Cryo-EM structure of constitutively active human Frizzled 7 in complex with heterotrimeric Gs

Dear Editor, The ten mammalian Frizzleds (FZD 1 – 10 ) belong to the class F of G protein-coupled receptors (GPCRs) and mediate WNT signaling through interaction with transducer proteins including Dishev-elled (DVL) or heterotrimeric G proteins. 1 Their involvement in human disease has put FZDs at the forefront of drug targets, especially anti-cancer therapy. 2 However, no drugs have been developed for ef ﬁ cient pharmacological modulation of FZDs, partially owing to the limited understanding of FZD structure and activation mechanisms. 1,3 Among class F, FZD 7 is intensively pursued due to its relevance in various tumor models, particularly in intestinal cancers. 4 Detailed structures of the receptor complexes would allow for structure-guided discovery of new drug candidates. FZD 1 – 10 share structural similarity with the related class F member Smoothened (SMO), which mediates Hedgehog signaling and is a validated target for cancer therapy. 2 In an effort to understand

a total of 5872 movies were recorded at 130k nominal magnification (calibrated pixel size: 1.04 Å/pixel) and super resolution mode by SerialEM using the beam image shift acquisition method with one image near the edge of each hole and saved as non-gain normalized TIFF files. A 50-m C2 aperture was always inserted during the data collection period. The defocus ranged from -0.7 to -2.2 µm. For each movie stack, a total of 40 frames were recorded at a dose rate of 8 e -/pixel/sec for a duration of 8.1s, yielding a total dose of 60 e -/A 2 .

Data processing
Both datasets were motion corrected with MotionCor2 6 using 5 × 5 patches and no frame grouping.
Both the dose weighted and non-dose weighted averages were saved, and the CTF parameters were estimated based on the non-dose-weighted averages using CTFFIND4 6,7 . Only images with the highest resolution less than 4 Å were selected for further processing. Moreover, images with visible contamination or large carbon regions by manual examination were also removed. For the FZD7-mGs-Nb35 dataset, a total of 4462 movies were finally chosen for particle picking. To avoid potential bias about the structural conformation in the dataset, a Laplacian-of-Gaussian blob picker was firstly applied to pick particles. Good 2D class averages with randomized orientations and clear secondary features were selected as the 2D templates for another round of auto-picking process, yielding an initial particle stack of 2,074,920 particles. Another round of 2D classification was applied to eliminate bad quality particles by Relion 3.1 8 , yielding a dataset containing 700,220 good quality particles with randomized orientations and clear secondary features. These particles were imported into cryoSPARC 2.15 9 to generate five de novo initial models, and one model out of five from which 2D projection matches the majority of 2D class averages was selected as the initial model for further processing by Relion3.1 8 . For the first round of 3D classification, this initial model was firstly low-pass filtered by 20 Å and used to divide the dataset into 3 different 3D classes. To overcome problems with conformational heterogeneity, the particles with reasonable 3D reconstructed maps were then grouped before subsequent rounds of 3D classifications focusing on the transmembrane domains. The centering, clipping, binning, thresholding operations of the maps; and the generation, addition, multiplication of masks were performed by EMAN2 10 . Homemade python (https://www.python.org/) scripting was also applied to eliminate particles that could not converge to the correct Euler angles. Finally, a total of 307,000 particles were selected for homogeneous refinement and post-processing, yielding a map with resolution of 3.3 Å at a Fourier shell correlation of 0.143. Then the dataset was subjected to further CTF refinement and post-processing, and the final resolution was improved to 3.2Å. Subsequently, classification without alignment focusing on the transmembrane region was implemented to further separate particles into 3 different groups. 224,750 particles which have high resolution features at the transmembrane region were selected and were subject to final rounds of homogeneous refinement. The local resolution was estimated using the cryoSPARC v2.15 9 "local resolution estimation" function.

Model Building
The homology models of the human FZD7, mGαs were initially generated by Swiss model 11 (template PDB: 5V57 5 , 7D3S 12 ). For Gβ, Gγ and Nb35, the model 3CIK 13 , 6PCV 14 and 6GDG 1 were chosen. These models were then fitted into the density maps in UCSF Chimera 15 , and manually adjusted to fit the density maps in Coot software 16 . Subsequently, the generated model was automatically refined and manually adjusted in Coot 16 and Phenix 17 , respectively, for several iterations. The clashscores, Molprobity and Ramachandran analysis was performed using the MolProbity 18 . The final refinement statistics were generated using the "comprehensive validation The final model of FZD7-mGs complex revealed seven intact transmembrane helices (7TM), extracellular loops 1 and 2 (ECL1 and 2), the entire hinge domain 19 (aa 210-250), the three intracellular loops (ICLs), the helix 8 perpendicular to the core helices, the mGs heterotrimer and Nb35. The extracellular N N-terminus including the CRD (aa 47-168) and ECL3 (aa 509-525) were not resolved, likely owing to structural flexibility relative to the receptor core.

Molecular dynamics (MD) simulations
To obtain a simulation complex of suitable size, the mGαs-Gβ-Gγ-Nb35 complex was replaced by miniGs393 (mGαs) 19 . The FZD7-miniGs393 system was built using Schrödinger Maestro 2020-4 molecular modeling platform (Schrödinger, LLC, New York, NY, 2020). The unresolved ECL3 of FZD7 was modelled using the SMO structure (PDB ID: 5L7D) as a template and miniGs393 using the original mGs as a template. For unresolved amino acids at mGs structure, miniGs393 structure (PDB ID: 5G53) guided the model building.
The MD simulations were run using GROMACS 2020.4 20 . The receptor was oriented using the OPM database 21 and embedded in the POPC lipid bilayer (151 lipids / leaflet) by CHARMM-GUI server 22 with TIP3p water molecules and 0.15 M NaCl. The system was minimized for approximately 3000 steps and then equilibrated 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 atoms only. It should be noted that the simulations were run in absence of the G subunits and Nb35.
Ten independent isobaric and isothermic (NPT) ensemble production simulations of 50 ns each were initiated from random velocities using the CHARMM36m force field 23 and a 2-fs time step.
The temperature at 303.15 K was maintained with Nose-Hoover thermostat 24 and the pressure at 1 bar with Parrinello-Rahman barostat 25 . Potential-shift-Verlet was used for electrostatic and van der Waals interactions with 12 Å cut-off, and the bonds between hydrogen and other atoms were constrained by the LINCS algorithm 26 . The data were analyzed using VMD (visualization and measurement of RMSDs, distances and angles) 27 and visualized in PyMol. MD simulations trajectories will be available at GPCRmd (an open-access MD database for GPCRs; www.gpcrmd.org).

Plasmids and molecular cloning
The FRET-based cAMP sensor (H187) was from Kees Jalink (The Netherlands Cancer Institute, Amsterdam, The Netherlands) 28

Densitometric quantification of protein bands
The complex sample fractions after size-exclusion chromatography were mixed with protein loading buffer and prepared for SDS-PAGE (Genscript). Proteins were visualized with Coomassie blue. The gels were scanned and exported as image files. The bands were quantified by densitometry using image J software (NIH) (www.imagej.net). Data are presented as means ± SEM.