Endogenous oligomer formation underlies DVL2 condensates and promotes Wnt/ β -catenin signaling

Activation of the Wnt/ β -catenin pa thway crucial ly depends on polymerization of dishevelled 2 (DVL2) into biomolecular condensates. However, given the low affinity of known DVL2 self-interaction sites and its low cellular concentration it is unclear how polymers can form . H ere, we detect oligomeric DVL2 complexes at endogenous protein levels, using a biochemical ultracen trifugation assay. We identify a low - complexity region (LCR4) in the C - terminus whose deletion and fusion decreased and increased the complexes , respectively. Notably, LCR4 - induced complexes correlated with the formation of microscopically visible multimeric condensates. Adjacent to LCR4, we mapped a conserved domain (CD2) promoting condensates only. Molecularly, LCR4 and CD2 mediated DVL2 self-interaction via aggregating residues and phenylalanine stickers, respectively. Point mutations inactivating these interaction sites impaired Wnt pathway activation by DVL2 . Our study discovers DVL2 complexes with functional importanc e for Wnt/ β - catenin signaling. Moreover, we provide evidence that DVL2 condensates form in two steps by pre - oligomerization via high - affinity interaction sites, such as LCR4, and subsequent condensation via low - affinity interaction s ites, such as CD2.

Activation of Wnt/β-catenin signaling by DVL2 crucially depends on DVL2 polymerization via its Nterminal DIX domain (Kishida et al., 1999;Schwarz-Romond et al., 2007a).However, the low autoaffinity of the DIX domain in the mid-micromolar range and the low cellular concentration of DVL2 strongly disfavor polymerization, and only the pre-clustering of DVL2 at Wnt-receptor-complexes is suggested to overcome this problem (Bienz, 2014).Mechanistically, DVL2 polymerization may support DVL2 clustering, AXIN1 recruitment and/or signalosome formation (Bilic et al., 2007;Schwarz-Romond et al., 2007a;Schwarz-Romond et al., 2007b).When DVL2 is overexpressed, DIX-mediated polymerization gives rise to microscopically visible DVL2 assemblies, which are sphere-like, membranefree and highly dynamic (Schwarz-Romond et al., 2005).Recently, these DVL2 assemblies were characterized as phase-separated biomolecular condensates (Kang et al., 2022), which represent a functionally diverse class of membrane-free cell organelles with common biophysical properties (Banani et al., 2017;Shin and Brangwynne, 2017).In addition to the DIX domain, other parts of the protein contribute to DVL2 condensation and activity, such as the DEP domain or an intrinsically disordered region (Gammons et al., 2016;Kang et al., 2022;Vamadevan et al., 2022), suggesting that major regions of DVL2 are evolutionary optimized for condensation.Moreover, condensation of DVL2 appears to be a regulated process controlled through posttranslational modification, such as ubiquitination, and depending on its conformation, open versus close form (Lee et al., 2015;Vamadevan et al., 2022).However, there is still a controversial debate on whether DVL2 forms condensates at endogenous expression levels, as recent studies report only small DVL2 assemblies with less than 10 molecules (Kan et al., 2020), or only one big DVL2 condensate at the centrosome (Schubert et al., 2022), or about 100 condensates per cell with sizes of 0.2 to 0.5 µm (Kang et al., 2022).
In vertebrates, three DVL paralogs exist, DVL1, DVL2 and DVL3.Although overexpression of each paralog activates Wnt/β-catenin signaling, loss-of-function studies consistently report that DVL1, DVL2 and DVL3 exhibit different capabilities to transduce Wnt signals and that overexpression of one paralog does not compensate for loss of another (Lee et al., 2008;Paclikova et al., 2021).Thus, these studies point to non-redundant molecular functions of the DVL paralogs, yet, their molecular differences remain poorly understood.
Here, we report biochemical evidence for endogenous DVL2 complexes consisting of at least eight molecules, supporting the idea of DVL2 polymerization at endogenous expression levels.Using DVL2 deletion and point mutants, we mapped and characterized a low complexity region in the DVL2 Cterminus that promoted complex formation through mediating intermolecular DVL2 self-interaction.
Our data suggest that these complexes most likely represent underlying substructures of DVL2 biomolecular condensates, which precede and initiate condensation.Moreover, the discovered oligomeric DVL2 complexes were of functional importance because point mutations that impaired complex formation attenuated Wnt pathway activation by DVL2.
Interestingly, the DVL2 complexes appeared to be paralog-specific because complexes were almost absent for DVL1 and DVL3 (Fig. 1C, D).The persistence of the complexes at low protein concentrations in cellular extracts indicated that they form via interaction sites with rather high affinity.Although the DIX domain is the best-characterized polymerization domain in DVL2 (Bienz, 2014), its low auto-affinity suggested that it is not involved in the formation of these complexes (Schwarz-Romond et al., 2007a).

A low complexity region in the C-terminus promotes DVL2 complexes
Deletion of the DEP domain (construct ∆DEP) decreased DVL2 complexes, in line with its reported function in DVL2-DVL2 interaction (Gammons et al., 2016).However, additional deletion of the remaining C-terminus (construct 1-418) markedly reduced them further, demonstrating a strong contribution of the deleted residues 521-736 to complex formation (Fig. 2A-C).Importantly, decreased protein complexation of DVL2 ∆DEP and 1-418 in ultracentrifugation experiments conspicuously correlated with decreased formation of condensates in immunofluorescence-based assays (Fig. 2D, E) and decreased Wnt pathway activation in reporter assays (Fig. 2F).Therefore, we hypothesized that the DVL2 complexes may be important for signaling activity.To identify candidate regions for a more precise mapping within residues 521-736, we used the SEG algorithm predicting low-complexity regions (Wootton and Federhen, 1993), the TANGO algorithm predicting aggregation (Fernandez-Escamilla et al., 2004) and protein alignments (Sievers et al., 2011).We identified four low-complexity regions (LCR1-4), which are associated with protein assembly (Martin et al., 2020;Martin and Mittag, 2018), one potential aggregation site embedded in LCR4, and two domains whose evolutionary conservation may point to functional importance (CD1-2, Fig. S1D).Since deletion of residues 521-736 showed strong effects when combined with the DEP deletion (1-418 vs ∆DEP, Fig. 2), we performed the following mapping in the ∆DEP context.Given the good correlation between complexes and condensates (Fig. 2C, E), we decided to use immunofluorescence-based analysis of condensates for mapping, as it is more convenient than density gradient ultracentrifugation.Upon individual deletion of the six identified regions LCR1-4 and CD1-2, only deletion of LCR4 and CD2 decreased condensate formation and Wnt pathway activation of DVL2 ∆DEP (Figs.3A and S2A-C).Combined deletion of LCR4 and CD2 increased the effect (Figs. 3A and S2A-C).We, therefore, consider the two adjacent regions as one functional unit, hereafter referred to as condensate forming region (CFR).DVL2 ∆DEP-∆CFR exhibited a marked decrease in the number of cells with condensates (Fig. 3A-C), in the number of condensates per cell (Figs.3D and S2D-G) and in Wnt pathway activation (Fig. 3E), similar to 1-418.
Interestingly, the investigated DVL2 mutant proteins predominantly formed nuclear condensates in contrast to the cytosolic condensates of WT DVL2, most likely, because a nuclear export signal (Fig. 2A) was deleted in these mutants (Fig. 3A).However, investigating only cells with cytosolic condensates (Fig. S3E, F) revealed similar differences between the DVL2 mutants as were observed when investigating mainly cells with nuclear condensates (Fig. 3C and S3B), suggesting that the detected differences are not due to nuclear localization but reflect the overall condensation capacity of the DVL2 mutants.Moreover, fusion of CFR to the isolated DVL2 DIX or AXIN1 DAX domain sufficed to trigger condensation, which was prevented by M2/M3 mutation of the DIX/DAX domain (Fig. S3D, G, H).Thus, loss-and gain-of-function experiments identified CFR as the crucial region for condensation and Wnt pathway activation within the DVL2 C-Terminus, which functionally cooperates with the DIX domain.
Next, we investigated whether CFR contributes to the formation of DVL2 complexes detected by density gradient ultracentrifugation.Indeed, CFR deletion from ∆DEP (∆DEP-∆CFR) and fusion with 1-418 (1-418+CFR) decreased and increased complexes, respectively (Fig. 4A-D).Moreover, replacing the respective sequence in DVL1 with the CFR of DVL2 (DVL1-CFR DVL2 ) promoted complex formation (Fig. 4E, F).Surprisingly, only LCR4 but not CD2 was required for complex formation when deleted from ∆DEP or mediated complex formation when fused to 1-418 (Fig. 4A-D), although both parts were required for condensate formation (Fig. S2B).Of note, LCR4 is not well conserved in DVL1 and DVL3, which do not form complexes (Figs.S1D and 1C).Together, our data revealed a bipartite, 58 amino acid region at the very C-Terminus of DVL2 consisting of a low complexity region (LCR4) that promotes complexes and condensates and a conserved domain (CD2) that only promotes condensates and is dispensable for complexes.

LCR4 and CD2 cooperatively mediate DVL2 self-interaction
Given the role of the CFR parts LCR4 and CD2 in complex and condensate formation of DVL2, we speculated that they might directly mediate DVL2-DVL2 interaction.To analyze the role of a putative aggregation site (aggregon) that was predicted by the TANGO algorithm within LCR4 (Fig. S1D), we designed mutations of two key valine residues (VV-AA) predicted to prevent aggregation (Figs.5A and   S5A).LCR4 VV-AA mutation of DVL2 ∆DEP decreased condensate formation as efficiently as LCR4 deletion (Fig. 5B, C), and fusion of VV-AA-mutated LCR4 failed to increase condensate formation of 1-418, in contrast to WT LCR4 (Fig. S5B, C).In CD2, the TANGO algorithm did not predict aggregation sites.However, we detected interspersed phenylalanine residues in CD2 (Fig. 5A), which might promote interaction through stacking of their aromatic rings, acting as "stickers" of unstructured regions, as recently described (Martin et al., 2020).CD2 FF-AA mutation decreased condensate formation and Wnt pathway activation as efficiently as CD2 deletion (Figs.5B, C and S5D), and CD2 FF-AA fusion failed to increase condensate formation or Wnt pathway activation, in contrast to WT CD2 (Fig. S5E-G).Our data suggest that CFR is a bipartite protein interaction site for self-association of DVL2.
In line with this, the isolated CFR, which exhibited a homogeneous cellular distribution when expressed alone, accumulated within DVL2 condensates upon co-expression of both proteins (Fig. 5D, E), suggesting CFR-DVL2 interaction.CFR VV-AA FF-AA mutation markedly reduced this CFR-DVL2 colocalization (Fig. 5D, E).Moreover, DVL2 CFR did not co-localize with DVL1 or DVL3 condensates (Fig. S5H), consistent with the low conservation of the LCR4 aggregon in these paralogs (Fig. S1D).Notably, co-expression of the isolated CFR inhibited Wnt pathway activation by DVL2 in a dosage-dependent manner, which was attenuated by CFR FF-AA mutation (Fig. 5 F).In this experiment, free CFR might weaken DVL2-DVL2 interaction by saturating the interaction surface of the CFR within DVL2.Together, our data provide evidence that CFR mediates intermolecular DVL2 self-interaction via an aggregation site in LCR4 and phenylalanine stickers in CD2.Since individual deletion of LCR4 or CD2 was sufficient to impair condensates (Fig. S2B), both interaction sites most likely cooperate to drive condensate formation of DVL2.

DVL2 CFR promotes phase separation
While 1-418 and DVL2 DIX exhibited a homogenous cellular distribution, fusion of CFR induced spherical, microscopically visible condensates of 1-418+CFR and DIX+CFR (Figs. 3B and S3G).To study the nature of CFR-mediated condensates, we treated cells expressing DVL2, 1-418+CFR or DIX+CFR with a hypoosmolar buffer (osmotic shock) or with the bivalent alcohol 1,6-hexandiol, as previously done in biomolecular condensate research to challenge phase separation (Nott et al., 2015;Ribbeck and Gorlich, 2002).Both treatments significantly decreased condensates of all three studied proteins within 3 min for osmotic shock and 1 h for 1,6-hexandiol (Fig. 6A-C, and movies M1 and M2).These findings are consistent with the transition from a two-phase state of condensates with high protein concentration and surrounding spaces with low protein concentration to a one-phase state of homogenous protein distribution (Fig. 6A).We concluded that CFR indeed induced phase separation to promote 1-418+CFR and DIX+CFR condensates, in line with the fact that WT DVL2 was shown to undergo phase separation (Kang et al., 2022),.Moreover, fusion of an AXIN1-derived, sequencewisenonrelated condensate-forming region (CFR AX , see Fig. S4 for details) to DVL2 ∆DEP-∆CFR (∆DEP-∆CFR+CFR AX ) restored condensate formation to the level of ∆DEP (Fig. 6D-F), indicating that CFR AX compensates for loss of CFR DVL2 .More importantly, CFR AX fusion (∆DEP-∆CFR+CFR AX ) also rescued the decreased Wnt pathway activation of ∆DEP-∆CFR compared to ∆DEP (Fig. 6G), suggesting that it is indeed the CFR phase-separating activity that is crucial for signaling.

DVL2 CFR contributes to Wnt pathway activation
In order to determine the impact of CFR within full-size DVL2, we mutated the four crucial residues identified in CFR in DLV2 (DVL2 VV-AA FF-AA).This shifted the DVL2 complexes in ultracentrifugation assays from large to smaller sizes, similar to deletion of CFR (Fig. 7A, B), and reduced the formation of condensates by about 50% (Fig. 7C, D).Importantly, DVL2 VV-AA FF-AA exhibited more than 50% reduced Wnt pathway activation compared to WT (Fig. 7E), with similar expression of both constructs (Fig. S5I).The DVL2 variants were transiently expressed on top of endogenous DVL1/2/3 in this experiment.In addition, we used DVL1/2/3 knockout cells, as they represent an elegant system to study Wnt pathway activation upon re-expression of DVL2 variants without any interference of endogenous WT DVL (Paclikova et al., 2017).In these cells, overexpression of DVL2 VV-AA FF-AA almost completely failed to activate the pathway and was as inactive as the DIX domain M2 mutant (Schwarz-Romond et al., 2007a), which can be considered as the gold standard for DVL2 inhibiting point mutations (Fig. 7F).In addition, we used DVL1/2/3 knockout cells with additional knockout of RNF43 and ZNRF3 (DVL tKO+), which allowed higher pathway activation upon DVL2 overexpression (Fig. S5J), as the DVL2-activating receptors were no longer degraded through RNF43 and ZNRF3 (Hao et al., 2012;Paclikova et al., 2021).Also in these cells, DVL2 VV-AA FF-AA exhibited markedly impaired pathway activation as compared to WT (Fig. S5J).Finally, we re-expressed DVL2 variants at close to endogenous levels in DVL1/2/3 knockout cells to rescue Wnt-induced pathway activation, which was disrupted through DVL knockout (Figs.7G and S5K).While re-expression of WT DVL2 resulted in a complete rescue as compared to WT cells, DVL2 VV-AA FF-AA was significantly impaired and as inactive as DVL2 M2 in this assay (Fig. 7G).The VV-AA FF-AA mutation inhibited complexation, condensation and Wnt pathway activation as efficiently as CFR deletion (Fig. 7A-F), strongly indicating that it is specifically the interaction activity of CFR through the aggregon and the phenylalanine stickers that is required for signaling.A comparison between the VV-AA FF-AA mutation and the established M2 mutation showed on average about 65% and 80% reduction of Wnt pathway activation as compared to WT, respectively (Figs. 7E-G and S5K), suggesting that DVL2 CFR markedly contributes to Wnt pathway activation.Consistently, we observed strong significant correlations of CFR-mediated condensation and complexation with Wnt pathway activation for the DVL2 deletion constructs used in our study (Fig. 7H).

DISCUSSION
Here, we provided strong biochemical evidence that endogenous DVL2 forms oligomeric complexes (Fig. 1A, B), supporting the idea that DVL2 assemblies exist at physiologic protein levels.Although we investigated several scaffold proteins with various endogenous interactors in this assay, we only observe such complexes for DVL2 and AXIN2 and they were specifically associated with aggregating protein sequences (Bernkopf et al., 2019;Miete et al., 2022), indicating that most types of proteinprotein interactions are not preserved in this assay.Furthermore, overexpressed DVL2 did not exhibit reduced complex sizes, as one would have expected, if limited endogenous interactors had been part of the complexes.Therefore, we think that the detected complexes most likely reflect homotypic DVL2 assemblies, which would then be about eight molecules in size (Fig. 1A, B).The size of the complexes was reminiscent of previously described endogenous DVL2 oligomers identified via TIRF imaging (Kan et al., 2020).Through deletion analysis, we discovered a 14 aa long, low complexity region (LCR4) in the DVL2 C-terminus mediating complex formation (Fig. 4A-D) and condensate formation (Fig. S2B).
An adjacent 38 aa long, evolutionary conserved domain (CD2) was required for the formation of DVL2 condensates (Fig. S2B), and we conceptually combined LCR4 and CD2 as condensate forming region (CFR).Molecularly, an aggregon in LCR4 and phenylalanine residues in CD2 mediated DVL2 assembly (Figs.7A-D and S5A-G).The latter may promote protein interaction via sticking of their aromatic rings, as previously described for phase-separating proteins (Martin et al., 2020).Co-localization of the isolated CFR with DVL2 indicated that CFR may mediate DVL2-DVL2 interaction, and this was prevented by specific point mutations targeting the aggregon in LCR4 and the phenylalanine stickers in CD2 (Fig. 5D, E).Treatments that challenge phase separation diffused CFR-induced condensates (Fig. 6A-C), in line with a recent report showing that DVL2 condensates form via phase separation (Kang et al., 2022).
Importantly, point mutations that inhibit CFR self-interaction markedly attenuated Wnt pathway activation by DVL2 (Fig. 7E-G and S5K).Especially in DVL1/2/3 triple knockout cells, the DVL2 CFR point mutant VV-AA FF-AA was as signaling deficient as the DIX domain M2 mutant (Fig. 7F, G), which is frequently used for DVL2 inactivation (Schwarz-Romond et al., 2007a).In these cells, mere DVL2 VV-AA FF-AA overexpression almost completely failed to activate the pathway (Fig. 7F) and its reexpression at close to endogenous levels only poorly rescued activation of the pathway through Wnt ligands, which was disrupted by the DVL knockout (Figs.7G and S5K).We thus identified a novel DVL2 region that promotes complex formation, condensate formation, and Wnt pathway activation.
Our study provides deeper insights into the assembly of multimeric DVL2 condensates.The complexes detected by ultracentrifugation showed sizes of about eight DVL2 molecules, which were irrespective of expression levels as they were similar between endogenous and overexpressed DVL2 (Fig. 1A-C).
Therefore, the complexes cannot be identical with the known microscopically visible condensates, containing thousands of molecules.These oligomeric complexes thus existed in parallel to the condensates or constituted a condensate substructure.Analysis of mutant DVL2 proteins revealed that DVL2 contains domains that promote both complexes and condensates, such as LCR4 and DEP (Figs. 2B-E, 4A-D and S2B), and domains that promote only condensates but not complexes, such as CD2 and DIX (Figs. 1C, 4A-D, 7C and S2B).However, all DVL2 mutations that reduced complexes also affected condensates.Therefore, we suggest that the oligomeric complexes are required for and possibly represent substructures of the multimeric condensates (Fig. 8).As the stability of complexes at low protein concentrations, e.g. in cellular extracts, indicated a rather high affinity of the underlying interaction sites (Fig. 1A), we hypothesize that complex formation via LCR4 and DEP precedes further assembly of these substructures into condensates via CD2 and DIX (Fig. 8).Our proposed two-step model of DVL2 condensate formation is in line with and supportive of the emerging stickers-model for the formation of biomolecular condensates (Choi et al., 2020).According to this model, oligomerization via one interaction site can drive subsequent condensate formation by increasing the valence of another interaction site (=sticker) of the oligomer compared to a monomer (Choi et al., 2020).The proposed two-step model would increase the options for regulating polymerization of DVL2 by modulating the oligomerization step or the condensate formation step (Fig. 8).DVL2 polymerization is crucial for Wnt signal transduction.However, the low DVL2 concentration and the low DIX-DIX affinity strongly disfavor polymerization, and clustering of DVL2 at activated Wnt-receptor-complexes is suggested to overcome this limitation (Bienz, 2014).Pre-oligomerization of DVL2 via high-affinity interactions as identified in the ultracentrifugation assay might facilitate Wnt-induced polymerization.
Furthermore, the discovered C-terminal self-interaction site CFR may contribute to the regulation of DVL2 condensates through conformational changes.Binding of the very C-terminus of DVL2 to its PDZ domain results in a closed protein conformation (Lee et al., 2015).It has been suggested that this closed conformation limits the accessibility of the N-terminal intrinsically disordered region that promotes condensate formation, thereby suppressing DVL2 condensates (Kang et al., 2022;Vamadevan et al., 2022).It is very intriguing to speculate that condensate formation will be additionally suppressed in the closed conformation by limiting the accessibility of CFR because only one amino acid separates the identified crucial phenylalanine residues of CFR from the PDZ binding motif in the DVL2 C-terminus.Notably, Wnt ligands induce the open DVL conformation (Harnos et al., 2019), which will increase the CFR accessibility and, according to our model, would allow LCR4 and CD2 to contribute to pre-oligomerization and subsequent condensate formation of DVL2, respectively.
Several studies suggest functional differences between the DVL paralogs (Gentzel et al., 2015;Lee et al., 2008;Paclikova et al., 2021), while the underlying molecular differences remain unclear.Notably, complex formation as revealed by ultracentrifugation was only observed for DVL2 and not for DVL1 or DVL3, revealing a marked difference between the paralogs (Fig. 1C).In addition, CFR did not co-localize with DVL1 or DVL3 in contrast to DVL2, indicating that CFR-mediated DVL2-DVL2 interaction, which most likely drove DVL2 complexation, is absent from DVL1 and DVL3 (Figs. 5B and S5H).Consistently, the crucial CFR aggregon located in LCR4 was not conserved in DVL1 or DVL3 (Fig. S1D).Moreover, replacing the complementary part of DVL1 with the DVL2 CFR promoted complex formation (Fig. 3H,   I).Although DVL1 and DVL3 lack the discovered CFR, all three DVL paralogs are able to form condensates (Fig. 2D and S5H) and to activate Wnt signaling to some extent (Lee et al., 2008;Paclikova et al., 2021), which can be potentially explained by the other interaction sites (DIX, DEP, intrinsically disordered region).However, quantitative studies suggest functional differences between the paralogs and that they have to cooperate at a certain molar ratio for optimal Wnt pathway activation, with DVL2 being the most abundant (Lee et al., 2008;Paclikova et al., 2021).In this context, the DVL2 condensates with their underlying stable complexes may function as a kind of superscaffold for integration of DVL1 and DVL3.Our findings may help to understand the functional differences between the paralogs in the future.

MATERIALS AND METHODS
Cell culture, transfections and treatments HEK293T, HeLa and U2OS cells were grown in low glucose DMEM supplemented with 10% fetal calf serum and antibiotics at 37°C in a 10% CO2 atmosphere, and passaged according to ATCC recommendations.T-REx DVL1/2/3 triple knockout cells (DVL tKO) and T-REx DVL1/2/3, RNF43 and ZNRF3 penta knockout cells (DVL tKO+) were generated in the Bryja lab and have been previously described (Paclikova et al., 2017;Paclikova et al., 2021).They were grown in high glucose DMEM supplemented with GlutaMAX, 10% fetal calf serum and antibiotics at 37°C in a 10% CO2 atmosphere.
Transfections of plasmids and siRNA were performed with polyethylenimine and Oligofectamine (Invitrogen) according to manufacturer's recommendations, respectively.Wnt3a medium was prepared as originally described (Willert et al., 2003).For the osmotic shock treatment, cell culture medium was diluted 1:1 with sterile water to reduce the osmolarity by 50%.1,6-Hexandiol (240117) was obtained from Sigma-Aldrich.

Sucrose gradient ultracentrifugation
Cells were lysed about 24 h after transfection, when required, or 48 h after seeding in a Triton X-100based buffer (150 mM NaCl, 20 mM Tris-HCl pH 7.5, 5 mM EDTA, 1% Triton X-100, Roche protease inhibitor cocktail).A linear sucrose density gradient was prepared in 13x51 mm centrifuge tubes (Beckman Coulter) by overlaying 2 ml of a 50% (v/w) sucrose solution with 2 ml of a 12.5% (v/w) sucrose solution followed by horizontal incubation of the tube for 3 h at RT (Stone, 1974), bevor loading a 200 µl cell lysate sample on top.After centrifugation in a Beckman Coulter Optima MAX Ultracentrifuge (217100 g, 25°C, 18 h), 20 fractions à 200 µl were collected from top to bottom and analyzed by Western blotting, as indicated.The commercially available size markers thyroglobulin (669 kDa, T9145) and albumin (66 kDa, A8531) were obtained from Sigma Aldrich.In case of thyroglobulin and albumin, fractions were analyzed by silver staining of the proteins in polyacrylamide gels.

Western Blot
Proteins in cell lysates or in fractions of sucrose gradients were denatured, separated by gel electrophoresis in polyacrylamide gels under denaturing conditions (SDS-PAGE), and transferred onto a nitrocellulose membrane (VWR).The proteins were detected using suitable combinations of primary and HRP-conjugated secondary antibodies (see above) via light emission upon HRP-catalyzed oxidation of luminol in a LAS-3000 with Image Reader software (FUJIFILM).Intensities of protein bands were quantified with AIDA 2D densitometry.

Immunofluorescence
Cells were fixed in a 3% paraformaldehyde solution, permeabilized with 0.5% Triton X-100 and blocked with cell culture medium to reduce unspecific antibody binding, before proteins of interest were stained with suitable combinations of primary and fluorochrome-conjugated secondary antibodies (see above).Analysis and image acquisition was performed at an Axioplan II microscope system (Carl Zeiss) using a Plan-NEOFLUAR 100x/1.30NA oil objective and a SPOT RT Monochrome camera (Diagnostic Instruments).Cells were categorized in a blinded fashion as "cell with condensates" when exhibiting more than three distinct sphere-like structures, to reduce the number of false positives.The Spot Detector tool of the Icy open source bio-imaging software (Institut Pasteur, version 2.2.1.0)was used to objectively quantify numbers of condensates per cell (Olivo-Marin, 2002).

Live-cell imaging
For live-cell imaging of the osmotic shock treatment, the culture medium of cells expressing indicated GFP-tagged proteins was replaced with a 50% hypoosmolar phosphate buffered saline solution.Images were acquired at constant exposure times every 15 s over the next 3 min at an Axiovert25 microscope system (Carl Zeiss) using a LD A-Plan 40x/0.50Ph2 objective and a SPOT Insight QE camera with the SPOT Basic software (Diagnostic Instruments, version 4.0.1).Movies were rendered using Photoshop 19.1.5(Adobe).

Luciferase reporter assay
Cells were transfected with a luciferase reporter plasmid either with a β-catenin-dependent promoter (TOP, Tcf optimal) or with a β-catenin-independent control promoter (FOP, far from optimal), a constantly active β-galactosidase expression plasmid and expression plasmids for the proteins of interest, as indicated in the figures.After lysis (25 mM Tris-HCl pH 8, 2 mM EDTA, 5% glycerol, 1% Triton X-100, 20 mM DTT), the luciferase activity was measured via light emission upon luciferin decarboxylation in a Centro LB 960 Microplate Luminometer (Berthold technologies) and the βgalactosidase activity was assessed as release of yellow ortho-nitrophenol upon ortho-nitrophenyl-βgalactoside hydrolysis using a Spectra MAX 190 (Molecular Devices).The luciferase activities were first normalized to the respective β-galactosidase activities to correct for minor variations in transfection efficiency, before the TOP values were normalized to the respective FOP values to correct for unspecific β-catenin-independent changes.TOP/FOP reporter assays were performed in technical duplicates.

Statistical analysis
Data sets were probed for statistical significance using two-tailed Student's t-tests in a non-paired (Figs.3D and S2E-G) or paired (all others) fashion, depending on the experimental setup.Statistical significance is indicated by asterisks in the figures (* p<0.05, ** p<0.01, *** p<0.001), when required, and n values of biological replicates are stated in the figure legends for all experiments.We assumed normal distribution of the data based on the nature of the assays and graphical assessment, which, however, was not formally tested owing to the small sample sizes.P-values for the correlation analyses in Fig. 7H were calculated by the test statistic t=R*square root((n-2)/(1-R 2 )), with n-2 degrees of freedom.and the p value, respectively.Note that condensation correlates with whether CFR is intact (LCR4 and CD2 intact, green), partially intact (either LCR4 or CD2 intact, grey) or lost (neither LCR4 nor CD2 intact, red), and that the presence of LCR4 determines complexation.The plots summarize data that were shown before within this study.and of sticker domains mediating condensate formation, such as CD2 and the DIX domain (magenta circles).Oligomerization into complexes (1) (light green background, right) increases the valence of stickers in the complexes as compared to the monomers, which allows to overcome the low affinity of the isolated stickers and drives subsequent formation of condensates (2) (purple, bottom) by the multivalent stickers, according to the emerging stickers-model (Choi et al., 2020).Fraction numbers are indicated above the blots according to Fig. 1E.C Western blotting for indicated endogenous proteins in lysates of HEK293T cells, which were transfected with a control or a DVL2targeting siRNA for 48 h.A-C Two different antibodies were used for DVL2 detection (3224 and 3216, CellSignaling).D Clustal Omega alignment of indicated dishevelled protein sequences, with specific UniProt identifiers provided in brackets.Identity (*) and conservation between amino acid groups of strongly (:) and weakly (.) similar properties are indicated (Sievers et al., 2011).Marks show conserved domains (CD1 and CD2), low-complexity regions (LCR1, LCR2, LCR3 and LCR4) according to the SEG algorithm (Wootton and Federhen, 1993) and the aggregation propensity of individual residues of mouse DVL2 (yellow-orange heat map) according to the TANGO algorithm (Fernandez-Escamilla et al., 2004).LCR4 and CD2 are highlighted in green and magenta, respectively, as they were identified as crucial regions for condensate induction later on.Fig. S4 Identifying a phase-separating region in AXIN1.A Phase separating (CFR) propensity score (yaxis) of amino acid residues of AXIN1 (x-axis) according to the catGRANULE algorithm, which predicts liquid-liquid phase separation propensity (Bolognesi et al., 2016).Red arrow points to a potential phase separating region in AXIN1 (CFR AX ).B Left column: To scale schemes of artificial fusion proteins between the polymerizing AXIN1 DAX domain (grey) with CFR AX (different shades of blue).CFR AX was fused up to three times to the DAX domain (3xCFR AX -DAX).M3 indicates an inhibiting point mutation of DAX-mediated polymerization, which is illustrated by the red X (Fiedler et al., 2011).Middle column:

Fig. 2
Fig. 2 The DVL2 C-terminus promotes complexes, condensates and activity.A To scale schemes of

Fig. 3 A
Fig. 3 A 58 aa C-terminal region promotes DVL2 condensates and activity.A To scale schemes of DVL2

Fig. 4
Fig. 4 LCR4 mediates complex formation.A, C, E Western blotting for indicated transiently expressed

Fig. 6
Fig. 6 CFR-induced condensates form via phase separation.A, E Immunofluorescence of indicated HA-

Fig. 7
Fig. 7 DVL2 CFR is crucial for Wnt signaling activity.A Western blotting for indicated proteins, which

Fig. 8
Fig. 8 Two-step model of DVL2 condensate formation.Schematic illustration of DVL2 domains

Fig.
Fig. S1 Endogenous DVL2 forms complexes.A, B Western blotting for indicated endogenous proteins