Dact1 induces Dishevelled oligomerization to facilitate binding partner switch and signalosome formation during convergent extension

Introduction

Convergent extension (CE) is a universal morphogenetic engine that shapes tissues and organs in diverse organisms ranging from insects to mammals^1,2. During germband extension in flies and neurulation in vertebrates, for example, oriented cell intercalation leads to multicellular reorganization to simultaneously elongate and narrow tissues along the anteroposterior (AP) and mediolateral (ML) axis, respectively^3,4,5,6. At the subcellular level, CE is regulated by directional extension of cell protrusions and contraction of cell-cell junctions via conserved mechanisms including polarized distribution and turnover of actomyosin and cadherins^7,8,9.

Despite deep conservation of subcellular, cellular and tissue behaviors, CE in insects and vertebrates rely on different genetic and signaling pathways as polarity input to guide directional activities. Spatially patterned expression of pair-rule genes and Toll receptors provide polarity cues for CE in insects but not vertebrates^10,11; whereas CE in vertebrates is orchestrated by core members of the planar cell polarity (PCP) pathway and non-canonical Wnt ligands^1,2,12,13. In humans, mutations in PCP genes and non-canonical Wnt gene WNT5A contribute to diverse congenital birth defects related to impaired CE, including neural tube closure and skeletal defects^14,15,16, highlighting the importance of PCP-mediated CE in human biology and diseases and the need to understand its operational mechanisms in vertebrate models.

The PCP pathway was originally discovered in flies as a mechanism that coordinates cellular polarity in the plane of an epithelium. PCP shares some components with the canonical Wnt pathway, including the receptor protein Frizzled (Fz) and the cytoplasmic signal transducer Dishevelled (Dvl), but does not act through β-catenin. Instead, Fz and Dvl function together with a set of distinct “core” proteins such as the tetraspanin protein Van gogh (Vang) in the PCP pathway^17,18,19. In static epithelial cells, inter- and intra-cellular interactions among PCP proteins establish feedback loops that lead to partitioning of Fz/Dvl and Vang into two distinct complexes on opposing cell cortexes to set up a molecular asymmetry that in turn coordinates and propagates cellular polarity across the epithelium^20,21,22,23.

Although the core PCP proteins display highly conserved action in coordinating epithelial cell polarity in both flies and vertebrates, they are required only for CE in vertebrates but not flies^11,13. Furthermore, the asymmetric distribution of core PCP proteins characteristic in static epithelial cells has not been consistently observed during all the dynamic processes of CE in vertebrates²⁴. Finally, whereas non-canonical Wnts, including Wnt5a and Wnt11, are essential for PCP-mediated CE in vertebrates^12,25,26, Wnt ligands are dispensable for PCP in flies^27,28,29. Collectively, these differences emphasize that adopting core PCP proteins for non-canonical Wnt signaling to regulate CE is a vertebrate specific adaptation, and there are likely unique mechanisms that modify PCP protein action during vertebrate CE.

This idea led us to investigate Dact1 (Dishevelled-binding antagonist of beta-catenin 1), first identified as a Dvl-binding protein belonging to a chordate specific gene family believed to evolve late in the deuterostome lineage³⁰. Early studies in zebrafish and frogs provided conflicting evidence that Dact1 could either positively or negatively regulate canonical Wnt signaling^31,32,33,34. More recent mouse genetic studies, however, revealed that loss of Dact1 caused only morphogenetic but no canonical Wnt signaling defects³⁵. Furthermore, Dact1 displayed a functional antagonism against Vangl2 (Vang-like 2, an ortholog of fly PCP gene Vang) in the mouse: the morphogenetic defects in mice carrying one copy of the semi-dominant negative Vangl2^Looptail allele (Vangl2^Looptail/+) can be rescued by loss of Dact1, and those in Dact1^-/- mutants can be reciprocally rescued by heterozygous Vangl2^Looptail mutation³⁵. Although the detailed mechanism behind this genetic interaction was not known, this study found that Dact1 can bind to not only Dvl but also Vangl2 in vitro.

Direct binding between Vang/Vangl and Dvl was also reported by a number of early studies^36,37,38,39. This finding was intriguing since Vang/Vangl and Dvl are often observed to localize to opposing cell cortexes during epithelial PCP signaling. A new fly study demonstrated that a Vang point mutation specifically abolishing interaction with Dvl causes PCP defects, and proposed that a transient Vang-Dvl binding might be crucial for PCP signaling⁴⁰. We previously also studied functional and biochemical interactions between Dvl2 and Vangl2 in mice and frogs, and proposed a model in which the Vangl2-Dvl2 interaction acts as a key bi-functional switch to regulate CE: Vangl2 binding recruits and poises Dvl2 at the plasma membrane for non-canonical Wnt signaling, but also keeps Dvl2 inactive to prevent ectopic signaling. In the presence of non-canonical Wnt ligands, Dvl2 transitions from Vangl2 to Fz to activate signaling required for CE⁴¹.

While this model provides a foundation to understand the central logic of PCP protein action during CE, a major question remaining is how Dvl2 transition between Vangl2 and Fz is regulated intracellularly. Given the reported role of Dact1 in vertebrate CE, its functional antagonism against Vangl2 and its ability to bind to Dvl and Vangl2 in vitro, we postulated that Dact1 may regulate Vangl2-Dvl interaction during CE. Our current study revealed that Dact1 positively regulates non-canonical Wnt signaling by helping Dvl2 to disengage from Vangl2 and transition to Fz to form signalosome-like clusters. Most interestingly, Dact1 functions by promoting multivalent interactions among Dvl2 proteins to facilitate their oligomerization. Our findings uncover a mechanism where the oligomeric state of Dvl2 simultaneously facilitates binding partner changes and enhances non-canonical Wnt signaling output to regulate the dynamic process of CE.

Results

Dact1 synergizes with Dvl2 and antagonizes Vangl2 during CE

In gastrulating Xenopus embryos, both loss and over-expression of core PCP proteins such as Dvl2 and Vangl2 can block CE-mediated elongation of the A-P axis, a phenotype that can be quantified by measuring the length-to-width ratio (LWR) of the embryos^42,43,44. To study the action of Dact1 during Xenopus CE, we first injected mRNA encoding either GFP-tagged Xenopus Dact1.L (GFP-Dact1.L, formerly known as Dapper³¹) or mScarlet-I-tagged human DACT1 (mSc-hDACT1) into the dorsal marginal zone (DMZ) of four-cell stage embryos. We found that DMZ injection of GFP-dact1.L and mSc-hDACT1 mRNA can both dose-dependently block CE, causing significant reduction of the LWR in the injected embryos by the mid-tailbud stages (~St.25-26) (Fig. 1A, B). Therefore, when over-expressed human DACT1 functions similarly to its Xenopus ortholog to perturb CE.

**Fig. 1: DACT1 synergizes with Dvl2 but antagonizes Vangl2 during CE.**

To test whether loss of endogenous dact1 could also block CE, we used two previously reported morpholinos (MO)^31,32 to knockdown dact1.L and its homolog dact1.S (formerly known as Frodo). Dorsal injection of dact1.S/L MOs was previously reported to cause loss of head structures and abnormal morphogenesis reminiscent of CE defects³³. We found that DMZ injection of the MOs perturbed CE, resulting in embryos with reduced LWR (Supplementary Fig. 1A, A’). Using animal cap explant as an additional assay for CE, we found that dact1.S/L MOs could also abolish activin induced elongation of the explants (Supplementary Fig. 1B, B’). Co-injection of mSc-hDACT1 can restore elongation in the morphant explants (Supplementary Fig.1B, B’), confirming that the CE defect is due to loss of endogenous Dact1 rather than non-specific effect of the morpholinos.

The above data indicate that, similar to core PCP members like Dvl2 and Vangl2, both gain- and loss-of-function of Dact1 can disrupt CE in Xenopus. Our previous work pointed to the important balance of Dvl-mediated activation of non-canonical Wnt signaling and Vangl2-mediated inhibition of Dvl for proper CE morphogenesis. For instance, we found that whereas over-expression of Dvl2 or Vangl2 individually can each disrupt CE, their co-overexpression can neutralize each other’s effect to restore CE⁴¹. Given that Dact1 has been reported to interact with both Dvl2 and Vangl2 biochemically³⁵, we studied how Dact1 would functionally interact with Dvl2 and Vangl2. Injecting 0.2 ng mSc-hDACT1 or 0.25 ng Flag-Dvl2 mRNA individually into the DMZ causes only mild CE defect, but their co-injection results in significantly more severe CE defect (Fig. 1C, C’). Similarly, co-injection of 15 ng dact1.S MO with 0.25 ng Xdd1 mRNA (encoding a dominant negative mutant of Dvl⁴⁵) into the DMZ results in significantly more severe CE defect than separate injection of either (Supplementary Fig. 2A, A’). These gain- and loss-of-function results therefore support that Dact1 and Dvl2 synergize during CE. Conversely, 0.2 ng myc-Vangl2 mRNA induced severe CE defects can be rescued significantly by co-injecting mSc-hDACT1 in a dose-dependent manner, suggesting that Dact1 antagonizes Vangl2 during CE (Fig. 1C, C’, Supplementary Fig. 2B, B’). The over-expression result in our Xenopus study is also consistent with the previous mouse genetic study where partial loss of Vangl2 and Dact1 can reciprocally rescue each other to restore normal morphogenesis³⁵. Taken together, these data indicate that Dact1 synergizes with Dvl2 and antagonizes Vangl2 to regulate CE in vertebrates.

DACT1 removes Dvl2 from Vangl2 by promoting DIX-mediated oligomerization

To understand the mechanism underlying Dact1’s synergy with Dvl2 and antagonism against Vangl2, we studied how Dact1 may impact Dvl2-Vangl2 interaction. We and others reported previously that Vangl2 binds to Dvl2 via its cytoplasmic tail^36,39 and cell-autonomously recruits Dvl2 to the plasma membrane in DMZ cells undergoing CE⁴¹. In the current study, we used animal cap explant as another model to study Dvl2-Vangl2 interaction. We injected varying doses of Dvl2-EGFP mRNA (0.03-0.6 ng) into two-cell stage Xenopus embryos and prepared animal cap explants. In naïve animal cap explant without Activin treatment, Dvl2-EGFP remains diffusely distributed in the cytoplasm (Fig. 2K; Supplementary Fig. 3A-C). In contrast, in explants that have been injected with low-dose Dvl2-EGFP (0.03 ng) and treated with Activin overnight to induce the onset of CE, Dvl2-EGFP displays largely even distribution on the plasma membrane (Supplementary Fig. 3E-G), similar to the previous report¹³.

**Fig. 2: Vangl2 binding is not required for Dact1 to detach Dvl2 from Vangl2.**

Our western blot analysis with an anti-Vangl2 antibody⁴⁶ revealed that Activin treatment upregulated endogenous Xenopus Vangl2 (XVangl2) by ~two folds (Supplementary Fig. 3Q, R), suggesting to us that increased XVangl2 level may be responsible for the observed Dvl2-EGFP plasma membrane recruitment. In support of this idea, an MO that diminishes endogenous XVangl2 also abolishes Activin-induced plasma membrane recruitment of Dvl2-EGFP (Supplementary Fig. 3I-K; Q, R), whereas 50 pg mouse Vangl2 (mVangl2) mRNA injection can restore Dvl2 plasma membrane recruitment in XVangl2 morphant. Moreover, in naïve explant over-expression of mVangl2 is sufficient to recruit Dvl2-EGFP from the cytosol to the plasma membrane (Fig. 2L-N). Together, these data demonstrate conclusively that in animal cap explant, Vangl2 is both necessary and sufficient to recruit Dvl2 to the plasma membrane for the initiation of CE.

When we continued to examine the Activin treated animal cap explants six hours later, where CE has resulted in active extension of the explants, we found that Dvl2-EGFP formed many distinct puncta on the plasma membrane (Supplementary Fig. 3S). Immunostaining showed that endogenous XVangl2 remains evenly distributed on the plasma membrane and does not co-cluster with Dvl2-EGFP into puncta (Supplementary Fig. 3U, V). By contrast, small dose of Fz7-EGFP (20 pg mRNA) injected into the animal region also displays distinct puncta formation on the plasma membrane like Dvl2 (Supplementary Fig. 3W). These observations led us to postulate that with the progression of CE, Dvl2 may disengage from Vangl2 and form signalosome-like clusters with Fz7, a model that is supported by our subsequent studies (see below).

Having established the animal cap as a valid model to study the interaction between Dvl2 and Vangl2, we used it to investigate how their interaction could be modulated by Dact1. First, we found that over-expressed Vangl2 can recruit not only Dvl2 but also hDACT1 from the cytoplasm to the plasma membrane (Fig. 2B-E, Supp. Fig. 4D-F, J-L), consistent with the reported Vangl2-Dact1 binding in vitro³⁵. When we co-expressed Vangl2, Dvl2 and hDACT1 together, we found that Vangl2 remains on the plasma membrane but Dvl2 displays reduced co-localization with Vangl2 with simultaneous formation of puncta (Fig. 2O-R). Close examination of these puncta revealed that they are primarily in the cytosol (Fig. 2O’-R’, arrows) and contain only Dvl2 and hDACT1 but not Vangl2. A few puncta can also be observed on the plasma membrane (Fig. 2O’-R’, arrowheads), and in those cases they display enrichment of only Dvl2 and DACT1 but not Vangl2, resembling the Dvl2 puncta found in the elongating animal cap explants after Activin treatment (Supplementary Fig. 3S-Z). To quantitatively assess Dvl2-Vangl2 colocalization we calculated Pearson’s coefficient using ImageJ plug-in JACoP and found that hDACT1 over-expression indeed significantly reduced Dvl2 co-localization with Vangl2 (Fig. 2W).

We considered two possible mechanisms through which Dact1 disengages Dvl2 from Vangl2. First, Dact1 and Dvl2 may bind to overlapping region on Vangl2 such that Dact1 disrupts Dvl2-Vangl2 interaction directly through competitive binding. We tested this possibility by creating a hDACT1-ΔSR mutant in which the serine-rich region (SR) required for Vangl2 interaction³⁵ is deleted (Fig. 2A). Indeed, hDACT1-ΔSR can no longer be recruited to the plasma membrane by Vangl2 (Fig. 2F-I, Supp. Fig. 4M-R). When co-expressed with Dvl2 and Vangl2, however, hDACT1-ΔSR retains the ability to reduce Dvl2 co-localization with Vangl2 and simultaneously induce Dvl2 to form puncta (Fig. 2S-W). Therefore, competitive binding on Vangl2 does not appear to be the mechanism through which Dact1 disengages Dvl2 from Vangl2.

We then considered the second possibility that interaction with Dact1 may induce a change on Dvl such that the binding affinity with Vangl2 is reduced. In this scenario the most significant change that we noted is puncta formation, so we postulate that Dvl2 oligomerization may be promoted by Dact1 to trigger its dislodging from Vangl2. As the Dact1-induced Dvl2 puncta are reminiscent of the proteinaceous bodies formed through Dvl oligomerization via its N-terminal DIX (Dishevelled and Axin) domain⁴⁷, we tested whether the DIX domain could be required for Dvl2 to undergo Dact1 induced puncta formation and disengagement from Vangl2. To this end we test two DIX domain mutants: Dvl2-ΔDIX in which the entire DIX domain is deleted⁴⁸, and Dvl2-M2M4 in which triple point mutations (Y27D/V67A/K68A) disrupt DIX mediated oligomerization⁴⁷. When injected with Vangl2, both mutants can still be recruited to the plasma membrane (Fig. 3F-H, Supp. Fig. 5H-J), but does not display either puncta formation or reduced co-localization with Vangl2 upon hDACT1 co-injection (Fig. 3M-Q, Supp. Fig. 5K-N). Instead, both DIX domain mutants and hDACT1 are able to remain co-localized with Vangl2 on the plasma membrane. Together these data suggest that Dact1 promotes Dvl2 to undergo DIX-mediated self-oligomerization to reduce Dvl-Vangl2 interaction.

**Fig. 3: DACT1 removes Dvl2 from Vangl2 by promoting DIX-mediated oligomerization.**

To confirm the imaging results, we performed in vivo and in vitro co-IP (co-immunoprecipitation) assays. To first test whether endogenous Dvl2 and Vangl2 could bind to each other and how their binding could be impacted by Dact1, we used SH-SY5Y cell line that expresses high levels of DVL2 and VANGL2 but little hDACT1. By pulling down endogenous VANGL2 using the anti-Vangl2 antibody⁴⁶, we could also co-IP endogenous DVL2 (Fig. 3R). Transfection of hDACT1, but not GFP control, can significantly reduce DVL2 co-IP with VANGL2 (Fig. 3R, R’), supporting that endogenous VANGL2-DVL2 interaction can be reduced by increased level of hDACT1. We then tested how Dact1 may impact Vangl2-Dvl2 binding in the Xenopus model by co-injecting Vangl2 and wild-type Dvl2 or Dvl2-M2M4 mutant, either with or without hDACT1. Our co-IP results show that hDACT1 can reduce co-IP between Vangl2 with wild-type Dvl2 but not Dvl2-M2M4 mutant (Fig. 3S, S’), indicating that the oligomerization ability of the DIX domain is required for Dvl2 to undergo Dact1 induced dissociation from Vangl2. Curiously, we notice that Dact1 over-expression also appears to induce Dvl phosphorylation in an oligomerization dependent manner, as evidenced by the mobility shift of endogenous Dvl2 and wild-type Dvl2-EGFP, but not Dvl2-M2M4 mutant, upon hDACT transfection or co-injection (Fig. 3R, S, input lanes). Only the faster migrating, presumably unphosphorylated form of Dvl2, was precipitated by Vangl2 (Fig. 3R, S, IP lanes of Dvl2 co-immunoprecipitated by anti-Vangl2 or anti-Myc antibodies), similar to our recent report⁴⁹.

Lastly, we used in vitro binding and co-IP assay to test whether Dact1 is able to dissociate Dvl2 from existing Dvl2-Vangl2 complexes. We used wheat-germ extract for in vitro synthesis of Flag tagged wild-type Dvl2 or ΔDIX mutant, Myc tagged Vangl2 cytoplasmic tail (Myc-Vangl2C) and V5 tagged mSc-hDACT1. We first pre-incubated Myc-Vangl2C with Flag-Dvl2 or Flag-ΔDIX to allow complex formation, and then added either hDACT1 or wheat-germ extract control to incubate for another hour before performing co-IP with anti-Myc antibody (Supplementary Fig. 5P). Our data show that compared to the wheat-germ controls, the subsequent incubation with hDACT1 significantly reduced the amount of Flag-Dvl2 pulled down by Myc-Vangl2 (Supplementary Fig. 5Q, Q’). In contrast, Flag-ΔDIX pull-down by Myc-Vangl2 is not significantly affected by the addition of hDACT1 (Supplementary Fig. 5R, R’). Together, our imaging and biochemical assays strongly support that Dact1 dissociates Dvl2 from Vangl2 by inducing DIX-mediated oligomerization of Dvl2.

Finally, to test the functional significance of the ability of Dact1 to promote Dvl2 oligomerization and disengagement from Vangl2 during CE, we performed DMZ injection and found that whereas hDACT1 can synergize with wild-type Dvl2 to disrupt CE, the synergy is not observed with either form of Dvl2-DIX domain mutants (Supplementary Fig. 6A, B). Moreover, the severe CE phenotype induced by Xdd1 can be significantly rescued by wild-type Dvl2 but not Dvl2-M2M4 mutant (Supplementary Fig. 6C), indicating that the oligomerization function of the DIX domain is necessary for Dvl2 to promote CE. Collectively, these data support our idea that Dact1 promotes Dvl2 oligomerization to overcome Vangl2 sequestration, and in turn to facilitate activation of non-canonical Wnt signaling during CE.

The leucine zipper and PDZ binding domains are required for Dact1 to induce Dvl2 oligomerization and disengagement from Vangl2

We next investigated which domains of Dact1 are required to promote Dvl2 oligomerization and disengagement from Vangl2. Besides the SR domain involved in Vangl2 binding (Fig. 2A), Dact1 has two other highly conserved domains: the leucine zipper (LZ) and PDZ-binding (PDZb) domain that mediates its homodimerization and binding to Dvl, respectively^35,50. Using previously reported GFP-tagged Dact1.L wild-type and mutants lacking the LZ or PDZb domains (GFP-Dact1.LΔLZ and GFP-Dact1.LΔPDZb; Fig. 4A³¹;), we performed various co-injection and imaging experiments. First, we found that like hDACT1, wild-type GFP-Dact1.L can induce co-injected Dvl2 (tagged with mCherry or mScarlet-I; Dvl2-mCh or Dvl2-mSc) to form puncta, and co-clusters with Dvl2 in these puncta (Supplementary Fig. 7M-P). In contrast, neither GFP-Dact1.LΔLZ nor GFP-Dact1.LΔPDZb mutant can induce Dvl2 to form puncta, or to form puncta themselves (Supplementary Fig. 7Q-X). These results indicate that, in order for Dact1 to promote Dvl2 oligomerization, it has to be able to 1) bind directly to Dvl via its PDZb domain, and 2) homodimerize via its LZ domain (see model in Fig. 9A and Discussion). Secondly, we found that both wild-type Dact1.L and the two mutants can be recruited to the plasma membrane upon co-injection with Vangl2 (Fig. 4E-M), indicating that LZ and PDZb are dispensable for Vangl2-mediated plasma membrane recruitment.

**Fig. 4: The homodimerization and Dvl binding domains are required for Dact1 to induce Dvl2 detachment from Vangl2.**

To further test whether Dact1 induced Dvl2 oligomerization may underlie its ability to disengage Dvl2 from Vangl2, we co-expressed wild-type or the two mutant Dact1.L with both Vangl2 and Dvl. Similar to hDACT1, wild-type Dact1.L does not alter the plasma membrane localization of Vangl2 but induces Dvl2 to form puncta that are mostly detached from Vangl2 (Fig. 4N-Q, Z). Dact1.LΔLZ and Dact1.LΔPDZb mutants, however, do not induce Dvl2 to form puncta or detach from Vangl2. Instead, they co-localize with Dvl2 and Vangl2 on the plasma membrane (Fig. 4R-Y). Pearson’s coefficient analyses further revealed that Dvl2 co-localization with Vangl2 is diminished by wild-type Dact1.L but not Dact1.LΔLZ or Dact1.LΔPDZb mutants (Fig. 4Z). Together with the analyses of hDACT1-ΔSR (Fig. 2) and Dvl2-M2M4 (Fig. 3) mutants, these data indicate that, via binding to Dvl2 and self-homodimerization, Dact1 facilitates formation of high-order complex of Dact1-Dvl2 to detach Dvl2 from Vangl2.

hDACT1 induces Dvl2 clustering on the plasma membrane with Fz7

Dvl is known to interact with Fz, and plasma membrane recruitment of Dvl by Fz is crucial for PCP signaling⁵¹. Given that Dact1 induced oligomerization causes Dvl2 detachment from Vangl2 on the plasma membrane, we wondered how Dact1 may impact Dvl2 when recruited to the plasma membrane by Fz. When co-injected with Fz7 into animal cap cells, Dvl2-EGFP becomes evenly distributed on the plasma membrane as expected (Fig. 5D), but mSc-hDACT1 remains diffusely localized in the cytoplasm (Fig. 5F), indicating that unlike Dvl2, Dact1 cannot interact directly with Fz. We then co-injected Fz7 with both mSc-hDACT1 and Dvl2-EGFP, and found that hDACT1 can still induce Dvl2 to form puncta and co-localize with Dvl2 in these puncta. In this case, however, majority of the puncta appeared to be retained on the plasma membrane, presumably with Fz7 (Fig. 5G-I’).

**Fig. 5: Dact1 induced Dvl2 oligomers attach to the plasma membrane with Fz.**

To directly assess whether the Dact1-induced Dvl puncta is indeed on the plasma membrane with Fz7, we injected Dvl2-mCh with GFP-tagged Fz7, either with or without hDACT1, into animal cap cells. In the absence of hDACT1 co-injection, Dvl2-mCh displays even plasma membrane distribution with Fz7 (Supplementary Fig. 8A-C). In contrast, with hDACT1 co-injection, Dvl2-mCh formed puncta that are largely retained on the plasma membrane with Fz (arrowheads, Supplementary Fig. 8D-F). To quantify the impact of hDACT1 on Dvl2 co-localization with Fz, we performed Pearson’s coefficient analyses and found that Dvl2 co-localization with Fz7 is not reduced by hDACT1 despite the induced puncta formation (Supplementary Fig. 8G).

To test whether formation of Dvl2 puncta on the plasma membrane with Fz also occur through DIX-mediated oligomerization, we injected Dvl2-M2M4 mutant with Fz7 and hDACT1. We found that, like wild-type Dvl2, Dvl2- M2M4 can be recruited to the plasma membrane by Fz7 (Fig. 5E). Unlike wild-type Dvl2, however, Dvl2- M2M4does not form puncta upon hDACT1 co-injection; and in this case both Dvl2-M2M4 and hDACT1 display even distribution on the plasma membrane (Fig. 5J-L). Since hDACT1 cannot be recruited by Fz directly (Fig. 5F), we interpret this result to mean that Dvl2-M2M4 is able to bind hDACT1and recruit it to the plasma membrane via Fz7. Collectively, these data indicate that in the presence of Fz, Dact1 also promotes Dvl2 to undergo DIX-mediated oligomerization. But in contrast to Vangl2, the Dact1-induced Dvl2 oligomers remain bound to Fz on the plasma membrane.

Dact1 mediates non-canonical Wnt induced clustering of Fz and Dvl

In the absence of non-canonical Wnt ligands, Dact1-induced Dvl2 oligomerization is not sufficient to induce co-clustering of Fz (Supplementary Fig. 8D, E), whereas signalosomes with increased coupling of receptors with their cytoplasmic signaling transducers are observed in the presence of Wnt ligands to enhance signaling output⁵². We therefore tested whether Dact1-mediated Dvl oligomerization may also promote Fz/Dvl response to non-canonical Wnt. In Xenopus and zebrafish, non-canonical Wnt such as Wnt11 and Wnt5a can induce formation of Fz-Dvl complexes that cluster into patches at the cell-cell contacts^53,54, providing a cellular readout for Fz/Dvl response to non-canonical Wnt signaling. We found that in animal cap explants, Xenopus Wnt11 can induce mouse Dvl2-EGFP or Dvl2-mSc to form distinct patches along cell-cell contacts (Fig. 6A-D; Supplementary Fig. 9A-C). When EGFP-tagged Xenopus Fz7 is co-injected, it exclusively forms membrane patches that overlap completely with Dvl2 (Supplementary Fig. 9D-G), consistent with clustering of Fz-Dvl complexes. Similarly, mSc-hDACT1 also forms membrane patches under Wnt11 induction (Fig. 6E-H), and when co-expressed with Dvl2-EGFP it completely co-localizes with Wnt11-induced Dvl2-EGFP patches (Fig. 6I-L’), indicating that Dact1 is a component of the non-canonical Wnt induced Fz-Dvl clusters. On the other hand, partial knock down of endogenous dact1.L/S with morpholino significantly abolished Wnt11-induced patch formation of Dvl and Fz, based on quantification of Dvl2 patch length and number per cell border (Fig. 6M-O, T, U, Supplementary Fig. 9I-Q). The defect in dact1 morphants can be rescued by injecting wild-type mSc-hDACT1, indicating that it arises from loss of endogenous Dact1 rather than non-specific effect of the morpholino (Fig. 6P-S, T, U). Together, these data support the notion that Dact1, and in turn Dact1-induced Dvl oligomerization, is necessary for clustering of Fz-Dvl complexes induced by non-canonical Wnt.

**Fig. 6: Dact1 is a necessary component of the Fz/Dvl cluster induced by non-canonical Wnt signaling.**

Previously Yamanaka et al. reported that Xenopus Dvl lacking the DIX domain is unable to participate in Wnt11 induced Fz-Dvl patch formation⁵⁴, supporting the notion that Dvl oligomerization underlies clustering of Fz-Dvl complexes upon Wnt induction. We confirmed the prior result, and found that Dvl2-ΔDIX and Dvl2-M2M4 mutations either completely blocked or severely diminished Wnt11 induced patch formation on the plasma membrane (Supplementary Fig. 10B, C). Unexpectedly, co-injection of hDACT1 enabled both Dvl2 DIX domain mutants to form patches in response to Wnt11 (Supplementary Fig. 10 E-J). Our further studies revealed that Dact1 promotes Dvl to oligomerize via multivalent interactions (see below), providing a possible explanation for this unexpected result.

Lastly, since both the homodimerizing LZ and Dvl-interacting PDZb domains are required for Dact1 to promote Dvl2 oligomerization and the resulting detachment from Vangl2 (Fig. 4; Supplementary Fig. 7), we tested their role in Wnt11-induced Dvl patch formation. To our surprise, when co-injected with Dvl2 and Wnt11 into otherwise wild-type Xenopus embryos, both Dact1.L-ΔLZ and Dact1.L-ΔPDZb mutants are still able to form patches with Dvl2 on the plasma membrane. Closer examination, however, revealed that the patches formed by the two mutants appear less distinct than those formed by wild-type Dact1 with Dvl2 (Supplementary Fig. 11 D-E’). Quantification of signal heterogeneity using the Mean to Sigma Ratio revealed a significant decrease of Dvl2 enrichment in the patches formed with the two Dact1.L mutants than with wild-type Dact1.L (Supplementary Fig. 11K). In dact1 morphant, Wnt11-induced Dvl2 patch formation defect can be significantly rescued by wild-type Dact1.L, but is only marginally rescued by Dact1.L-ΔLZ and completely failed to be rescued by Dact1.L-ΔPDZb mutant (Supplementary Fig. 11F-J”’, L). These findings indicate that both the Dvl-binding and homodimerization function of Dact1 are required for it to mediate Dvl2 clustering into patches in response to non-canonical Wnt signaling.

Distinct Dvl2 binding modes likely underlie differential oligomer association with Fz vs. Vangl

We next addressed how Dact1-induced Dvl2 oligomers largely detach from Vangl2 but remain bound to Fz7. To understand this differential binding preference, we examined the behaviors of various Dvl2 mutants. Besides the N-terminal DIX domain, Dvl has highly conserved central PDZ (Postsynaptic density protein-95, Disk large tumor suppressor, Zonula occludens-1) and C-terminal DEP (Dishevelled, Egl-10, and Pleckstrin) domains. We first created Dvl2-ΔPDZ mutant lacking the entire PDZ domain (aa250-348). Compared to wildtype Dvl2, Dvl2-ΔPDZ displays severely reduced plasma membrane recruitment by Vangl2 when co-injected (Fig. 7A-H), suggesting that the PDZ domain is critical for Dvl2 interaction with Vangl2. To our knowledge, this is the first in vivo evidence that PDZ mediates Dvl-Vangl interaction, consistent with in vitro studies suggesting a requirement of PDZ for Dvl binding to Vangl^35,38,39. On the other hand, when co-injected with Fz, Dvl2-ΔPDZ displays similar plasma membrane recruitment as wild-type Dvl2 (Fig. 7M-P, Q and T), consistent with the reports that the PDZ domain is largely dispensable for Dvl-Fz interaction in flies and mammalian cells^51,55,56.

**Fig. 7: The PDZ and DEP domain mediates Dvl2 binding to Vangl2 and Fz, respectively.**

We next created Dvl2 DEP domain mutants to investigate its function. Since deleting this domain will remove a stretch of polybasic residues that stabilize Dvl at the plasma membrane by binding to negatively charged phospholipids⁵⁷, we took advantage of several known point mutations. E499 and C501 are two conserved residues mapped to the loop region between the anti-parallel β strands in the C-terminal portion of the DEP domains; and their respective mutations, E499G (EG) and C501R (CR), abolish PCP/ non-canonical Wnt signaling but maintain the overall structural integrity of the DEP domain^58,59. We created a Dvl2-EG/CR double mutant and found that its plasma membrane recruitment by Vangl2 is largely maintained (Fig. 7I-L) but that by Fz7 is significantly reduced (Fig. 7R, T). We also studied the dsh1 mutant, where a missense mutation of a conserved lysine residue (K446) in the “DEP finger” region within a loop between α-helix 1 and 2 specifically disrupts PCP signaling in flies and CE in vertebrates^51,60,61. We made a Dvl2-446KA mutant by substituting K446 with alanine (KA) and found that, like Dvl2-EG/CA, its membrane recruitment by Fz7 is significantly weakened (Fig. 7S, T). Collectively, these results indicate that while the PDZ domain is responsible for Dvl-Vangl interaction, the DEP domain is involved in Dvl binding to Fz.

We then investigated the roles of the PDZ and DEP domains in Dact1-induced oligomerization, and retention of oligomerized Dvl with Vangl or Fz. When we co-injected Dvl2-ΔPDZ with hDACT1 and Vangl2, we found that Dvl2-ΔPDZ failed to form puncta, and hDACT1 remained on the plasma membrane with Vangl2 (Fig. 8E-H). This is consistent with the idea that Dact1, via its PDZb, binds to the PDZ of Dvl to induce oligomerization. To confirm this idea, we co-injected Dvl2-ΔPDZ mutant with hDACT1 and Fz7 and again found no puncta formation of either ΔPDZ or hDACT1 (Fig. 8P-R). In this case, although Dvl2-ΔPDZ can strongly co-localize with Fz7 on the plasma membrane, it fails to mediate hDACT1 recruitment to the plasma membrane like wild-type Dvl2 or DIX domain mutants (compare Fig. 8Q to Fig. 5H’ and K), again arguing diminished binding to Dact1. Pearson’s coefficient analyses revealed that, compared to wild-type Dvl2, co-localization between Dvl2-ΔPDZ and hDACT1 was significantly reduced, irrespective of co-injection with Vangl2 or Fz (Fig. 8Y). These data indicate that the PDZ domain of Dvl mediates not only its binding to Vangl2, but also its interaction with Dact1 and consequently, its oligomerization.

**Fig. 8: Requirement of the PDZ and DEP domains for Dvl2 to undergo Dact1-induced oligomerization.**

We then co-injected Dvl2-EG/CR mutant and hDACT1, either with Vangl2 or Fz7. Similar to Dvl2-ΔPDZ mutant, Dvl2-EG/CR mutant also failed to form puncta in either case (Fig. 8I-L and S-U). But unlike Dvl2-ΔPDZ, which strongly co-localizes with Fz yet fails to recruit hDACT1 to the plasma membrane, Dvl2-EG/CR mutant weakly co-localizes with Fz but is still able to enrich hDACT1 around the plasma membrane (Fig. 8S-U). Quantification of the colocalization of the Dvl2-EG/CR mutant and hDACT1 showed a Pearson’s coefficient similar to that of wild-type Dvl2 and hDACT1 (Fig. 8Y). We therefore reason that Dvl2-EG/CR mutant can still interact with Dact1, and that its failure to form puncta with hDACT1 is most likely due to the proposed role of DEP dimerization, and specifically the E499 residue, to cross-link DIX-mediated oligomers into higher order signalosome-like structure⁵⁰. We envision a model where Dact1 acts as a “super cross-linker” to facilitate multivalent interactions of both DIX-DIX and DEP-DEP domains to help Dvl form signalosome-like clusters (Fig. 9A).

**Fig. 9: Models for Dact1 induced Dvl oligomerization and binding partner switch during non-canonical Wnt signaling.**

Finally, when we co-expressed another DEP domain mutant, Dvl2-KA, with hDACT1 and Fz7, we found that, unlike Dvl2-EG/CA, Dvl2-KA was able to form puncta with hDACT1. The Pearson’s coefficient of Dvl2-KA with hDACT1 was similar to that of wild-type Dvl2 (Fig. 8V-X, Y). However, the puncta formed by Dvl2-KA and hDACT1 largely fail to remain attached to Fz7 and are instead localized in the cytoplasm (Fig. 8V-X). These data are consistent with the report that K446 is required for DEP binding (in its monomeric form) to Fz, but not for DEP dimerization⁶². Therefore, attachment of Dvl2 oligomers to Fz still requires binding via the DEP finger.

Together, these data demonstrate that whereas the PDZ domain mediates Dvl binding to both Vangl2 and Dact1, the DEP domain is primarily involved in binding to Fz only. Therefore, binding of Dact1 to the PDZ domain, as well as its subsequent induction of Dvl oligomerization, may occlude the availability of PDZ for Vangl2 interaction, leading to the eventual detachment of Dvl2 oligomers from Vangl2. Conversely, some of the DEP domain within Dvl oligomers may remain in the monomeric form to preserve the DEP finger conformation for interaction with Fz. In essence, we propose a model where Dact1 induces Dvl oligomerization to block PDZ-mediated interaction yet allow DEP-mediated binding mode, biasing Dvl2 oligomers to favor Fz over Vangl2 binding (Fig. 9B, C).

Discussion

To uncover potential vertebrate-specific mechanisms that evolved to regulate “core” PCP protein action for polarized cell re-arrangement during CE, we carried out cellular, molecular and functional analyses of a chordate-specific protein Dact1. Our data uncover an unexpected role of Dact1 in regulating Dvl oligomerization to simultaneously accomplish two goals: releasing Dvl from sequestration by Vangl and helping Dvl to cluster with Fz for activation of non-canonical Wnt signaling (Fig. 9). The combined effect of these two actions may provide the dynamic control of non-canonical Wnt signaling required for rapid and polarized cell behaviors underlying CE. Our findings and model establish a framework to decipher how core PCP proteins are adapted to control the dynamic process of CE.

Dact1 promotes Dvl oligomerization to regulate CE in vertebrates

Dact family of genes are considered “a chordate innovation” as they arose in the deuterostome lineage during evolution³⁰. Syntenic and phylogenetic analyses suggest that four Dact genes were generated by two rounds of whole genome duplication in the vertebrate ancestor, but subsequent gene loss left the extant species with varied numbers between one to four. Dact1 and 3 are thought to arise from one ancestral gene generated from the first genome duplication and function primarily in Wnt signaling modulation, whereas Dact2 and 4 arose from the other ancestral gene with additional role in Tgfβ signaling³⁰.

As the best studied Dact family member, Dact1 was initially proposed to act as a negative regulator of canonical Wnt signaling based on cell culture studies^31,63, but studies in frogs and zebrafish led to the opposite conclusion that Dact1 positively regulates canonical Wnt signaling, and may also regulate morphogenesis in a β-catenin independent manner^32,33,34. Mouse genetic studies demonstrated conclusively that loss of Dact1 did not alter canonical Wnt signaling but caused only CE defect due, at least in part, to elevated Vangl2 activity^35,64. But how Dact1 functions in CE or counters Vangl2 function remains unclear. Our studies in Xenopus embryos support the findings from the loss of function mouse genetic studies to indicate a crucial role of Dact1 in CE (Fig. 1), but more importantly our data helped to delineate a detailed mechanism where Dact1 promotes Dvl oligomerization to facilitate Dvl transition from Vangl to Fz to activate non-canonical Wnt/PCP signaling (Fig. 9).

A previous study, based on transient transfection in cell culture, proposed that Dact1 promotes degradation of aggregated Dvl via autophagy to suppress canonical Wnt signaling⁶⁵. The fact that Dact1^-/- knockout cells display no increased canonical Wnt signaling and only modest change in Dvl protein levels^35,64, however, raises the question whether the function of endogenous Dact1 is to induce Dvl degradation. Our data in this study provide several lines of evidence to support Dact1-induced Dvl oligomerization in activating non-canonical signaling during CE: 1) Dact1 dissociates Dvl from Vangl2 by promoting DIX-mediated oligomerization (Fig. 3); 2) Dact1 induced-Dvl oligomers cluster with Fz, and endogenous Dact1 is required for Dvl clustering with Fz in response to Wnt11 (Figs. 5, 6; Supplementary Fig. 9); and 3) Dact1 synergizes with Dvl during CE (Fig. 1) and their synergy depends on the ability of the DIX domain to mediate Dvl oligomerization (Supplementary Fig. 6). Therefore, we propose that, at least in vertebrate tissues undergoing CE, the role of endogenous Dact1 is to promote Dvl oligomerization to activate non-canonical Wnt/PCP signaling. In support of this view, a recent genetic study in zebrafish also demonstrated the requirement of Dact1 and 2 during CE⁶⁶. We note that an important distinction between our data and the previous report⁶⁵ is that in our Xenopus system expression of mouse Dvl2 alone does not often cause autonomous puncta formation/ aggregation (Fig. 2A; Supplementary Fig. 2A), presumably because the moderate dosage we used is closer to the physiological level. We verified this by injecting varying doses of Dvl2-EGFP, and found that Dvl2-EGFP displays diffuse distribution within the cytoplasm of uninduced animal cap cells injected with up to 600 pg of Dvl2-GFP mRNA at two-cell stage, and only forms puncta autonomously with 1200 pg injection (Supplementary Fig 12).

Dvl oligomerization in non-canonical Wnt/PCP signaling

Dvl oligomerization is best known in activating canonical Wnt signaling. A large body of work demonstrated unequivocally that, through both DIX-mediated polymerization and DEP-mediated dimerization, Dvl forms puncta-like signalosomes to interrupt Axin from assembling of the destruction complex^47,62,67. By contrast, whether Dvl forms oligomers and the potential role of Dvl oligomerization in PCP/ non-canonical Wnt signaling have been less clear. In flies, some studies found that mutations in the DIX domain could perturb both canonical Wnt and PCP signaling, implying a requirement of Dvl oligomerization in both pathways^51,58. Other studies, however, found the DIX domain dispensable for PCP⁶⁸. In Xenopus, deletion of the DIX domain was reported to diminish the ability of over-expressed Dvl to activate canonical Wnt signaling as well as to disrupt CE¹³, suggesting that DIX-mediated Dvl oligomerization could be required for CE in addition to canonical Wnt signaling. Our data showing that the functional synergy between Dvl and Dact1 relies on the oligomerization activity of the DIX domain (Supplementary Fig. 6) further supports a functional role of Dvl oligomerization in CE.

Cellular studies provided additional evidence for Dvl oligomerization in PCP/ non-canonical Wnt signaling. In flies, quantitative imaging studies revealed that in asymmetrically segregated stable PCP clusters, the ratio of Dvl to Fz is normally 2:1⁶⁹, suggestive of Dvl oligomerization in relation to Fz. In Xenopus, the DIX domain is normally required for Dvl to participate in the formation of non-canonical Wnt induced Fz/Dvl membrane patches, implying that the ability to oligomerize is necessary for Dvl to form these Fz/Dvl clusters (ref. ⁵⁴; Supplementary Fig. 10B, C). Our data revealed that Dact1 is also a necessary component of the Wnt11-induced Fz/Dvl clusters (Fig. 6, Supplementary Fig. 9), providing additional evidence for Dvl oligomerization as an underlying mechanism for Fz/Dvl clustering. The inclusion of Dact1 in the Dvl clusters may distinguish Dvl oligomers functioning in canonical Wnt versus non-canonical Wnt signaling (see below).

An interesting observation we made in this study is that over-expression of Dact1 allowed Dvl2-DIX domain mutants to participate in Wnt11-induced patch formation with Fz (Supplementary Fig. 10E-J). This observation could be explained by Dact1’s ability to enhance Dvl-oligomerization. The current literature on Dvl oligomerization during canonical Wnt signaling indicates the involvement of two domains: the DIX domain that undergoes head-to-tail polymerization to form fibrils, and the DEP domain that dimerizes to stabilize the weak interaction among DIX domains and to crosslink linear DIX polymers into “punctate phase-separated assemblies”^47,62. The central PDZ domain is not involved in Dvl oligomerization during canonical Wnt signaling, and is functionally dispensable for canonical Wnt signaling in mammalian cells and flies^51,55,70. Our data, in contrast, showed that Dact1, through binding to PDZ and self-dimerizing, provides a third means for Dvl to oligomerize (Figs. 4, 7, 8, 9; Supplementary Fig. 7). Therefore, in the presence of Wnt11, which may trigger local enrichment of Fz and Dvl to favor oligomerization, Dact1 over-expression may increase PDZ-mediated oligomerization to compensate for the normal requirement of the DIX domain, allowing Dvl2-ΔDIX and Dvl2-M2M4 mutants to participate in patch formation with Fz.

How does Dact1-dependent Dvl oligomerization promote non-canonical Wnt signaling for CE? Our data collectively suggest a model where Dact1-mediated oligomerization triggers 1) a binding mode switch to facilitate Dvl transition from Vangl to Fz; and 2) co-clustering of Dvl-Fz in response to non-canonical Wnt (Fig. 9B). Our previous work in the DMZ found that during CE, Vangl functions as a relay to cell-autonomously recruit Dvl to the plasma membrane: it helps to enrich Dvl at the plasma membrane to facilitate rapid Fz signaling upon binding of non-canonical Wnt ligands, but at the same time sequesters Dvl from Fz (in the absence of Wnt) to prevent ectopic activation^41,49. In the current study, we verified the prior finding using animal cap (Supplementary Fig. 3) and further found that the PDZ domain contributes to Dvl binding with Vangl2 in vivo (Fig. 7). Since Dact1 also binds to Dvl via the PDZ domain (Fig. 8; ref. ³⁵), and the PDZ of Dvl has only one cleft to accommodate binding of a small client peptide⁷¹, Dvl binding to Vangl2 or Dact1 is expected to be mutually exclusive. Intriguingly, however, competitive binding to PDZ alone does not seem to be sufficient for Dact1 to dissociate Dvl from Vangl2: our data show that although wild-type Dvl2 and the oligomerization deficient DIX or Dvl2-EG/CR mutants can all bind to Dact1, only wild-type Dvl2 but not DIX or Dvl2-EG/CR mutant can be detached from Vangl2 by Dact1 over-expression (Figs. 3, 8). Likewise, the dimerization deficient Dact1-ΔLZ mutant (which cannot induce Dvl oligomerization) also fails to dissociate Dvl from Vangl2 (Fig. 4R-U) despite its ability to bind Dvl³¹. Therefore, we postulate that Dvl oligomerization may either reinforce Dact1 binding to PDZ so that it can outcompete Vangl2, or occlude the PDZ to reduce its accessibility for Vangl2 interaction (Fig. 9B, left panel). Lastly, we note that Dact1 induced Dvl oligomerization also appears to promote Dvl phosphorylation (Fig. 3R and S), and we reported recently that Vangl2-bound Dvl2 tends to be the unphosphorylated form⁴⁹. Therefore, Dvl oligomerization induced by Dact1 may additionally result in post-translational changes such as phosphorylation to favor dissociation from Vangl2. These mechanisms are not mutually exclusive, and may act together to remove oligomerized Dvl from Vangl.

In contrast to Vangl2, Dact1-induced Dvl2 oligomers remain bound to Fz7 on the plasma membrane (Fig. 5). One explanation for this difference, as suggested by our data, relates to the fact that Dvl binds to Fz primarily via the C-terminal DEP domain, which is likely to be more accessible than the central PDZ domain in the oligomer form (Fig. 9B, right panel). The DEP domain is a small globular structure where three α-helices (H1-3) interact to form a hydrophobic core^62,72,73. An extended loop between helix1 (H1) and H2 gives rise to the so called “DEP finger” that interacts with the cytoplasmic domains of Fz. Mutations at the DEP finger, including L445 and K446, compromise Dvl binding to Fz (ref. ⁶², Fig. 7).

Importantly, DEP also undergoes dimerization, which is required for Dvl signalosome-like structure formation because DEP dimerization helps to stabilize the relatively weak head to tail oligomerization of DIX to form fibrils, and cross-links the fibrils into punctate assemblies⁶². DEP dimerization happens via an unusual process known as “domain swap”, where two DEP monomers exchange their H1 with each other to form two almost identical hydrophobic cores, and the “DEP finger” loops are extended to form two antiparallel β-strands to secure the dimer conformation. The E499G mutation known to disrupt DEP dimerization⁶² also abolished Dact1-induced Dvl puncta formation (Fig. 8), indicating that domain swap mediated-DEP dimerization is also required for Dact1-induced Dvl oligomerization (Fig. 9A). The action of Dact1, therefore, seems to be reinforcing both DIX-oligomerization and DEP-dimerization to help Dvl to form puncta, and it does so by binding to the PDZ of Dvl with the PDZ-b domain and homodimerizing via the LZ domain. Essentially, besides DIX and DEP, Dact1 enables PDZ to be used as a third domain for Dvl to oligomerize (Fig. 9A).

A natural consequence of DEP dimerization is losing the ability to bind Fz due to unfolding of the “DEP fingers” to accommodate H1 domain swap between two DEP monomers⁶². Therefore, one would predict that Dvl puncta, which form via DEP-dimerization, are prone to detaching from Fz. Indeed, in canonical Wnt signaling, Dvl binding to Fz is transient, and the rapid detachment of Dvl signalosomes from Fz is thought to ensure unidirectionality of Wnt signaling cascade^62,74. In contrast, PCP signaling in flies requires Dvl to be bound to Fz as “stable, membrane-localized signalosome-like complexes for many hours, even days”^69,74,75. Therefore, Dvl signalosome-like complex during PCP was speculated to have either attenuated level of DEP dimerization or use adaptors to accommodate persistent Fz binding⁷⁴. In our study, Dact1-induced Dvl puncta remain bound to Fz (Fig. 5), and only detach from Fz with mutation of K446 that normally mediates binding of monomeric DEP to Fz (Fig. 8). We therefore interpret the data to mean that in Dact1-induced Dvl puncta, whereas a portion of DEP domains have to dimerize for Dvl oligomerization, some DEP domains must remain as monomers to preserve Fz binding (Fig. 9B). Presumably, since Dact1 enables Dvl to utilize PDZ as an additional means to oligomerize, it lessens the threshold requirement for DEP dimerization such that Dact1-induced Dvl oligomers can tolerate a higher level of monomeric DEP to maintain strong binding with Fz. This may provide a key regulatory mechanism to maintain Dvl clustering on Fz in the non-canonical Wnt/ PCP pathway, in contrast to rapid dissociation of Dvl clusters from Fz during canonical Wnt signaling.

Lastly, although Dact1-induced Dvl oligomers can stay bound to Fz, they are insufficient to induce co-clustering of Fz (Fig. 5). This contrasts the effect of non-canonical Wnt, which induces Fz/Dvl co-clustering that manifests as membrane patches at cell-cell contact (Fig. 6^53,54;). These membrane patches were reported to increase cell contact persistence, and may therefore contribute to the dynamics of junction remodeling that underlies CE⁵³. Both Dact1 and the DIX domain of Dvl are required for the Dvl/Fz patch formation in response to Wnt11 ((Fig. 6; Supplementary Fig. 9; ref. ⁵⁴), indicating Dvl oligomerization as a necessary (but not sufficient) mechanism for Dvl/Fz co-clustering. On the other hand, Ror2 (Receptor Tyrosine Kinase Like Orphan Receptor 2), a co-receptor required for non-canonical Wnt signaling during CE⁷⁶, was also reported to be a component of the Fz/Dvl patches⁷⁷, and we recently found that Ror2 is indeed required for Wnt11 induced Fz/Dvl patch formation⁴⁹. Therefore, non-canonical Wnt induced coupling of Fz and Ror2 may initiate additional signaling events besides Dvl oligomerization to trigger co-clustering of Fz and Dvl. Uncovering these additional events in the future will provide a more thorough understanding of the regulation of non-canonical Wnt signaling during CE.

Another question remaining to be addressed in the future is how Dact1 induced Dvl oligomerization is regulated beyond the domain requirement defined in this study. Co-overexpression of Dvl2 and Dact1 is sufficient to cause autonomous puncta formation (Figs. 2–4; Supplementary Fig. 7), suggesting that the local concentration of Dvl and Dact1 could be one of the main determinants that impact oligomerization. Since Dvl and Dact1 can each be recruited to the plasma membrane by Vangl2 via the PDZ and SR domains, respectively (Fig. 2, Supplementary Figs. 4 and 12), we postulate that in the inactive state (without non-canonical Wnt), endogenous Dvl and Dact1 may each be sequestered by Vangl2 to prevent Dact1 from inducing Dvl oligomerization (Fig. 9B). Non-canonical Wnt is the most likely trigger to initiate Dact1-induced Dvl oligomerization, because it induced dissociation of Dvl from Vangl2⁴¹ and formation of Dvl/Fz patches that relies on Dvl oligomerization. The detailed mechanism, however, will need to be further elucidated in the future by tagging and tracking endogenous Dvl and Dact1 during non-canonical Wnt signaling activation.

In summary, we identified Dact1 as a potent crosslinker that enables Dvl to oligomerize using not only the DIX and DEP domains at the N and C terminus, respectively, but also the central PDZ domain. Dact1-induced Dvl oligomerization orchestrates a binding partner switch from Vangl to Fz and co-clustering of Dvl with Fz in the presence of Wnt ligands to efficiently activate non-canonical Wnt signaling during CE. We propose that Dact1 may have evolved as a vertebrate innovation to enhance Dvl oligomerization, thereby increasing the dynamics of core PCP protein cross-regulation to control rapid and polarized cell movement during CE.

Methods

Animal experiments were performed in agreement with the National Institutes of Health. Xenopus laevis adults were maintained according to the established protocols by the Institutional Animal Care and Use Committee at the University of Alabama at Birmingham, under Animal Project Number IACUC-22388. There is no evidence for sexual dimorphism in gastrulating Xenopus embryos, which were used as the model organism in the study, so sex was not considered as a biological variable in the study design.

Xenopus husbandry and embryo manipulations

Female frogs were induced to ovulate via injection of human chorionic gonadotropin (800 units, Sigma-Alrich) the night before use, and eggs were collected for in vitro fertilizations. Embryos were dejellied with 3% cysteine solution, and micro-injected with 10 nL of mRNAs or antisense MOs into either the animal regions of both blastomeres at the two-cell stage or the marginal zone regions of the two dorsal blastomeres at the four-cell stage (as indicated in the text), using a Warner Instruments PLI-100A injector. For functional studies, DMZ injected embryos were raised in 0.1x MMR (Marc’s modified Ringer’s), fixed in 4% paraformaldehyde (PFA), and staged according to Nieuwkoop & Faber⁷⁸. For length-to-width ratio (LWR) measurement, fixed embryos were first positioned with dorsal side facing up and imaged with a Leica DFC 490 camera mounted on a Leica M205 FCA stereomicroscope. A Fiji macro was utilized to process the images and obtain the projected area and the length of each embryo. Briefly, the macro first extracts individual embryos from the images, measures the area of each embryo, and generates the smallest rectangle that fully encloses each embryo. The length of the rectangle is considered the length of the embryo, while the width is calculated by dividing the embryo’s area by its length. The LWR is then calculated in Excel. The length of the embryos with significant curved shape was correct manually by drawing a line along the anteroposterior axis to measure the maximal distance from the head to the tail of each embryo using the Leica LAS software with Interactive Measurement module. The LWR for each group in a biological replicate was normalized to the mean value of the control group within the same batch before being combined for statistical analyses. For animal cap elongation assay, embryos were dissected to collect the animal caps and the explants were cultured in 0.5× MMR with or without 10 ug/ mL Activin A for ~6 hours at 29 °C to elongate. Explants were fixed in 4% PFA and imaged as above, and the length of their elongation was determined by drawing a line to measure the elongated portion of the explants.

Cell culture and transfection

SH-SY5Y cells were purchased from ATCC and cultured in DMEM/F-12 (1:1) media (glibco) supplemented with 10% fetal bovine serum (gibco) and 50 mg/mL penicillin/streptomycin (gibco) at 37 °C in a 5% CO2 atmosphere in a Thermo Forma incubator (Thermo Fisher, Waltham, MA). The cells were seeded at a density of 2 × 10⁶ cells per 100 mm plate and grown to approximately 80% confluency for all experiments. For transfection, the 4D-Nucleofector system (Lonza) was used following the provided instructions. Briefly, 3 × 10⁶ cells were mixed with 2 µg of either pCS105-GFP-hDACT or pCS105-mem-GFP plasmid in SF Cell Line 4D-Nucleofector solution, and electroporated using program CA-137. The mixture was then incubated at room temperature for 10 minutes before being transferred to a 60 mm plate and cultured for 24 hours before harvesting.

Morpholinos, plasmids and cloning

The XVangl2 (5’-CGTTGGCGGATTTGGGTCCCCCCGA-3’), dact1.L and dact1.S morpholinos were described previously^31,32,33,41, and ordered from Gene Tools. All constructs were made in the pCS105 vector except Dvl2-EGFP, which was made in pCS2 + . Correct sequences were confirmed via Sanger sequencing at the UAB Heflin Genomics Center or Eurofins. Dvl2-EGFP, Dvl2-mCherry, Flag-Dvl2, myc-Vangl2, myc-Vangl2C (cytoplasmic tail aa238-561) and miRFP670-Vangl2 were reported previously⁴¹. Dvl2-ΔDIX-EGFP mutant, which removes aa27-159 of mouse Dvl2, was made according to Rothbacher et al.⁴⁸ with PCR (template: Dvl2-EGFP; primers: Dvl2-D1 F 5’ -AGGCGAGACCGACCTAGG-3’ and Dvl2-D1 R 5’-AGGAGTCTCTTCTTCATCCAG-3’) and KLD reaction (NEB M0554). Dvl2-M2M4 mutant (Y27D/V67A/K68A) was made in two steps, first making Y27D with mutagenesis PCR using primers Y27D F 5’ -GAAGAGACTCCTGACCTGGTGAAGATCCCTG and Y27D R 5’ -TCTTCACCAGGTCAGGAGTCTCTTCTTCATCCAG; the Y27D mutants were then used as template to perform a second round PCR followed by KLD reaction (using primers V67A/K68A F 5’ -GCGGAAGAGATCTCCGATGA and V67A/K68A R 5’ -GGCCACCCCAAAATCCTGAT). Dvl2-ΔPDZ-EGFP mutant, which removes aa250-348 of mouse Dvl2, was made similarly as -ΔDIX but with primers ΔPDZ-KLD F 5’- CTCACCGTGGCCAAGTGTTGGG-3’ and ΔPDZ-KLD R 5’-GGTCCTCTCCATGCGTGGC-3’. Dvl2-E499G/C501R mutant was made according to⁵⁹ and⁵⁸ with site directed mutagenesis PCR using primers Dvl2EGCR F 5’-ACAAGATTACTTTCTCTGGGCAGCGCTATTATGTCTTCGGGGAC-3’ and Dvl2EGCR R 5’-CCCCGAAGACATAATAGCGCTGCCCAGAGAAAGTAATCTTGTTGACGG-3’. Dvl2-K446A mutant was made with site directed mutagenesis PCR using primers Dvl2KA F 5’-ACCGCATGTGGCTCGCGATCACCATCCCAAACGCCTTTCTAGGC-3’ and Dvl2KA R 5’-GTTTGGGATGGTGATCGCGAGCCACATGCGGTCCCGGACTTCG-3’. mScarlet-I-tagged human DACT1 (mSc-hDACT1) was made via Gibson Assembly (NEB E5510S) to fuse three PCR fragments together: pCS105 (template: pCS105; primers p105 F 5’-TCGAATTCGTCGACAGGCC-3’ and p105 R 5’-ATCGATGGGATCCTGCAAAA-3’), V5-G2SG2 linker-mScarlet-I (template: mScarlet-I⁷⁹; primers V5-Sca F1: 5’-TTGCAGGATCCCATCGATACCATGGGTAAGCCTATCCCTAACCCTCTCCTCGGTCTCGAT-3’, V5-Sca F2: 5’-AACCCTCTCCTCGGTCTCGATTCTACGGGTGGCTCTGGTGGCGTGAGCAAGGGCGAGGCAGTG-3’ and GFP-hDACT R 5’-ACCGCCAGAACCGCCCTTGTACAGCTCGTCCATGC-3’), and hDACT1 (template: hDACT1 cDNA; primers hDACT F 5’-GGCGGTTCTGGCGGTAAGCCGAGTCCGGCCGGG-3’ and hDACT R 5’-CCTGTCGACGAATTCGATCAAACCGTCGTCATCAGTTTC-3’). hDACT1ΔSR mutant, which removes aa 748-808 of human DACT1, was made according to³⁵ using mutagenesis PCR (F: 5’-GAGGACGAGCAGAGCTTTGTCAAAATTAAGGCCTCAC-3’ and R 5’-GGCCTTAATTTTGACAAAGCTCTGCTCGTCCTCACT-3’). The Xenopus Dact1 constructs, GFP-Dact1.L, GFP-Dact1.L-ΔLZ, and GFP-Dact1.L-ΔPDZ were reported previously³¹ and acquired from Addgene.

Imaging and analyses

For imaging, 0.03-0.6 ng of mRNA encoding mSc-hDACT1, Dvl2-EGFP, Dvl2-ΔDIX-EGFP, Dvl2-ΔPDZ-EGFP, Dvl2-EG/CA-EGFP, Dvl2-KA-EGFP or miRFP670-Vangl2 were injected in various combination into the animal regions at the two-cell stage and dissected at ~St.9. Alternatively, 0.5 ng of mRNA encoding GFP-Dact1.L, GFP-Dact1.L-ΔLZ or GFP-Dact1.L-ΔPDZ was injected with 1 ng Dvl2-mCh mRNA and 0.6 ng miRFP670-Vangl2. To examine Dvl2 and Fz7 localization in activin-treated animal caps, 0.03 ng of Dvl2-EGFP mRNA, 0.02 ng of Fz7-EGFP mRNA, and 50 ng of xVangl2 MO were injected into the animal regions at the two-cell stage. Animal cap explants were dissected at ~ St. 9 and cultured in 0.5x MMR with 10 μg/ mL Activin A until imaging. For Wnt11-induced patch formation analyses, 0.5 ng Wnt11 mRNA, 0.3 ng Frizzled7-flag or Fz7-EGFP mRNA, 0.3 ng mem-miRFP670 mRNA, 0.3 ng Dvl2-EGFP mRNA, 0.6 ng mSc-hDACT1 mRNA or 20 ng dact1.S/L MO (10 ng each) were injected in various combinations. Dissected animal caps were imaged on an Olympus FV1000 with a 20x water immersion objective or Zeiss LSM 900 with a 20x air objective.

Images were imported into ImageJ (NIH). Co-localization with Pearson’s coefficient analyses were done using the ImageJ plug-in, JACoP⁸⁰. The colocalization data was compiled into graphs using GraphPad Prism 8. Wnt11-induced Dvl2 patch length and number per cell border were analyzed in ImageJ using the ‘Otsu’ threshold filter to isolate and identify the patches and cell borders. Length was measured by drawing a line along each patch or cell border. Length of each patch was normalized to the average cell border length prior to graphing in a violin plot. Number of patches per cell border was calculated by counting the patches on each cell border and calculating the average number of patches normalized to the total number of cell borders within the image. If no patches were present on a cell border, patches/border was listed as zero. The Mean to Sigma Ratio was calculated by dividing the average signal intensity within the patches by the variance of the membrane signal intensity outside the patches. Images from at least three different embryos collected on different days were analyzed per injection group.

Co-immunoprecipitation and western blots

For in vitro Co-IP proteins were synthesized with the TnT® SP6 High-Yield Wheat Germ Protein Expression System (Promega). 6 μg Myc-Vangl2C, 5 μg mSc-hDACT1, 3 μg Flag-Dvl2, or 3 μg Flag-Dvl2-ΔDIX plasmid DNAs were added to 30 μL of wheat germ extract mix and MilliQ water for a total reaction volume of 50 μL. Synthesis reactions were run at 25 °C for 2 hours. Co-IPs were performed by mixing 6 μL of myc-Vangl2C with 2 μL of Flag-Dvl2 or Flag-Dvl2-ΔDIX in TBST (TBS (150 mM NaCl, 50 mM Tris, PH 7.5) with 0.1% Tween-20) buffer (400 μL total volume) for 1 hours at room temperature (RT), followed by addition of 5 μl of mSc-hDACT1 or blank wheat germ extract control and incubation for another hour. 15 μL of pre-conjugated anti-Myc magnetic beads were washed 3 times with TBST at 4 °C before adding to the protein mixtures for immunoprecipitation at 4 °C for 3 hours. Afterwards, the anti-Myc conjugated beads were washed 3 times with 5x TBS (750 mM NaCl, 250 mM Tris, PH 7.5) + 0.25% Tween-20 at 4 °C. Beads were resuspended in 1x LDS sample buffer (ThermoFisher NP0007) with 5% β-mercaptoethanol and boiled for 10 minutes at 70 °C. For the Co-IP experiment conducted with the SH-SY5Y cell line, 3 × 10⁶ cells were lysed with 400 µL of RIPA buffer on ice for 20 minutes. The lysate was then centrifuged at 16,000 g at 4 °C to remove cell debris, and the supernatant was recovered. The lysate was pre-cleaned with 25 µL of pre-washed protein G beads (Pierce™ Protein G Magnetic Beads) at 4 °C for 1 hour. After pre-cleaning, the supernatant was recovered, and 4 µg of either anti-Vangl2 antibody(36E3)⁴⁶ or anti-GFP antibody (GFP Antibody (B-2): sc-9996, SANTA CRUZ) was added. The mixture was incubated at 4 °C overnight. For immunoprecipitation, 25 µL of pre-washed protein G beads was added to the lysate and incubated at 4 °C for 3 hours. The protein G beads were then washed three times with RIPA buffer at 4 °C, resuspended in 1x LDS sample buffer with 5% β-mercaptoethanol, and boiled for 10 minutes at 70 °C. Samples were run on 8% SDS-PAGE and transferred to PVDF membranes for western blots. Membranes were blocked for 30 minutes at RT in TBS Protein-free Blocking Buffer (Licor). For western blots, primary antibodies used were Rabbit anti-Myc (1:1000; CST Cat.# 2278), Rabbit anti-Dvl2 (1:1000; CST Cat.# 3224), Rat anti-Vangl2 (36E3 or 2G4; 0.5 ug/ ml, ref. ⁴⁶) and Rabbit anti-V5 (1:1000; CST Cat.# 13202) diluted in TBS with 0.1% Tween-20. Secondary antibodies were Licor anti-Rabbit 680 (Cat.# 926-68071) and 800 (Cat.# 926-32211) diluted 1:10,000 in TBS with 0.1% Tween-20. Stained membranes were imaged on Licor M scanner. We followed the protocol by Burckhardt et al.⁸¹ and used FIJI to quantify the relative amount of Dvl2 co-IP with Vangl2. Briefly, the intensity value from the co-IP of Dvl2 was divided by that from the IP of Vangl2 (immunoprecipitated by either anti-Vangl2 or anti-Myc antibody) under each control or experimental condition; the mean ratio from the controls was then used to normalize the ratio under each experimental condition to determine the “Relative Co-IP Dvl2 amount”.

Statistics

Statistics were performed in GraphPad Prism 8. Datasets were determined to be distributed normally via histogram, Shapiro-Wilk, and Kolmogorov-Smirnov tests. The parametric, two-tailed, unpaired T-test was used to compare the LWR among different groups for CE phenotype studies and Pearson’s coefficient colocalization studies. Linear regression was used to show a correlative relationship between increased overexpression of hDACT1 and increasingly severe CE defects. Two-tailed paired T-test was used to compare groups in co-IP studies. Wnt11-induced Dvl2 patch size and number per cell border were compared between injection groups using an unpaired, two-tailed T-test. For all statistical analyses, p values < 0.05 were deemed significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.