The phosphatase Pgam5 antagonizes Wnt/β-Catenin signaling in embryonic anterior-posterior axis patterning

Wnt signaling pathways play a major role in axis determination, patterning, morphogenetic movements and differentiation during embryonic development. In Xenopus laevis, the relocalization of maternal components of the Wnt/β-Catenin pathway at fertilization defines the future dorsal side of the embryo and induces the formation of Spemann's organizer (reviewed by Hikasa and Sokol, 2013). During gastrulation, an anterior-posterior (a-p) gradient of Wnt/β-Catenin activity that patterns the a-p axis is established by posterior expression of Wnt ligands and anterior endomesoderm-derived, secreted antagonists such as Dickkopf1 (Dkk1) and Cerberus, as well as ectodermal intracellular antagonists including Amer2 (Glinka et al., 1997, 1998; Kazanskaya et al., 2000; Kiecker and Niehrs, 2001; Pfister et al., 2012; Piccolo et al., 1999). Wnt/β-Catenin signaling is functionally linked and tightly balanced with the activity of β-Catenin-independent Wnt pathways, which are key regulators of mesodermal and neural convergent extension movements that position cells within the tissue and shape the a-p body axis (reviewed by Wallingford, 2012).

Most Wnt signaling cascades share a subset of effector proteins, including Frizzled (Fzd) receptors, Dishevelled (Dvl) proteins, β-Arrestins (Arrb) and protein kinases such as Casein Kinase 1 (CK1) and Glycogen Synthase Kinase 3β (GSK3β; Niehrs, 2012; Schulte et al., 2010; Seitz et al., 2014). In unstimulated cells, free β-Catenin is rapidly degraded. The core β-Catenin destruction complex consists of β-Catenin, Adenomatous Polyposis Coli protein (APC), Axin, the kinases CK1 and GSK3β and the Ubiquitin ligase βTrCP. In the absence of a Wnt stimulus, N-terminal phosphorylation of β-Catenin targets it for ubiquitylation and subsequent proteasomal degradation (reviewed by Stamos and Weis, 2013). The Wnt/β-Catenin pathway is typically activated by binding of Wnt ligands to a receptor complex composed of Fzd and Lrp5/6 as a co-receptor. Receptor activation triggers phosphorylation of Lrp5/6 and the recruitment of Axin, Dvl, casein kinases and GSK3β to the receptor complex. These events lead to inhibition of β-Catenin degradation, and the accumulation of β-Catenin in the cytoplasm and translocation into the nucleus (Hsu et al., 1998; Schneider et al., 1996). β-Catenin then binds to LEF/TCF transcription factors, acts as a transcriptional transactivator and thus regulates expression of β-Catenin-responsive genes (reviewed by Clevers and Nusse, 2012; Nusse, 2012).

Dvl becomes highly phosphorylated in response to Wnt stimuli (Bryja et al., 2007; Lee et al., 1999; Takada et al., 2005; Yanagawa et al., 1995), which is considered a key event in Wnt signal transduction. Multiple kinases have been found to phosphorylate Dvl and to play a role in Wnt signal transduction, including but not limited to the serine/threonine kinase CK1, Casein kinase 2 (CK2), Protein Kinase C (PKC), Hypoxia-induced protein kinase (HIPK), Receptor-interacting protein kinase 4 (RIPK4), MARK2/Par1b and MARK3/Par1a (Huang et al., 2013; Kühl et al., 2001; Ossipova et al., 2005; Peters et al., 1999; Shimizu et al., 2014; Sun et al., 2001; Willert et al., 1997). Despite the large number of kinases that phosphorylate Dvl, little is known about Dvl dephosphorylation, except for the inhibitory effects of the ubiquitous protein phosphatase 1 (PP1C; Shimizu et al., 2014) and putatively more context-specific regulation by protein phosphatase 2 (PP2) regulatory subunits (Hannus et al., 2002; Yamamoto et al., 2001). Here, we have identified the protein phosphatase Pgam5 as a novel binding partner of Dvl2. Pgam5 belongs to the phosphoglycerate mutase family, but in contrast to other members of this family, Pgam5 acts as protein serine-threonine phosphatase (Takeda et al., 2009). We demonstrate that Pgam5 is a potent antagonist of Wnt/β-Catenin signaling and we have identified Dvl2 as one target of Pgam5-mediated dephosphorylation. Pgam5 inhibits β-Catenin stabilization and interaction of β-Catenin, Tcf1 and Dvl2. In vivo, Pgam5 is required for the formation of an a-p gradient of β-Catenin activity, a-p patterning and head formation in Xenopus laevis.

We hypothesized that an Arrb2/Dvl2 protein complex might provide a hub for protein-protein interactions, which are distinct from interactions of the single proteins. Therefore, we have analyzed the interactome of Arrb2 as bait in the presence of co-expressed Dvl2 by quantitative mass spectrometry. Besides known Dvl interactors, such as Casein Kinases or CAMK2 (Bernatik et al., 2011; Cong et al., 2004a; Gentzel et al., 2015; Kishida et al., 2001; Klein et al., 2006), we have identified the phosphatase Pgam5 as a novel protein interaction partner. In the reverse co-immunoprecipitation experiment with Dvl2 as bait and Arrb2 co-expression, Pgam5 recovery relative to the bait was increased 66-fold when compared with the Arrb2 bait, indicating that Pgam5 binds to Dvl2 rather than Arrb2. Similarly, more interacting kinases were detected at higher abundances. The relative abundances of the kinases in each experiment were within the same order of magnitude as Pgam5 abundance and support the assumption that Pgam5 interacts with Dvl2 and only indirectly with Arrb2 (Fig. 1A).

Pgam5 is highly conserved among vertebrates. Owing to its allotetraploidy, the Xenopus laevis genome comprises two pgam5 genes designated pgam5.S and pgam5.L. These denominations should not be confused with human PGAM5-S and PGAM5-L (Lo and Hannink, 2008), which describe two protein isoforms derived from the same gene by alternative splicing. The last exon of the shorter human transcript variant (Ensemble accession number ENST00000317555.6) is not present in Xenopus tropicalis or Xenopus laevis pgam5 genes. Consistently, all Xenopus Pgam5 proteins correspond to the longest human PGAM5 isoform: PGAM5-L (Fig. S1). In this study, either Xenopus laevis Pgam5.S or human PGAM5-L were used and in the following are designated Pgam5 or PGAM5, respectively.

The Pgam5 family shares characteristic features, including an N-terminal mitochondrial targeting sequence and the Pgam domain with a catalytic histidine residue in the active center, which are also present in Xenopus Pgam5 proteins. Overall, Xenopus laevis Pgam5 and Xenopus tropicalis Pgam5 are 91% identical. Xenopus laevis Pgam5 shares 66%, 65% and 60% identical residues with its mouse, human and zebrafish orthologs, respectively (Fig. S1).

We validated the functional proteomics results by co-immunoprecipitation and western blot analysis (Fig. 1B; Fig. S2A). Pgam5 co-precipitated with Dvl2 irrespective of the co-expression of Arrb2 (Fig. 1B). If present, Arrb2 also strongly co-precipitated with Dvl2, as expected from our previous work (Bryja et al., 2007). By contrast, only low amounts of Pgam5 co-precipitated with Arrb2 (Fig. S2A), which recapitulated our results from quantitative mass spectrometry. Together, these findings support the conclusion that Pgam5 interacts with Dvl2 and only indirectly with Arrb2. We further confirmed this interaction by co-immunoprecipitation of endogenous PGAM5 and ARRB2 with endogenous DVL2 from HEK293T cell lysates (Fig. 1C). Relative quantification of band intensities for PGAM5 showed that PGAM5 was enriched fivefold in DVL2 immunoprecipitates when compared with control IgG. Next, we have mapped the region within the Dvl2 protein that is required for Pgam5 binding to amino acids 334-428 between the PDZ and DEP domains. The DIX and DEP domain as well as the C-terminus were dispensable for Pgam5 binding. Interestingly, the strongest binding was observed when the PDZ domain and the interdomain region between DIX and PDZ domain were also present, indicating that these contribute to the Pgam5 binding interface (Fig. 1D; Fig. S2B).

Dvl is highly phosphorylated upon Wnt stimulation (Bryja et al., 2007; Lee et al., 1999; Takada et al., 2005; Yanagawa et al., 1995); therefore, we evaluated whether Dvl2 might be a substrate of Pgam5. Knockdown of PGAM5 was sufficient to induce hyperphosphorylation of endogenous DVL2 in human cells (Fig. 2A), indicating that PGAM5 indeed dephosphorylates DVL2. Consistently, overexpression of PGAM5, but not the phosphatase-deficient mutant PGAM5-H105A (Takeda et al., 2009), suppressed phosphorylation of overexpressed DVL2 (Fig. 2B) and reduced WNT3A-induced phosphorylation of endogenous DVL2 (Fig. S2C). To further confirm that DVL2 is a substrate of PGAM5, we added recombinant GST-PGAM5 to whole-cell lysates and observed dose-dependent dephosphorylation of DVL2 (Fig. 2C). Moreover, affinity-purified DVL2 was dephosphorylated by recombinant PGAM5 to the fully dephosphorylated state (Fig. S2D), confirming that PGAM5 directly dephosphorylates DVL2.

To investigate the role of Pgam5 in embryonic development, we have first analyzed the temporal and spatial expression pattern of pgam5 in Xenopus laevis embryos. We detected predominantly one mRNA corresponding to the pgam5.S homeolog, whereas pgam5.L was expressed only very weakly or not at all (Fig. S3A). Accordingly, we have focused on pgam5.S for the following experiments.

pgam5 was expressed maternally and zygotically throughout embryonic development (Fig. 3A; Fig. S3B-D). In cleavage and blastula stages, pgam5 mRNA was detected in the entire animal hemisphere (Fig. 3Ai,ii; Fig. S3C). Expression was restricted to the ectodermal layer in the late blastula (Fig. 3Aii′). During gastrulation, pgam5 expression extended vegetally with progressing blastopore closure and was detected in the entire ectoderm by mid-gastrulation, albeit with a bias towards the animal-dorsal region (Fig. 3Aiii,iii′; Fig. S3C,D). Ectodermal expression remained biased towards the anterior neuroectoderm throughout neurulation, which became more prominent at later stages (Fig. 3Aiv,iv′,v,v′). Post-neurulation, the highest levels of pgam5 mRNA were detected in the central nervous system, the eye and the otic vesicle, as well as in the migrating neural crest and later in branchial arches (Fig. 3Avi,vii; Fig. S3C,D).

Pgam5 was knocked-down using two translation-blocking morpholino-oligonucleotides (MOs) that target non-overlapping sequences in pgam5.S mRNA (Fig. S4A,C). MO1 is predicted to bind to pgam5.L mRNA with two mismatches and therefore supposedly also blocks translation of pgam5.L (Fig. S4B). Injection of either Pgam5 MO1 or Pgam5 MO2 in both blastomeres at the two-cell stage resulted in strongly reduced or absent head structures in ∼40% of the embryos, which was not observed in embryos injected with a control MO (Fig. 3B; Fig. S4D). The phenotype was rescued by co-injection of a morpholino-insensitive Xenopus pgam5 mRNA or by human PGAM5 mRNA (Fig. 3B; Fig. S4D), which confirmed the functional conservation of Pgam5 between human and frog, and the specificity of the observed knockdown phenotype. Moreover, co-injection of mRNA encoding the phosphatase-deficient mutant PGAM5-H105A failed to rescue the Pgam5 MO phenotype, suggesting that phosphatase activity of Pgam5 is required for its developmental function (Fig. 3B). Co-injection of PGAM5 mRNA with the control MO in both dorso-animal blastomeres of eight-cell stage embryos resulted in targeted overexpression. About 20% of the embryos showed a moderate enlargement of anterior structures, consistent with Pgam5 being a positive regulator of head development (Fig. 3B). However, 18% also showed a small-head phenotype, similar to PGAM5 morphant embryos. Targeted overexpression of PGAM5-H105A also resulted in 15% of embryos with smaller heads and, in addition, overexpression of wild-type or mutant PGAM5 caused shortened anterior-posterior body axes in a subset of the embryos (Fig. 3B).

Pgam5 has been functionally implicated in apoptosis, necroptosis and mitophagy, although it apparently acts as positive or negative regulator, depending on the experimental context (Chen et al., 2014; Ishida et al., 2012; Lenhausen et al., 2016; Lu et al., 2016; Moriwaki et al., 2016; Wang et al., 2012). Increased cell death in the anterior ectoderm might lead to a loss of head structures, while inhibition of cell death in early Xenopus embryos results in an expansion of neuroectoderm (Offner et al., 2005; Yeo and Gautier, 2003). After Pgam5 MO injection, we neither observed overall increased cell death assayed using Acridine Orange, nor increased apoptosis indicated by cleaved-Caspase 3-positive cells (Fig. S5A,B). At the dose used here, overall survival rates were also comparable with control injections (Fig. S5C). Therefore, we concluded that the Pgam5 loss-of-function phenotype described above was neither caused by increased cell death in general nor by specific loss of anterior cells.

Head induction and anterior patterning of the neuroectoderm in vertebrates require the inhibition of Wnt/β-Catenin signaling (reviewed by Niehrs, 2010). Notably, the Pgam5 morphant phenotype was highly reminiscent of impaired head induction reported after blocking of the Wnt-antagonist Dkk1 (Glinka et al., 1998). Anterior overexpression of Dkk1 in Pgam5 morphant embryos restored head development similar to co-injection of MO-insensitive pgam5 mRNA (Fig. 4A), indicating that elevated Wnt/β-Catenin signaling underlies the Pgam5 morphant phenotype. Indeed, we observed higher amounts of dephosphorylated, active β-Catenin as well as increased Dvl2 phosphorylation in Pgam5 morphant embryos (Fig. 4B). We next analyzed the expression levels of the Wnt/β-Catenin target genes xnr3, msgn1 and hoxd1 by quantitative RT-PCR in Pgam5 morphant embryos. All three genes were upregulated, confirming overall increased Wnt/β-Catenin activity in Pgam5-deficient embryos (Fig. 4C,D). Notably, mRNA levels of dkk1 were not changed (Fig. 4D), indicating that the Pgam5 knockdown phenotype was not due to repression of dkk1 expression. PGAM5 knockdown in human cells enhanced the activation of the TOP-Flash reporter (Korinek et al., 1997) by WNT3A in HEK293T and C2C12 cells (Fig. 4E; Fig. S6A), and induced β-Catenin stabilization and expression of the feedback target AXIN2 (Leung et al., 2002; Lustig et al., 2002) (Fig. S6B). Taken together, these results supported a role of Pgam5 as antagonist of Wnt/β-Catenin signaling in Xenopus development and human cells.

The a-p axis is patterned by a gradient of Wnt/β-Catenin signaling activity, which is established at mid-gastrula stages (Kiecker and Niehrs, 2001). We have visualized this gradient in dorsal explants of stage 11.5 embryos that were dissected into the anterior, midpart and posterior thirds. Active, dephoshorylated β-Catenin (ABC) was low in the anterior third and increased in mid and posterior parts of control explants (Fig. 5A), reflecting graded Wnt/β-Catenin signaling activity. By contrast, in Pgam5-depleted embryos the anterior levels of active β-Catenin were significantly increased and no gradient of active β-Catenin was observed (Fig. 5A). These results strongly suggested that the disruption of graded Wnt/β-Catenin activity underlies the head phenotype observed after Pgam5 knockdown. Consistently, anterior-posterior patterning of the neuroectoderm was modulated by Pgam5 knockdown or overexpression. Pgam5 knockdown shifted the expression of the forebrain marker otx2 and more pronouncedly of the hindbrain marker krox20 anteriorly (Fig. 5B). Again, co-injection of human PGAM5 mRNA rescued the phenotype, whereas PGAM5-H105A mRNA did not. When PGAM mRNA was co-injected with the control MO, resulting in anterior overexpression of PGAM5, both markers were shifted posteriorly (Fig. 5B), as would be expected by stronger inhibition of Wnt/β-Catenin signaling. By contrast, overexpression of the phosphatase-dead mutant PGAM5-H105A in embryos injected with control MO induced an anterior shift of both markers (Fig. 5B). In addition, we observed downregulation of krox20 in about 15% of the embryos injected with PGAM5-H105A mRNA, indicating more severe defects induced by this mutant.

Next, we investigated whether overexpression of Pgam5 suppressed Wnt/β-Catenin activity. In Xenopus embryos, ectopic ventral activation of Wnt/β-Catenin signaling induces a secondary body axis (McMahon and Moon, 1989). Here, we injected Xenopus wnt8 mRNA into both ventral blastomeres of four-cell stage embryos, which resulted in full or partial axis duplication in more than 90% of the embryos (Fig. 6A). Co-expression of human PGAM5 partially suppressed axis duplication induced by Wnt8. In particular, the percentage of complete secondary axes was reduced from 54% to 18%. PGAM5-H105A showed only weak inhibition of axis duplication (Fig. 6A), confirming that phosphatase activity is required for Pgam5 activity as a Wnt/β-Catenin antagonist.

Dorsal inhibition of Wnt/β-Catenin activity in early Xenopus embryos prevents formation of Spemann's organizer and therefore results in ventralized embryos characterized by the lack of dorsal structures such as the dorsal fin, the neural tube, notochord or somites (Heasman et al., 2000; Montross et al., 2000). After injection of PGAM5 mRNA into both blastomeres of two-cell stage embryos, i.e. global overexpression of PGAM5, we observed different degrees of ventralization marked by the progressive loss of the dorsal fin, somites and dorsal head structures (Fig. 6B). In contrast to wild-type PGAM5, PGAM5-H105A overexpression caused small or absent heads in around 40% of the embryos, whereas the dorsal fin and somites were well formed in the trunk (Fig. 6B). These phenotypes were reminiscent of the loss-of-function phenotypes (Fig. 3B), indicating that PGAM5-H105A might act as a dominant-negative mutant in Xenopus embryos. Next, we investigated whether overexpression of Pgam5 was sufficient to decrease nuclear β-Catenin levels in Xenopus embryonic fibroblasts. In cells stimulated with human WNT3A, nuclear β-Catenin was clearly visible. Overexpression of GFP-tagged Pgam5 resulted in a significant decrease of nuclear β-Catenin (Fig. 6C), which further confirmed our findings obtained in Xenopus embryos.

In order to further characterize the mechanism underlying Pgam5-mediated inhibition of Wnt/β-Catenin signaling, we asked whether Pgam5 inhibited Wnt/β-Catenin signaling upstream or downstream of β-Catenin stabilization. Overexpression of either human PGAM5 or Xenopus Pgam5 antagonized TOP-Flash activation induced by Lrp6, Dvl2 or stabilized β-Catenin in HEK 293T cells (Fig. 7A; Fig. S7A,B). PGAM5 also suppressed Wnt/β-Catenin signaling in the APC-deficient SW480 colon carcinoma cell line, in which β-Catenin signaling is constitutively active (Korinek et al., 1997), to a similar extent to ectopic AXIN (Fig. S7C). Consistently, Pgam5 inhibited β-Catenin-induced axis duplication in Xenopus laevis embryos (Fig. S7D). These results indicated that PGAM5 interferes with Wnt/β-Catenin signaling downstream of β-Catenin stabilization and suppresses β-Catenin-dependent transcription.

Transcriptional regulation of Wnt/β-Catenin target genes requires the interaction of β-Catenin with transcription factors of the Lef/Tcf family (Behrens et al., 1996; Molenaar et al., 1996). Overexpression of Pgam5 suppressed human WNT3A-induced activation of the TOP-Flash reporter in the presence of Tcf1-Flag (Fig. 7B), indicating that Pgam5 might interfere with Tcf1-β-Catenin binding or transcriptional activation. Immunoprecipitation of Tcf1-Flag from unstimulated cells showed a basal interaction with endogenous β-Catenin, which was increased by WNT3A stimulation. Pgam5 overexpression reverted the effect of WNT3A and reduced Tcf1-β-Catenin interaction back to basal levels (Fig. 7C). Notably, we also observed co-precipitation of Pgam5 with Tcf1, indicating that Pgam5 was able to bind to one or more components of the transcription factor complex. Next, we overexpressed the stabilized mutant β-Catenin 4ST/A (Yost et al., 1996) together with Tcf1. Binding of β-Catenin 4ST/A to Tcf1 was not further enhanced by WNT3A stimulation. Interestingly, the amount of Pgam5 that was co-precipitated with Tcf1 and β-Catenin 4ST/A was reduced by WNT3A stimulation. However, overexpression of Pgam5 was still sufficient to reduce interaction between β-Catenin 4ST/A and Tcf1 (Fig. 7D) both in the presence and absence of exogenous Wnt3a. Together with the observation that Pgam5-mediated inhibition of reporter gene activation by stabilized β-Catenin requires its phosphatase activity (Fig. 7A), these results suggested that a substrate of Pgam5 is required to form or stabilize the β-Catenin/Tcf complex.

Above, we have identified Dvl2 as a substrate of Pgam5. In addition to its role in the Frizzled receptor complex, Dvl has also been found in the nucleus, where it interacts with the Tcf/β-Catenin transcription complex and has been shown to play a role in transcriptional regulation (Gan et al., 2008; Itoh et al., 2005; Wang et al., 2015).

Therefore, we analyzed whether PGAM5 affected nuclear localization of DVL2. As expected, treatment of HEK293T cells with WNT3A induced stabilization of cytosolic β-Catenin and an increase of β-Catenin levels in the nuclear fraction (Fig. 8A). In agreement with a putative nuclear function of DVL2, we detected endogenous DVL2 in the nuclear fraction. Interestingly, phosphorylated DVL2 was observed in the nuclear fraction only after WNT3A stimulation (Fig. 8A). PGAM5 overexpression blocked the WNT3A-induced increase of β-Catenin and phospho-DVL2 in both the cytoplasmic and nuclear fractions. By contrast, PGAM5-H105A increased cytoplasmic β-Catenin and DVL2 phosphorylation already in unstimulated cells, but seemed to simultaneously inhibit nuclear accumulation of β-Catenin and phospho-DVL2 (Fig. 8A).

The nuclear function of Dvl2 is still somewhat controversial, but evidence is accumulating that Dvl2 can form complexes with Tcf, β-Catenin and FOXK transcription factors (Gan et al., 2008; Itoh et al., 2005; Wang et al., 2015). Therefore, we asked whether Pgam5 also regulated the interaction between Dvl2 and Tcf1 or β-Catenin. Dvl2 was co-immunoprecipitated with Tcf1 and endogenous β-Catenin, irrespective of WNT3A stimulation (Fig. 8B; see also Fig. 7C). Only one band of Dvl2 was detected in the co-precipitates, whereas additional faster-migrating bands were present in the lysate, suggesting that Tcf1 preferentially interacted with phosphorylated Dvl2. The interaction between β-Catenin and Tcf1 was not further stimulated by WNT3A in the presence of exogenous Dvl2, whereas Pgam5 strongly reduced binding of both Dvl2 and endogenous β-Catenin to Tcf1. When cells overexpressing Pgam5 were stimulated with WNT3A, we observed a moderate increase of Tcf1-Dvl2 interaction and a concomitant decrease of Tcf1-Pgam5 binding compared with unstimulated Pgam5-overexpressing cells (Fig. 8B). This was in agreement with a similar effect shown in Fig. 7D and indicated that WNT3A stimulation at least partially reverted the inhibitory effect of Pgam5. In the inverse experiment, when Dvl2 was immunoprecipitated, an interaction of Dvl2 with β-Catenin but not with Tcf1 was detected in unstimulated cells (Fig. 8C). WNT3A stimulation induced binding of Tcf1 to Dvl2, which was again inhibited by Pgam5 overexpression, whereas the interaction of Dvl2 with β-Catenin was apparently not affected (Fig. 8C). Together, these results confirmed the interaction of Dvl2 with Tcf1 and β-Catenin, and further suggested that this interaction is phosphorylation dependent and therefore sensitive to Pgam5-mediated dephosphorylation of Dvl2.

Various different kinases have been identified that phosphorylate Dishevelled proteins (Bernatik et al., 2011; Cong et al., 2004a; Gentzel et al., 2015; Kishida et al., 2001; Klein et al., 2006; Kühl et al., 2001; Ossipova et al., 2005; Peters et al., 1999; Shimizu et al., 2014; Sun et al., 2001; Willert et al., 1997). As counterparts, to date only the rather universal phosphatases PP1 and PP2 have been described (Carmen Figueroa-Aldariz et al., 2015; Shimizu et al., 2014). Here, we have identified the phosphatase Pgam5 as an interactor of Dvl2-Arrb2 protein complexes and negative regulator of Wnt/β-Catenin signaling.

In gastrula and post-gastrula stage Xenopus laevis embryos, pgam5 was expressed throughout the ectoderm but biased towards the anterior-dorsal ectoderm and the anterior neural tube. Consistent with this expression pattern, we showed that Pgam5 plays a role in head formation and a-p patterning. Low anterior Wnt/β-Catenin activity is required for the induction of anterior fates in the dorsal ectoderm and, in combination with the inhibition of BMP signaling, for head induction (Glinka et al., 1997, 1998). Considering that Pgam5 interacts with Dvl2, a role of Pgam5 in Wnt signaling could be assumed. Consistently, heads were small or absent in Pgam5 morphant embryos, indicating elevated Wnt/β-Catenin signaling. However, these phenotypes would also be explained by a loss of anterior cells (Offner et al., 2005). Necroptosis-promoting and -protective roles, as well as pro- and anti-apoptotic functions of Pgam5 have been described (Ishida et al., 2012; Lenhausen et al., 2016; Lu et al., 2016; Moriwaki et al., 2016; Zhou et al., 2012). However, we have not detected any local or global changes in cell death or apoptosis in Pgam5 morphant Xenopus embryos, indicating that regulation of cell death does not represent the major early developmental function of Pgam5 in Xenopus.

Our results demonstrate that Pgam5 indeed acts as an antagonist of Wnt/β-Catenin signaling in vivo. We observed increased amounts of active β-Catenin, as well as elevated expression levels of maternal and more pronouncedly of zygotic Wnt/β-Catenin target genes in Pgam5 morphant embryos. Moreover, the Pgam5 knockdown phenotype was rescued by overexpression of the Wnt antagonist Dkk1 (Glinka et al., 1998; Mao et al., 2001; Semënov et al., 2001). Notably, dkk1 expression was not affected in Pgam5 morphant embryos, which excluded an indirect effect via transcriptional regulation of dkk1.

Patterning of the a-p axis and head induction depend on a gradient of Wnt/β-Catenin activity present at mid-gastrula stages (Kiecker and Niehrs, 2001). Indeed, this gradient was flattened and elevated to the higher posterior levels of active β-Catenin in Pgam5 morphant embryos. Consistently, Pgam5-depletion shifted the expression of forebrain and hindbrain markers anteriorly, further confirming a role for Pgam5 in a-p patterning and head induction in Xenopus development.

Unlike our observations in Xenopus development, no significant developmental phenotype has been reported for Pgam5 knockout mice, except for a general reduction in size and body mass, and increased postnatal lethality (Lu et al., 2014; Moriwaki et al., 2016). Interestingly, we also observed increased lethality when higher doses of Pgam5 MO were injected, which might correspond to the observations in Pgam5 knockout mice. The apparent lack of developmental phenotypes in the mouse is likely due to redundant mechanisms, such as intracellular inhibition of Wnt/β-Catenin signaling by ICAT (Inhibitor of β-Catenin and TCF) (Satoh et al., 2004), the contribution of other signaling pathways and transcriptional control of head inducers (reviewed by Arkell and Tam, 2012), or other compensatory mechanisms. Although it must be assumed that redundant players are also active in Xenopus a-p patterning and head induction, the contribution of individual proteins may vary, as also observed, for example, with the contribution of individual Wnt ligands to neural crest induction in Xenopus and mouse (reviewed by Barriga et al., 2015).

Antagonism of Wnt/β-Catenin signaling by Pgam5 was further supported by the suppression of and TOP-Flash reporter activity in human cells and inhibition of axis duplication, as well as the overexpression phenotypes in Xenopus embryos. In Xenopus, dorsoventral axis determination and formation of the dorsal signaling center, Spemann's organizer, depends on maternal Wnt/β-Catenin signaling (reviewed by Hikasa and Sokol, 2013). Interfering with this early Wnt/β-Catenin function, e.g. by knockdown or overexpression of β-Catenin, impairs or augments organizer formation and consequently the development of dorsal structures. Pgam5 morphant embryos were not dorsalized, most likely due to the presence of maternal Pgam5 protein, which is not depleted by morpholino injections. By contrast, early and global overexpression of Pgam5 led to the loss of dorsal structures reminiscent of the β-Catenin knockdown phenotype (Heasman et al., 2000), which clearly demonstrated the ability of Pgam5 to suppress endogenous Wnt/β-Catenin signaling. Targeted injections of Pgam5 RNA at the eight-cell stage resulted in enlarged heads, consistent with augmented anterior inhibition of Wnt/β-Catenin signaling. However, we also observed smaller heads with similar frequencies, which appears contradictory. Injections into the dorsoanimal blastomeres of eight-cell stage embryos targets predominantly the anterior ectoderm, but partially also the marginal zone and thereby the future organizer. Therefore, this phenotype might be explained by effects of Pgam5 gain-of-function on organizer formation or function, which have also been reported for dorsovegetal β-Catenin depletion (Heasman et al., 2000). In addition, pgam5 is expressed in the cranial neural crest. Accordingly, anterior overexpression might also affect head morphogenesis and therefore indirectly head size.

We have further investigated the molecular mechanism underlying this novel Pgam5 function and identified Dvl2 as a substrate of Pgam5. Dvl associates with the Frizzled/Lrp5/6 receptor complex upon Wnt stimulation and becomes highly phosphorylated in response to Wnt signaling, which is required for intracellular signal transduction together with Arrb2 (Bryja et al., 2007; Lee et al., 1999; Tauriello et al., 2012; Tolwinski et al., 2003; Wong et al., 2003). In the context of Wnt/β-Catenin signaling, CK1-mediated phosphorylation of Dvl appears particularly crucial (Cong et al., 2004a; Kishida et al., 2001; Klimowski et al., 2006; Peters et al., 1999). Dvl contributes to inhibition of the β-Catenin destruction complex and therefore to β-Catenin stabilization (Cong et al., 2004b; Fiedler et al., 2011; Metcalfe et al., 2010; Wong et al., 2003; Zeng et al., 2008). Interestingly, an additional nuclear function of Dvl has been reported by several studies (Barry et al., 2013; Gan et al., 2008; Itoh et al., 2005) and, recently, Wang and colleagues showed that nuclear import of Dvl was phosphorylation dependent (Wang et al., 2015). Therefore, dephosphorylation of Dvl by Pgam5 might inhibit Dvl function on multiple levels and our data support this hypothesis.

We observed stabilization of β-Catenin, an increase in dephosphorylated, active β-Catenin and phospho-Dvl2 after Pgam5 depletion in human cells and in Xenopus embryos, as well as decreased nuclear β-Catenin levels in Pgam5-overexpressing cells. We have identified Dvl2 as a substrate of Pgam5 and a phosphatase-deficient mutant of Pgam5 was less effective in antagonizing Wnt/β-Catenin signaling. Thus, the data presented here support the conclusion that Pgam5 inhibits phosphorylation-dependent Dvl function in Wnt/β-Catenin signaling upstream of β-Catenin degradation. However, we cannot rule out the possibility that other substrates of Pgam5, such as the protein kinase ASK1 (Takeda et al., 2009), which was found to promote β-Catenin degradation (Noh et al., 2010), or additional, so far unknown, substrates also contribute to the inhibition of Wnt/β-Catenin signaling.

Notably, Pgam5 was still able to inhibit reporter gene activation and axis duplication in Xenopus caused by overexpressed β-Catenin or by the stabilized β-Catenin 4ST/A or S33Y mutants. These observations suggested an additional role for Pgam5 downstream of β-Catenin stabilization. Indeed, we observed less binding between β-Catenin and Tcf1 in cells that overexpressed Pgam5. Although this effect might be explained by failure of these cells to stabilize β-Catenin in response to a Wnt signal, the same was true for stabilized β-Catenin 4ST/A. The latter indicated that Pgam5 modulated the composition or stability of Tcf1 transcription complexes. Interestingly, we also found that Pgam5 itself co-precipitated with Tcf1 and β-Catenin, and thus could directly dephosphorylate components of this complex.

We observed a marked decrease in phosphorylated Dvl2 levels in the nuclear fraction of HEK 293T cells overexpressing Pgam5, indicating Pgam5-mediated regulation of nuclear Dvl2 function. Moreover, only one, slow-migrating band was co-precipitated with Tcf1, indicating that Tcf1 preferentially interacts with phosphorylated Dvl2, and we found that interaction of Dvl2 with Tcf1 was strongly impaired by Pgam5. Therefore, we conclude that Pgam5-mediated dephosphorylation of Dvl2 inhibits its function both upstream of β-Catenin stabilization and downstream of its nuclear function in the Tcf transcription complex. Again, we cannot exclude the possibility of additional targets of Pgam5 because the nuclear localization, interaction and transcriptional activity of β-Catenin and Tcf/Lef transcription factors is also regulated by site-specific phosphorylation (Fang et al., 2007; Lee et al., 2001; Mahmoudi et al., 2009; Shitashige et al., 2010; Wang and Jones, 2006; Wu et al., 2008) and thus might be influenced by Pgam5. Moreover, the interaction of Pgam5 with Tcf1 was attenuated in WNT3A-stimulated cells, indicating that it is dynamically regulated by Wnt signaling. One might speculate that the regulation of Pgam5 protein-protein interactions, and potentially also enzymatic activity, could be part of the machinery that fine-tunes the cellular response to Wnt signals; however, this needs to be addressed experimentally.

A phosphatase-deficient mutant of Pgam5, PGAM5-H105A, was not able to rescue the Pgam5 morphant phenotype and was less efficient in inhibiting Wnt/β-Catenin signaling, indicating that the developmental function of Pgam5 depends on its phosphatase activity. Moreover, when overexpressed either globally or targeted to the dorsoanimal blastomeres of eight-cell stage embryos, this phosphatase-deficient mutant phenocopied Pgam5 loss of-function, which pointed towards a dominant-negative activity. Interestingly, PGAM5-H105A increased the amount of phospho-DVL2 in the cytoplasm, even in the absence of a Wnt stimulus, but decreased nuclear phospho-DVL2. DVL2 is not dephosphorylated by PGAM5-H105A. However, the point mutation is unlikely to affect binding to DVL2, which would result in sequestering of phospho-DVL2 in the cytoplasm. Elevated levels of phosphorylated Dvl2 in the cytoplasm have also been observed in PGAM5-deficient cells and in Pgam5 morphant embryos, which could explain the dominant-negative function of PGAM5-H105A. However, we also observed low residual activity of PGAM5-H105A in repressing axis duplication and activation of the TOP-Flash reporter. The simultaneously lower amounts of phospho-Dvl in the nucleus would explain this residual activity.

It is certain that Dvl function is regulated by post-translational modifications, namely phosphorylation by multiple kinases. However, one would expect that the activity of modifying enzymes was counteracted and balanced, in this case by phosphatases, to allow dynamic regulation of cellular signaling and function. In this respect, the number of putative candidates involved in Wnt/β-Catenin signaling was and still is very limited. With the identification of the phosphatase Pgam5 as a Wnt/β-Catenin antagonist in Xenopus development and human cells, one element was added to the kinase and phosphatase network that regulates Dvl function in Wnt signaling.

Xenopus embryos were generated and cultured according to general protocols and staged according to the normal table of Nieuwkoop and Faber (Nieuwkoop and Faber, 1994). All procedures were carried out according to the German animal use and care law (Tierschutzgesetz) and approved by the local authorities and committees (animal care and housing approval: I/39/EE006, Veterinäramt Erlangen; animal experiments approval: 54-2532.2-8/08, German state administration Bavaria/Regierung von Mittelfranken).

RNA for microinjection was prepared from the respective plasmids (Table S1) using the mMessage mMachine Kit (Ambion). Embryos were injected as indicated and cultivated until they reached the desired stage. In situ hybridizations were carried out as described by Harland (1991).

HEK293T cells were transiently transfected with Arrb2-Flag and Dvl2-GFP. Bait proteins were affinity isolated on anti-Flag beads (Sigma-Aldrich) or on anti-GFP beads prepared by coupling an anti-GFP antibody (Acris Antibodies) covalently to agarose beads (Direct IP Kit, Thermo Scientific). Eluates were digested with trypsin and subsequently with Lys-C in solution. After addition of protein digests as quantitative standards, the samples were desalted on C-18 stage tips (Nest Group) and analyzed by nanoflow HPLC-MS/MS (2D-NanoLC; Orbitrap Velos MS, Thermo Fisher Scientific; Vasilj et al., 2012). Protein identification was carried out using Mascot V2.2 (Matrixscience) and the determination of peptide ion signals for label-free quantification was performed using Progenesis LCMS V2.6 (Nonlinear Dynamics) and relative quantification was based on the most intense three signals (MI3) (Groessl et al., 2012; Silva et al., 2006). Isoform specificity of peptides was obtained from the proteomicsdb database (www.proteomicsdb.org) (Wilhelm et al., 2014).

HEK293T human embryonic kidney cells (Leibniz Institute Collections of Microorganisms and Cell Culture, DSMZ, Germany), SW480 colorectal cancer cells and C2C12 cells were cultured in DMEM supplemented with 10% fetal calf serum (Life Technologies) at 37°C in a humidified atmosphere of 10% CO₂ and transfected as indicated using Rotifect (Carl Roth), polyethylenimine (Sigma-Aldrich) or Oligofectamine (ThermoFisher), according to the manufacturer's instructions.

Xenopus embryonic fibroblasts (XTC, a kind gift from Ana Losada, CNIO, Madrid, Spain) were cultured in 67% DMEM/H₂O supplemented with 10% fetal calf serum at 25°C in a humidified atmosphere of 5% CO₂ and transfected as indicated. Information on plasmids used for transfection is summarized in Table S1.

For immunofluorescence staining, cells were transfected and stimulated as indicated, and fixed with 4% paraformaldehyde 24 h post-transfection. Cells were permeabilized with 0.1% Triton X-100, blocked with 10% horse serum and endogenous β-Catenin was detected using rat anti-β-Catenin and anti-rat-Cy3 (Dianova, 1:400) antibodies; nuclei were stained with DAPI. Immunofluorescence imaging was carried out using a Zeiss Apotome imaging system (Zeiss) or a Leica SP5 confocal laser-scanning microscope (Leica). Quantification of fluorescence intensity was carried out using the ImageJ software package. Briefly, nuclei were identified based on DAPI staining and β-Catenin intensity was measured in individual nuclei of Pgam5-GFP-positive and -negative cells.

Statistical differences in phenotype frequencies between two conditions were determined using the χ²-test. For graded phenotypes such as ventralization or axis duplication, pairwise comparisons using a two-sample Wilcoxon rank sum test were performed. For quantitative measurements, Student's t-tests were applied as indicated. All statistical calculations were carried out using either built-in functions or user-defined calculations in Microsoft Excel.

Dorsal tissue explants were prepared at stage 11.5, sectioned into anterior, mid and posterior parts, and snap-frozen until analysis. Embryos, cells or tissue explants were lysed in 20 mM HEPES (pH 7.4), 150 mM NaCl, 0.5% NP-40 and 10% glycerol supplemented with protease inhibitor and phosphatase inhibitor cocktails (Roche) at 4°C. Lysates were cleared at 16,000 g for 10 min, and the protein concentration was determined using a BCA Assay, according to the manufacturer's instructions (Applichem). For co-immunoprecipitation, lysates were incubated for 4 h at 4°C with the appropriate antibody and protein G beads (Life Technologies). Immunoprecipitates were collected, washed four times with lysis buffer and eluted with SDS sample buffer. Samples were separated on polyacrylamide gels, transferred to PVDF membranes and probed with the indicated antibodies using chemoluminescence or colorimetric detection as described previously (Pfister et al., 2012). For a complete list of primary antibodies see Table S2.

Cytoplasmic fractions were obtained by hypotonic lysis and cleared by centrifugation at 16,000 g in 20 mM Tris-HCl (pH 7.5) and 1 mM EDTA; expression levels of non-cytoplasmic proteins were determined in Triton X-100 extracts of the fraction insoluble in hypotonic lysis buffer. Nuclei were prepared and extracted as described previously (Dignam et al., 1983).

Total RNA was extracted from Xenopus embryos at the indicated developmental stages (High Pure RNA Isolation Kit) and reverse transcribed using MMLV reverse transcriptase (New England Biolabs). Quantitative RT-PCR (qRT-PCR) was carried out using Brilliant III Sybr Green Master Mix and the Agilent AriaMX system (Agilent Technologies). Primer sequences are listed in Table S3.

Activation of Wnt/β-Catenin activity was analyzed using HEK293T, C2C12 or SW480 cells, as indicated, transfected with the TOP-Flash or FOP-Flash reporter plasmid (Korinek et al., 1997) together with constitutively expressed Renilla luciferase (TK Renilla, Promega) or β-Galactosidase for normalization as indicated. Overexpression or knockdown was achieved by co-transfection of the respective expression plasmids or siRNAs. Cells were lysed 24 h post-transfection (plasmid transfection) or 72 h post-transfection (siRNA transfection) and reporter gene activity was measured in the lysates and normalized. All samples were measured as technical duplicates and at least three biological replicates.

GST-PGAM5Δ lacks amino acids 13 to 50, thereby removing the mitochondrial membrane targeting sequence for facilitated purification. GST-PGAM5Δ was expressed upon IPTG induction in Rosetta (DE3) bacteria. Bacteria were lysed in Triton X-100-based buffer containing lysozyme and GST-PGAM5Δ was precipitated on glutathione beads, washed on beads and eluted by addition of soluble glutathione.

The authors thank A. Losada, G. Weidinger, K. Takeda, A. Kikuchi, E. Arenas and R. T. Moon for providing plasmids and reagents; the Tcf1-Flag construct and the rat anti-β-Catenin antibody were kind gifts from K. Mansperger and R. Rupp. We are grateful to J. Ernesti and T. Hübler for their help with Xenopus experiments, and for helpful discussions and support from Andrej Shevchenko.