Structure of the Wnt-Frizzled-LRP6 initiation complex reveals the basis for co-receptor discrimination

Wnt morphogens are critical for embryonic development and tissue regeneration. Canonical Wnts form ternary receptor complexes composed of tissue-specific Frizzled receptors together with the shared LRP5/6 co-receptors to initiate β-catenin signaling. The structure of a ternary complex of an affinity-matured XWnt8-Frizzled8-LRP6 complex elucidates the basis of co-receptor discrimination by canonical Wnts by means of their N-termini and linker domains that engage the LRP6 E1E2 domain funnels. Chimeric Wnts bearing modular linker ‘grafts’ were able to transfer LRP6 domain specificity between different Wnts and enable non-canonical Wnt5a to signal through the canonical pathway. Synthetic peptides comprising the linker domain serve as Wnt-specific antagonists. The structure of the ternary complex provides a topological blueprint for the orientation and proximity of Frizzled and LRP6 within the Wnt cell surface signalosome.


Introduction
Wnt morphogens play essential roles in embryonic development and tissue regeneration by orchestrating stem cell proliferation and differentiation (Clevers and Nusse, 2012;MacDonald et al., 2009;Nusse and Clevers, 2017;Rim et al., 2022). To initiate signaling, Wnt simultaneously engages the extracellular cysteine-rich domain (CRD) of a seven-transmembrane receptor Frizzled Due to its crucial function in cell fate decisions, aberrant Wnt activation is linked to various diseases, including multiple types of cancers, making it an attractive therapeutic target by Wnt signaling antagonists (Anastas and Moon, 2013). On the other hand, Wnt agonism can be 15 harnessed for regenerative medicine upon tissue injury or in bone and hair losses (Clevers et al., 2014). Engineered bispecific dimerizers that induce Fzd-LRP6 proximity can act as surrogate agonists in order to therapeutically exploit Wnt's powerful functions in stem cell expansion and tissue renewal (Chen et al., 2020;Janda et al., 2017;Miao et al., 2020;Tao et al., 2019). These molecules, as well as Norrin (Chang et al., 2015;Xu et al., 2004), demonstrate the induced 20 proximity between Fzd and LRP5/6 is the principal mechanism of signal activation. However, the signaling strength of such agonists appears highly dependent on the relative orientation and proximity of the Fzd-LRP6 heterodimer, and structure-activity relationships are not fully understood due to the absence of the natural Wnt-Fzd-LRP6 ternary complex.
The previous Wnt-Fzd crystal structures revealed the binding mode of canonical Wnts,25 Xenopus Wnt8 (XWnt8) and Wnt3a to their primary receptor Fzd, in which the hand-shaped Wnt pinches the Fzd CRD (FzdCRD) with "thumb and index" fingers (Hirai et al., 2019;Janda et al., 2012). However, these structures fell short of recapitulating the fully competent cell surface signaling complex because they lack LRP5/6 that interact with Wnt through their ECD consisting of four YWTD β-propellers flanked by EGF-like domains. Furthermore, it has been shown that 30 Wnts can be clustered into different families based on their binding preferences to distinct LRP6 domains (Bourhis et al., 2010;Gong et al., 2010). It remains to be determined how different Wnts, which are highly conserved can bind to different regions in LRP6 in contrast to the fact that Wnts bind to the FzdCRD in a very conserved manner. Furthermore, it is also unclear how non-canonical Wnts have structurally evolved to bypass LRP5/6 co-receptor requirements for their function. For these reasons, structural information on a Wnt-Fzd-LRP6 ternary signaling complex is important for understanding the geometric requirements (i.e., orientation and proximity) of the receptors for 5 signaling initiation, how Wnts differentially bind to LRP6, and finally, for rationally optimizing the surrogate Wnts to achieve the natural topology of the endogenous Wnt signaling complex. 10 To reconstitute a stable Wnt-Fzd-LRP6 complex for cryo-electron microscopy (cryo-EM) analysis, we first sought to engineer a Wnt that binds with increased affinity to LRP6. However, since Wnt proteins are lipidated and insoluble in the absence of detergents and also require the dedicated cargo receptor Wntless (WLS) for secretion (Bänziger et al., 2006;Nile and Hannoush, 2016;Nygaard et al., 2021;Zhong et al., 2021), they are problematic for conventional cell surface  Figure 1A bottom panel). The cell-surface XWnt8-hFzd5CRD complex robustly bound human LRP6 E1E2 (hLRP6E1E2), the Wnt8 binding module (Gong et al., 2010;Ren et al., 2021), which was tetramerized to increase the affinity through avidity effect for detecting by flow cytometry ( Figure 1B). Deleting the NC-linker of XWnt8 25 (XWnt8ΔNC), which has previously been reported to mediate LRP6 binding in the context of Wnt3 signaling (Chu et al., 2013;Hirai et al., 2019), substantially decreased, but did not completely abolish, hLRP6E1E2 binding ( Figure 1B). These indicated that the wild-type XWnt8-Fzd5CRD complex displayed on the cell surface is functional regarding LRP6 recognition, and that the NClinker is a primary LRP6 binding site, but additional interactions beyond the NC-linker may 30 contribute to LRP6 binding. We then created 'soft-randomized' libraries of XWnt8 variants that harbored mutations in the NC-linker (residues 222-234) with an experimental diversity of approximately 1 million unique sequences (Figure 1-figure supplement 1A). The libraries were designed to introduce sparse mutations within the NC-linker that would exhibit enhanced binding energetics to LRP6 while maintaining the overall structural integrity of the linker.

Wnt engineering
After three rounds of fluorescence-activated cell sorting (FACS) selection for hLRP6E1E2 binding ( Figure 1A and Figure 1-figure supplement 1B), we isolated five XWnt8 clones with 5 improved hLRP6E1E2 tetramer binding relative to wild-type XWnt8 (Figure 1-figure supplement   2A). After screening for recombinant expression, we tested the XWnt8 variant with the highest expression (high-affinity XWnt8, haXWnt8) for hLRP6E1E2 monomer binding on cells, which had acquired six amino acid substitutions ( Figure 1D). The haXWnt8 variant had at least 6-fold improved binding affinity (EC50) relative to wild-type XWnt8 ( Figure 1D Figure 1E). haXWnt8 showed more potent β-catenin signaling than the wild-type XWnt8 both in 15 the absence and presence of R-spondin 2. In addition, we noticed that the surrogate Wnt showed a substantially better EC50 than the Wnts due to its high affinity but with a lower maximum effect (Emax), which is likely due to a different, less optimal Fzd-LRP6 heterodimer receptor geometry than induced by Wnt. 20

Reconstitution of the soluble Wnt ternary complex and structural analysis
To prepare the soluble ternary complex, we co-cultured S2 cells expressing haWnt8-mFzd8CRD and hLRP6E1E2, and the complex was purified by a series of affinity and size-exclusion chromatography steps. The peak fractions corresponding to the 1:1:1 haXWnt8-mFzd8CRD-hLRP6E1E2 complex were subjected to cryo-EM analysis ( Figure 2C). hLRP6E1E2 appears to "cap" XWnt8 at the top of the structure by engaging two sites on XWnt8 distal from the Fzd 30 binding site. Site A is formed by the XWnt8 N-terminal loop, and Site B is formed by the NClinker, which insert into the E1 and E2 funnels in the center of the hLRP6E1E2 β-propellers, respectively ( Figure 2D). Both binding motifs in XWnt8 were not resolved in the previously reported XWnt8-mFzd8CRD crystal structure (Janda et al., 2012), suggesting they are disordered in the absence of LRP6.
The presumed flexibility of the unusual interaction mode seen, as compared to a more typical rigid protein-protein interface that buries large amounts of surface area, has limited the 5 quality of our cryo-EM map, precluding placement of amino acid side chains for the XWnt8 Nterminus and NC-linker. While the N-terminal interaction is particularly tenuous, the NC-linker mainchain clearly shows that the central region of the loop inserts into the funnel. It seems likely that even when in complex, hinge flapping at the XWnt8-hLRP6E1E2 junction is an inherent feature of this interaction.

10
The observed Site A interaction mediated by the XWnt8 N terminus was surprising, so we asked if it was merely an adventitious, but energetically inert, interaction by virtue of its proximity to the NC-linker interaction with E2, or if it contributed to the overall binding free energy in forming the complex. There is precedent for Wnt N-termini being important for function in that Tiki family metalloproteases are Wnt inactivating enzymes that cleave N-terminal loops of Wnts 15 (Zhang et al., 2016(Zhang et al., , 2012. As a matter of fact, Tiki inactivates both Wnt3a and XWnt8. Using our mammalian Wnt display system, we find that deleting the hydrophobic N-terminal loop (XWnt8ΔN) reduced LRP6 binding similar to XWnt8ΔNC ( Figure 2E). Since the effect was less pronounced than deleting the NC-linker, the NC-linker clearly plays the major role in driving the Wnt-LRP6 association, with the N-terminal loop being supportive, consistent with the less defined, 20 discontinuous N-terminal cryo-EM density observed. Unlike most Wnts, Wnt8s have shorter Nterminal sequences preceding the N-domain, lacking a subdomain containing a disulfide bond (Hirai et al., 2019;Nygaard et al., 2021;Zhong et al., 2021). However, with a few exceptions, Wnts have comparably long disordered loops with hydrophobic residues, reinforcing the shared Wnt-LRP6 binding mechanism in canonical β-catenin signaling.

Generality of NC-linker to mammalian Wnts
We wished to determine the generality and modularity of the NC-linker of XWnt8 and mammalian Wnts. The NC-linker is one of the least conserved amino acid stretches across different Wnt proteins, but it is highly conserved across species for a particular Wnt (Figure 3- figure   30 supplement 1). Using human Wnt1 as a model system, we performed alanine scanning mutagenesis on the NC-linker analogous to that of XWnt8 to assess its functional role in mediating Wnt/β-catenin signaling. We transfected HEK293 TOPbrite luciferase reporter cells (Nile et al., 2018) with plasmids encoding Wnt1 mutants with pairs of alanine mutations. Despite comparable expression, most Wnt1 variants were less active in inducing β-catenin signaling compared to wildtype Wnt1, with mutants 65, 70, 73, and 76 having the most impaired signaling activities ( Figure   3A and Figure 3-figure supplement 2). Thus, the NC-linker of human Wnt1 also appears to be a 5 critical mediator of Wnt signaling activity, and this likely extends to other Wnt family members and species.

The NC-linker controls Wnt-LRP6 domain specificity
To investigate the modularity of the NC-linker in mediating LRP6 interactions by different Wnts, 10 we designed three chimeric Wnt constructs in which the NC-linker was swapped between different human Wnt proteins ( Figure 3B). We chose Wnt1, Wnt3a, and Wnt5a as templates for NC-linker swapping because they represent canonical (Wnt1 and Wnt3a) and non-canonical (Wnt5a) Wnts and interact with LRP6E1E2 (Wnt1) and LRP6E3E4 (Wnt3a) (Bourhis et al., 2010;Gong et al., 2010).
Of note, since no structures of Wnt1 and Wnt5a were available, the designs of the chimeras, 15 particularly the NC-linker boundaries, were not informed by structure and are therefore approximations. We expect that these structural imprecisions of the NC-linker grafts in the chimeric molecules will quantitatively influence their activities, and also that some loops will resist grafting due to structural differences between the NC-linker attachment sites to the Wnt globular core. With these caveats in mind we undertook a range of grafts and assessed their activities. As 20 expected, the canonical Wnt proteins Wnt1 and Wnt3a induced β-catenin signaling in TOPbrite assays, whereas the non-canonical Wnt5a did not induce reporter expression above background (Grumolato et al., 2010) ( Figure 3B). Grafting the NC-linker from the E1E2-binding Wnt1 into the E3E4-binding Wnt3a (termed Wnt3a_1) did not markedly alter the activity of wild-type Wnt3a, suggesting that we successfully transferred LRP6E1E2 E1E2 specificity from Wnt 1 to Wnt3a.

25
However, the reverse engineered Wnt1_3a chimera was inactive. This could reflect the fact that the NC-linker was disordered in the XWnt8-mFzd8CRD crystal structure, whereas the Wnt3a NClinker was ordered into a β-hairpin. As a result, grafting the Wnt3a NC-linker may have less tolerance for different structural contexts. Consistent with this, grafting the Wnt1 NC-linker into Wnt5a rendered the Wnt5a_1 chimera able to induce β-catenin signaling ( Figure 3B), functionally 30 turning a classical non-canonical Wnt into a canonical Wnt. Consistently, Wnt1, Wnt3a, Wnt3a_1, and Wnt5a_1 but not Wnt5a and Wnt1_3a induced LRP6 phosphorylation, the accumulation of active (non-phosphorylated) β-catenin ( Figure 3C) and Axin2 transcription ( Figure 3D) (Bourhis et al., 2011;Gong et al., 2010) ( Figure 3E). As expected, YW210 inhibited Wnt1-induced signaling and YW211 inhibited Wnt3a-induced signaling in luciferase assays, consistent with the reported LRP6 domain specificity ( Figure 3F and Figure 3- figure   10 supplement 3). However, the activities of the Wnt3a_1 and Wnt5a_1 chimeras were inhibited by YW210, and not by YW211, demonstrating that the Wnt1 NC-linker directs the binding of these chimeras to LRP6E1E2, mimicking Wnt1 signaling characteristics. In summary, these results suggest a critical role of the Wnt NC-linkers in determining canonical and non-canonical Wnt signaling activity and LRP6 domain specificity, and that the signaling characteristics of Wnts can 15 be re-engineered by transferring the NC-linker. While not all grafts were successful, we think the modularity of the NC-linker demonstrated for Wnt1 is likely representative for most Wnts.

Synthetic peptides mimicking NC-linker inhibit Wnt activity
To further support the role of the NC-linker as a critical determinant of LRP6 domain specificity, 20 we investigated the binding specificities of the isolated NC-linkers and their ability to interfere with Wnt activity. Superposition of the three available wild-type Wnt structures (Chu et al., 2013;Hirai et al., 2019;Janda et al., 2012) revealed two loosely structurally conserved 'interface motifs (Bazan et al., 2012;Hirai et al., 2019) Table 2). This is consistent with previously reported Wnt- weaker interaction between W3a.cys and LRP6E1E2 can be attributed to "sticky" non-specific binding as it could not be outcompeted by anti-LRP6E1 YW210 ( Figure 4C). Despite the inherent flexibility of isolated peptides, W1.cys, W2b.cys and W7a.cys, but not W3a.cys, inhibited Wnt1mediated β-catenin signaling in luciferase assays without any observed toxicity on cells ( Figure   5 4D and Figure 4-figure supplement 2B), presumably by competing with Wnt1 for LRP6E1E2 binding. The lack of activity of the Wnt3a peptide is consistent with its binding ability to LRP6E3E4.
Taken together, the overall consistency in LRP6 domain specificity of isolated NC-linker peptides and full-length Wnt proteins, as well as the ability of the peptides to interfere with Wnt activity, further supports the critical role of the NC-linker in determining LRP6 domain specificity.

10
To further map the binding region of the NC-linker peptides on LRP6, we performed a competition ELISA with YW210, YW211, and H07, a single-domain antibody fragment (VHH) that binds to the LRP6 E3 domain (Fenderico et al., 2019). YW210 competed strongly with W1.cys and W2b.cys, and partially with W7a.cys for LRP6E1E2 binding ( Figure 4C). Given the relatively large size of Fabs, it likely sterically masked neighboring binding epitopes on LRP6E2.

15
Furthermore, YW211 and H07 VHH both competed with W3a.cys for binding to LRP6E3 ( Figure   4E), suggesting that the Wnt3a NC-linker binds to the E3 domain of LRP6.
hLRP6E1E2 and hLRP6E3E4 are relatively conserved regarding their funnel-like cavity at the center of the β-propellers, hydrophobicity and electrostatic property of β-propellers, and the ~40 Å distance between the proximal funnels ( Figure 5-figure supplement 1), consistent with a shared 20 two-site Wnt binding mechanism by both LRP6E1E2 and LRP6E3E4. Our H07 VHH blocking data shows that E3 domain blockade prevents Wnt3a binding, localizing the NC-linker to the LRP6 E3 domain, while the NC-linker of XWnt8 and Wnt1 binds to the LRP6 E2 domain. This structurefunction data now allows us to orient how E1E2-binding versus E3E4-binding Wnts engage LRP6 through their NC-linkers and N termini. We propose that, like XWnt8, E1E2-binding Wnts use 25 their N termini to bind E1 and NC-linker to bind E2, whereas E3E4-binding Wnts, represented by Wnt3, use their N termini to bind E4 and their NC-linker to bind to E3. Thus E1E2-binding versus E3E4-binding Wnts appear to bind to LRP6 domains in 'reverse' orientations to accommodate the NC-linker specificity ( Figure 5A). Whether all Wnts use their N termini to bind to the neighboring propeller funnel and orient themselves is not addressed by our data, but the perfect match of 40 Å 30 distance between the N-termini and NC-linker on Wnts with the 40 Å distance between funnels on neighboring propellers suggests this is likely.

Discussion
The structure of the ternary haXWnt8-mFzd8CRD-hLRP6E1E2 complex resolves the basis for coreceptor binding by Wnts, revealing a modular determinant that not only controls LRP5/6 domain specificity, but also appears to differentiate the non-canonical Wnt5a and explain its co-receptor 5 independence. The signaling complex structure further elucidates the mechanism of LRP5 gainof-functions mutations (including D111Y, R154M, G171V/R, N198S, A214V/T and T253Iwhich correspond to D98, R141, G158, N185, A201, and T240 in LRP6) that are associated with high bone mass (HBM) disease and the significant increase in bone strength and thickness in affected patients (Babij et al., 2003;Gregson and Duncan, 2020). These HBM mutations are 10 located on the top surface of the E1 β-propeller of LRP5, and overlap with the binding epitopes of Wnt antagonists sclerostin (SOST) (Winkler et al., 2003) or Dickkopf (DKK) (Glinka et al., 1998;Krupnik et al., 1999) ( Figure 5B). Consistent with the notion that the Wnt NC-linker-LRP6 E2 interaction is the main driver for binding, HBM mutations impaired the binding of SOST and DKK1, but with a minimal effect on Wnt9b in functional assays, resulting in the selective loss in 15 affinity for Wnt signaling antagonists (Ai et al., 2005;Bourhis et al., 2011;Semenov and He, 2006).
The ECD of LRP5/6 consists of four YWTD β-propellers flanked by EGF-like domains, E1-E4 ( Figure 5B left panel), followed by three low-density lipoprotein receptor type A domains, and exists in a range of bend conformations (Matoba et al., 2017). The ECD interacts with Wnt 20 proteins and antagonists, including SOST and DKK, to regulate signaling. While SOST antagonizes LRP6E1E2-binding Wnts including Wnt1, Wnt2, and Wnt9b, but not LRP6E3E4-binding Wnts including Wnt3a (Kim et al., 2020), DKK1 inhibits both the Wnt groups (Bourhis et al., 2010;Sima et al., 2016). Our cryo-EM structure of the ternary haXWnt8-mFzd8CRD-hLRP6E1E2 complex now provides insights into the precise mechanisms of differential Wnt inhibition by 25 SOST and DKK.
SOST is a secreted glycoprotein predominantly expressed by osteocytes and inhibits bone formation and regeneration by antagonizing Wnt/β-catenin signaling (Winkler et al., 2003). SOST engages the top surfaces of the E1 and E2 β-propellers with its loop 2 and the C-terminal tail (Bourhis et al., 2011;Kim et al., 2020), respectively, reminiscent of the tandem binding mode of terminus might dislodged from the Wnt core upon binding to FzdCRD, which may result in being more accessible by LRP6. However, this allosteric "priming" of Wnt by FzdCRD binding is only supported by indirect structural evidence.
The ternary XWnt8-mFzd8CRD-hLRP6E1E2 complex structure, together with the structure of the Fzd transmembrane regions (Tsutsumi et al., 2020;Yang et al., 2018), and previous mechanistic insights into heterodimeric Wnt activation (Miao et al., 2020;Tsutsumi et al., 2020) 5 allow us to propose a structural model of the transmembrane Wnt signaling complex for both LRP6 E1E2 and E3E4 binding Wnts ( Figure 5C). Given the different topologies of these signaling complexes it is tempting to speculate that this could affect signaling strength. We have recently shown that bispecific Wnt "surrogates" composed of a FzdCRD binding domain fused to DKK1C can activate β-catenin signaling by simply heterodimerizing Fzd and LRP6. However, the 10 surrogates with different binding domains and linkers show great variations in signaling strength, suggesting that the geometry (i.e., orientation and proximity) of the two receptor chains within the signaling complex is a critical determinant of signaling characteristics. Furthermore, the geometry of the signaling complex induced by the Wnt surrogates and natural Wnt ligands differs greatly, possibly yielding the differential Emax. Thus, the haWnt8-mFzd8CRD-hLRP6E1E2 structure provides

Mammalian cell line for Wnt display
To display functional Wnts on cells, we engineered HEK293F cells (Thermo Fisher) to saturate the cell surface with hFzd5CRD (residues 27-155) in a manner analogous to scFv displayed on mammalian cells (Chesnut et al., 1996). The N-terminal HA-tagged hFzd5CRD was fused to the

Wnt display
To test Wnt display on the cell surface, Xenopus laevis Wnt8 (XWnt8) was expressed in the parental HEK293F cells and the hFzd5CRD-1TM cells before FACS sorting. Full-length XWnt8 25 was cloned into pcDNA3.1+ (Thermo Fisher) vector with endogenous signal peptide and Cterminal Protein C epitope. The cells were transiently transfected with the pcDNA3.1+ XWnt8 vector and stained by Alexa Fluor 647 conjugated Protein C antibody (PrC647; the antibody produced in-house from HPC-4 hybridoma, ATCC HB-9892, and labeled with Alexa Fluor 647 To confirm that the lipidated Wnt stays on the cell which produced the molecule, the sorted hFzd5CRD-1TM cell line was transfected either with the XWnt8 or green fluorescent protein (GFP)coding mammalian expression vector, and 24 hrs after transfection, the Wnt and GFP induced cells were mixed at 1:1 ratio and further shaken at 37 o C for 24 hrs. Cell-surface Wnt was labeled with PrC647, and Wnt/GFP expression was checked by the flow cytometry. Wnt transfer to the GFP 5 positive cells was estimated to be ~5%.
To prepare stable cell lines expressing XWnt8 or hWnt8 with or without the deletion of the NC-linkers (residues 222-234 for XWnt8, and 221-233 for hWnt8, that were replaced by GSGS), and N-terminal loops (residues 23-32 for XWnt8), the constructs were cloned into the pLV-EF1a-IRES-Puro vector (Addgene 85132) with N-terminal HA signal peptide and C-terminal FLAG tag.  FreeStyle 293 media, and once the total cell number reached >10 million for both library, 5 million cells each of FLAG-FITC+ library 1 and 2 were mixed to make the naïve XWnt8 library.

XWnt8 library selection and the NC-linker chimera of hWnt8a
The naïve XWnt8 library was stained with 100 nM hLRP6E1E2-SA647 tetramer and 1:50 FLAG- The engineered NC-linkers from the XWnt8 library were grafted to hWnt8 to test if they 30 improve the hLRP6E1E2 binding. The wild-type pLV-EF1a-IRES-Puro hWnt8 plasmid was modified to have the engineered NC-linkers from haXWnt8, haXWnt8c, and haXWnt8d synthesized by IDT, making hahWnt8, hahWnt8c, and hahWnt8d, respectively. The hahWnt8sexpressing cell lines were prepared and hLRP6E1E2 tetramer staining was confirmed as described above in comparison with the wild-type hWnt8, showing superior binding ( Figure 1D and Figure   1-figure supplement 2).
The hLRP6E1E2 monomer binding was confirmed by incubating the hFzd5CRD-1TM cells 5 that display Wnt8 variants with biotinylated hLRP6E1E2 at 30, 100, 300, 1000, 3000 nM concentrations, followed by washes with PFE and labeling with 50 nM SA647 (binding) and 1:50 FLAG-FITC (Wnt expression). The cells were again washed with PFE twice and analyzed by the flow cytometer. Two independent experiments were performed in triplicate, and the representative experiment was shown with Prism 9 (GraphPad).

Preparation of recombinant XWnt8 and haXWnt8s
The five haXWnt8 variants (residues 22-338) were cloned into the pActinCD8 vector as previously described (Janda et al., 2012). Drosophila S2 cells were maintained in Schneider's Drosophila The cells were co-transfected with wild-type or variants XWnt8, the pActinCD8 mFzd8CRD-Fc vector (Janda et al., 2012), and pCoBlast using the calcium phosphate precipitation kit (Invitrogen) according to the manufacturer's protocol. Each cell line was bulk selected for three weeks with a gradual increase of Blasticidin (Invitrogen) concentrations; 5 μg/ml, 10 μg/ml, 20 μg/ml, for the first, second, and the third week, respectively. Expression levels were checked by western blotting and 500 mM NaCl, while the mFzd8CRD-Fc remained bound to the resin. Wnts were further purified by Con A agarose (Tsutsumi et al., 2020).

TOPflash signaling assay
To confirm the affinity enhancing mutations do not interfere with XWnt8 activity, SuperTopFlash 5 HEK293 (293STF) cells were stimulated with a gradient concentration of haXWnt8, wild-type XWnt8, and next-generation surrogate (NGS) Wnt in the presence or absence of R-spondin 2 (Janda et al., 2017;Miao et al., 2020). The agonist concentrations were 0.39 nM to 50 nM for XWnt8s and 0.016 nM to 2 nM for NGS Wnt, respectively, with two-fold serial dilution. 293STF were seeded into 96-well plates and maintained in DMEM media supplemented with 10% (v/v) 10 FBS one day before Wnt stimulation. Cells were stimulated on the next day and lysed after 24 hrs following manufacturer's protocol in Dual Luciferase Assay kit (Promega). Luminance signal readings were performed using or SpectraMax i3x (Molecular Devices). Two independent experiments were performed in triplicate, and the representative experiment was plotted and analyzed by Prism 9.

Preparation of the ligand-receptor complex for structural study
The S2 cell line expressing hLRP6E1E2 was prepared as described above with the pActinCD8 and pCoBlast vectors. The S2 haXWnt8-mFz8CRD-Fc cells and hLRP6E1E2 cells were mixed at the cell ratio of 1:3, and co-cultured for 5 days. The complex was captured by rProtein A Sepharose, and 20 eluted by adding in-house 3C protease, while the cleaved Fc remained bound to the resin and hLRP6E1E2 was affinity-purified by haXWnt8 bound to mFzd8CRD. The elution was concentrated down to 500 uL and injected to a Superdex 200 10/300 GL column equilibrated with HBS containing 2 mM CaCl2. The peak fraction with a retention volume corresponding to the 1:1:1 complex was concentrated to 0.5 mg/ml for cryo-EM analysis.

25
Cryo-EM specimen preparation and data collection 3 µL sample was applied onto glow-discharged 300 mesh gold grids (UltrAuFoil R1.2/1.3). Excess sample was blotted to a filter paper for 3 sec with botting force of 3, before plunge-freezing with a Vitrobot Mark IV (Thermo Fisher Scientific) at 4°C and 100% humidity. The cryo-EM movies 30 were collected using a Titan Krios (Thermo Fisher Scientific) operated at 300 kV equipped with Gatan K3 camera in counting mode. Nominal magnification was set to 29,000x, corresponding to a pixel size of 0.8521 Å and a calibrated magnification of 58,680x. Movies were recorded using SerialEM (Mastronarde, 2005) for 2.5 sec with 0.05 sec exposure per frame at an exposure rate of ~16 electrons/pixel/sec at the specimen, and the nominal defocus range between -1.0 and -2.0 μm on gold support. A beam-image shift was used with an active calibration to collect 9 movies from 9 holes per stage shift and autofocus.  Tan et al., 2017). As expected from the 2D class averages with preferred particle orientation ( Figure 2B and Figure 2-figure supplement 1C), the map has a large variation in the directional resolutions, as well as highly diverged local resolutions, with the LRP6E1E2 density better defined (Figure 2-figure supplement 1D-F). The 3D map also indicates exposure to the air-water interface around XWnt8's lipid (Figure 2-figure supplement 1E).

Model building and refinement
The crystal structures of XWnt8-mFzd8CRD (PDB 4F0A) and hLRP6E1E2 (PDB 3S94) were manually placed into the cryo-EM map using UCSF ChimeraX (Goddard et al., 2018). To facilitate map interpretation beyond the docking model, the missing N-terminal loop and NC-linker were modeled in the cryo-EM density using Coot (Emsley and Cowtan, 2004). Clearly distorted regions from the initial model, represented by the Wnt index finger, were flexibly fitted. We note that, based on the docked initial structures, the regions with weak cryo-EM density such as the Wnt thumb were kept in the final model but with lower occupancy (0.2). To avoid overinterpretation, all sidechains were truncated to Cβ except for prolines, cysteines and the palmitoleate (PAM)modified serine. The PAM acyl torsion angles were also manually adjusted from the initial model.

Peptide synthesis
Peptide synthesis was outsourced to ChemParter (Shanghai, China); all the linker peptides were 20 synthesized using standard 9-fluorenylmethoxycarbonyl protocol as described earlier (Stanger et al., 2012;Zhang et al., 2009). They were purified by reverse-phase HPLC. Peptide quality (>90% purity) was verified by liquid chromatography coupled to mass spectroscopy (LC-MS).

25
Human LRP6 extracellular fragments, hLRP6E1E2 (residues 20-631) and hLRP6E3E4 (residues 631-1253), carrying a C-terminal hexahistidine-tag were secreted from Trichoplusia ni cells. Secreted proteins were isolated with an affinity column of immobilized anti-His-tag mAbs and eluted with buffer containing 50 mM sodium acetate pH 4.8 and 300 mM NaCl. The eluted proteins are further purified by SEC on a Superdex S200 column. The SEC column was equilibrated with buffer 30 containing 10 mM sodium citrate pH 5.6 and 300 mM NaCl for hLRP6E1E2, and 10 mM sodium cacodylate pH 6.5 and 300 mM NaCl for hLRP6E3E4. Monomeric fractions were pooled, concentrated to ~10 mg/ml, flash-frozen, and stored at -80°C until further usage.
LRP6-binding Fabs YW210 and YW211 were expressed as follows; 10 ml of inoculation culture was grown from E. coli 67A6 transformed with a Fab-encoding plasmid in Luria Broth (LB) media with 50 μg/ml carbenicillin overnight at 30°C. The inoculation culture was added to 5 500 ml of Soy CRAP media with 50 μg/ml carbenicillin and incubated overnight at 30°C with shaking. Expression cultures were spun down at 10,000 g for 10 min. Cell pellets were resuspended in buffer containing 200 ml PBS, 25 mM EDTA and Complete Protease Inhibitor Cocktail tablets (Roche, 1 tablet per 50 ml buffer). The mixture was homogenized and then passed twice through a microfluidizer. The suspension was centrifuged at 20,000 g for 45 min. The protein was loaded  (Table 2) was used. A serial dilution of competitors was added, and the bound His-tagged LRP6 domains were detected as described.

TOPbrite dual-luciferase Wnt reporter assays
HEK293 cells with stably integrated firefly-luciferase-based Wnt reporter (TOPbrite) (Zhang et al., 2009) and pRL-SV40 Renilla luciferase (Promega) were maintained in a 5% CO2 humidified incubator at 37°C in DMEM with nutrient mixture F12 (50:50), 10% (v/v) FBS, 2 mM GlutaMAX 5 (Gibco) and 40 μg/ml hygromycin (Cellgro). Cells were grown for at least 24 hrs before any experiments. 20,000-40,000 cells/well in 50 μl medium were seeded in each well of clear-bottom white polystyrene 96-well plates (Falcon) and incubated for 24 hrs. Cells were then transfected with 0-25 ng of Wnt or Wnt-chimera-expressing constructs mixed with FuGENE HD in 10 μl OptiMEM, followed by incubation for 24-48 hrs. Treatment with peptides or LRP6 Fabs was done 10 for 6 h before the assay measurement. Readout was obtained with 50 μl of Dual-Glo Luciferase Assay system (Promega) according to the manufacturer's instructions on a Perkin Elmer EnVision multilabel reader. The ratios of firefly luminescence to Renilla luminescence were calculated, and in some instances normalized to control (non-treated) samples. Cell lines were tested for mycoplasma contamination and authenticated by single-nucleotide polymorphism (SNP) analysis.

Western blot analysis
For Wnt expression assay, 4.0 x 10 6 HEK293 luciferase reporter cells were seeded onto 10-cm dishes and transfected on the following day with 4 μg DNA encoding Wnt1, Wnt3a, Wnt5a, Wnt1_3a, Wnt3a_1 and Wnt5a_1 using FuGENE HD. Empty vector (pcDNA3.1) was used as 20 control. After 48 hrs of incubation, culture medium was collected and cells were gently washed twice with 5 ml cold PBS and lysed in 0.5 ml of lysis buffer (PBS, 1% (v/v) Triton X-100 and protease inhibitor cocktail (Roche)). Collected lysate was centrifuged for 5 min at 18,400 g to remove cell debris, and supernatant was used to determine total protein concentration by BCA assay (Thermo Fisher Scientific). 10 μg of total protein was loaded onto an SDS-PAGE (4-12%) 25 to assess Wnt protein levels. To evaluate secreted Wnt protein levels, the collected Wntconditioned medium (6 ml) was incubated with 50 μl Blue Sepharose beads (GE) on a rocking platform overnight at 4°C. The beads were washed three times with 500 μl PBS containing 0.1% (v/v) Tween 20 by centrifugation at 2,400 g for 5 min. 50 μl of 1x XT sample buffer including reducing reagent (Bio-rad) was added to the beads and the sample was boiled for 5 min at 95°C 30 before loading onto an SDS-PAGE (4-12%). Proteins were then transferred to nitrocellulose membrane and probed with antibodies against Wnt1 (Abcam), Wnt3a (Cell Signaling), Wnt 5a/b (Cell signaling) and HSP90 (Cell Signaling).
For the measurement of LRP6/β-catenin levels, 2.0 x 10 5 HEK293 luciferase reporter cells were seeded onto 24-well plates and transfected next day with 250 ng of DNA encoding Wnt proteins and chimeras as described above using FuGENE HD. After 24 hrs incubation in regular 5 condition, cells were gently washed twice with 500 μl cold PBS and directly lysed in 50 μl of lysis buffer (PBS, 1% (v/v) Triton X-100 and protease inhibitor cocktail (Roche)). Collected lysate was centrifuged for 5 min at 18,400 g to remove cell debris and supernatant was used to determine total protein concentration by BCA assay. 15 μg of total protein was loaded onto an SDS-PAGE (4-12%) to assess LRP6 and β-catenin protein levels using phospho-and total LRP6 (Cell Signaling),  20 4.0 x 10 5 HEK293 cells were seeded onto 12-well plates and transfected next day with 500 ng of DNA encoding Wnt proteins and chimeras, Wnt1, Wnt3a, Wnt5a, Wnt1_3a, Wnt3a_1 and Wnt5a_1, using FuGENE HD. After 24 hrs incubation, cells were gently washed twice with 500 μl cold PBS and RNA was isolated using the RNeasy kit (Qiagen), as instructed by the manufacturer's manual. Briefly, cells were lysed in RLT buffer including β-mercaptoethanol,

Material availability
The unique materials generated in this study will be available from the corresponding authors by request.

Data availability
The cryo-EM map has been deposited in the Electron Microscopy Data Bank under accession code 5 EMD-26989, and the model coordinate has been deposited to Protein Data Bank under accession number 8CTG.

Acknowledgments
We thank members of the Garcia and Hannoush Labs for thoughtful discussion and helpful feedback. Cryo-EM data were collected at the Stanford cryo-EM center (cEMc). We thank Drs.                           Hydrophobicity of LRP6 funnels. The AlphaFold2 hLRP6E1E4 model in surface representation were colored as indicated in the panel, and the figure is prepared with UCSF ChimeraX. Red, more 5 acidic; Blue, more basic; Brown, more hydrophobic; Green, more hydrophilic.   . However, the binary complex was reported to be insoluble without this truncation, indicating that the hydrophobic N-terminal loop is structurally isolated from the Wnt core and exposed to the solvent when bound to mFzd8CRD.