Structural insight into small molecule action on Frizzleds

WNT-Frizzled (FZD) signaling plays a critical role in embryonic development, stem cell regulation and tissue homeostasis. FZDs are linked to severe human pathology and are seen as a promising target for therapy. Despite intense efforts, no small molecule drugs with distinct efficacy have emerged. Here, we identify the Smoothened agonist SAG1.3 as a partial agonist of FZD6 with limited subtype selectivity. Employing extensive in silico analysis, resonance energy transfer- and luciferase-based assays we describe the mode of action of SAG1.3. We define the ability of SAG1.3 to bind to FZD6 and to induce conformational changes in the receptor, recruitment and activation of G proteins and dynamics in FZD–Dishevelled interaction. Our results provide the proof-of-principle that FZDs are targetable by small molecules acting on their seven transmembrane spanning core. Thus, we provide a starting point for a structure-guided and mechanism-based drug discovery process to exploit the potential of FZDs as therapeutic targets.

G protein-coupled receptors (GPCRs) are membrane proteins, which constitute as much as 30% of all drug targets 1,2 . However, of the~800 GPCRs in human only a small fraction is targeted by FDA-approved drugs leaving a large untapped, therapeutic potential in the remaining receptors 1 . The Class F (Frizzled; FZD) of GPCRs, which consists of ten FZD paralogues (FZD [1][2][3][4][5][6][7][8][9][10] and Smoothened (SMO) is critically involved in embryonic development, organogenesis, stem cell regulation, and in the development of diverse pathologies, such as different forms of tumors, fibrosis, bone disease, cardiovascular conditions, and neurological disease 3 . While there are several small molecules available that target SMO as agonists (SAG1.3, SAG1.5, and purmorphamine), inverse agonists (cyclopamine-KAAD), and neutral antagonists (vismodegib and SANT-1), no small molecules with clear-cut pharmacology have been identified targeting any FZD. Given their involvement in pathology, FZDs harbor a huge therapeutic potential and therefore, drugging FZDs has attracted substantial attention [4][5][6] . Interestingly, the crystal structure of FZD 4 , which presents a ligand-free receptor inferred that development of small molecules targeting the core of FZDs can be virtually impossible given the hydrophilic nature of the binding pocket 7 , a notion that has previously been challenged 8 . In addition, the concept of allosteric modulators has been explored with the small molecules FzM1 and FzM1.8, which were characterized as negative and ago-positive allosteric modulators, respectively, acting on the third intracellular loop (ICL3) of FZD 4 with low degree of selectivity 9,10 .
The WNT family of lipoglycoproteins constitutes endogenous agonists for FZDs, activating the receptor through interactions with its extracellular cysteine-rich domain (CRD) 11 . Intracellularly, FZDs interact with Dishevelled (DVL), which is a signaling hub to mediate β-catenin-dependent and planar cell polarity (PCP)-like WNT signaling 12 . Furthermore, heterotrimeric G proteins interact with FZDs to initiate a network of G proteindependent signaling pathways 5,13 . One of the explanations for the absence of FZD-targeting small molecule compounds is the lack of high-throughput assays that monitor FZD activation more directly than the T-cell factor/lymphoid enhancer-binding factor (TCF/ LEF) transcriptional reporter (TopFlash) assay can do. The Top-Flash assay has the clear disadvantages that (i) not all FZDs, particularly not FZD 3 and FZD 6 , mediate WNT/β-catenin signaling and (ii) it does not cover all signaling pathways that branch off from activated FZDs, such as PCP or G protein-dependent signaling 5,14 . Recently developed resonance energy transfer-based methods (bioluminescence resonance energy transfer, BRET and Förster resonance energy transfer, FRET) can be advantageous to obtain more direct insight into FZD activation manifested in receptor conformational changes, FZD-G protein interaction, G protein activation, and FZD-DVL interactions [15][16][17] .
Based on the sequence homology between SMO and FZD 6 , and the possibility that SMO ligands could act on closely related FZDs, we show here that the small molecule SMO agonist SAG1.3 targets the transmembrane core of FZD 6 as a partial agonist with limited subtype selectivity. SAG1.3 binds FZD 6 and evokes a conformational change reminiscent of that seen in other agonistbound GPCRs. Moreover, SAG1.3 stimulates FZD 6 -dependent DVL membrane recruitment arguing that SAG1.3 stabilizes distinct receptor conformations accommodating G protein or DVL, supporting pathway-dependent functional selectivity 15 . In summary, our data indicate that FZDs can be targeted by small molecules.

Results
The SMO ligand binding pocket is similar in FZD 6 . Phylogenetic analysis of Class F receptors 3,7,15 and sequence alignment of human SMO, FZD 6 , and FZD 4 indicate substantial homology among these receptors. However, FZD 6 shows a higher degree of sequence similarity with SMO than it does with FZD 4 at regions corresponding to the small-molecule binding pocket within the 7TM core (as observed in SMO; Fig. 1a and Supplementary  Fig. 1), suggesting also functional similarities. Here, we chose to compare FZD 6 to FZD 4 and SMO because crystal structures of FZD 4 and SMO allow comparison on the atomistic level. Despite the differences that SMO mediates hedgehog signaling and FZD 6 mediates WNT signaling, both are characterized by their ability to couple to and activate heterotrimeric G i proteins 15,[18][19][20][21] , and their inability to signal via the WNT/β-catenin pathway 22,23 . FZD 6 and SMO are both characterized by a long TM6 extending above the plasma membrane toward the CRD 15,[24][25][26] and the longest C-terminal domains across Class F receptors (SMO: 250 aa; FZD 6 : 211 aa). Thus, the similarities of FZD 6 and SMO compared to FZD 4 provided the basis of our efforts of reprofiling SMO agonists for FZD 6 .
In silico analysis of SAG1.3-FZD 6 interactions. As the putative small-molecule binding pockets in the transmembrane domain of FZD 6 are unknown, we built 15 homology models of FZD 6 using the ΔCRD SMO-taladegib (PDB ID: 4JKV) complex as a template 27 . Of these models, we selected the one with the best DOPE score for further studies 28 . The selected FZD 6 model (inactive FZD 6 ) subsequently underwent molecular dynamics (MD) simulations for 200 ns (in two independent replicas) in the ligand-free state in order to relax the structure. Subsequently, SAG1.3 (Fig. 1b) was docked to the binding site in the transmembrane core of the receptor, defined by the location of the cocrystallized SAG1.5 in the SMO crystal structure (PDB ID: 4QIN 29 ; Fig. 1c). To compare SAG1.3-FZD 6 interactions with those present in a SAG1.3-SMO complex, we used the SAG1.5-SMO crystal structure, in which we modified the agonist by substituting the fluorine atoms for hydrogen atoms. Subsequently, the MD simulations were run for additional 3 µs (1 µs in three independent replicas) and 600 ns (200 ns in three independent replicas) with SAG1.3-FZD 6  Provided the recent insight into SMO activation in a ternary complex of ligand, receptor, and heterotrimeric G i protein (PDB ID: 6OT0 21 ), we built also a FZD 6 model based on the active SMO structure and ran MD simulations with SAG1.3 docked to the same binding site as described above (Fig. 1d, e, Supplementary  Fig. 3; active-like FZD 6 ). The MD data (three independent replicas of 1 µs each) were then used for a retrospective analysis of the binding site and interactions of SAG1.3 in complex with activelike FZD 6 . To avoid misleading interpretations, we consider only one of the 1 µs replica of the simulations of the SAG1.3-bound inactive FZD 6 in all analyses; the N-terminus and extracellular loops in the other two replicas started to undergo rapid and noisy fluctuations after 400 ns and 600 ns of the simulations, respectively (most probably due to the instability in the homology model).
With the active-like FZD 6 model, such behavior did not occur, and all 3 µs of data are considered in all analyses (Fig. 1d, e, Supplementary Fig. 3).
The overall binding location of SAG1.3 remained robustly similar in both studied proteins throughout different simulations (Fig. 1, Supplementary Figs. 2-4), suggesting that FZD 6 has a binding site for SAG1.3 in the transmembrane core of the receptor between the TM5, TM6, TM7, and the extracellular loop 2 (ECL2), similar to SMO. When comparing this binding area to the structure of FZD 4 (PDB ID: 6BD4, after 3 × 200 ns MD simulations), the extracellular portion of TM6 in FZD 4 together with the ECL3 clash heavily with the suggested SAG1.3 binding site underlining a structural basis for ligand-receptor selectivity (Fig. 1c) 7 . Sequence alignment of these receptors supports this observation, since the parts of TM6 and ECL3 that construct the binding site of SAG1.3 in FZD 6 and SMO are for the most part missing from the sequence of FZD 4 (Fig. 1a, Supplementary  Fig. 1).
In complex with the inactive FZD 6 , N2 of SAG1.3 interacted quite robustly with E438 6.54 throughout the 1 µs simulation (Fig. 1d, e, Supplementary Figs. 3a and 4). The simulation with active-like FZD 6 strengthens this observation further. E438 6.54 remains at a hydrogen-bonding distance to N2 over 75% of all 3 µs of these MD trajectories (Fig. 1d, Fig. 3e and 4). Interestingly, the active-like FZD 6 simulation produced two clear clusters of binding poses, whereas in the inactive FZD 6 simulation the SAG1.3 poses-apart from the fact that they maintain N2-E438 6.54 interaction-were notably more deviant (Fig. 1d, Supplementary Fig. 3f, g). Unlike SMO, the FZD 6 binding site contains only these two negatively charged amino acid residues able to interact with N2 of SAG1.3; thus SAG1.3 does not change the charge-assisted hydrogen-bonding partner to amino acid 7.38 in FZD 6 as it does in SMO (Fig. 1a, d, Supplementary Fig. 4).
In the active-like FZD 6 simulations, the average distance between N187 of FZD 6 (corresponding to N219 of SMO) and the amide oxygen of SAG1.3 is~5 Å ( Supplementary Figs. 3h and 4). Even though the distance is most of the time too long to suggest direct hydrogen bonding, a water-mediated hydrogen bond is highly possible (Supplementary Fig. 3i). Unlike the active-like FZD 6 model, the inactive model rarely poses N187 in the vicinity of the amide oxygen of SAG1.3, but R442 6.58 remains there instead (Supplementary Fig. 3b, c). In the active-like model, R442 6.58 interacts rather with the chlorine atom of SAG1.3 (Fig. 1d, e).
In the SAG1.5-SMO crystal structure, the corresponding MD simulations, and the SAG1.3-SMO and SAG1.3-FZD 6 MD simulations, the two aromatic ends of the SAG derivatives form a stacked π-π complex located at a sub-pocket lined by aromatic amino acid residues of the TM6 and the ECL2 (Fig. 1a, d, Supplementary Fig. 4). Due to the different sizes of these residues (F484 6.65 in SMO vs. W449 6.65 in FZD 6 ), the aromatic end of SAG1.3 occupies a slightly different space and obtains a slightly different orientation in these two receptors (Fig. 1d). As SAG1.3 is a relatively rigid molecule (only five rotatable bonds, which contribute to its overall shape), the orientation of the aromatic part of the molecule restricts the available locations of the hydrogen-bonding functional groups of SAG1.3 at its binding site. Even though the SAG1.3-SMO (based on the MD data), SAG1.5-SMO (based on the crystal structure and the MD data), and the SAG1.3-FZD 6 (based on the MD data) complexes share similar hydrogen-bonding characteristics , the apparent fit of the aromatic ends of SAG derivatives to these receptors may be one of the factors contributing to differences in affinity to SMO and FZD 6 . Please see Supplementary Figs. 6 and 7, and Supplementary Data files 1-30 for the details of all the MD simulation runs. 6 . Pharmacological analysis of ligandreceptor interactions is best studied using direct ligand binding experiments. Here, we employed a recently established assay format based on NanoBRET detection between a BODIPY-tagged ligand and a nanoluciferase (Nluc)-tagged receptor ( Fig. 2a; summary of FZD 6 constructs in Supplementary Fig. 8) 33 . The interaction of BODIPY-cyclopamine with SMO allows thorough characterization of SMO-binding ligands 34 . Given the similarities of SMO and FZD 6 in the ligand binding pocket, we used BODIPYcyclopamine as a probe for FZD 6 . Moreover, in order to exclude endogenously expressed SMO as a confounding factor in the BODIPY-cyclopamine-based binding assay, we generated a ΔSMO HEK293 cell line devoid of this GPCR using CRISPR/ Cas9 genome editing ( Supplementary Fig. 9). BODIPYcyclopamine binding to Nluc-FZD 6 resulted in monophasic and saturable concentration-dependent BRET signals ( Fig. 2b; BODIPY-cyclopamine pK d ± s.d. = 6.3 ± 0.1; refer to Supplementary Fig. 10a for the assessment of cell membrane expression of the constructs and Supplementary Fig. 10b/Supplementary Data file 31 for FZD 6 -BODIPY-cyclopamine docking poses). Additionally, BRET was dependent on donor expression levels and acceptor:donor ratio and was not directly proportional to the acceptor levels arguing for specificity of BODIPY-cyclopamine to Nluc-FZD 6 binding ( Supplementary  Fig. 10c). In competition experiments, increasing concentrations of unlabeled SAG1.3 decreased BODIPY-cyclopamine (300 nM) binding to Nluc-FZD 6 in a concentration-dependent manner (Fig. 2c, SAG1.3 pK i ± s.d. = 5.6 ± 0.1). Similarly, a fixed concentration of SAG1.3 (10 µM) right shifted the BODIPY-cyclopamine binding curve ( Supplementary Fig. 10d, BODIPY-cyclopamine with SAG1.3 (10 µM) pK d ± s.d. = 5.8 ± 0.2) and the BODIPY-cyclopamine binding curve was right shifted in the presence of 10 µM unlabeled cyclopamine (Supplementary Fig. 10e). Importantly, the SAG1.3-induced reduction in BODIPY-cyclopamine/Nluc-FZD 6 NanoBRET was not due to a nonspecific effect on luminescence or fluorescence Fig. 1 The binding pocket of FZD 6 accommodates SAG1.3. a Sequence alignments of the binding pockets of the human SMO, FZD 6 , and FZD 4 (Supplementary Fig. 1 and Supplementary Data file 34). Red squares indicate residues in close proximity (<4 Å) between SAG1.3 and the receptor from the SMO and FZD 6 molecular dynamics (MD) simulations. b Structure of SAG1.3. The bold nitrogen represents the N2 referred to in the MD simulations below. c Comparison of the SAG1.3 binding sites of SMO and FZD 6 (inactive model; upper panel), and SMO and FZD 4 (lower panel) underlining the inability of FZD 4 to accommodate a ligand-like SAG1.3 in this binding space because of the short TM6 (red arrow). d The last frames from the selected MD simulations of the SAG1.3-SMO (left panel) and the SAG1.3-FZD 6 complexes (inactive model: middle panel, active-like model: right panel) with the important residues of the binding site depicted as sticks. Different positions of SAG1.3 throughout the time of simulation are indicated by transparent SAG1.3 molecules in the binding pocket. e Distance plots over simulation time between SMO D473 6.54 -SAG1.3, SMO E518 7.38 -SAG1.3, FZD 6 E438 6.54 -SAG1.3, and R442 6.58 -SAG1.3 (inactive and active-like models), which are predicted to form H-bonding interactions important for stabilizing the SAG1.3 binding conformation. The dotted line (red) indicates the maximum distance (4 Å) that is still likely to allow polar interactions. Thick traces indicate the moving average smoothed over a 2 ns window and thin traces represent raw data. The origin of the y-axis for all graphs e is 0 Å. Fig. 10f). In addition, the NanoBRET-based binding assay has the advantage of a low contribution of nonspecific binding in general as seen for other GPCRs and in particular for Class F receptors, as we recently characterized during the establishment of the assay for Nluc-SMO and BODIPY-cyclopamine binding 34,35 . In conclusion, we provide here experimental data arguing that BODIPY-cyclopamine interacts with FZD 6 and that SAG1.3 interferes with BODIPYcyclopamine interaction.

(Supplementary
A FZD 6 -FRET probe monitors SAG1.3 binding and efficacy. In order to obtain a functional measure for ligand-FZD 6 interactions , we designed an intramolecular FRET probe to study conformational changes of the receptor upon ligand binding. On the basis of conformational changes in active Class A and B GPCRs, the previously validated probes for other GPCRs and particularly FZD 5 16,36 and the information from the recently published active SMO structures 21,37 (Fig. 3a), we created an intramolecular FZD 6 -FRET probe. The probe, which was designed to monitor agonist-induced conformational changes of FZD 6 in living cells, consists of the FRET donor (TFP) at the C-terminus and the FRET acceptor FlAsH (fluorescein arsenical hairpin binderethanedithiol, FlAsH-EDT 2 )-binding motif inserted between G404 and R405 in the ICL3 (Fig. 3b).
The FZD 6 -FRET sensor was detectable on the cell surface using confocal microscopy assessing TFP fluorescence, and we therefore conclude that it is efficiently trafficked to the cell membrane ( Supplementary Fig. 11a). Basal energy transfer between the fluorophores was determined as FRET efficiency of the sensor (4.4 %) by using BAL (2,3-dimercapto-1-propanol) as an antidote ( Supplementary Fig. 11b). In order to exclude energy transfer between individual receptors, for example in a FZD 6 dimer 25 , we assessed intermolecular FRET between FZD 6 -TFP and FZD 6 -FlAsH-PK and detected no measurable energy transfer ( Supplementary Fig. 11c).
To quantify the efficacy of the endogenous ligand of FZD 6 in this assay, we analyzed the effect of increasing concentrations of WNT-5A using the FZD 6 -FRET probe. The maximal response to WNT-5A defines the full agonist in this assay with a log 10 EC 50 ± s.d. (ng ml −1 ) = 2.2 ± 0.1 and the maximum efficacy at 1000 ng ml −1 reaching 4.9% (FRET ratio = 95.1% of basal; Fig. 3c). Since WNT-5A is not membrane permeable, the effect of ligand stimulation on the FZD 6 -FRET probe further corroborates efficient trafficking of FZD 6 -FlAsH-TFP to the plasma membrane. Stimulation of HEK293 cells transiently expressing the FZD 6 -FRET probe with SAG1.3 resulted also in a sigmoidal, concentration-dependent decrease in the FRET ratio by 1.7 % of the basal FRET ratio with a pEC 50 ± s.d. (M) = 6.5 ± 0.9 (Fig. 3d).
Our data demonstrate that (i) SAG1.3 binds to FZD 6 , (ii) that the polar residues D351, E438 6.54 , and R442 6.58 have an important role in small-molecule binding, (iii) that agonist binding to FZD 6 evokes conformational changes that are detectable by the FZD 6 -FRET sensor reminiscent of movements observed in activated Class A/B GPCRs and SMO, and (iv) that SAG1.3 acts as a FZD 6 partial agonist in this assay. 6 . In order to further explore the mode of action of SAG1.3 on FZD 6 , we made use of Venus-tagged mini G (mG) proteins, which serve as BRETcompatible, conformational sensors of the ligand-bound, active state of GPCRs 15,38,39 (Fig. 4a). Similar to what we have shown before for FZD 6 and other Class F receptors, we used SNAP-FZD 6 -Rluc8 and FLAG-FZD 6 -Nluc (BRET donor; see Supplementary Fig. 12 for analysis of membrane expression of FLAG-FZD 6 -Nluc) in combination with Venus-mGsi (BRET acceptor) transiently overexpressed in HEK293 cells to monitor WNT-5A-induced Venus-mGsi recruitment to FZD 6 , thereby defining the assay response with the physiological, full agonist (Fig. 4b, c). Further, we established the concentration-response relationship for both FZD 6 constructs in combination with Venus-mGsi using SAG1.3 (Fig. 4d, e). Interestingly, SAG1.3 induced a biphasic concentration-response curve similar to what was previously reported for SAG-SMO responses in the same assay format as well as in other assays 15,[40][41][42] . In order to exclude a functional role of the WNT-binding CRD for the SAG1.3-induced and FZD 6 -mediated Venus-mGsi recruitment, we compared ΔCRD and full-length FLAG-FZD 6 -Nluc constructs (  6 is independent of SMO. Since SAG1.3 was designed as SMO agonist, it appeared crucial to exclude a contribution of endogenously expressed SMO to the observed SAG1.3-induced and FZD 6 -mediated effects. SMO is also a G i/ocoupled receptor and it is expressed in HEK293 cells 15,20,43 . Thus, BRET mGsi recruitment assays were performed in the ΔSMO HEK293 cells using SNAP-FZD 6 -Rluc8 and Venus-mGsi in combination with increasing concentrations of SAG1.3. A similar biphasic concentration-response curve was observed (Fig. 4f). Furthermore, we compared ΔCRD and full-length FLAG-FZD 6 -Nluc with regard to their ability to recruit Venus-mGsi in ΔSMO HEK293 cells (Fig. 4g), further supporting the concept that SAG1.3 targets the receptor core. In line with the results of the indirect binding assay with the FZD 6 intramolecular FRET sensor, SAG1.3 elicited a smaller maximum Venus-mGsi recruitment compared to the highest WNT-5A concentration used, underpinning the partial agonist nature of SAG1.3. In order to define subtype selectivity of SAG1.3 toward FZD 6 over FZD 4 , we also assessed SAG1.3-induced Venus-mG recruitment to FZD 4 -Nluc using Venus-mG13 15,44 . In agreement with the in silico structural analysis, which suggested that SAG1.3 would not bind this FZD subtype, we did not detect any SAG1.3induced Venus-mG13 recruitment (Fig. 4h). On the other hand, we tested FZD 7 , a Class F receptor from the FZD 1,2,7 homology cluster, and the ability of SAG1.3 to induce Venus-mGs recruitment to SNAP-FZD 7 -Rluc8 15 . SAG1.3 induced a biphasic concentration-response curve similar to what we observed for FZD 6 , indicating that SAG1.3 does not only act at FZD 6 but also on other FZD subtypes ( Supplementary Fig. 13a). Indeed, the comparison of models of FZD 6 and FZD 7 on the atomistic level revealed large similarities in their SAG1.3 binding site (Supplementary Fig. 13b,c), in contrast to the one of FZD 4 . MD simulation of FZD 7 bound to SAG1.3 further underlined that the receptor-ligand interaction is stable for the time of the simulation ( Supplementary Fig. 13d).

SAG1.3 action on FZD
SAG1.3 promotes G protein and ERK1/2 activation. In order to further validate that SAG1.3 acts as a functional FZD 6 agonist, capable of initiating downstream signaling in a G proteindependent manner, we made use of heterotrimeric NanoBiT G proteins 45 . For this purpose, ΔSMO HEK293 cells were transfected with receptor or pcDNA, the Gα i1 and Gβ 5 subunits fused to complementary parts of a modified Nluc (LgBiT and SmBiT) and the untagged Gγ 2 (Fig. 5a). First, we used the muscarinic M 2 receptor as a prototypical G i -coupled receptor with acetylcholine (ACh), to demonstrate that we can detect a ligand-induced decrease in Nluc luminescence (Nluc lum ) indicative of the dissociation of the heterotrimeric G i protein ( Supplementary  Fig. 14, pEC 50  Receptor activation is predicted to result in a loss of FRET due to conformational rearrangement in accordance to previous data obtained for FZD 5 16 . www.guidetopharmacology.org/GRAC/ObjectDisplayForward? objectId = 14). Then we used SAG1.3 to monitor its ability to induce G i heterotrimer dissociation in a FZD 6 -dependent manner, excluding the contribution of endogenous SMO by using ΔSMO HEK293 cells. Similar to the Venus-mGsi protein recruitment assay, SAG1.3 elicited a bell-shaped concentration response only when SNAP-FZD 6 was coexpressed (Fig. 5b).  GPCR-mediated activation of heterotrimeric G i/o proteins leads to phosphorylation and activation of extracellular signalregulated kinases 1/2 (ERK1/2) 46 and we have previously shown that FZD 6 mediates ERK1/2 phosphorylation 15,25 . To further support the positive efficacy of SAG1.3 acting on FZD 6 resulting in G protein-dependent signaling, we quantified ERK1/2 phosphorylation in lysates of ΔSMO HEK293 cells transfected with SNAP-FZD 6 and stimulated with SAG1.3. These experiments were performed in the presence of endogenous G proteins. Further the autocrine stimulation by endogenously produced WNTs was blocked by pretreatment with the porcupine inhibitor C59 (5 nM). In agreement with our data so far, SAG1.3 induced a biphasic concentration-dependent ERK1/2 phosphorylation only when FZD 6 was transiently overexpressed in ΔSMO HEK293 cells (Fig. 5c).
SAG1.3 affects FZD 6 -DVL2 interaction. DVL is a central mediator of the β-catenin-dependent and PCP-like WNT signaling pathways and its recruitment to FZD is an initial step in DVL-dependent signaling 12,47,48 . Simultaneous overexpression of DVL and FZD leads to FZD-dependent membrane recruitment of DVL even in the absence of a ligand [49][50][51]   colocalization and recruitment of cytosolic DVL present in punctate aggregates to membrane-expressed FZDs 47,49-51 . However, quantification of recruitment and measurement of ligandinduced dynamics were not possible. Employing direct BRET, it was recently shown that the FZD 4 -selective agonist Norrin enhances FZD 4 -DVL interaction 17 . In order to assess agonistinduced effects on FZD 6 -DVL2 interactions, we used WNT-5A and SAG1.3 in two different experimental paradigms of BRETbased assays. First, we assessed the proximity of Nluc-DVL2 to SNAP-FZD 6 or FLAG-FZD 6 -His indirectly in a bystander BRET assay 15 . Nluc-DVL2 membrane recruitment was quantified by co-expressing a membrane-bound Venus-tagged CAAX domain of KRas (termed Venus-KRas 52 ), and assessment of bystander BRET between Nluc and Venus (Fig. 6a-d; Supplementary  Fig. 15a) 15,39 . Second, we measured BRET between coexpressed Nluc-DVL2 and FLAG-FZD 6 -Venus (Fig. 6e-g; Supplementary Fig. 15b; see Supplementary Fig. 12 for analysis of membrane expression of ΔCRD and full-length FLAG-FZD 6 -Venus). The settings of the direct BRET assay are reverse to those employed recently 17 , where the authors used YFP-DVL2 (BRET acceptor) and FZD 4 -Rluc (BRET donor). However, it is envisaged that fusing the BRET acceptor to FZD circumvents the potential analysis issues arising from DVL polymerization at high expression levels required for validation of the assay by a saturation curve 53 . In order to assess ligand-induced effects on DVL-FZD BRET, we chose an acceptor:donor ratio corresponding to the plateau part of the saturation curves for both setups. In addition, we did not treat the cells with the porcupine inhibitor C59 to block secretion of endogenous WNTs as their presence had no significant effect on the basal recruitment of the overexpressed DVL2 to the overexpressed FZD 6 as recently reported 51 and presented in Supplementary Fig. 15a, b. To avoid any input of endogenous FZDs or SMO, we used ΔFZD 1-10 HEK293 cells to study WNT-5A-induced effects and ΔSMO HEK293 cells to study SAG1.3induced effects. As shown in Fig. 6b, c, f, g, both ligands increased BRET between Nluc-DVL2 and Venus-KRas or FZD 6 -Venus in a concentration-dependent manner. Interestingly, SAG1.3 evoked FZD 6 -DVL2 BRET changes with lower potency than SAG1.3induced G protein-related events. Moreover, SAG1.3 did not show the bell-shaped concentration-response when monitoring FZD-DVL recruitment. SAG1.3 also displayed a positive efficacy on ΔCRD FZD 6 ( Fig. 6d, g). Further, we have validated the assays and SAG1.3-selectivity using pcDNA-and SNAP-FZD 4 -transfected ΔSMO HEK293 cells (Fig. 6h). Biochemically, we were able to detect a SAG1.3-induced electrophoretic mobility shift of the endogenous DVL2 indicative of its phosphorylation and activation in the SNAP-FZD 6 -but not control-transfected ΔSMO HEK293 cells (Supplementary Fig. 15c). Finally, we confirmed that SAG1.3 (10 µM) does not activate FZD 4 -specific TopFlash activity in ΔFZD 1-10 HEK293 cells (Fig. 6i), arguing for the subtype selectivity of this small-molecule ligand.
WNT and SAG1.3 stimulation increased BRET in both experimental paradigms. However, given the ratio of receptor-DVL expression (Supplementary Fig. 14) and the nature of BRET as a readout, we cannot differentiate clearly between an increase in FZD-DVL recruitment in a 1:1 ratio, DVL polymerization in close proximity to the receptor or a rearrangement of the FZD-DVL complex in response to agonist affecting distance or dipole orientation. Nevertheless, a change in FZD-DVL BRET can serve as a functional readout of FZD ligands keeping the caveats of this technique in mind.
Mutational analysis of the SAG1.3 binding site. Having established a diverse set of functional readouts for FZD 6 activation by SAG1.3 allowed now a mutagenesis analysis of residues involved in SAG1.3 interactions. The MD simulations in SAG1.5-and SAG1.3-bound SMO, and SAG1.3-bound FZD 6 using the inactive-and active-like models provided detailed insight into the engagement of residues in SAG-derivative interactions over time (Fig. 1d, Supplementary Fig. 4). In this analysis, D351, E438 6.54 , K479 7.41 , and R442 6.58 emerged as the most relevant, polar residues, which were included in a mutagenesis approach (Supplementary Fig. 16). SNAP-tagged receptor mutants were tested for their cellular and membranous expression in comparison to pcDNA-and SNAP-FZD 6 -transfected ΔSMO HEK293 cells (Supplementary Fig. 16a). While these proteins are indeed translated, mutation of these residues dramatically affects receptor maturation and cell surface expression. Only E438D 6.54 , R442A 6.58 , and R442K 6.58 were detectable at the membrane albeit at lower levels compared to wild-type SNAP-FZD 6 (Supplementary Fig. 16a). Nevertheless, surface expression mirrored the ability of the receptor mutants to recruit Nluc-DVL2 to the membrane assessed by bystander BRET using Nluc-DVL2 and Venus-KRas ( Supplementary Fig. 16b). In order to provide biologically sound and meaningful data, we only used three mutants showing surface expression and DVL recruitment to assess SAG1.3-induced effects. Mutation of the SAG1.3 binding site in FZD 6 affected the ability of SAG1.3 to induce mGsi and Nluc-DVL2 recruitment and G i protein activation (Supplementary Fig. 16c, d, e). In general, SAG1.3-induced responses of E438D 6.54 and R442K 6.58 mutants showed lower efficacy and potency when compared with the wild-type receptor (Supplementary Figs. 16c, d, e and 17). Furthermore, SAG1.3 stimulation of the R442A 6.58 mutant resulted in hardly detectable responses corroborating even weaker SAG1.3 interactions. Thus, mutating residues to their chemically conserved counterparts allows preserving a somewhat functional SAG1.3 binding site, whereas alanine mutation does not. While functional assays are directly affected by the fraction of the receptor protein that is trafficked to the cell membrane, we reasoned that assessment of ligand affinity could be a suitable complement to quantify the direct involvement of key residues in the FZD 6 binding site. We focused on FZD 6 R442 6.58 and its alanine mutation comparing the affinity of BODIPY-cyclopamine to FZD 6 and FZD 6 R442A 6.58 in the presence and absence of 10 µM SAG1.3 ( Supplementary Fig. 18). Since the presence of SAG1.3 right shifted BODIPY-cyclopamine binding only in the case of wild-type FZD 6 , we concluded that R442 6.58 -in agreement with the in silico predictions and functional assessment-is a key component of the SAG1.3 binding site.
In order to further support our findings concerning the importance of N2-E438 6.54 interaction, we provide MD data investigating the likelihood of interaction for inactive FZD 6 with a mock SAG1.3 (mock ligand), where we introduced a carbon (C7) instead of the N2 nitrogen ( Supplementary Fig. 19a). MD simulations of FZD 6 with SAG1.3 in comparison to mock ligand (3 × 200 ns) and distance plots between the E438 6.54 and either N2 of SAG1.3 or the C7 of the mock ligand indeed argue that SAG1.3 binds closer to E438 6.54 than the mock ligand over the time course of the simulation (Supplementary Fig. 19b).
Purmorphamine is also a FZD 6 agonist. In addition to SAG derivatives, purmorphamine presents another, structurally unrelated SMO agonist that is surmountable by the inverse agonist cyclopamine-KAAD ( Supplementary Fig. 20a) 54,55 . In order to support the broader applicability of our findings, we also examined purmorphamine-FZD 6 interaction by in silico docking and performed a pharmacological characterization of purmorphamine activity using two key assays presented in this work. In silico docking indicated that the purmorphamine binding site overlaps substantially with that of SAG1.3 ( Supplementary  Fig. 20b), in agreement with previous pharmacological characterization 55 . Compared to SAG1.3 purmorphamine promoted a similar, concentration-dependent recruitment of Venus-mGsi to SMO-Rluc8 and SNAP-FZD 6 -Rluc8, albeit with lower potency and without the distinct bell-shaped pattern (Supplementary Fig. 20c, d). Since the mGsi protein serves as sensor of the active FZD 6 conformation feeding into heterotrimeric G protein signaling, we conclude that purmorphamine binding to  Fig. 20e) arguing for a distinct functional selectivity of this ligand.

Discussion
Here, we provide the proof-of-principle that FZDs are druggable with small-molecule ligands targeting the 7TM core of the receptors. This stands in stark contrast to previous claims that the FZD binding pocket might be unfavorable for accommodating small-molecule ligands 7 . Our discovery opens the door for the development of FZD-targeting small molecules interacting with the receptor at a site reminiscent of that of Class A GPCRs and SMO ligands. Based on the data monitoring receptor binding, FZD 6 conformational changes, mG protein association as conformational sensors of the active GPCR state of FZD 6 , heterotrimeric G protein dissociation, and FZD 6 -DVL2 recruitment, we argue that SAG1.3 acts as a partial agonist with functional selectivity toward G proteins over DVL. SAG1.3-induced effects on FZD 6 were generally moderate but statistically significant calling for medicinal chemistry efforts to expand on our proof-ofconcept study (Supplementary Fig. 21). We also provide evidence that SAG1.3 acts at FZD 7 but not FZD 4 and that purmorphamine acts through FZD 6 , albeit with lower potency, indicating that different scaffolds exist to initiate a medicinal chemistry optimization. Furthermore, we use BODIPY-cyclopamine for assessing ligand binding introducing another, sterol-based moiety interacting with FZD 6 . While small-molecule agonists will provide an exciting tool to understand FZD activation mechanisms and receptor pharmacology, ligands with negative or no efficacy would be more suitable for anticancer therapy. Inverse agonists or neutral antagonists could provide a useful therapeutic approach in tumors that are driven by high levels of WNTs or constitutively active FZDs 3,4,15 .
In line with what we have previously proposed 5,15 , our data suggest that SAG1.3 can stabilize at least two distinct FZD 6 conformations feeding into FZD-G protein and FZD-DVL signaling. This is supported by the finding that an active SAG1.3-FZD 6 -mGsi complex is stabilized at lower nanomolar concentrations, whereas micromolar SAG1.3 concentrations are required to affect FZD 6 -DVL2 interaction. Additional support of this concept is provided by the purmorphamine data, showing a positive efficacy toward mGsi protein but not DVL. Furthermore, SAG1.3 is not merely an allosteric modulator of FZD 6 amplifying basal WNT input as it significantly increases the phosphorylation of ERK1/2 in the absence of endogenous WNTs.
Employing in silico analysis, we predict that D351, E438 6.54 , and R442 6.58 in FZD 6 are key residues involved in polar interactions with SAG1.3 (Fig. 1d). We then generated mutants of these residues (together with K479 7.41 ) and assessed their cellular and membranous expression as well as SAG1.3-induced effects. SAG1.3-induced and receptor-mediated effects were reduced as shown by mGsi and Nluc-DVL2 BRET recruitment and G i NanoBiT assays, however, only few mutants are folded and trafficked to the cell membrane albeit not as efficiently as the wildtype FZD 6 . This obviously needs to be taken into consideration when interpreting these mutagenesis experiments. BRET experiments are ratiometric, i.e., they do not rely on expression levels per se because the signal is detected only when donor and acceptor come into close proximity (up to 100 Å). Thus, BRET experiments can be used to evaluate and compare ligand-induced effects on non-equally expressed receptors. However, in the case of G i -dissociation measured by the NanoBiT assay, it cannot be ruled out that-given the cell permeability of SAG1.3-events from inner membranes contributes to the response. Furthermore, we have identified a polar network of interactions involving ECL2, TM6, and TM7 that is a part of the proposed FZD 6 ligand binding site. According to our mutagenesis data, this network is crucial to maintain proper protein folding and cell surface trafficking of FZD 6 ( Supplementary Fig. 19c) 34,56,57 .
The pharmacological profiles of SAG1.3 as an agonist on FZD 6 differ depending on the experimental readout. While SAG1.3-induced conformational changes in the FZD 6 -FRET probe and FZD 6 -DVL2 interaction assays follow a sigmoidal concentration-response relationship with an EC 50 in the lower micromolar range, all G protein-related readouts follow a bellshaped pattern with higher potency. Remarkably, the shape of the concentration-response curves is similar to what has been reported for SAG1.3-induced and SMO-mediated effects on Gli-1 reporter, mG recruitment, and inositol phosphate accumulation assays 15,32,40,41 . However, it remains obscure what the descending part of these SAG1.3 concentration-response curves represents in the context of SMO and FZD 6 signaling 40,58 , and interestingly, purmorphamine, which occupies a similar binding site, does not exert a bell-shaped concentration-response curve as SAG1.3 in the G protein-dependent assays. Similarly to what we demonstrate for SAG1.3-FZD 6 , it can also be seen for SMO that SAG1.3 potencies differ depending on the assay type and the cell line used (Supplementary Fig. 22). While this could merely represent the differential functionality and sensitivity of the engineered assay probes, it seems more likely that they underline the functional selectivity or ligand bias of the agonist. Similar discrepancies in the pharmacological profiles have been reported for the β 2 adrenergic receptor 59 .
The subtype selectivity of SAG1.3 toward FZD 6 and FZD 7 over FZD 4 is mostly determined by the length of TM6, which in case of a shorter connection to TM7 in FZD 4 traverses through the SAG1.3 binding pocket. While more experiments are necessary to map the SAG1.3 selectivity for all FZD 1-10 , it is likely that FZDs with a long TM6 could accommodate this ligand, whereas the homology cluster FZD 4,9,10 might not.
In summary, repurposing of a SMO agonist led to the identification of a partial agonist of FZD 6 that acts in the 7TM core of the receptor. Medicinal chemistry efforts are now required to define structure-activity relationships of FZD small-molecule ligands to better understand subtype selectivity, and to promote the discovery of diverse pharmacological probes for FZDs as a starting point for the development of therapeutic compounds.

Methods
Computational modelling and MD simulations. The homology modeling of the inactive FZD 6 has been described previously 25 . Briefly, the taladegib-bound structure of SMO was used as a template (PDB ID: 4JKV) 29 and the sequence of FZD 6 (UniProt ID: O60353) was aligned to that of SMO (UniProt ID: Q99835) with ClustalX2 60 . The N-and C-termini were excluded due to a lack of suitable template, and the alignment was manually edited to ensure the proper alignment of transmembrane domains and conserved motifs present in Class F GPCRs. Fifteen homology models were generated with MODELLER 9.11 61 and a representative model was selected based on DOPE score and visual inspection.
Active-24,25(S)-epoxycholesterol bound and G i bound-SMO structure (PDB ID: 6OT0) 21 , was published during the revision of our study. To implement these new structural data, we built 20 new FZD 6 and FZD 7 models with MODELLER 9.19 61  The docking study of purmorphamine was conducted with AutoDock Vina similarly to SAG1.3, with both target proteins (inactive FZD 6 model or SMO crystal structure) relaxed by 200 ns of MD (see below) prior to docking. Docking study of BODIPY-cyclopamine was conducted with Glide, and LigPrep (Schrödinger Release 2018-4) with Epik was used for generating ligand conformations and protonation states (at pH 7 ± 2). The conformational complexity of BODIPY-cyclopamine was reduced by restraining the cyclopamine core to a similar conformation with the cyclopamine that is cocrystallized with SMO (PDB ID: 4O9R) 65 . The BODIPY-cyclopamine was docked to the same FZD 6 model as purmorphamine, to a 20 × 20 × 20 Å 3 box located based on the SAG1.5 binding site in SMO as described above.
The best representatives of the active-like SAG1.3-FZD 6 and SAG1.3-FZD 7 docking complex was also used for initiating MD simulations. SAG1.3 was parametrized with AmberTools18 package (University of California San Francisco) using GAFF2 force field and AM1-BCC charges.
MD simulations were performed on the models of the inactive and active-like FZD 6 , active-like FZD 7 , the crystal structures of FZD 4 , and SMO (PDB IDs: 6BD4 7 and 4QIN 29 , respectively) using GROMACS 66 . The missing residues in the SMO structure (aa 434-440 and aa 494 6.75 -505) were modelled using the SMO structures with PDB IDs: 4JKV and 5L7D 67 , respectively. The missing residues in the FZD 4 structure (aa 420 5.76 -427) were modelled using the SMO structure (PDB ID: 4JKV). The protonation states were assigned at pH = 7.4 in Chimera 68 . The OPM database was used to correctly orientate the proteins and the CHARMM-GUI server 69 was used to embed them in the phosphatidylcholine lipid bilayer, add water molecules and 0.15 M NaCl. Typically, the system was minimized in 1500 steps and was subsequently subjected to equilibration with gradually decreasing position restraints on protein and lipid components. In the last 50 ns of the equilibration run, the harmonic force constants of 50 kJ mol −1 nm −2 were applied on the protein and ligand atoms only. Lastly, the independent isobaric and isothermic (NPT) ensemble production simulations for each system were initiated from random velocities. In these simulations, the CHARMM36m force field was used with a 2 fs time step. The temperature at 310 K was maintained with Nose-Hoover thermostat and the pressure at 1 bar was maintained with Parrinello-Rahman barostat. Particle-mesh Ewald for electrostatic interactions and a 9 Å cutoff for van der Waals interactions were used. All the bonds between hydrogen and other atoms were constrained using the LINCS algorithm. The data files were saved every 100 ps. The MD simulation data (~12.5 µs combined) were analyzed using VMD (analysis extensions "RMSD Trajectory Tool", "VolMap Tool", and "Hydrogen Bonds") and PyMol (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC). The distance plots were produced with distance.tcl script (https://www.ks.uiuc.edu/Training/Tutorials/vmd/vmd-tutorialfiles/distance.tcl). Please see the Supplementary Fig. 7 for more detailed information on the MD simulations. Snapshots of the MD simulations are provided as Supplementary Data files 1-34.
HEK293A cells (female origin; Thermo Fisher Scientific) were seeded in a 12well culture plate at a density of 5 × 10 4 cells ml −1 in 1 ml per well 1 day before transfection. The SMO sgRNA-encoding plasmid vector was transfected into the HEK293A cells using Lipofectamine 2000 (Thermo Fisher Scientific) according to a manufacturer's protocol. Forty eight hours later, the cells were harvested and processed for isolation of GFP + cells using a fluorescence-activated cell sorter (BD FACSDiva). The cells were sorted directly onto a cell-culture grade 96-well plate and the colonies were expanded for 20 days. Subsequently, the cells were lysed and genomic DNA was isolated with NaOH/Tris-HCl. The clones were analyzed for mutations in the targeted genes by a restriction enzyme digestion as described previously 73 . To amplify the sgRNA-targeting sites, the following pair of PCR primers was used: 5′-AAACAAGAGGCTCGTCCCTG-3′ and 5′-TAGCTGTG CATGTCCTGGTG-3′. Seven candidate clones that harbored restriction enzymeinsensitive PCR fragments were assessed for their genomic DNA alterations by direct sequencing. The two resulting, selected, sequence-determined candidate clones were further assessed for absence of SMO protein by immunoblotting ( Supplementary Fig. 9). ΔSMO HEK293A cell line 3 was used in the experiments presented in this study (referred to as ΔSMO HEK293 cells).
In order to create FRET-based sensors, the FZD 6 coding sequence from FZD 6 -GFP 74 was subcloned into pcDNA3-TFP1 (Allele Biotechnology and Pharmaceuticals) between the HindIII and EcoRI restriction sites. Subsequently, the FlAsH binding sequence (FLNCCPGCCMEP) was inserted between G404 and R405 of the FZD 6 ICL3 using GeneArt site-directed mutagenesis kit. FZD 6 -FlAsH-PK was generated by subcloning the FZD 6 insert from FZD 6 -GFP into PK-vector (DiscoverX) with BglII and HindIII, and subsequently the FlAsH binding sequence was inserted as above. The primer sequences can be found in the Supplementary Fig. 23. All the constructs were confirmed by sequencing (GATC-Eurofins, Konstanz, Germany).
FlAsH labeling and FRET efficiency measurements. HEK293 cells were seeded onto coverslips. Cells were transfected 18-20 h later using Effectene (Qiagen), according to the manufacturer's instructions. Cell culture medium was replaced 24 h later and the analysis was done 48 h after transfection. For analysis of the cellular expression and FRET efficiency determination, the cells were transfected with 0.5 µg per well of the corresponding receptor construct, either FZD 6 -FlAsH-TFP or FZD 6 -TFP. For control experiments of basal energy transfer, cells were cotransfected with 0.3 µg per well FZD 6 -TFP and 0.3 µg per well FZD 6 -FlAsH-PK. FlAsH labeling was performed as previously described 75 . In brief, transfected cells were washed once with labeling buffer (10 mM HEPES, 150 mM NaCl, 25 mM KCl, 2 mM MgCl 2 , 4 mM CaCl 2 , 10 mM Glucose, pH = 7.3) and then incubated for 1 h at 37°C with labeling buffer supplemented with 12.5 μM 1,2-ethanedithiol (EDT) and 1 µM FlAsH. In order to reduce nonspecific labeling, cells were rinsed once with labeling buffer and incubated at 37°C for 10 min with labeling buffer containing 250 µM EDT. Cells were then washed twice with labeling buffer and maintained in DMEM prior to measurements.
Fluorescence imaging was performed as previously described 75 . Briefly, coverslips with FlAsH-labeled cells were mounted using an Attofluor holder and placed on a Zeiss inverted microscope (Axiovert200), equipped with an oil immersion 63× objective lens and a dual-emission photometric system (Till Photonics). Cells were maintained in imaging buffer (10 mM HEPES, 140 mM NaCl, 5.4 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , pH = 7.3), and 5 mM of BAL was added to the cells 20-30 s after the recording started. Cells were excited at 436 ± 10 nm using a frequency of 10 Hz with 40 ms illumination time out of a total of 100 ms. Emission of TFP (480 ± 20 nm) and FlAsH (535 ± 15 nm), and the FRET ratio (FlAsH/TFP) were monitored simultaneously over time. Fluorescence signals were detected by photodiodes and digitalized using an analogue-digital converter (Digidata 1440 A, Axon Instruments). FRET efficiency was calculated by inputting the maximum and minimum values of TFP into the following formula: FRET efficiency = (ΔE/Emax) × 100, as previously described 75,76 . Fluorescence intensities data were acquired using Clampex software. Data were analyzed using the software GraphPad Prism 6.
Ligand-induced changes in FZD 6 -FRET probe. To investigate the ligand-induced conformational changes in FZD 6 in populations of cells, HEK293 cells were transfected in suspension using Lipofectamine 2000 (Thermo Fisher Scientific). For the experiments 4 × 10 5 cells ml −1 were transfected with 1000 ng of FZD 6 -FlAsH-TFP plasmid DNA and 100 µl of the suspension was seeded onto a poly-D-lysine (PDL)-coated black 96-well cell culture plate with solid flat bottom (Greiner Bio-One). Analysis of the cells was done 48 h after transfecting/seeding the cells using a CLARIOstar microplate reader (BMG). Following the labelling procedure described above, the cells were excited at 440-15 nm, and emission was detected at 490-20 nm and 530-20 nm. During measurements, the cells were maintained in Hanks' balanced salt solution (HBSS) containing 0.1% BSA. Recombinant WNT-5A or SAG1.3 were added to the cells 5 min after the reading started. Fluorescence changes were recorded for an additional 20 min. Data from the FRET ratio measurements obtained 2 min after the ligand addition were analyzed using GraphPad Prism 6.
Live-cell imaging. Confocal microscopy experiments were performed on a Leica TCS SP2 system, equipped with a HCX PL APO CS 63.0 × 1.32 oil objective. Coverslips with cells expressing the desired constructs were mounted using an Attofluor holder (Molecular Probes) and cells were maintained in the imaging buffer. TFP was excited at 458 nm and fluorescence intensities were recorded from 465-550 nm. Images were taken with 512 × 512 pixel format, 400 Hz, line average 4, frame average 3.
NanoBRET binding assay. ΔSMO HEK293 cells were transiently transfected in suspension using Lipofectamine 2000 (Thermo Fisher Scientific). A total of 4 × 10 5 cells ml −1 were transfected with a total amount of 1000 ng of plasmid DNA using Nluc-FZD 6 : 10 ng or 100 ng for low donor condition or 1000 ng for high donor condition, and the remaining plasmid amount of pcDNA. The cells (100 μl) were seeded onto a PDL-coated black 96-well cell culture plate with solid flat bottom (Greiner Bio-One). Forty eight hours post transfection, cells were washed once with HBSS (HyClone) and maintained in the same buffer. For BODIPY-cyclopamine/10 µM SAG1.3 competition experiments at Nluc-FZD 6 and Nluc-FZD 6 R442A 6.58 , and BODIPY-cyclopamine/10 µM cyclopamine competition experiments 1 ng of donor plasmid DNA was used and the experiments were performed 24 h post transfection. In the saturation experiments, the cells were incubated with different concentrations of BODIPY-cyclopamine (80 μl) for 90 min at 37°C before the addition of the luciferase substrate coelenterazine h (5 μM final concentration, 10 μl; Biosynth #C-7004) for 6 min prior to the BRET measurement. In the competition experiments, the cells were either preincubated with different concentrations of SAG1.3 (70 μl) for 30 min at 37°C followed by the addition of BODIPY-cyclopamine (300 nM, 10 μl); or the cells were preincubated with SAG1.3 (10 µM, 70 µl) for 30 min at 37°C followed by the addition of the different concentrations of BODIPY-cyclopamine (10 μl). The cells were then incubated for additional 90 min at 37°C before the addition of the luciferase substrate colenterazine h (5 μM final concentration, 10 μl) for 6 min prior to the BRET measurement. The BRET ratio was determined as the ratio of light emitted by BODIPY-cyclopamine (energy acceptor) and light emitted by Nluc-tagged receptor (energy donor). The BRET acceptor (bandpass filter 535-30 nm) and BRET donor (bandpass filter 475-30 nm) emission signals were measured using a CLARIOstar microplate reader (BMG). ΔBRET ratio was calculated as the difference in BRET ratio of cells treated with SAG1.3 and cells treated with vehicle. BODIPY fluorescence was measured prior to reading luminescence (excitation: 477-14 nm, emission: 525-30 nm). Data were analyzed using GraphPad Prism 6.
BRET assays. HEK293, ΔSMO HEK293, or ΔFZD 1-10 HEK 293 cells were transiently transfected in suspension using Lipofectamine 2000 (Thermo Fisher Scientific). For the mG BRET assays, 4 × 10 5 cells ml −1 were transfected with 800 ng of mG plasmid DNA, 100 ng of the Rluc8/Nluc-tagged receptor plasmid DNA, and 100 ng of pcDNA. For the DVL2 recruitment bystander BRET assays, 4 × 10 5 cells ml −1 were transfected with 780 ng of Venus-KRas plasmid DNA, 200 ng of the receptor plasmid DNA, and 20 ng of Nluc-DVL2 plasmid DNA. For the direct DVL2-FZD recruitment BRET assays, 4 × 10 5 cells ml −1 were transfected with 800 ng of Venus-tagged FZD 6 plasmid DNA, 20 ng of Nluc-DVL2 plasmid DNA, and 180 ng of pcDNA plasmid DNA. The cells (100 µl) were seeded onto a PDL-coated black 96-well cell culture plate with solid flat bottom (Greiner Bio-One). Forty eight hours post transfection, cells were washed once with HBSS (Gibco or Thermo Fisher Scientific) and maintained in the same buffer. The cells were stimulated with ligands 6 min after the addition of the luciferase substrate coelenterazine h (5 µM final concentration; Biosynth #C-7004). The BRET signal was determined as the ratio of light emitted by Venus-tagged biosensors (energy acceptors) and light emitted by Rluc8/Nluc-tagged biosensors (energy donors). The BRET acceptor (535-30 nm) and BRET donor (475-30 nm) emission signals were measured using a CLARIOstar microplate reader (BMG). In the saturation BRET experiments, Net BRET was calculated as the difference in BRET ratio between cells expressing both donor and acceptor, and cells expressing only donor. Venus fluorescence was measured prior to reading luminescence (excitation 497-15 nm, emission 540-20 nm) and calculated as average fluorescence from each control well. Data presented in this study come from the ligand-induced BRET measurements obtained 5 min after the ligand addition (11 min after the coelenterazine h addition), and the saturation BRET measurements obtained 7 min after the coelenterazine h addition. Data were analyzed using GraphPad Prism 6.
NanoBiT luciferase assay. ΔSMO HEK293 cells were transiently transfected in suspension using Lipofectamine 2000 (Thermo Fisher Scientific). For the experiments 4 × 10 5 cells ml −1 were transfected with 100 ng of Gα i1 -LgBiT plasmid DNA, 500 ng of SmBiT-Gβ 5 , 500 ng of Gγ 2 , and 200 ng of receptor plasmid DNA. The cells (100 µl) were seeded onto a PDL-coated white 96-well cell culture plate with solid flat bottom (Greiner Bio-One). Forty eight hours post transfection, the cells were washed once with 0.1% BSA/HBSS (Gibco or Thermo Fisher Scientific) and maintained in the same buffer. The cells were stimulated with ligands 30 min after the addition of the luciferase substrate coelenterazine h (10 µM final concentration). Nluc lum (470-80 nm) was measured using a CLARIOstar microplate reader (BMG). Data from the luminescence measurements obtained 5 min after the ligand addition were analyzed using GraphPad Prism 6.