Endocytic Adaptor Protein Tollip Inhibits Canonical Wnt Signaling

Many adaptor proteins involved in endocytic cargo transport exhibit additional functions in other cellular processes which may be either related to or independent from their trafficking roles. The endosomal adaptor protein Tollip is an example of such a multitasking regulator, as it participates in trafficking and endosomal sorting of receptors, but also in interleukin/Toll/NF-κB signaling, bacterial entry, autophagic clearance of protein aggregates and regulation of sumoylation. Here we describe another role of Tollip in intracellular signaling. By performing a targeted RNAi screen of soluble endocytic proteins for their additional functions in canonical Wnt signaling, we identified Tollip as a potential negative regulator of this pathway in human cells. Depletion of Tollip potentiates the activity of β-catenin/TCF-dependent transcriptional reporter, while its overproduction inhibits the reporter activity and expression of Wnt target genes. These effects are independent of dynamin-mediated endocytosis, but require the ubiquitin-binding CUE domain of Tollip. In Wnt-stimulated cells, Tollip counteracts the activation of β-catenin and its nuclear accumulation, without affecting its total levels. Additionally, under conditions of ligand-independent signaling, Tollip inhibits the pathway after the stage of β-catenin stabilization, as observed in human cancer cell lines, characterized by constitutive β-catenin activity. Finally, the regulation of Wnt signaling by Tollip occurs also during early embryonic development of zebrafish. In summary, our data identify a novel function of Tollip in regulating the canonical Wnt pathway which is evolutionarily conserved between fish and humans. Tollip-mediated inhibition of Wnt signaling may contribute not only to embryonic development, but also to carcinogenesis. Mechanistically, Tollip can potentially coordinate multiple cellular pathways of trafficking and signaling, possibly by exploiting its ability to interact with ubiquitin and the sumoylation machinery.


Introduction
Adaptor proteins act as molecular scaffolds in various intracellular processes [1]. Usually lacking enzymatic activities, adaptors mediate protein-protein and protein-lipid interactions Tollip can function as an inhibitor of canonical Wnt signaling in cultured mammalian cells, both normal and cancerous, acting independently of dynamin-mediated endocytosis. We also reveal a role of Tollip in Wnt signaling in embryonic development of zebrafish: interfering with Tollip function in fish results in phenotypes reminiscent to those of canonical Wnt signaling mutants. Cumulatively, our results point to a previously unknown function of Tollip in intra-and intercellular signaling.

Ethics Statement
No specific ethics approval under Polish, Swiss and EU guidelines was required for this project, as all zebrafish used in this study were between 0 and 5 days old. Fish embryos were obtained in-house, as both the International Institute of Molecular and Cell Biology and the University of Geneva have institutional licenses to house and breed fish.

Cell transfection
Transient DNA transfections were done using Lipofectamine2000 (Life Technologies) or FuGene (Promega). SureFECT (SABiosciences) or Attractene Transfection Reagent (Qiagen) were used for co-transfection of DNA and siRNA. HiPerFect Transfection Reagent (Qiagen) was used for siRNA transfection. Transfections were performed according to manufacturers' instructions and analyzed 24-48 h after DNA transfection or 72 h after siRNA transfection.

Luciferase reporter assays
In overexpression experiments HEK293 cells were transfected with 200 ng of firefly luciferase reporter (Super8xTOPFlash, Super8xFOPFlash, CCND1-Luc or AXIN2-Luc), 50 ng of pRL-SV40 reporter vector and maximum 1 μg of the tested plasmid in a 24-well format (plasmids amounts were scaled down for experiments performed in a 96-well format). The amount of transfected DNA in all tested samples was kept constant by addition of an empty vector. In silencing experiments, RNAi at a final concentration of 33 nM were co-transfected with 50 ng of Super8xTOPFlash and 10 ng of pRL-SV40 reporter vectors in a 96-well format. In case of Wnt3a stimulation, cells were treated for the last 18 h before lysis, unless otherwise indicated. 24-72 hours post transfection cells were lysed in Passive Lysis Buffer (Promega) and either luciferase assay or Western blot analysis was performed. Luciferase reporter activity was measured in Firefly Buffer [50 mM Tris pH 7.8, 12 mM MgCl 2 , 10 mM DTT, 0.2 mM ATP (Sigma-Aldrich), 0.5 mM D-luciferin (Lux Biotechnology), 0.25 mM coenzyme A (Sigma-Aldrich)] or Renilla Buffer [50 mM Tris pH 7.8, 100 mM NaCl, 2.5 nM coelenterazine (Lux Biotechnology)]. The amount of light emitted in the reaction was determined in a microplate luminometer Centro XS3 LB960 (Berthold Technologies). The firefly luciferase activity derived from Super8xTOPFlash or Super8xFOPFlash reporter was normalized to its respective Renilla luciferase activity as a control for the transfection efficiency. Results are presented as the fold of untreated control in cells without stimulation and transfected with an empty vector or non-targeting RNAi, along with the luciferase reporter plasmids. If esiRNA was used, the pathway activity was normalized to the average of the four non-targeting controls (esiRNAs targeting EGFP and 3 different regions of β-galactosidase). All values are mean ± SEM from at least three independent experiments, each with three or four independent transfections performed in parallel.
Setup of the esiRNA screen esiRNAs targeting 80 selected human genes encoding endocytic proteins, esiRNAs against βcatenin, EGFP and 3 different regions of β-galactosidase were prepared as described above. HEK293 cells were transfected in a 96-well format with 33 nM esiRNA and reporter plasmids for 72 h, stimulated with Wnt3a-conditioned medium for the last 18 h prior to cell lysis and the luciferase assays were performed and analyzed as described above.
Cell stimulation and inhibition of endocytosis HEK293 cells were treated with Wnt3a-conditioned medium or control-conditioned medium for 1.5-18 h (depending on the assay) prior to cell lysis. LiCl was used at the final concentration of 50 mM for 5 h prior to cell lysis. Stimulation with Wnt1, Dvl2 and β-catenin S33Y was obtained by 24 h co-transfection of 25 ng of the respective plasmids in a 96-well format. To inhibit dynamin-dependent endocytosis cells were either transfected for 48 h with plasmid encoding the dominant-negative dynamin2-K44A mutant or treated with 30 μM dynasore (Merck) for 30 min and then for 5 h with Wnt3a-conditioned medium and dynasore.

Immunoblotting
Samples of 10-50 μg total protein were subjected to SDS-PAGE. Resolved proteins were transferred to nitrocellulose membrane (Whatman), which was blocked in 5% skim milk, probed with specific antibodies and detected by enhanced chemiluminescence on ImageQuant LAS 4000 (GE Healthcare Life Sciences) or Odyssey infrared imaging system (LI-COR Biosciences).

Immunoprecipitation
Cells were seeded on 100 mm culture plates and the next day transfected with plasmids encoding Tollip and HA-tagged ubiquitin (2.5 μg and 0.5 μg DNA, respectively). For immunoprecipitation of β-catenin cells were grown for 24 h and treated for the last 4 h before lysis with 5 μM MG132 (Enzo Life Sciences). For immunoprecipitation of Tollip, transfected cells were grown for 48 h and incubated for the last 18 h prior to lysis with Wnt3a-conditioned medium in the presence or absence of 5 μM MG132. After 24 or 48 h cells were washed once with PBS and lysed using RIPA buffer (1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris pH 8.0, 150 mM NaCl, 0.5 mM EDTA) containing 5 mM N-ethylmaleimide (Sigma-Aldrich) at 4°C. Lysates were clarified from cell debris by centrifugation and DNA was sheared using Qiashredder columns (Qiagen). After preclearance for 3 h with nonspecific immunoglobulins, bound proteins were depleted using Protein A or G agarose (Roche) for 1 h. Precleared cell lysates were then incubated overnight at 4°C with the indicated antibodies (anti-β-catenin from Santa Cruz sc-7199, anti-Tollip from Santa Cruz sc-27315) coupled to Protein A or G agarose. The agarose beads were washed four times using IP buffer (50 mM HEPES, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 10% glycerol) before elution with 0.1 M glycine of pH 2.5. Finally samples were analyzed by SDS-PAGE and immunoblotting.

Immunofluorescence and microscopy
HEK293 cells were seeded on 12-mm gelatin-coated coverslips in a 24-well plate (5x10 4 cells/ well for overexpression experiments; 10 4 cells/well for silencing experiments). The coverslips were incubated in 0.2% gelatin (Sigma-Aldrich, G9391) for 30 min at 4°C prior to plating. The next day cells were transfected with DNA using Lipofectamine2000 or with siRNA using HiPerFect (Qiagen) according to the manufacturer's instructions. After 24 h (overexpression) or after 72 h (silencing) cells were stimulated with Wnt3a-conditioned medium or controlconditioned medium for the indicated time, rinsed with PBS containing 1 mM CaCl 2 and 0.5 mM MgCl 2 and fixed with 3.6% paraformaldehyde (PFA) in PBS for 15 min at room temperature. Next, cells were permeabilized with 0.1% Triton X-100 and blocked using 10% fetal bovine serum in PBS for 30 min. Cells were further incubated overnight with primary antibodies against: β-catenin (BD Transduction Laboratories, #610154), Tollip (Sigma-Aldrich, HPA038621) or myc tag (Cell Signaling, #2278). Next day, cells were incubated with appropriate Alexa-conjugated secondary antibodies (30 min) and DAPI in 5% fetal bovine serum in PBS. For quantitative analysis at least ten randomly selected regions were imaged for each experimental condition. Single confocal plane images were acquired with Zeiss LSM710 microscope using 40x/1.30 oil immersion objective and 1024x1024 pixel resolution in ZEN 2009 software. Images were exported as TIFF and for visual presentation arranged in Adobe Photoshop CS4 Extended software with only linear adjustments. For quantification of β-catenin in the nucleus, 8-bit images were analyzed in ImageJ. Nuclei were identified as regions of interest based on DAPI staining. Fluorescence of β-catenin in the nuclei was measured and normalized to the nucleus area. Average β-catenin intensity in control cells plus standard deviation served as a cut-off for identification of β-catenin-positive nuclei. On average, over 200 nuclei per condition were analyzed in each experiment. The data are average percentage of β-catenin-positive nuclei or mean fluorescent intensity of nuclear β-catenin from 3 independent experiments ± SEM.

Developmental staging and maintenance of zebrafish embryos
General maintenance, collection, and staging of the zebrafish were carried out according to the Zebrafish Book [34]. Wild-type (the AB strain) zebrafish embryos were maintained in Danieau zebrafish medium and grown at 28°C. The developmental stages were estimated based on time post-fertilization (hours or days; hpf or dpf) at 28°C as described [35].

Immunohistochemistry (IHC)
Zebrafish embryos collected at different stages post fertilization were manually dechorionated, fixed with 4% paraformaldehyde and permeabilized in a graded series of methanol dilutions: 30%, 60%, 100% for 5 min each at RT and in 100% methanol for at least 1 h at -20°C. One dpf embryos were additionally treated with proteinase K at the concentration of 10 μg/ml for 5 min. Activity of endogenous peroxidases was inhibited by treatment with 0.3% H 2 O 2 for 30 min. Embryos were blocked in 5% fetal bovine serum for 1 h and incubated with anti-Tollip antibody (R&D Systems, #MAB4678) diluted 1:250 overnight. Primary antibody was detected using Peroxidase Vectastain ABC Kit (Vector), according to the manufacturer's recommendations. Secondary antibodies were detected by incubation in DAB (3,3'-diaminobenzidine) solution for 15 min and next in DAB solution plus 0.01% H 2 O 2 . The embryos were washed to stop the reaction and mounted in glycerol.

Morpholino injections
Translation-blocking morpholino (MO) oligonucleotides complementary to 5'UTR and ATG regions of ztollip mRNA were obtained from Gene Tools: ATG1 5'-GTGCTTATTGTTG TCGCCATTCTGC-3', UTR1 5'-TTATGATATGACGGACAGCGCCTGT-3' (ATG1 and UTR1 target the first, longer transcript variant), ATG2 5'-CTGCTGCTGAGTCGGCAT GATCCTC -3' (targeting the second, shorter transcript variant). Dose-response experiments were carried out for all MOs to determine the highest nontoxic dose for use in further experiments. Up to 4 ng MO was microinjected into the yolk of 1-cell stage embryos (0.75 ng ztollip ATG1 MO, 4 ng ztollip UTR1 MO, 2 ng ztollip ATG2 MO). In double knock-down ISH experiments half of the MO amount was used. To determine the effects of gene silencing, a combination of the ATG2 and the UTR1 MO was used. MO-injected embryos were then manually dechorionated and either their phenotype was analyzed or they were subjected to ISH. To knock down Rab5 function, 1-cell stage embryos were injected with 4 ng of morpholino oligonucleotide directed against the coding region of the zebrafish rab5a (5'-GACAGTTGTCAAT CACCCCGTCTTC-3'). The β-catenin2 (ctnnb2) morpholino was previously described: 5'-CCTTTAGCCTGAGCGACTTCCAAAC-3' [36].

In situ hybridization (ISH)
In situ hybridization was carried out according to standard procedures [37]. For anti-sense cmlc2, fgf8, goosecoid and ntl RNA synthesis the appropriate plasmids were linearized and in vitro transcribed. Zebrafish embryos were fixed with 4% paraformaldehyde. One dpf embryos were additionally treated with proteinase K at the concentration of 10 μg/ml for 20 min. Prehybridization was performed at 65°C for 1 h. Embryos were incubated with digoxigenin-labeled RNA probes at 65°C overnight and blocked in 2% Blocking Reagent (Roche) for 5 h. Incubation with anti-digoxigenin alkaline phosphatase-conjugated antibodies (Roche #11093274910) diluted 1:5000 was performed at 4°C overnight. Antibodies were detected by incubation in the alkaline phosphatase substrate BM Purple (Roche). Embryos were washed to stop the reaction and mounted in glycerol.

Statistical analysis
Statistical analyses were carried out using the software STATISTICA 8.0 (StatSoft). Comparisons among groups were made by nonparametric Mann-Whitney U test or Wilcoxon signedrank test. Data are expressed as mean ± SEM from 3-4 independent experiments. Differences were considered statistically significant for P0.05.

RNAi screen identifies Tollip as a negative regulator of canonical Wnt signaling
We set out to screen a subset of soluble endocytic proteins for a role in the regulation of canonical Wnt signaling, related to or independent of their function in endocytosis. The setup of a small-scale RNAi screen involved targeting 80 genes encoding adaptor and accessory proteins regulating endocytic internalization and cargo sorting. We employed a library of endoribonuclease-prepared short interfering RNAs (esiRNAs) to downregulate expression of the selected genes in HEK293 cells stimulated with Wnt3a. As readout, we used the Super8xTOPFlash luciferase reporter measuring the activity of TCF/LEF transcription factors [38]. The screen revealed a panel of genes whose knockdown decreased or increased the pathway activity (S1 Fig). Some of the hits, such as Tsg101 [39] and APPL1 previously reported by us [40], were known to positively regulate canonical Wnt signaling, arguing for the correct setup of the screen.
Among the potential inhibitors of the pathway (i.e. increasing the reporter activity upon knockdown), Tollip knockdown consistently increased the reporter activity with three independent RNAi reagents (an esiRNA pool and two single commercial siRNA oligonucleotides; Fig 1A and 1B). This effect was specific since Tollip silencing did not affect the control FOP-Flash reporter bearing mutated TCF/LEF binding sites (Fig 1A, right bars). Furthermore, knockdown of Tollip did not increase basal Super8xTOPFlash reporter activity in HEK293 cells without Wnt3a stimulation (Fig 1A, middle bars). Conversely, overexpression of Tollip inhibited the reporter activity up to 5-fold in Wnt3a-treated cells, thus resulting in an opposite effect than silencing of Tollip expression (Fig 1C). Consistently, the observed inhibition was specific for the ligand-induced, but not basal pathway activity and for the wild-type, but not the mutated reporter ( Fig 1C).
Overexpression of Tollip decreased the activity of promoters of two Wnt target genes: AXIN2 [41] and CCND1 (encoding cyclin D1) [25,42] (Fig 1D). Similarly, the levels of expression of two other Wnt target genes, FGF9 and NRP1 [43][44][45], were reduced upon Tollip overproduction, as measured by qPCR ( Fig 1E). However, expression of some other Wnt target genes, such as NKD1 or DKK1 [46], was not changed under these conditions, arguing that Tollip affects a specific subset of transcriptional targets of the Wnt pathway. Overall, gene expression analysis validated the results obtained in the Super8xTOPFlash reporter screen and allowed concluding that Tollip can exert an inhibitory activity on canonical Wnt signaling.

The function of Tollip in Wnt signaling is independent of dynaminmediated endocytosis
We then verified whether the function of Tollip in Wnt signaling is due to its activity in endocytic processes. Endocytosis was shown to regulate the canonical Wnt signaling, although its role seems to be either stimulatory or inhibitory depending on the cell type [22,[47][48][49]. In Wnt3a-stimulated HEK293 cells, expression of dominant-negative K44A mutant of dynamin2 [50] reduced the reporter activity (Fig 2A), without affecting the levels of Tollip protein (S2A Fig). This effect was independently confirmed by treating cells with a pharmacological inhibitor of dynamin, dynasore [51], which also efficiently blocked the reporter stimulation ( Fig 2C). Inhibition of endocytosis occurred in both cases (overexpression of dynamin mutant and dynasore treatment), as internalization of transferrin was strongly reduced (data not shown). However, when Tollip was overexpressed together with the dynamin-K44A mutant or along with the dynasore application, its negative effects on the TCF/LEF reporter were still clearly visible (Fig 2B and 2C). Conversely, Tollip depletion could potentiate the reporter activity in dynasore-treated cells (Fig 2D). These data indicate that the inhibitory activity of Tollip in canonical Wnt signaling is independent of dynamin-mediated endocytosis, thus likely representing a moonlighting function of this protein.

Ubiquitin binding is important for the function of Tollip in Wnt signaling
To address possible mechanisms underlying the moonlighting action of Tollip, we first studied which of its domains (engaged in well-known interactions with other molecules and pathways) might be important for Wnt signaling. Of the three domains of Tollip, the N-terminus is responsible for association with TOM1 protein, the middle C2 domain binds phosphoinositides, while the C-terminal CUE domain interacts with ubiquitin [7,11].
Deletion mutagenesis was performed and the resulting five mutants (Fig 3A) were tested for their effects in the reporter assay ( Fig 3B). The mutants devoid of the CUE domain (del.1 and del.2) lost the ability to inhibit the Wnt pathway activity. Instead, inhibition was retained in a mutant lacking the N-terminal TOM1-binding motif (del.3). Interestingly, the C2 domain  alone (del.4) also inhibited the reporter activity, in contrast to a minor effect of the C-terminal fragment containing the CUE domain (del.5). We therefore concluded that the interaction with TOM1 via the N-terminus of Tollip is not important for the regulation of Wnt signaling. Instead, both the C2 and the CUE domains mediate this function of Tollip, suggesting that binding of phosphoinositides and/or ubiquitin may be involved. Consistent with this idea, both types of molecules are implicated in Wnt signaling. Phosphatidylinositol (4,5)-bisphosphate (PI4,5P 2 ) is produced at the plasma membrane upon Wnt3a stimulation to drive formation of LRP6 signalosomes [52][53][54], while ubiquitin plays a number of roles, either proteolytic or regulatory, at various stages of the canonical Wnt cascade [55].
We confirmed these results with a panel of previously characterized point mutations of Tollip ( Fig 3A). A double mutant M240A/F241A within the CUE domain perturbs an association with ubiquitin [15]. Six individual point mutations within the C2 domain (R78A, R123A, H135A, K150E, R157A, K162A) inhibit binding of Tollip to phosphatidylinositol 3-phosphate (PI3P) and two of them (R78A, R123A) additionally destabilize an interaction with PI4,5P 2 [9,10]. When tested in the reporter assays, the ubiquitin binding-deficient M240A/F241A mutant did not inhibit the pathway activity as efficiently as the wild-type protein (Fig 3C). This confirms the results of deletion mutagenesis and indicates that ubiquitin binding is important for the repressive function of Tollip in Wnt signaling. Of note, we detected K48-and K63-linked ubiquitin chains in immunoprecipitates of Tollip (S2B Fig) which indicates its association with ubiquitinated proteins, as reported [17], and/or a direct modification of Tollip itself.
Of the six phosphoinositide-binding mutants, two (R78A, R123A) were unable to inhibit the reporter activity, although the levels of R78A overexpression were the lowest (Fig 3D). The other mutants (expressed at levels comparable to each other and to the wild-type form) exhibited either the wild-type (H135A, K162A, K150E) or slightly lower (R157A) activity. These results indicate that, if at all, association with PI4,5P 2 may be of some importance for the function of Tollip in Wnt signaling, consistent with the role of this lipid in Wnt signaling [52][53][54]. However, binding of PI3P appears not to be required for the Tollip function in Wnt inhibition. Considering that PI3P is predominantly enriched in endosomal membranes [56], this would further support our conclusion that the endocytic and Wnt regulatory functions of Tollip are independent of each other. Consistent with the reporter data, when expression of Tollip target genes FGF9 and NRP1 was tested, the ubiquitin-binding M240A/F241A mutant failed to inhibit expression of the two targets, in contrast to the wild-type and the K150E mutant ( Fig  3E). Finally, increased K48-linked ubiquitination of β-catenin and/or its associated proteins was detected in β-catenin immunoprecipitates upon overexpression of the wild-type Tollip but not of its mutants deficient in ubiquitin binding (M240A/F241A and del.1 devoid of the CUE domain; Fig 3F). These data are in agreement with an inhibitory activity of Tollip on Wnt signaling because enhanced polyubiquitination of β-catenin favors its degradation, thus restraining the pathway. Cumulatively, these data show that the ability of Tollip to bind ubiquitin is required for its function in Wnt signaling.

Tollip regulates a pool of active β-catenin in Wnt-stimulated cells
We wished to identify components of the Wnt transduction pathway affected by Tollip. First, we assessed the impact of Tollip on the level of β-catenin, both total and its active (non-phosphorylated on Ser37/Thr41) form. In Wnt3a-treated cells, overexpression of Tollip reduced the pool of active β-catenin without altering its total levels (Fig 4A, left panel). Conversely, depletion of Tollip increased amounts of active β-catenin but not its overall pool (Fig 4A, right  panel). These observations are consistent with the role of Tollip as a negative regulator of the pathway.
As activation of β-catenin leads to its translocation to the nucleus, we verified whether Tollip influenced this process, employing quantitative immunofluorescence microscopy. As expected, silencing of Tollip expression potentiated accumulation of β-catenin in nuclei of Wnt3a-treated cells (Fig 4C-4F). This was manifested by increased percentage of β-cateninpositive nuclei (Fig 4D) and higher mean fluorescence intensity of nuclear β-catenin (Fig 4E). No such changes were observed in cells without Wnt stimulation (Fig 4B, 4D and 4E), consistent with the notion that Tollip affects the ligand-induced, but not basal pathway activity (Fig 1A and 1C). We could not, however, measure unequivocal effects of Tollip overexpression on β-catenin nuclear localization in Wnt3a-treated cells (S3 Fig), possibly due to variable levels of protein overproduction between individual cells upon transient transfection. Cumulatively, the biochemical and microscopical data suggest that in Wnt3a-treated cells Tollip can counteract the activation of β-catenin and its subsequent nuclear translocation, thus inhibiting the pathway.

Tollip can also inhibit ligand-independent, constitutive Wnt signaling
The Wnt pathway can also be induced in the absence of extracellular ligands by alterations in intracellular signaling components. We thus verified whether Tollip can also inhibit ligandindependent Wnt signaling. First, we employed also two intracellular stimuli such as overexpression of a pathway activator Disheveled 2 (Dvl2) [57] and treatment of cells with LiCl which inhibits GSK3β [58,59], in comparison to another canonical ligand (Wnt1). In all three cases, overexpression of Tollip was still able to inhibit the TCF/LEF reporter activity (Fig 5A), as initially observed for cells stimulated with Wnt3a (Fig 1C). In case of Wnt1, this inhibition was also accompanied by a visible decrease in the levels of active β-catenin without changes in its total amounts (Fig 5A), similarly to effects exerted by Wnt3a (Fig 4A). These data argue that ubiquitin-binding deficient M240A/F241A mutant; D, phosphoinositide-binding deficient point mutants). Increasing amounts of mutant-encoding plasmids were transfected. Ctrl, cells transfected with an empty pcDNA plasmid instead of a Tollip-encoding construct. All values are expressed as fold of untreated control, i.e. cells incubated with control-conditioned medium (CM) and transfected with an empty plasmid and the Super8xTOPFlash reporter. Data are mean ± SEM from 3 (C) or 4 (B, D) independent experiments; *P0.05, **P<0.01, ***P<0.001 (Mann-Whitney U test). Expression of mutated Tollip proteins was verified by Western blotting (WB) using anti-myc antibodies, with α-tubulin as a loading control (lower panels). (E) Expression of FGF9 and NRP1 genes upon overexpression of Tollip wt, K150E and M240A/F241A mutants in Wnt3a-stimulated HEK293 cells (Ctrl), measured by qPCR. All values are relative expression levels, compared to controls from cells incubated with control-conditioned medium and transfected with an empty plasmid (normalized to 1 for each gene; not shown). Data are mean ± SEM from 3 independent experiments; *P0.05 (Mann-Whitney U test). (F) Immunoprecipitation of β-catenin from lysates of HEK293 cells transfected with an empty plasmid (Ctrl) or with constructs expressing wild-type Tollip (wt), M240A/F241A mutant or deletion 1 (del.1) mutant. Cells were either untreated or incubated with MG132 for 4 h before lysis. Upper panel; immunoprecipitates were probed with antibodies against β-catenin, total ubiquitin (Ub), its K48-linked (Ub-K48) or K63-linked (Ub-K63) chains. Lower panel; 10% of starting lysates taken for immunoprecipitation (input) were blotted against total ubiquitin (Ub), β-catenin and Tollip, with actin as a loading control.  upon ligand stimulation, Tollip counteracts the activation of β-catenin without affecting its stabilization. However, no changes in β-catenin status were apparent in lysates of cells overexpressing Dvl2 or treated with LiCl (Fig 5A), suggesting that Tollip can exert its inhibitory action at yet another step in the pathway, occurring after β-catenin stabilization.
Ligand-independent stabilization of β-catenin leads to permanent activation of the Wnt pathway and is a common pathological mechanism of numerous diseases, including cancer [23,24]. To test whether Tollip could inhibit constitutive Wnt signaling, we first overproduced a degradation-resistant β-catenin mutant S33Y [60] in HEK293 cells. Under these conditions, overexpression of Tollip inhibited, while its silencing potentiated the TCF/LEF reporter activity (Fig 5B). These effects argued that Tollip can interfere with ligand-independent Wnt signaling.
This conclusion was further supported by testing colorectal cancer cell lines in which canonical Wnt signaling is constitutively active due to mutations either in β-catenin (HCT116 line) or in APC (DLD1 line) [60,61]. Indeed, the levels of total and active β-catenin were increased in these lines, compared to Wnt3a-treated HEK293 cells (Fig 5C). In both lines, overexpression of Tollip inhibited the TCF/LEF reporter activity, with the strongest effects observed in better transfectable HCT116 cells (Fig 5D). However, overproduced Tollip did not change the levels of total or active β-catenin (Fig 5E), arguing that it inhibits the pathway at a step downstream of β-catenin stabilization.
Overall, our results suggest that an inhibitory action of Tollip in Wnt signaling can be exerted at two steps of the pathway. In ligand-stimulated cells, Tollip lowers the pool of active (non-phosphorylated) β-catenin without altering its total levels. Interestingly, in macrophages Tollip was reported to participate in the activation of GSK3β [62]. GSK3β inhibits Wnt signaling by promoting phosphorylation and degradation of β-catenin [23]. Therefore, Tollip could potentially impair activation of β-catenin via increased GSK3β activity. In parallel, Tollip can also inhibit the Wnt pathway acting downstream of β-catenin stabilization, as observed under conditions of ligand-independent, constitutive Wnt signaling. This could possibly involve regulation of nuclear activities of β-catenin and its associated proteins by Tollip which can also localize to the cell nucleus [18].
In general, our data support a widespread role of Tollip-mediated regulation of Wnt signaling, both ligand-induced and ligand-independent, in different cell types, including cancer cells. Consistently, decreased expression of Tollip was reported during oncogenic progression from normal colon mucosa to adenomas and carcinomas [63]. Activation of canonical Wnt signaling is a well-documented pathological mechanism in colon cancer [64] and lower levels of an inhibitory protein such as Tollip could additionally contribute to this process.

The function of Tollip in embryonic development of zebrafish
In the adult, Wnt signaling drives gene expression to maintain stem cell homeostasis and when aberrant leads to carcinogenesis. However, the Wnt pathway controls also body patterning during early vertebrate development [65]. To address whether Tollip-dependent regulation of Wnt signaling is evolutionarily conserved, we analyzed the developmental role of Tollip in zebrafish embryos. In this organism, the canonical Wnt signaling pathway is essential for the specification of ventral and posterior fates (for review see [66]). Indeed, inhibition of the canonical Wnt signaling by either Wnt3/Wnt8 downregulation or missexpression of the Wnt signaling inhibitor Dickkopf1 (Dkk1) consistently reduces the formation of the posterior body [67][68][69]. The genome of Danio rerio harbors a Tollip ortholog (GenBank accession number NM_207061.1) which encodes a well-conserved protein of 276 amino acids sharing the same domain structure [13] and 82% overall amino acid identity with its human counterpart ( Fig  6A). Indeed, antibody against human Tollip recognized a band of an appropriate size in western blots from fish embryo lysates (Fig 6B). Immunohistochemistry staining during early development showed that zebrafish Tollip (zTollip) was maternally expressed and distributed throughout the blastoderm during the blastula stage (Fig 7A-7D). zTollip expression was also Overexpression of Tollip (left panel) or its siRNA-mediated silencing (right panel) alters the activity of Super8xTOPFlash luciferase reporter. All values are expressed as fold of untreated control, i.e. the reporter activity in cells without β-catenin-S33Y overexpression which was normalized to 1 (Ctrl NS). Ctrl denotes the reporter activity in cells treated with β-catenin-S33Y overexpression and transfected with an empty plasmid instead of Tollip (left panel) or nontargeting siRNA control (right panel). Data are mean ± SEM from 3 independent experiments; *P0.05 (Wilcoxon signed-rank test). (C) Levels of active and total β-catenin, and of Tollip were investigated in HEK293 cells treated with Wnt3a-or control conditioned (CM) medium for 6 h, and in unstimulated DLD1 and HCT-116 cells. Immunoblotting was performed with the indicated antibodies, with actin as a loading control. (D) Overexpression of Tollip suppresses the activity of Super8xTOPFlash luciferase reporter in cancer cell lines DLD1 and HCT116 without exogenous stimulation (upper panel). Increasing amounts of Tollip-encoding plasmid were tested. The reporter activity in cells transfected with an empty plasmid was normalized to 1 for each line (Ctrl). Data are mean ± SEM from 3 independent experiments; *P0.05 (Mann-Whitney U test). Lower panel: the levels of Tollip overexpression were verified by Western blotting (WB), with α-tubulin serving as a loading control. (E) Levels of active and total β-catenin, and of Tollip were investigated in DLD1 and HCT-116 cells overexpressing Tollip or transfected with an empty plasmid (Ctrl). Immunoblotting was performed with the indicated antibodies, with actin as a loading control.
doi:10.1371/journal.pone.0130818.g005 uniform until the end of gastrulation and beginning of somitogenesis (Fig 7E and 7F). During somitogenesis, particularly high expression was observed in the developing brain, eyes, and in the intersomitic regions containing blood vessels (Fig 7I). At 3 days post-fertilization (3 dpf), zTollip was also present around the heart (Fig 7J).
To investigate the role of zTollip during development and its possible contribution to Wnt signaling, we injected synthetic ztollip mRNA in wild-type embryos. Upon injection of 200 pg of ztollip mRNA into 1-cell stage embryos (48 hours post-fertilization, hpf), embryos displayed a reduction of the posterior body, a strong reduction (even lack) of somites, and partial fusion of the eyes (Fig 8A and 8B).
Developmental defects caused by overexpression of ztollip mRNA were further analyzed by whole-mount in situ hybridization with mesendodermal, dorsal and neural markers. In ztollip mRNA-injected embryos, expression domains of mesendodermal markers such as goosecoid (gsc) appeared slightly enhanced compared to control embryos (Fig 8F and 8G). Moreover, the Fig 7. zTollip is broadly expressed during early zebrafish development. Immunocytochemistry analysis of zTollip protein localization in early embryonic development of zebrafish. Embryos fixed at various stages (indicated below the images) were incubated with primary antibodies against human Tollip, followed by secondary antibodies and chromogenic peroxidase-based detection. As controls for staining specificity, embryos at tail bud stage were processed omitting the primary antibody (G) or secondary antibody (H). zTollip expression is uniform from 1-cell stage until tail bud (A-F), but its expression increases in the head (arrowhead in I) and intersomitic regions (black arrows in I) at 24 hpf. At 3 dpf, zTollip is maintained in the intersomitic regions (arrows in J), but it is also expressed in the tissue surrounding the heart (arrowhead in J). Lateral views (A-H). Anterior to the left (I, J).
doi:10.1371/journal.pone.0130818.g007 notochord was also slightly wider as visualized by the expression of notail (ntl) (Fig 8H and 8I). However, the expression of fgf8 in the tail was reduced or even absent (Fig 8L and 8M). These phenotypes are reminiscent of those observed in zebrafish embryos where inhibitors of the canonical Wnt signaling such as Dkk1 were overexpressed [70] or when the canonical Wnt signaling was inhibited by wnt8/wnt3 morpholino injection [71]. Indeed, downregulation of β-catenin2 by morpholino injection resulted in a decrease in fgf8 expression ( Fig 8N) and in defects similar to those observed in ztollip-overexpressing embryos (S4 Fig), indicating that zTollip could also function as an inhibitor of Wnt signaling during vertebrate development. Supporting this idea, overexpression of ztollip in HEK293 cells resulted in a reduction in the Super8xTOPFlash reporter activity (Fig 6C), as observed in the case of the human protein ( Fig 1C). These data point to the conservation of Tollip function during vertebrate evolution.
Next, we knocked down ztollip by using various non-overlapping, independent morpholinos (MOs). We capitalized on the existence of two zebrafish mRNA variants, which differ by their 5' UTR and ATG regions (http://zfin.org/ZDB-GENE-030131-8820, ZFIN IDs: ZDB-TSCRIPT-110325-2304 for a longer transcript and ZDB-TSCRIPT-110325-1396 for a shorter transcript). To block both mRNA variants, a combination of two morpholinos targeting each of them was used. Activation of Wnt signaling leads to pericardial oedema, smaller eyes, body curvature and failure of the heart to loop [72]. Similarly, morpholino-mediated downregulation of ztollip resulted in twisted tail, pericardial oedema and smaller eyes (Fig 8A  and 8C). Moreover, cmlc2 expression in the heart revealed that ztollip morpholino-injected embryos also failed to loop the heart (Fig 8J and 8K). In addition, and consistent with a role of Tollip in inhibiting Wnt signaling, ztollip morpholino-injected embryos showed an increase in fgf8 expression in both the tail and the brain (Fig 8L and 8O). Most importantly, the defects displayed by embryos with ztollip knockdown (Fig 8C) could be rescued by downregulation of β-catenin2 expression (Fig 8D and 8E). To address whether the function of zTollip during zebrafish development is independent of its role in endocytosis, we overexpressed ztollip, while we simultaneously blocked the endocytic cycle in embryos by overexpressing the dominant-negative dynamin2 K44A mutant or downregulating the expression of rab5a, one of the key regulators of early endocytic steps [73,74]. While fgf8 expression was decreased upon overexpression of ztollip (Fig 8M), it was unaffected by injection of dynamin-K44A or rab5a-targeting morpholino (Fig 8P and 8Q), suggesting that endocytosis does not play a significant role in regulating expression of fgf8. Moreover, inhibiting the endocytic pathway by injection of dynamin-K44A did not rescue the negative effect of ztollip overexpression on the expression of fgf8 (Fig 8R), further pointing to a role of zTollip in Wnt signaling independent of its endocytic activity.

Discussion
In summary, our results identify a novel inhibitory activity of Tollip towards canonical Wnt signaling in mammalian cells and in zebrafish embryonic development. In addition to the universal core components of Wnt signaling, an increasing number of its regulators are being identified which may act in a more tissue-restricted or specialized manner [75]. Based on our data, we propose such a modulatory role for Tollip. At the molecular level this function is unrelated to endocytosis, but it is possible that Tollip acts by linking Wnt signaling with other signal transduction pathways.
The spectrum of pathways reported to be regulated by Tollip is broad, including innate immunity and NF-κB signaling [13,15], TGF-β signaling [14], Rac1-dependent bacterial entry [16], sumoylation of nuclear and cytoplasmic proteins [18] and autophagic clearance of cytotoxic protein aggregates [17]. It is conceivable that the plethora of functions of Tollip could be linked via its role in autophagy and ability to bind ubiquitin (the latter being important also for the regulation of Wnt signaling, as we show here). On the one side, autophagy regulates various aspects of innate immunity, inflammation and pathogen elimination [76]. On the other side, selective autophagic degradation of cytoplasmic proteins affects a number of pathways [77], including Wnt signaling where autophagy-mediated elimination of Disheveled 2 [78] or of βcatenin [79,80] were reported.
We propose that Tollip can inhibit the Wnt cascade at two stages. Upon ligand stimulation, Tollip reduces the pool of active β-catenin, possibly due to its reported role in the activation of GSK3β [62]. Additionally, under conditions of ligand-independent Wnt constitutive signaling (e.g. as occurring in cancer cells), Tollip inhibits the pathway downstream of β-catenin stabilization. Tollip was reported to interact with the components of sumoylation machinery and to regulate nucleocytoplasmic shuttling of proteins [18]. This could suggest an involvement of Tollip in sumoylation and/or regulation of nucleocytoplasmic distribution of β-catenin-associated proteins in the nucleus. Modification by sumoylation has been reported for TCF4 and LEF1 transcription factors [81,82], as well as for their nuclear co-regulators such as the TBL1-TBLR1 co-repressor [83], Groucho [84], reptin [85], the CoCoA co-activator [86] or duplin/Chd8 [87], which in all cases modulates the output of Wnt signaling. In general, the involvement of Tollip at two steps of the Wnt cascade has already precedence in the NF-κB pathway, where Tollip functions at multiple stages and intracellular locations (plasma membrane, early and late endosomes) [8,13,15].
The Tollip ortholog in zebrafish has not been previously investigated. We show that zTollip is broadly expressed in early embryonic stages, and interfering with its function induces developmental phenotypes consistent with its proposed role in modulation of Wnt signaling. Thus, this function of Tollip is evolutionarily conserved between fish and mammals. Its physiological importance is further underscored by the fact that reduced expression of Tollip was reported in colon cancers [63]. It is therefore possible that the inhibitory activity of Tollip on canonical Wnt signaling could contribute to oncogenic processes in Wnt-dependent human malignancies.
Supporting Information S1 Fig. esiRNA screen of selected endocytic proteins for the regulation of Wnt3a-dependent transcription. HEK293 cells were transfected for 72 h with 33 nM esiRNA targeting 80 indicated genes encoding soluble proteins regulating endocytosis. Cells were stimulated with Wnt3a-conditioned medium for 18 h before lysis and Super8xTOPFlash luciferase reporter assays were performed. Depletion of an established pathway activator β-catenin (CTNNB1) served as a positive control. esiRNAs targeting EGFP or 3 different regions of β-galactosidase served as negative controls. The pathway activity is presented relatively to the average of four non-targeting controls, which is normalized to one arbitrary unit. Average of non-targeting controls in cells without Wnt3a treatment (Ctrl NS) is shown to demonstrate the extent of stimulation. Data represent average of 3 independent experiments, error bars are SEM. Arrow indicates the results for Tollip knockdown.