|Home | About | Journals | Submit | Contact Us | Français|
Canonical Wnt signaling requires inhibition of Glycogen Synthase Kinase 3 (GSK3) activity, but the molecular mechanism by which this is achieved remains unclear. Here we report that Wnt signaling triggers the sequestration of GSK3 from the cytosol into multivesicular bodies (MVBs), so that this enzyme becomes separated from its many cytosolic substrates. Endocytosed Wnt co-localized with GSK3 in acidic vesicles positive for endosomal markers. After Wnt addition, endogenous GSK3 activity decreased in the cytosol, and GSK3 became protected from protease treatment inside membrane-bounded organelles. Cryoimmuno electron microscopy showed that these corresponded to multivesicular bodies. Two proteins essential for MVB formation, HRS/Vps27 and Vps4, were required for Wnt signaling. The sequestration of GSK3 extended the half-life of many other proteins in addition to β-Catenin, including an artificial Wnt-regulated reporter protein containing GSK3 phosphorylation sites. We conclude that multivesicular endosomes are essential components of the Wnt signal transduction pathway.
Canonical Wnt signaling plays a crucial role in development, tissue regeneration, stem cells and cancer (Logan and Nusse, 2004; Clevers, 2006; MacDonald et al., 2009; Angers and Moon, 2009). A cytoplasmic destruction complex consisting of Glycogen Synthase Kinase 3 (GSK3, which has α and β isoforms), Casein Kinase 1 (CK1), Adenomatous Polyposis Coli (APC) and Axin mediates the phosphorylation of β-Catenin. Phosphorylation targets β-Catenin for polyubiquitinylation and degradation in proteasomes. In the presence of Wnt, the destruction complex becomes inactivated in ways that are incompletely understood. Wnt triggers signaling by binding to Frizzled and LDL-receptor related protein 6 (LRP6), causing the aggregation of Dishevelled (Dvl) and Axin on the plasma membrane (Bilic et al., 2007; Zeng et al., 2008). The key step in the canonical pathway is the inactivation of GSK3, for pharmacological inhibition of this enzyme elicits a typical Wnt signal. The molecular mechanism of GSK3 inhibition remains one of the main open questions in the Wnt field (Wu and Pan, 2010).
Internalization of receptor complexes is an absolute requirement for Wnt signaling (Blitzer and Nusse, 2006; Yamamoto et al., 2006). Bilic et al. (2007) discovered that cytoplasmic particles designated LRP6-signalosomes - containing aggregates of phospho-LRP6, Frizzled, Dvl, Axin, and GSK3 - are formed at and under the plasma membrane 15 min after Wnt addition. Activated Wnt receptors recruit Axin and GSK3, which phosphorylates five critical PPPS/TP sequences in the intracellular domain of LRP6 (Zeng et al., 2008; Niehrs and Shen, 2010). A number of mechanisms have been proposed to explain the inhibition of GSK3 (Kimelman and Xu, 2006). For example, the LRP6 tail may act as a direct inhibitor of GSK3 (Mi et al., 2006; Cselenyi et al., 2008; Piao et al., 2008; Wu et al., 2009). The LRP6 PPPSP repeats serve both as substrates and binding sites for GSK3 and may act as competitive inhibitors of this enzyme, although at low affinity (Ki of 1.3 × 10−5 M; Cselenyi et al., 2008).
GSK3 has many substrates in addition to β-Catenin, including Dvl, Axin and APC (Jope and Johnson, 2004). This promiscuous enzyme phosphorylates Serine or Threonine at position minus 4 of sites primed by phosphorylation (S/TXXXS/T[PO3]) (Cohen and Frame, 2001). We reported that the transcription factor Smad1 is polyubiquitinylated and degraded after GSK3 phosphorylation and is stabilized by canonical Wnt signaling, resulting in the integration of BMP and Wnt signaling (Fuentealba et al., 2007). Additional substrates destabilized by GSK3 phosphorylation have since been identified (Kim et al., 2009).
During our investigations on signaling integration, we measured GSK3 enzyme activity in Wnt-treated cell extracts (prepared in the presence of Triton X-100), and were surprised to find that Wnt did not change GSK3 activity (data not shown), even though in the direct GSK3 inhibition model one would have predicted inhibition. How could this be? Upon reflection, we realized that following ligand binding and endocytosis, growth factor receptors are incorporated into multivesicular endosomes within 15 minutes (Gruenberg and Stenmark, 2004). Multivesicular body (MVB) formation is an obligatory step before degradation in lysosomes can take place (Katzmann et al., 2002). As first discovered for EGF receptor, the topology is such that the cytoplasmic side of the plasma membrane corresponds to the lumen of the small MVB vesicles (McKanna et al., 1979). Therefore, Wnt-induced MVB formation would cause GSK3 bound to phosphorylated LRP6 cytoplasmic tails (and other GSK3 substrates such as Axin, APC, β-Catenin and Dvl) to become sequestered from its cytosolic substrates by two layers of membrane (see model in Figure 7 below), effectively inhibiting its activity.
In this study, we tested the GSK3 sequestration hypothesis of Wnt signaling. Fluorescence microscopy showed that Wnt signaling caused the relocalization of cytoplasmic GSK3 to vesicles that co-localized with endocytosed xWnt8-Venus protein, and with the MVB and lysosomal markers Rab7 and LysoTracker. Wnt signal transduction was blocked by depletion of Hrs/Vps27 or expression of dominant-negative Vps4, two proteins essential for intraluminal vesicle formation in MVBs (Katzman et al., 2002; Wollert and Hurley, 2010). Moreover, Wnt treatment decreased cytosolic GSK3 activity levels (measured in Digitonin-permeabilized cells), yet full enzyme activity was recovered after solubilizaton of all membranes with Triton X-100. Similarly, Wnt caused endogenous GSK3β to become partially protected from Proteinase K digestion, but only in the absence of Triton X-100. Finally, cryoimmuno electron microscopy showed that GSK3 was indeed translocated from the cytosol into MVBs after Wnt pathway activation. Bioinformatic analyses revealed that 20% of the human proteome contains multiple putative GSK3 phosphorylation sites. Total protein half-life was extended by Wnt treatment or GSK3 inhibition. The addition of GSK3 phosphorylation sites was sufficient to place the stability of a Green Fluorescent Protein (GFP) biosensor under the control of Wnt. We conclude that canonical Wnt signaling sequesters GSK3 inside MVBs, reducing its cytosolic levels and extending the half-life of many GSK3 substrates.
We first asked whether Wnt treatment changed the subcellular localization of GSK3. Human 293T cells expressing xWnt8-Venus (Mii and Taira, 2009) were cultured together with mouse 3T3 cells transfected with GSK3-RFP. Remarkably, endocytosed Wnt-Venus and GSK3-RFP accumulated in the same vesicular structures (arrows in 1F), while GSK3-RFP levels decreased in the cytosol (Figures 1A–1F). Relocalization of endogenous GSK3β was also observed in these co-cultures (Figure S1A–S1C). Wnt-Venus puncta were counted in each responding cell (n=80), and 56±9% of them co-localized with GSK3-RFP puncta (see histogram in Figure 1C′). Thus, cytosolic GSK3 decreases and becomes relocalized to the same endosomes as the internalized Wnt ligand.
Transfection of constitutively-active LRP6 lacking its extracellular domain, designated CA-LRP6, causes a potent Wnt signal (Tamai et al., 2004). CA-LRP6 cytoplasmic signalosome formation (Bilic et al., 2007) required Dynamin, and caused a striking relocalization of GSK3β from the cytosol into prominent cytoplasmic puncta (Figures 1G–1L′ and S1D–S1O). Treatment of 3T3 cells with LysoTracker, a dye that becomes concentrated in acidic organelles such as MVBs and lysosomes, showed that endogenous GSK3β relocated to these compartments (Figure 1G–1L′). In addition, GSK3β co-localized with Dvl, Axin and LysoTracker, as well as the PI3P probe FYVE-GFP, when Wnt signaling was activated by overexpression of Dishevelled (Figure S2A-S2O). Remarkably, a protein consisting of the DIX domain of Dvl fused to the LRP6 cytoplasmic tail (DIX>Ctail-GFP), which triggers a very strong Wnt signal (Metcalfe et al., 2010), also formed signalosomes co-localizing with acidic vesicles that sequestered cytosolic GSK3β (Figure S2P–S2Aa).
CA-LRP6 signalosomes co-localized with the late endosomal marker Rab7-GFP (Figure 1M 1R) in 62±7% of the GSK3 puncta, indicating that GSK3-RFP relocated to MVBs or lysosomes (Bucci et al., 2000). About 40% of GSK3-RFP vesicles co-localized with Vps4-GFP (see Figure 3L–3N′ below), a marker of the final stages of intraluminal vesicle formation in endosomes (Bishop and Woodman, 2000; Gruenberg and Stenmark, 2004). In untransfected 3T3 cells, the number and size of endogenous GSK3β puncta increased after treatment with Wnt3a conditioned medium (Figure 1S–1U) in cells permeabilized with Digitonin (which facilitates the visualization of intracellular organelles, Bishop and Woodman, 2000).
Taken together, these results strongly support the hypothesis that Wnt signaling causes the relocalization of cytosolic GSK3 to the endosomal compartment.
To determine whether Wnt3a treatment sequesters GSK3 kinase activity from cytosol, endogenous cytosolic kinase activity was measured in untransfected L-cells permeabilized with Digitonin through the incorporation of phosphate from [γ32P]-ATP into a phospho-Glycogen Synthase peptide substrate (Ryves et al., 1998). Digitonin solubilizes Cholesterol-rich patches in the plasma membrane, leaving MVBs and other intracellular organelles intact (Dunn and Holz, 1983). Addition of Wnt3a for 4 hours decreased cytosolic GSK3 activity levels by 66±5% (Figure 2A, lanes 1 and 2), and the missing GSK3 activity was recovered when all membranes were dissolved with 0.1% Triton X-100 (Figure 2A, compare lanes 2 and 4).
The gold standard to determine the localization of a protein inside a membrane-bounded compartment is the protease protection assay. GSK3 protein became protected from Proteinase K digestion (Vanlandingham and Ceresa, 2009) after Wnt3a treatment (Figure 2B, compare lanes 3 and 4) in untransfected L-cells permeabilized with Digitonin. This Wnt-dependent protease protection of GSK3 was eliminated when membranes were solubilized with 0.1% Triton X-100 (Figure 2B, lanes 4 and 6). As a negative control, we used α-Tubulin, which is not contained in vesicular organelles and was not protected from Proteinase K digestion (Figure 2B). These experiments strongly suggest that GSK3 becomes sequestered within membrane-bounded organelles upon Wnt treatment.
The relocalization of GSK3 to multivesicular endosomes in Wnt3a-treated cells was visualized by cryoimmuno electron microscopy. In untransfected 3T3 cells treated with control conditioned medium, endogenous GSK3β was found almost exclusively in the cytosol, whereas in Wnt3a-treated cells a substantial fraction was found inside MVBs (Figure 2C and 2D). In HeLa cells co-transfected with CA-LRP6 and GSK3-GFP, an anti-GFP antibody revealed colloidal Gold particles in MVBs (Figure 2E). In some cases Gold particles were observed on the cytoplasmic surface of vesicles fusing with multivesicular endosomes (Figure 2F), as well as within the small vesicles that fill MVBs (arrows with asterisks in Figure 2F), supporting the topology shown in Figure 7 below. In the absence of CA-LRP6, Gold-labeled GSK3-GFP was located in the cytosol (Figure 2G).
To confirm the cryoimmuno localization results in a quantitative way, we resorted to an activated mutant of Rab5. Rab5-Q79L causes the formation of giant multivesicular endosomes containing large numbers of intraluminal vesicles (Wegener et al., 2010). These MVBs are so large that the outer membrane (outlined by Rab5-QL-DsRed) can be readily distinguished from its internal contents by light microscopy. As shown in Figure 2H-2M, GSK3β translocated from the cytoplasm into the interior of Rab5-QL multivesicular endosomes (arrows, 77±9% co-localization) when CA-LRP6 was co-expressed, but not in its absence. Cytosolic depletion of GSK3-GFP was very clear (compare Figure 2J to 2M).
Taken together, these results demonstrate that the GSK3 enzyme becomes translocated from the cytosol into membrane-bounded multivesicular endosomes when Wnt signals.
The molecular machinery that forms endosomal intraluminal vesicles has been well characterized (Wollert and Hurley, 2010). Several Endosomal Sorting Complexes Required for Transport, or ESCRTs, have been identified through yeast genetics (designated Vacuolar Protein Sorting, Vps, mutants) and biochemistry (Katzmann et al., 2002). The GSK3 sequestration hypothesis predicts that ESCRT proteins essential for vesicle invagination should be required for Wnt signaling. We therefore tested two ESCRT proteins essential for vesicle formation (Figure 3A).
HRS/Vps27 (Hepatocyte growth factor Regulated tyrosine-kinase Substrate) initiates formation of the ESCRT-0 complex. Depletion of HRS by siRNA blocked the accumulation of β-Catenin observed after 2 hours of Wnt3a treatment in 293T cells, while control scrambled siRNA had no effect (Figure 3B, lanes 1–4). Total GSK3 levels increased by about 70% when HRS was depleted (Figure 3B compare lanes 1 and 3, three independent experiments; see also Figure S3), suggesting that GSK3 is normally partially degraded by the endosomal machinery; however, Wnt addition did not significantly change GSK3 levels in these experiments. We were expecting a decrease in total GSK3 levels in Western blots, because MVBs are usually targeted to the lysosome and degradation. Many reasons might explain this result. For example, GSK3 may have a long half-life, it may be replenished by translational regulation, or GSK3 may have a long time of residency in MVBs induced by Wnt signaling before lysosomal degradation takes place (Wnt might even affect the rate of MVB processing globally). In addition, there is recent evidence that intraluminal MVB vesicles may be recycled back into the cytosol by “back-fusion” to the late endosome limiting membrane (Falguieres et al., 2009). Despite the lack of change in total levels, the relocation of GSK3 from the cytosol into MVBs is clearly documented in the Figures above, providing the basis for the sequestration model for Wnt signaling presented in Figure 7.
The requirement of HRS for Wnt signal transduction was demonstrated directly in these cultures by measuring the expression of the SuperTopFlash-Luciferase reporter, which contains multiple TCF binding sites (Figure 3C, brackets). In addition, we observed that relocalization of GSK3 into vesicle-like structures in CA-LRP6 transfected cells was blocked by HRS siRNA, and was not affected by control scrambled siRNA (Figure 3D-3E′). In animal cap explants, the activation of TCF-Luciferase by CA-LRP6 mRNA was blocked by HRS antisense morpholino (Figure 3F, brackets). In Xenopus embryos, injection of CA-LRP6 mRNA into a single ventral blastomere at the 8-cell stage caused complete axis duplications, which were eliminated by co-injection of HRS MO (Figures 3G-3I and S4N). HRS depletion did not affect cell viability or proliferation until neurula stage (Figure S4A–S4M). The effects of HRS MO could be partially rescued by human HRS mRNA (Figure 4C and 4O below).
Vps4 is an ATPase required for the disassembly of ESCRT-III complexes, the last step in the pinching off of intraluminal vesicles in multivesicular endosomes (Gruenberg and Stenmark, 2004). A point mutation in the ATPase site (Vps4-EQ) creates a potent dominant-negative form that inhibits MVB formation (Bishop and Woodman, 2000). Vps4-EQ blocked canonical Wnt3a signaling (Figure 3J, brackets). The control in this experiment was Vps4-WT, which differs by only one amino acid yet was without effect in TCF-Luciferase assays. Vps4-EQ also inhibited the transcriptional effects of CA-LRP6 (Figure 3K, brackets). Vps4-EQ co-transfection inhibited the relocalization of endogenous cytosolic GSK3β into CA-LRP6 signalosomes (Figure 3L-3Q). In Xenopus embryos, Vps4-EQ, but not Vps4-WT, co-injected with CA-LRP6 mRNA into a ventral blastomere inhibited the formation of complete secondary axes containing head structures (Figures 3R–3T and S4O). Importantly, we also tested the requirement of the ESCRT machinery for axis induction by Siamois, a homeobox gene activated by Wnt signaling. We found that Vps4-EQ mRNA was unable to inhibit Siamois secondary axes (see Figures 4Ba–4Da and S5A below). This epistatic experiment placed the requirement for Vps4 early in the Wnt pathway, upstream of its transcriptional effector Siamois.
These experiments show that two components of the ESCRT machinery, HRS/Vps27 and Vps4, are required for canonical Wnt signaling.
While investigating the requirements for GSK3 sequestration, we made an unexpected finding: the translocation of GSK3 into CA-LRP6 signalosomes and its depletion from the cytosol was inhibited by β-Catenin siRNA (Figures 4A–4B and S3C). Using SuperTopFlash reporter in animal caps, we found that HRS MO blocked signaling by microinjected β-Catenin mRNA and was partially rescued by human HRS mRNA co-injection (Figure 4C). We also noticed that endogenous β-Catenin, especially its GSK3-phosphorylated form (i.e., targeted for degradation), co-localized with CA-LRP6-GFP in signalosomes (Figure 4D–4F). Taken together, these results suggest that β-Catenin is required for the sequestration of GSK3 in multivesicular endosomes.
We next asked whether β-Catenin overexpression was sufficient to trigger GSK3 sequestration in endosomes. Stabilized β-Catenin-GFP (in which its three GSK3 sites were mutated into alanines) accumulated in the nucleus as expected, but also in cytoplasmic vesicle-like structures which sequestered GSK3-RFP (Figure 4G–4I). Wild-type β-Catenin-GFP accumulated inside giant Rab5-QL MVBs (Figure 4J–4L) and in LysoTracker-positive vesicles (Figure S3D–S3F). Injection of β-Catenin mRNA into a ventral cell causes twinning in Xenopus embryos. Co-injection of β-Catenin mRNA with HRS-MO or Vps4-EQ showed that MVB formation was required for β-Catenin axis induction in Xenopus (Figures 4M–4R, S4P and S4Q).
This new function of β-Catenin in sequestering GSK3 in MVBs occurs upstream of the well-established transcriptional role of β-Catenin and Tcf3 in Wnt signaling (Clevers, 2006). Depletion of β-Catenin with MO generated ventralized Xenopus embryos lacking all neural structures (Figure 4S and 4T). Although GSK3 inhibition by DN-GSK3β mRNA (a dominant-negative catalytically inactive form) greatly expanded neural structures, this effect was blocked by β-Catenin or xTcf3 depletion (Figure 4U–4W). However, a construct fusing the transactivation domain of Xenopus β-Catenin to DN-xTcf3 was able to signal in β-Catenin-depleted embryos (Figure 4X). In secondary axis induction assays, the β-Catenin-DN-xTcf3 fusion was active even when MVB formation was inhibited by Vps4-EQ mRNA microinjection (Figures 4Y-4Aa and S5B). These experiments show that the classical transcriptional role of β-Catenin/Tcf3 is distinct from its new function in facilitating GSK3 sequestration.
We conclude that β-Catenin protein is required and sufficient to cause GSK3 sequestration in acidic cytoplasmic endosomes. It has been proposed that the entire destruction complex (which includes β-Catenin) is recruited to phosphorylated LRP6 receptors (Zheng et al., 2008). Once β-Catenin levels begin to rise during Wnt signaling, β-Catenin would enhance the signal, forming a feed-forward loop by facilitating GSK3 sequestration.
Phosphorylation of β-Catenin or Smad1 by GSK3 causes recognition by E3 polyubiquitin ligases and protein degradation (Logan and Nusse, 2004; Fuentealba et al, 2007). To investigate whether GSK3 sequestration causes the stabilization of other proteins in addition to β-Catenin and Smad1, we first designed computer algorithms to identify potential phosphate-primed GSK3 sites in the human proteome (Figure 5A) and whether they were evolutionary conserved. A large number of proteins (20% of the proteome) were found to contain three or more consecutive potential GSK3 sites (Tables S1 and S3), significantly more than expected from random distribution (Table S1). These included many novel putative GSK3 targets with known regulatory functions (Table S2),
To determine whether Wnt/GSK3 signaling regulates overall protein stability, we carri out radioactive pulse-chase experiments with 35S Methionine in untransfected 293T cells. Cultured cells were radioactively labeled for 30 min, washed, and then treated with Wnt3a or control conditioned medium containing a 4-fold excess of unlabeled Methionine. As shown in Figure 5B, the total half-life of labeled cellular proteins increased from 9.3 to 11.8 hours in Wnt3a medium. Electrophoresis and autoradiography after 6 hours of cold chase showed that a wide range of proteins were stabilized by Wnt treatment (Figure 5C, lanes 1 and 2), even when protein synthesis was inhibited with Cycloheximide (Figure 5C, lanes 3 and 4; Figure S6A). Thus, the effect of Wnt on radioactive protein stability is direct and not due to the expression of genes activated after Wnt is added.
Wnt-stabilized proteins were detected mostly in the 35–150 kDa range (Figure 5D and 5F). The stabilization of radioactive proteins by Wnt was eliminated by depletion of β-Catenin with siRNA (Figure 5D–5H). This requirement for β-Catenin in protein stabilization is explained by our finding that β-Catenin is essential for the sequestration of GSK3 (Figure 4). Axin is required for LRP6 signalosome formation and signaling (Bilic et al., 2007) and also for overall protein stabilization by Wnt (Figure S6F).
Protein stabilization by Wnt3a mimicked the effects of GSK3 inhibition. In 293T cells, depletion of GSK3α had little effect on protein stability, but depletion of GSK3β, or GSK3α and β, stabilized the same range of proteins as Wnt3a treatment (Figure 5I, lanes 1–5). A chemical inhibitor of GSK3, BIO, caused a similar stabilization, while the proteasome inhibitor MG132 generated a different profile in which proteins below the 35 kDa range were also strongly stabilized (Figure 5I–5M). Thus, although Wnt signaling stabilized many proteins, these represented only a subset of the total proteome.
These results suggest that Wnt signaling and GSK3 activity are regulators of the half-life of many cellular proteins. Some of the novel putative GSK3 targets listed in Table S2, HDAC4 and JunB, were tested and found to be stabilized by Wnt treatment (Figure S6B–S6E). In addition, the half-life of Smad4 was also regulated by Wnt, resulting in increased TGFβ signaling (H. Demagny and E.M.D.R., in preparation). Testing all possible Wnt/GSK3 targets would be a daunting task, so we took a different approach, generating an artificial protein that serves as a biosensor of Wnt/GSK3activity.
What would be the effect of adding GSK3 sites to a protein that lacked them? An artificial GFP construct was generated in which three GSK3 sites primed by a canonical MAPK phosphorylation site (PXSP) were added to its carboxy-terminus via a row of alanines (Figure 6A). A Flag-tag and a PPAY site for the binding of E3 polyubiquitin ligases of the HECT family (mimicking Smad1; Fuentealba et al., 2007) were also added. The HECT E3 ligase site helped, but was not essential (Figure S7B); even in its absence the biosensor protein could be destabilized by another E3 ligase, βTrcp (Figure S7C).
Cells were transfected with GFP-GSK3-MAPK construct, split into two cultures and treated with control or Wnt3a conditioned medium for 6 hours. The GFP-GSK3-MAPK reporter protein was stabilized by Wnt3a treatment (Figure 6C, compare lanes 2 and 3; Figure 6E and 6F). Remarkably, the degree of stabilization of this biosensor protein was comparable to that of endogenous β-Catenin, while α-Tubulin remained unchanged (Figure 6C). The stabilization caused by Wnt could be mimicked by co-transfection of DN-GSK3 (Figure 6C, lane 4), or by treatment with the GSK3 inhibitor BIO (data not shown). To demonstrate that stabilization by Wnt indeed occurred at a post-translational level, we mutated the GSK3 sites into alanines, generating GFP-GSK3mut-MAPK (Figure 6B). This GSK3 phosphorylation-resistant protein became stabilized and was no longer affected by Wnt3a treatment (Figure 6D, 6G and 6H). Thus, the presence of GSK3 phosphorylation sites is sufficient to destabilize GFP. The type of priming kinase was not critical, because constructs primed by CK1 were similarly stabilized by Wnt (Figure S7E).
These studies with a GFP reporter protein indicate that multiple consecutive GSK3 phosphorylation sites make a protein less stable, presumably by favoring recognition by E3 polyubiquitin ligases of the βTrcp or HECT types (Logan and Nusse, 2004; Fuentealba et al., 2007). The results indicate that Wnt signaling is able to deplete GSK3 from the cytosol to levels low enough to stabilize a biosensor of cellular GSK3 activity. Since the proteome contains a multitude of GSK3 phosphorylation targets, we expect that canonical Wnt signaling will have many previously unsuspected metabolic effects.
We investigated the cellular mechanism by which Wnt signaling causes GSK3 inhibition. The results support a new model for canonical Wnt signal transduction, in which the GSK3 enzyme becomes sequestered inside multivesicular endosomes triggered by activation of the Frizzled and LRP6 receptors, decreasing GSK3 levels in the cytosol (Figure 7). The key insight that led to the GSK3 sequestration hypothesis was the realization that during the normal endocytosis process early endosomes containing activated receptors are packaged within multivesicular endosomes (McKanna et al., 1979; Katzmann et al., 2002; Gruenberg and Stenmark, 2004). GSK3 is recruited to the cytoplasmic side of Wnt receptor complexes and phosphorylates LRP6 and other substrates such as Dvl, APC, Axin and β-Catenin (MacDonald et al., 2009; Table S2), to which it normally binds. The key to transcriptional activation by Wnt signaling is the nuclear accumulation of β-Catenin. Although β-Catenin protein initially translocates into MVBs together with GSK3, once cytosolic levels of GSK3 are sufficiently depleted, newly translated β-Catenin is not phosphorylated and becomes stabilized, accumulating in the nucleus. Many cellular proteins contain putative GSK3 phosphorylation sites (Table S3) and are also candidates for stabilization by canonical Wnt signaling.
The results strongly support the GSK3 sequestration hypothesis, providing a solution to the longstanding question of how GSK3 activity is inhibited by Wnt. In co-culture experiments, GSK3 accumulated in endosomes containing endocytosed xWnt8-Venus, and was depleted from the cytosol. Protease protection studies showed that Wnt3a treatment causes the relocalization of GSK3 inside membrane-bounded organelles. These corresponded to MVBs in cryoimmuno electron microscopy. Remarkably, Wnt pathway activation by overexpression of CA-LRP6, Dvl or of Dvl DIX>Ctail construct (Tamai et al., 2004; Capelluto et al., 2002; Metcalfe et al., 2010), relocalized GSK3 into acidic endosomes, decreasing levels of this enzyme in cytosol (Figures 1 and S2). We also observed reductions in cytosolic GSK3 by addition of Wnt3a protein acting on its endogenous receptors (Figures 1A–1C, S1A–S1C, 1S–1U, 2A–2D, 3B, 3C, ,55 and and6).6). Finally, two components of the molecular ESCRT machinery required for the formation of MVBs, HRS and Vps4, were shown to be essential for canonical Wnt signaling in TCF transcriptional reporter and Xenopus embryo essays. In epistatic experiments, ESCRT activity was not required for axis induction by Siamois, a transcriptional target downstream of Wnt signaling.
How does this GSK3 sequestration model fit in with the present state of knowledge concerning Wnt signaling? Receptor internalization is required for Wnt to signal (Blitzer and Nusse, 2006; Yamamoto et al., 2006). The proportion of Frizzled receptor destined for degradation versus recycling back to the membrane is regulated by a ubiquitinylation/deubiquitinylation cycle (Mukai et al., 2010), a process that was not investigated here. Bilic et al (2007) found that aggregates containing Dvl, phosphorylated LRP6, Axin and GSK3, are induced by Wnt treatment. Designated as LRP6-signalosomes, these aggregates are very prominent in cells in which the Wnt pathway is activated by CA-LRP6, Dvl, or DIX-Ctail overexpression (Bilic et al., 2007, Metcalfe et al., 2010). LRP6-signalosomes were proposed to provide a platform for Wnt signaling on the cytosolic side of early endosomes. The requirement for HRS and Vps4 reported here demonstrates that Wnt signaling requires the formation of multivesicular endosomes (Katzman et al., 2002; Gruenberg and Stenmark, 2004). There is no contradiction between LRP6-signalosome model and our results. Sequestration of GSK3 in MVBs is not alone responsible for Wnt signaling. In the early steps of the process, formation of a vesicular platform facing the cytosol recruits the various components to the plasma membrane and early endosomes. In the presence of DN-Dynamin, GSK3-RFP strongly accumulates in the plasma membrane when Wnt signaling is activated (Figure S1F), although Wnt signaling does not take place (Blitzer and Nusse, 2006). Wnt signaling requires the completion of the next step in the endocytic process. Once the neck of the intraluminal vesicles of MVBs is closed, GSK3 will be prevented from equilibrating with the cytosol, explaining why Wnt signaling requires a functional ESCRT machinery.
There is a discrepancy between the requirement of HRS in Wnt signaling described here and studies with HRS mutants in Drosophila. Seto and Bellen (2006) reported enhanced expression of Wg target genes in HRS clones (proposed to be due to decreased lysosomal degradation of Wg), although other authors reported no differences (compared to wild-type tissue) in similar clones (Rives et al., 2006). Perhaps in Drosophila some degree of MVB formation is still achieved in the absence of HRS through redundancy in the ESCRT machinery. In the studies presented here, we obtained significant inhibition of canonical Wnt signaling with HRS siRNA, HRS MO, and Vps4-EQ, both in cultured cells and in vivo in Xenopus.
Our observations help understand the paradoxical positive and negative effects of GSK3 in early Wnt signaling (Zeng et al., 2008). GSK3 phosphorylation of LRP6 initiates Wnt signaling (MacDonald et al., 2009; Niehrs and Shen, 2010), yet genetic and pharmacological studies show that GSK3 inhibition is sufficient to trigger the canonical Wnt signal (Logan and Nusse, 2004). The sequestration hypothesis helps resolve these apparently contradictory functions, for LRP6 phosphorylation by GSK3 not only initiates signaling but also causes the sequestration of GSK3 from the cytosol.
The key to canonical Wnt signaling is GSK3 inhibition, which results in β-Catenin stabilization (Kimelman and Xu, 2006; Wu and Pan, 2010). Several studies have shown that phosphorylated LRP6 intracellular domain can inhibit GSK3 activity (Mi et al., 2006; Cselenyi et al., 2008; Piao et al., 2008; Wu et al., 2009), although with low affinity (Ki in the 10−5 M range). However, GSK3 inhibition has been difficult to demonstrate biochemically. One exception was a study using hypotonic cell lysis, which reported a transient 40% decrease in GSK3 activity that peaked 10 min after Wnt addition (Ding et al., 2000). This early inhibition is likely explained via direct binding of GSK3 to phospho-LRP6. However, this decrease in GSK3 activity disappeared after 1 hour (Ding et al., 2000), while stabilized β-Catenin protein continues to accumulate for at least 6 hours after Wnt treatment (Blitzer and Nusse, 2006). We now found that Wnt3a treatment reduced cytosolic GSK3 activity levels by 66% after 4 hours of Wnt3a addition in cells permeabilized with Digitonin (Figure 2A).
Wnt signaling requires the sustained inhibition of GSK3, which is achieved by sequestration inside MVBs. The initial recruitment of the destruction complex (Zeng et al., 2008) would ensure that the GSK3 responsible for β-Catenin degradation is depleted first (explaining in part the insulation from other signaling pathways), followed by other cytosolic GSK3 molecules that bind to additional phosphorylated docking sites provided by GSK3 phosphorylation sites in LRP6, Axin, β-Catenin, Dishevelled and APC (Jope and Johnson, 2004; Table S2). Thus, although GSK3 is in stoichometric excess to Axin, there are many additional potential GSK3 binding sites present in the Wnt receptor signaling complex (Figure 7). Catalytically inactive DN-GSK3β is not concentrated in MVBs (compare Figure S1G to S1M), suggesting that GSK3 accumulates in MVBs because it binds to its own substrates. Conversely, a GSK3 form that does not bind Axin but is catalytically active can accumulate in LRP6 signalosomes (Figure S1J–S1L), demonstrating that GSK3 molecules not directly associated with the Axin destruction complex can still be sequestered by Wnt signaling. By binding to its own substrates, GSK3 can be depleted from cytosol even when LRP6 levels are stoichiometrically lower.
Did precedents for a role for the MVB pathway in Wnt signaling exist in the literature? Early on, van der Heuvel et al. (1989) reported that endocytosed Wg accumulates in the MVB matrix in cells responding to the Wg signal in Drosophila blastoderm. Acidification by vacuolar/H+-ATPase is required for canonical Wnt signaling (Cruciat et al., 2010). Dvl overexpression triggers a canonical Wnt signal, and it was reported that this activity correlated with its co-localization with phospholipids in vesicular membranes via its DIX-domain (Capelluto et al., 2002). Recently, a fusion protein between the Dvl DIX domain and the Ctail of LRP6 was found to be a very potent activator of Wnt signaling (Metcalfe et al., 2010). Dvl and DIX>Ctail form signalosomes that were initially interpreted as cytoplasmic protein aggregates. However, as shown in Figure S2, the Dvl and DIX>Ctail particles co-localized with FYVE-GFP (a PI3P marker) and also with the acidic dye LysoTracker. These endosomal vesicles depleted GSK3 from the cytoplasm. Their effect was specific, since point mutations in the DIX domain (Metcalfe et al., 2010) prevented co-localization with LysoTracker and GSK3. Moreover, inhibiting MVB formation prevented induction of a TCF-Luciferase reporter by the Dvl DIX/LRP6 fusion protein (Figure S2Ba). In conclusion, previous observations were consistent with a role for membrane vesicles in Wnt signaling, but their identity as MVBs that sequester GSK3 was not recognized.
An unexpected finding was that Wnt prolongs the overall rate of cellular protein degradation (Figure 5). The addition of three consecutive GSK3 sites was sufficient to convert GFP into a Wnt-responsive protein (Figure 6). Thus, cytosolic levels of GSK3 are sufficiently depleted by Wnt treatment to stabilize other proteins in addition to β-Catenin. Thus, GSK3 emerges as a protein kinase with a central role in regulating protein catabolism.
Bioinformatic analyses showed that about 20% of the human proteome contains three or more possible primed GSK3 sites, suggesting that many putative Wnt-regulated proteins exist. The principal transcriptional output of Wnt signaling is the interaction between TCF/LEF and β-Catenin (Clevers, 2006). However, many additional transcription factors interact with β-Catenin (reviewed by MacDonald et al., 2009, see their Table S1). It is intriguing that some of the proteins potentially stabilized by Wnt in our list (Table S2) also interact with β-Catenin (e.g., FoxO, HIF1α, Mitf, RAR). Therefore, Wnt signaling may have unforeseen regulatory complexities. While Wnt signaling can cross-talk with other signal transduction pathways in which GSK3 phosphorylations are involved, such as the BMP/Smad1 pathway (Fuentealba et al., 2007), insulation between pathways is also necessary. The degree of insulation will depend on the relative affinities of the various substrates and their resistance to decreases in levels of free GSK3. In addition, cross-talk versus insulation will depend on the type (and activity level) of priming kinase used by the various signaling pathways. The effect of Wnt on overall protein half-life had an absolute requirement for β-Catenin (Figure 5I). This is likely explained by the finding that β-Catenin is required, and its overexpression sufficient, for GSK3 sequestration in MVBs (Figure 4).
Our findings indicate that canonical Wnt provides a signal instructing cells to slow down degradation of many proteins. The recent discovery that LRP6 is phosphorylated at G2/M phase independently of Wnt (Davidson et al., 2009) suggests that the half-life of a multitude of GSK3 target proteins could be under cell cycle control. The list of putative GSK3/Wnt targets identified by our proteomic analyses suggests that Wnt/GSK3 signaling may have wide metabolic effects. In addition, it provides a rich resource for discovering possible novel nodes of signaling pathway integration (Table S3).
Targeting of membrane receptors to MVBs and lysosomes negatively regulates signaling by many growth factor receptors (McKanna et al., 1979; Katzmann et al., 2002). A possible positive role for multivesicular endosomes in the release of Notch intracellular domain to the nucleus has been reported (Coumailleau et al., 2009). In the case of canonical Wnt, the positive role of endosomal trafficking is of a different nature, for it is the sequestration of GSK3 in MVBs that generates the signal (Figure 7). This raises the question of whether other plasma membrane receptors might use multivesicular endosomes to sequester proteins bound to their cytoplasmic tails. In this view, specific intracellular proteins could be targeted to MVBs in a growth factor-dependent manner. All eukaryotic cells contain an elaborate ESCRT machinery that generates multivesicular endosomes. Further studies on how the cell biology of endocytic trafficking intersects with signaling by Wnt and other growth factors may have implications for understanding human disease, including cancer.
The following GFP or RFP fusion proteins were generated for this study: GSK3-RFP, GSK3-GFP, DN-GSK3-GFP, CA-LRP6-GFP, xDvl-RFP, stabilized mutant β-Catenin-GFP, human HRS-RFP, GFP-GSK3-MAPK reporter, GFP-GSK3mut-MAPK, GFP-GSK3-MAPK-E3mut and GFP-GSK3-CK1-E3. A pCS2 vector containing amino-terminal Flag-EGFP (GB#U55763) or HA-RFP (GB#AF506027) was used to clone in-frame fusions generated by PCR from ESTs or from constructs generously provided by colleagues listed in the acknowledgments. The siRNAs targeting human β-Catenin, HRS and Axin1 were ON-TARGETplus SMARTpool from Thermo Scientific #L-003482, #L-016835 and #L-009625, respectively. The siRNAs used to knock-down human GSK3α and GSK3β were Validated Stealth RNAi DuoPak from Invitrogen #45-3210 and #45-1488, respectively.
For immunostainings, primary antibodies were: anti-GSK3β monoclonal (BD Transduction #610201) at 1:350 and anti-phospho-β-Catenin (Cell Signaling #9561) at 1:350. For cryoimmuno electron microscopy, antibodies used were chicken IgY anti-GSK3β (Sigma #GW22779) at 1:800 and chicken IgY anti-GFP (Invitrogen #A10262) at 1:250. For Western blots, primary antibodies used were the anti-GSK3β monoclonal at 1:1000, anti-β-Catenin (Sigma #C2206) at 1:4000, anti-α-Tubulin monoclonal (Calbiochem #CP06) at 1:1500, anti-Flag monoclonal (Sigma #F1804) at 1:1500, and anti-Total-Erk (Cell Signaling #9102) at 1:1000. Secondary antibodies coupled to Infra Red Dyes (IRDye 680 and IRDye 800) at 1:3000 (LI-COR) were used, and Western blots were analyzed using a LI-COR Odyssey scanner system.
Results are given as the mean ± standard error of the mean (SEM). Statistical analyses were performed with Excel (Microsoft Co) applying the two-tailed t test, as appropriate. Differences of means were considered significant at a significance level of 0.05.
Detailed methods for cell culture, immunostainings, GSK3 enzyme activity measurements, protease protection, cryoimmuno electron microscopy, Xenopus embryo assays, bioinformatic proteomic analyses and radioactive pulse-chase experiments are provided in Supplemental Experimental Procedures.
We thank Drs. R. Moon for β-Catenin-GFP and SuperTopFlash reporter, X. He for CA-LRP6, B. van Deurs for dog Rab7-GFP, S. Sokol for Myc-xDvl, M. Bienz for DIX>Ctail-GFP, M. Taira for xWnt8-Venus, P. Woodman for Vps4-EQ-GFP, R. Pagano for Rab5-DsRed-WT, P. De Camilli for DN-Dynamin-GFP, H. Stenmark for FYVE-GFP, H. Clevers for DN-xTcf3, O. Wessely for β-Catenin-DN-xTcf3, P. Lemaire for Siamois, and R. Nusse for Wnt3a-producing L-cells. We thank three anonymous referees and members of our laboratories for improving the manuscript, H. Snitkin for her help in processing cultured cells for electron microscopy, U. Lendahl for suggesting the HRS depletion experiments, the Deutsche Forschungsgemeinschaft for supporting R.D. (DO1429/1-1), and the NIH (HD21502-24) for funding. E.M.D.R. is an Investigator of the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.