative feedback mode revealed here has specific consequences for the regulation of Wnt signalling. Replacement of axin by knock-in of conductin/axin2 cDNA leads to viable mice, suggesting that compensatory mechanisms can neutralise functional differences between axin and conductin in vivo (Chia and Costantini, 2005).

The separation of tasks between axin and conductin as constitutive and inducible negative regulators of Wnt signalling, respectively, is remarkable and contrasts with feedback modes in other pathways where negative regulators act in both constitutive and inducible modes (e.g. patched in hedgehog signalling). Although the reasons for this are not clear, we found that conductin was less active in β-catenin degradation than axin. It is conceivable that a gradual suppression of signalling by conductin is favoured over an abrupt block by axin to allow for temporal and spatial fine tuning of Wnt pathway activity.

Cells were grown in Dulbecco's modified Eagle's medium (DMEM) with 10% fetal bovine serum and 1% penicillin-streptomycin at 37°C under 10% CO₂. siRNAs against axin (Tanneberger et al., 2011), β-catenin and conductin (Hadjihannas et al., 2006) were transfected using Oligofectamine (Invitrogen), and plasmids were transfected using polyethylenimine (Sigma) (HEK293T, U2OS cells) or Lipofectamine 2000 (Invitrogen) (SW480 cells). Cells were treated with Wnt3a-conditioned medium (Willert et al., 2003) at 48 h after siRNA transfection. (2′Z,3′E)-6-Bromoindirubin-3′-oxime (BIO) was obtained from Sigma. TCF/β-catenin-dependent pBAR reporter activity (Major et al., 2007) was determined in HEK293T cells.

To generate GFP–axin^CdtDIX and GFP–Cdt^AxinDIX, amino acids 719–832 of rat axin and amino acids 700–840 of mouse conductin were exchanged. The M2 mutant of Dvl2 and M3 mutants of axin and conductin (Schwarz-Romond et al., 2007 and Fiedler et al., 2011, respectively) were generated using the Quikchange site-directed mutagenesis kit (Stratagene). Expression plasmids for Dvl1 and Dvl3 were provided by D. Sussman (University of Maryland, Baltimore, MD), Flag–axin (rat) by A. Kikuchi (Osaka University, Japan), GFP–axin (human) by M. Bienz (MRC, Cambridge, UK) and HA–Dvl2-CAAX by T. Dale (Cardiff University, UK).

Whole-cell extracts were prepared in Triton X-100-based lysis buffer or hypotonic buffer (experiments in Fig. 2C, upper panels, Lustig et al., 2002) or Passive Lysis Buffer (Promega) (experiments in Fig. 2D). Cytoplasmic and membrane fractions were obtained using the ProteoJET™ Membrane Protein Extraction Kit (Fermentas). Western blotting was performed as described previously (Tanneberger et al., 2011). Primary antibodies were: mouse anti-β-actin, rabbit anti-Flag, rabbit anti-HA, rabbit anti-PanCadherin (all from Sigma), rabbit anti-axin (Cell Signaling), mouse anti-conductin (C/G7; Lustig et al., 2002), mouse anti-active-β-catenin (ABC; Millipore), mouse anti-GFP (Roche) and rat anti-α-tubulin (Serotec) antibodies. Densitometric analysis was performed with AIDA 2D Densitometry software (raytest). For comparison of axin and conductin amounts in Fig. 1A,B, signal ratios obtained with anti-axin and anti-conductin antibodies were normalised to those obtained with anti-GFP antibodies.

Methanol-fixed cells were stained as described previously (Hadjihannas et al., 2006). Images were acquired on an Axioplan2 microscope using Metamorph (Zeiss). For intensity measurements, images were acquired at constant exposure times, and background-free intensities of the nuclear regions (Fig. 2) or of whole cells (Figs 1, 4) were determined.

P-values were determined using unpaired, two-tailed Student's t-tests (Fig. 1D; Fig. 2D and Fig. 4B,F,G) or significance was tested after ANOVA by post-hoc testing based on the Bonferroni method (Fig. 2A).

We thank A. Kikuchi, M. Bienz, T. Dale, and D. Sussman for reagents, and M. Brückner for technical assistance.