A direct NanoBRET assay to capture dynamic association of FZD5 and LRP6
To characterize the interactions between FZDs and LRP5/6 upon stimulation with WNTs, we established a direct NanoBRET assay using a C-terminally located Nanoluciferase (Nluc)/Venus system (Fig. 1A + B). For validation of the assay, we utilized FZD5, which robustly activates WNT/β-catenin signaling 14,19,20. Surface expression of all constructs used in BRET assays was analyzed by surface ELISA (Supp. Figure 1A + B). We could detect all FZD-Nluc/Venus and LRP-Nluc/Venus constructs at the cell surface with limited surface expression of LRP5 constructs and FZD5-Venus, whose expression levels were below the detection threshold of the ELISA. LRP6 constructs required the co-expression of the chaperone MESD 21 for significant surface expression levels. Therefore, MESD was co-transfected in all experiments that included LRP5/6. Notably, co-expression of LRP6-Venus and FZD5-Nluc decreased the surface expression of the latter, which nevertheless remained above the detection threshold (Supp. Figure 1C).
In acceptor titration NanoBRET experiments, we did not observe constitutive interactions between FZD5-Nluc and LRP6-Venus (Supp. Figure 2A). Given that WNT-3A is known to elicit stabilization of β-catenin by recruiting both FZD5 and LRP6, we next tested its effect in ligand stimulation experiments. Here, we observed a robust and dynamic BRET increase between FZD5-Nluc and LRP6-Venus upon addition of recombinant purified WNT-3A in HEK293 cells, and a notably lower BRET increase between FZD5-Nluc and LRP5-Venus, potentially due to lower surface expression of LRP5-Venus (Fig. 1C). The assay displayed similar results with an inverse probe setup using LRP5/6-Nluc and FZD5-Venus (Supp. Figure 2B). We decided to continue with FZD-Nluc/LRP5/6-Venus probes since FZD5-Venus showed low surface expression levels (Supp. Figure 1A). A dynamic BRET response was not observed upon stimulation of an unrelated GPCR (β2-adrenoceptor) and a CRD-truncated FZD5 construct (ΔCRD-FZD5) in combination with LRP6, confirming assay specificity (Supp. Figure 2C). A potential impact of endogenously expressed LRP5/6 was addressed by performing the experiments in HEK293T ΔLRP5/6 cells (Fig. 1D). Kinetic traces obtained here closely resembled those obtained in regular HEK293 cells. Along these lines, stimulation with a WNT surrogate9 resulted in a strong increase in ΔBRET between FZD5-Nluc and LRP6-Venus, and to a lesser extent between FZD5 and LRP5-Venus, comparable to WNT-3A (Fig. 1E).
LRP5 and LRP6 transduce WNT signals with different efficacies, which was pinpointed to be a property of their C-terminal portions 22. As LRP5 displayed lower ΔBRET values, we reasoned that exchanging the LRP5/6 C-termini might change the magnitude of the response to that of the respective other paralog. However, the traces observed in BRET experiments using FZD5-Nluc and an LRP6 chimera with an LRP5 C-tail (LRP6-5CT-Venus) or vice versa (LRP5-6CT-Venus) were practically unchanged compared to traces obtained with the respective wild-type LRP5/6-Venus (Fig. 1F). Notably, the cell surface expression of LRP5-Venus and LRP5-6CT-Venus did not surpass the detection threshold, while LRP6-5CT-Venus retained the approximate surface expression of LRP6-Venus (Supp. Figure 1). Our results suggest that the observed difference in ΔBRET between the traces of LRP5-Venus and LRP6-Venus did not result from a different conformational space obtained by their respective C-termini, but either from different expression levels or from their respective N-terminal/transmembrane domains.
Due to the lower surface expression of LRP5, we continued our experiments exclusively with LRP6. As WNT stimulation caused a dynamic BRET increase and constitutive interactions were not detectable, we conclude that our BRET assay measures WNT-induced association of FZD5 and LRP6 rather than conformational dynamics of a pre-formed complex. Our findings contrast a study employing similar BRET probes that claimed a constitutive, but not dynamic interaction between mouse FZD8 and LRP6 23.
LRP6 phosphorylation is not required for WNT-induced FZD-LRP6 association
WNT-induced association of LRP5/6 and FZD is a hallmark of WNT/β-catenin signaling. In the process of WNT/β-catenin signaling, five C-terminal PPP(S/T)P-motifs of LRP6 are phosphorylated by GSK3 and CK1 isoforms, components of the β-catenin destruction complex, presenting another proximal hallmark of the activation of the WNT/b-catenin pathway 24. To determine whether LRP6 phosphorylation has an impact on receptor association as measured in the direct BRET assay, we generated an LRP6-Venus mutant lacking all phosphorylation sites in the C-tail (LRP6-5A-Venus; Fig. 1G) 25. The mutant LRP6-5A-Venus expressed at the cell surface to a degree comparable to wild-type (WT) LRP6-Venus (Supp. Figure 1B). WNT-3A-stimulated BRET traces of FZD5 and LRP6-5A-Venus behaved identical to WT LRP6-Venus (Fig. 1H), emphasizing that FZD5-LRP6 association is independent of LRP6 phosphorylation.
To confirm signaling deficiency of the LRP6-5A mutant, we cloned a C-terminally untagged version of WT LRP6 and LRP6-5A. In agreement with the literature, LRP6-5A was in fact β-catenin signaling-incompetent both by overexpression and in response to WNTs, as assessed by TOPFlash reporter gene assays (Supp. Figure 3). In contrast to native LRP6, recombinant expression of LRP6-5A could not rescue TOPFlash reporter gene activity upon WNT-3A stimulation in ΔLRP5/6 cells and acted in a dominant-negative fashion when expressed in HEK293 cells, probably due to outcompeting endogenous LRP6. In summary, our data suggest that the phosphorylation status of the LRP6 C-terminus neither affected the WNT-3A-induced FZD5-LRP6 association nor its C-terminal conformation, despite significantly different signaling outputs.
DVL modulates the conformation of the FZD-LRP6 complex
The phosphoprotein DVL is the main intracellular transducer of WNT-induced and FZD-mediated signaling and binds to FZDs with high affinity mainly via its Dishevelled, Egl-10 and Pleckstrin (DEP) domain 17,20. DVL oligomers, polymerized by their Dishevelled/Axin (DIX) domains, are an integral part of the WNT receptor signalosome where they mediate interactions between the upstream WNT receptors and proteins of the β-catenin destruction complex. To assess the potential impact of DVL on the interaction between FZD5-Nluc and LRP6-Venus, we have analyzed WNT-3A-induced BRET responses in mutational paradigms from two angles (Fig. 2A): (i) a receptor-based angle, where we have employed the FZD5 R6.32A mutant, which prefers G protein- over DVL-mediated signaling 16. (ii) a DVL-based angle employing HEK293 ΔDVL1-3 cells in combination with non-functional mutants of DVL2, i.e. the oligomerization-deficient M2/M4 mutant26,27 as well as the L445E mutant incapable to bind to FZDs 17,28.
In an initial experiment, we compared ΔBRET traces between LRP6-Venus and WT FZD5-Nluc or FZD5-Nluc R6.32A upon WNT-3A stimulation and found no substantial difference between the traces (Fig. 2B). This suggested to us that DVL binding by FZD5 would not modulate the WNT-3A-induced complex formation between FZD5-Nluc and LRP6-Venus. However, when comparing the ΔBRET traces of WNT-3A-stimulated WT FZD5-Nluc and LRP6-Venus between regular HEK293 cells and ΔDVL1-3 cells, the ΔBRETmax increased substantially when DVL was absent (Fig. 2C). As DVL supposedly serves as a platform allowing clustering of multiple FZD5-WNT-LRP6 complexes into higher order complexes, the observed ΔBRET increase in ΔDVL1-3 cells came unexpected. We initially confirmed that DVL knockout and co-expression had no impact on FZD surface expression (Supp. Figure 4A). The increased ΔBRET in ΔDVL1-3 cells could, however, be lowered to a level similar to that observed in native HEK293 cells by co-transfection of DVL2. This suggested that the observed phenomenon is based on presence of DVL2 instead of a cell-line specific difference (Fig. 2C, Supp. Figure 4B). Transfection of the oligomerization-deficient DVL2 M2/M4 mutant into HEK293 ΔDVL1-3 cells showed the same effect as transfection of WT DVL2. DVL2 therefore modulates the interaction between FZD5-Nluc and LRP6-Venus in a DIX polymerization-independent manner. Similarly, co-transfection of DVL1 or DVL3 led to similar reductions in ΔBRETmax (Fig. 2D, Supp. Figure 4C) indicating that there are no DVL paralog-specific effects.
Similar observations were made in a combined approach using HEK293 ΔDVL1-3 cells and the FZD5 R6.32A mutant, whereby ΔBRET between FZD5 R6.32A-Nluc and LRP6-Venus was increased in ΔDVL1-3 cells and could be lowered by recombinant DVL expression (Supp. Figure 4D + E). The ΔBRETmax observed for WT FZD5 and the R6.32A mutant in ΔDVL cells was not significantly different from each other, suggesting that WNT-induced FZD5-LRP6 complex formation is similarly affected by the absence of DVL1-3 for both WT FZD5 and FZD5 R6.32A (Supp. Figure 4F). Lastly, we utilized the DVL2 L445E mutant, whose DEP domain cannot interact with FZDs 17,28, and a combined DVL2 M2/M4 L445E mutant. From this, we found that neither of the DVL2 L445E mutants could dampen the increased ΔBRET in ΔDVL1-3 cells, emphasizing that the FZD-DVL interaction is the main driver of the difference in WNT-3A-induced ΔBRETmax between HEK293 and ΔDVL cells (Fig. 2E, Supp. Figure 4G).
In conclusion, presence of FZD-binding DVL in the cell significantly decreases the dynamic BRET range between FZD5-Nluc and LRP6-Venus upon WNT-3A stimulation, independent of DIX-dependent DVL polymerization and the DVL paralog. We interpret this data as evidence for a role of DVL in orchestrating the arrangement of FZD5-Nluc and LRP6-Venus C-termini in a conformation disfavoring BRET (illustrated in Fig. 2F).
WNTs can stimulate FZD-LRP6 association without stabilizing β-catenin
To investigate FZD- and WNT-paralog selectivity, we transfected HEK293 cells with representative FZD-Nluc paralogs of the FZD homology clusters that are known to activate WNT/β-catenin signaling (FZD4/5/7-Nluc) 29,30 and LRP6-Venus, and stimulated them with diverse recombinant WNTs (WNT-3A, WNT-5A, WNT-10B, and WNT-16B) (Fig. 3A). FZD4/5/7 displayed a robust BRET increase when stimulated with WNT-3A with differences in ΔBRETmax (FZD5 > FZD4/7). When stimulated with WNT-5A, most BRET traces showed a transient increase upon ligand addition, followed by a decrease to baseline. The signal remained slightly above the baseline for FZD5 but displayed a markedly lower ΔBRETmax value compared to WNT-3A stimulation. WNT-10B stimulation yielded a weak signal for FZD5 and FZD7, but no reliable ΔBRET changes for FZD4. WNT-16B stimulation resulted in a monophasic ΔBRET increase to a plateau similar to that of WNT-3A stimulation for all tested FZDs (with ΔBRETmax values FZD5 > FZD4 > FZD7).
To determine whether the previously tested WNTs could activate WNT/β-catenin signaling, we investigated phosphorylation of LRP6 and DVL2 upon ligand stimulation in HEK293 cells by Western blotting (Fig. 3E-G). LRP6 phosphorylation was exclusively detected upon stimulation with WNT-3A or WNT surrogate, and not in response to WNT-5A, WNT-10B, or WNT-16B. Only WNT-3A, WNT-5A, and WNT-16B, however, led to a significant upward electrophoretic mobility shift of DVL2 indicative of WNT-induced DVL2 hyperphosphorylation (PS-DVL2) 31. In an orthogonal approach, we employed a TCF/LEF reporter gene assay (TOPFlash) as a measure of agonist-induced, β-catenin-dependent gene transcription 32,33, yielding essentially the same results in HEK293 cells and HEK293 ΔFZD1 − 10 cells specifically expressing FZD4/5/7 (Fig. 3H; notably, the employed WNT surrogate does not bind to the FZD4-CRD).
WNT-16B was of particular interest, as it induced association between FZD4/5/7-Nluc and LRP6-Venus but neither LRP6 phosphorylation nor reporter gene activity. To exclude that quantitative FZD degradation upon WNT-16B treatment masks a potential TOPFlash signal, we treated samples with R-spondin 1 (RSPO1), which prevents FZD ubiquitinylation and dramatically amplifies WNT efficacy 1,3. Co-treatment of HEK293 cells with WNT and RSPO1 greatly amplified the WNT-3A signal, but the WNT-16B signal remained indistinguishable from vehicle in TOPFlash reporter gene assays (Fig. 3I). Furthermore, we demonstrated that heat-inactivated WNT-16B (hi-WNT-16B) did not cause a ΔBRET increase between FZD5-Nluc and LRP6-Venus (Supp. Figure 5A). Next, we set out to show that WNT-16B behaves similarly to WNT-3A in previously used BRET assay paradigms: Like WNT-3A, WNT-16B displayed no altered ΔBRET trace when using a phosphorylation-insensitive LRP6-5A-Venus mutant (Supp. Figure 5B), and it also displayed an increased ΔBRET in absence of DVL (Supp. Figure 5C).
In summary, we observed that WNT-16B does not induce β-catenin-dependent signaling in our cell model but leads to a detectable ΔBRET increase between FZD-Nluc and LRP6-Venus probes. Stimulation with WNT-16B is also not accompanied by LRP6 phosphorylation characteristic for active WNT/β-catenin signaling. Our observations infer that WNT-induced FZD-LRP6 association as measured by BRET is not sufficient to induce β-catenin-stabilization. Despite the inability of WNT-16B to elicit β-catenin-dependent signaling, the positive efficacy of WNT-16B in the BRET assay and the electrophoretic mobility shift of DVL2 emphasize the protein’s ability to interact with its receptors and elicit functional downstream events.
WNT-3A and WNT-16B induce different transcriptome changes in HEK293 cells
The signaling activity of WNT-16B has remained an enigma for quite some time 34. To gain an unbiased perspective, we performed poly-A enriched bulk mRNA sequencing of HEK293 cells treated with vehicle control, WNT-3A, or WNT-16B. HEK293 cells were cultured in complete medium supplemented with 10 nM C59 to suppress endogenous WNT secretion for 24 h and were subsequently stimulated for another 24 h with recombinant WNTs. After variance stabilization transformation, the principal component analysis (PCA) of RNA-seq data demonstrated clear separation of cells by treatment (Supp. Figure 6). Untreated samples clustered quite closely to WNT-3A treated samples, which may hint at basal WNT-3A-like signaling in the sample cells. For differential gene expression analyses, thresholds were set to p-value < 0.05 and log2 fold-change of > 1.5. Volcano plots (Fig. 4A-C) depict significantly up- (red) and down-regulated (blue) transcripts. Stimulation with each WNT resulted in a unique mRNA transcriptome signature relative to vehicle-treated HEK293 cells (Fig. 4A,B; see Supp. Table 1 for datasets of differentially expressed genes). Direct comparison between WNT-3A and WNT-16B treatment also revealed a distinct mRNA signature (Fig. 4C). Genes regulated by one WNT sometimes correlated with genes regulated by the other WNT (Fig. 4D), indicating that some transcriptional responses were shared between the treatments, while the majority (1,206 out of 1,593 comparisons) did not meet significance criteria in both datasets (also see Fig. 4C). We conclude from this unbiased approach (i) that WNT-16B is transcriptionally active even though it was not inducing a TCF/LEF-dependent response in TOPFlash assays, and (ii) that despite some shared target genes, WNT-3A and WNT-16B mediate largely different transcriptional programs, reflecting the observed differences regarding the biophysical readouts and the TOPFlash data.
WNT-3A and WNT-16B induce FZD5-LRP6 protein complexes with distinct architecture
WNT-3A and WNT-16B can interact with the same set of cell surface receptors but mediate different cellular signaling programs with respect to the activation of β-catenin-dependent signaling. How does WNT-receptor activation differ to accomplish distinct cellular responses? We hypothesized that the different signaling outputs of WNT-3A and WNT-16B originate from differences in the molecular interaction between FZD and LRP6, which could not be captured by ensemble methods such as BRET. To this end, we employed dual-color single-molecule fluorescence microscopy to track FZD5 and LRP6 in live cells in response to stimulation with either WNT-3A or WNT-16B.
We transiently transfected N-terminally tagged SNAP-FZD5 and Halo-LRP6 constructs into Chinese Hamster Ovary K1 (CHO-K1) cells to achieve low physiological molecule density (FZD5: 0.42 ± 0.1, LRP6: 0.59 ± 0.15 molecules/µm2). Receptors were labeled with saturating concentrations of SNAP SiR-647 and Halo R110 fluorophores, respectively, and imaged with fast, multi-color total internal reflection fluorescence (TIRF) microscopy, combined with single-particle tracking (Fig. 5A + B). Data were acquired both under basal (Supp. Video 1 + 2 for raw movies and single-particle tracking with additional plotting of interactions) and after early (2–10 min) and late (11–25 min) stimulations with WNTs (Supp. Video 3–10). When analyzed by time-averaged mean squared displacement, FZD5 and LRP6 molecules explored a range of diffusion profiles on the plasma membrane, alternating between confined and random Brownian diffusion (Supp. Figure 7A). Both WNT-3A and WNT-16B stimulations increased the proportion of confined FZD5 and LRP6 molecules (Fig. 5C). To estimate the frequency and duration of FZD5 and LRP6 interactions, we applied previously developed methods based on deconvolution of apparent colocalization times with those of random colocalizations 35. Random colocalization times were estimated by imaging SNAP-β2-adrenoceptors (β2ARs), a prototypical GPCR with similar diffusion properties to that of FZD5 (expressed at similar densities 0.4 ± 0.08 molecules/µm2), and Halo-LRP6. In the absence of WNTs, FZD5 and LRP6 molecules did not colocalize for longer than what we measured for random colocalizations between SNAP-β2AR and Halo-LRP6 confirming our data from BRET acceptor titration experiments (Supp. Figure 2A, Supp. Videos 11 + 12).
Following stimulation, both WNT-3A and WNT-16B caused a substantial increase in FZD5 and LRP6 association rates at both early and late stimulation time-points. Notably, WNT-3A induced a ~ 1.5 and ~ 10-fold increase in association rate compared (kon) to WNT-16B at early and late stimulation time-points, respectively (Fig. 5D, left). We estimated that following early and late WNT-16B stimulation, FZD5 and LRP6 interactions lasted ~ 0.76 s and ~ 1.57 s, respectively (koff early = 1.32 s− 1, 95% confidence interval (CI): 0.81–1.84; koff late = 0.64 s− 1, 95% CI: 0.52–0.75). WNT-3A induced a ~ 2-fold (early and late) increase in FZD5-LRP6 interaction times, compared to WNT-16B (koff early = 0.65 s− 1, 95% CI: 0.37–0.92; koff late = 0.3 s− 1, 95% CI: 0.24–0.36) (Fig. 5D, right). Importantly, WNT-3A interaction times at late stimulation time-points were reaching the limit of the observation window in our experiments, which indicated that the true length of interactions following WNT-3A late stimulation are in fact longer than measured. A spatial confinement analysis revealed that both ligands induced receptor interactions in a co-confined state, where FZD5 and LRP6 are confined together, rather than a co-diffusing state, where both receptors move together across the plasma membrane. However, only WNT-3A showed statistically significant increase in receptor co-confinement. This shift from co-diffusion to co-confinement is reflected by the inverted pattern in the graph depicting the relative fraction of detected interactions (Fig. 5E). In addition, at the late stimulation time-point, we observed that WNT-3A induced aggregation of both FZD5 and LRP6 into clustered, higher-order complexes, which were not observed when cells were stimulated with WNT-16B (Fig. 5F). Furthermore, we exploited the single-molecule data to analyze the size of such complexes. Monomeric, fluorescent particles are expected to photobleach in a single step, whereas higher order oligomers follow a stepwise photobleaching pattern (Supp. Figure 7B). In response to WNT-16B-stimulation, stepwise photobleaching analysis showed that co-confined FZD5 and LRP6 molecules were largely monomeric (~ 60% and 80%, respectively). Conversely, WNT-3A-stimulated cells had a smaller fraction of monomeric co-confined FZD5 and LRP6 molecules (~ 20%) and higher numbers of molecules that photobleached in multiple steps (up to 15), indicating the presence of higher-order clusters (Fig. 5G). Notably, by the time receptors had aggregated in higher-order clusters in the single-molecule fluorescence microscopy experiments, ΔBRET traces had already reached their plateau (> 10 min).
While stimulation with both WNTs increased receptor confinement, co-confinement, and interaction duration, the characteristic formation of FZD5-LRP6 higher-order clusters was more prominent with WNT-3A than WNT-16B. These findings provide a mechanistic explanation for the lack of effective WNT/β-catenin signaling upon WNT-16B stimulation. Thus, ligand-induced receptor association and confinement, as seen with WNT-16B, is not sufficient for WNT/β-catenin signaling, which requires the formation of higher-order clusters.