Structural basis for FGF hormone signalling – Nature

Expression constructs

The ligation-independent In-Fusion HD cloning kit (no. 639648, Clontech) was used to construct neomycin and hygromycin resistant pEF1α-IRES-neo and pEF1α-IRES-hygro lentiviral vectors encoding wild-type full-length human FGFR2c, FGFR3c and FGFR4 according to the protocol previously described for the human FGFR1c lentiviral expression construct⁶. The pET-30a-based bacterial expression construct for N-terminally his-tagged human FGF23 (residues Tyr25 to Ile251) was described previously⁶. pHLsec expression vectors encoding the extracellular ligand-binding region (that is, D2–D3 region) of FGFR3c (residues Asp142 to Arg365; FGFR3c^ecto) and FGFR4 (residues Asp142 to Arg365; FGFR4^ecto) were made following the same strategy described previously for the human FGFR1c^ecto (ref. ⁶). Single and multiple site mutations and truncations were introduced into expression constructs encoding the wild-type proteins using a Q5 site-directed mutagenesis kit (no. E0554S, New England Biolabs Inc.). The expression constructs were verified by restriction enzyme digestion and DNA sequencing.

Recombinant protein expression and purification

To produce minimally glycosylated ectodomains of FGFR1c, 3c and 4, N-acetylglucosaminyltransferase I (GnTI⁻) deficient HEK293S cells were transiently transfected with respective pHLsec expression constructs via cationic polymer polyethyleneimine (PEI, no. 23966-1, Polysciences, Inc.) following a published protocol⁴⁰. Then, 24 h posttransfection, tryptone N1 (TN1, catalogue no. 19553, Organotechnie) was added to the medium to promote protein expression. At day three, secreted FGFR ectodomains from 1 l of conditioned medium were captured on a heparin affinity HiTrap column (no. 17040703, GE Healthcare) and eluted with 20 column volume of salt gradient (0–2 M). Fractions containing FGFR^ecto proteins were pooled, concentrated to 5 ml and applied to a Superdex-75 gel filtration column (no. 28989333, GE Healthcare). FGFR^ecto proteins were eluted isocratically in 25 mM HEPES pH = 7.5 buffer containing 1 M NaCl. Wild-type and mutated FGF23 proteins were expressed in E. coli as inclusion bodies, refolded in vitro and purified to homogeneity using sequential cation exchange and SEC following an established protocol⁶. Secreted αKlotho^ecto was purified from conditioned media of a HEK293S GnTI⁻ cell line stably expressing the entire extracellular domain of human αKlotho (residues Met1 to Ser981; αKlotho^ecto) using a heparin affinity HiTrap column followed by SOUCRE Q anion and Superdex 200 column chromatography, as described previously⁶.

Cryo-EM specimen preparation, image processing, model building and refinement

FGF23–FGFR1c–αKlotho–HS, FGF23–FGFR3c–αKlotho–HS or FGF23–FGFR4–αKlotho–HS quaternary complexes were prepared by mixing FGF23 with one of the FGFR ectodomains, αKlotho^ecto and a heparin dodecasaccharide 12 (HO12, Iduron Ltd) using a 1:1:1:1 molar ratio. The mixtures were concentrated to approximately 5 mg ml⁻¹, applied to a Superdex 200 column (no. 28989335, GE Healthcare) and eluted isocratically in 25 mM HEPES pH = 7.5 buffer containing 100 mM NaCl. Peak fractions were analysed by SDS–PAGE and top fractions containing the highest concentration and purity of the quaternary complex were used directly for grid preparation without further concentration to avoid protein aggregation. The final concentrations of FGF23–FGFR1c–αKlotho–HS, FGF23–FGFR3c–αKlotho–HS and FGF23–FGFR4–αKlotho–HS complexes for grid preparation were 1.5, 2.4 and 1.5 mg ml⁻¹, respectively.

To prepare the cryo-EM grids, 2–3 μl of purified protein complex at approximately1.5–2.5 mg ml⁻¹ was applied to glow discharged gold grid (UltrAuFoil). The grid was then blotted for 1–2 s under 0 or 1 force at 100% humidity using a Mark IV Vitrobot (FEI) before plunging into liquid ethane. Micrographs of the FGF23–FGFR1c–αKlotho–HS and FGF23–FGFR4–αKlotho–HS complexes were acquired on a Talos Arctica microscope with K2 direct electron detector at ×36,000 magnification (corresponding to 1.096 Å per pixel). Accumulated doses used were 50.37 e⁻/Ų and 53.84 e⁻/Ų for FGF23–FGFR1c–αKlotho–HS and FGF23–FGFR4–αKlotho–HS, respectively. Micrographs of the FGF23–FGFR3c–αKlotho–HS complex were collected on a Titan Krios microscope equipped with a K2 direct electron detector and an energy filter. The magnification used was ×130,000, with a pixel size of 1.048 Å, and an accumulated dose of 72.44 e^–/Ų. Leginon⁴¹ was used to target the holes with 5–100 nm of ice thickness, resulting in 10,186, 6,409 and 16,602 micrographs being collected for each of three quaternary complexes.

WARP⁴² was used for motion correction and contrast transfer function estimation for all three cryo-EM datasets. Micrographs with an overall resolution worse than 5.5 Å were excluded. The final number of micrographs used were 9,501, 5,164 and 15,049, respectively, yielding more than one million particles for each complex. Particle stacks were then imported to cryoSPARC⁴³ for two-dimensional classification, ab-initio reconstruction with three or four models and three-dimensional classification. Finally, 1,497,967 (FGF23–FGFR1c–αKlotho–HS), 291,540 (FGF23–FGFR3c–αKlotho–HS) and 856,877 (FGF23–FGFR4–αKlotho–HS) particles were used for heterogeneous refinement with C1 symmetry, resulting in 2.74, 3.20 and 3.03 Å resolutions, respectively. Components/domains from FGF23–FGFR1c–αKlotho X-ray structures (PDB: 5W21) were manually docked into cryo-EM density maps using Chimera⁴⁴ and the rigid body was refined. Initial models were then adjusted in Coot⁴⁵ and real-space refined in Phenix⁴⁶. Refinement and model statistics are shown in Extended Data Table 1. Representative cryo-EM images, two-dimensional class averages and three-dimensional maps of each quaternary complex are shown in Extended Data Fig. 1.

MD simulation

The cryo-EM structure of the FGF23–FGFR1c–αKlotho–HS quaternary complex was solvated in a water box of 16 × 16 × 16 nm³. The CHARMM-GUI server was used to generate the configuration, topology^47,48,49,50 and the parameter files with the CHARMM36m force field⁵¹. In addition to protein molecules, the simulation system included about 138,073 water molecules, 393 sodium and 393 chloride ions (mimicking the 150 mM NaCl present in the protein buffer), resulting in a total of 439,298 atoms. A 300 ns all-atom MD simulation trajectory was generated using GROMACS 2021 (ref. ⁵²) at 303 K using a time step of 2 fs. The cubic periodic boundary condition was used during the simulations and the van der Waals interaction was switched off from 1 nm to 1.2 nm. The long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) method. Energy minimization was carried out using the steepest descent algorithm, followed by a 0.4 ns constant particle number, volume and temperature (NVT) and a 20 ns constant particle number, pressure and temperature (NPT) equilibration simulation by gradually decreasing force restraints from 1000 kJ mol⁻¹ nm⁻² to 400 kJ mol⁻¹ nm⁻² (for the NVT stage) and 400 kJ mol⁻¹ nm⁻² to 40 kJ mol⁻¹ nm⁻² (for NPT stage). At the conclusion of the equilibration steps, all force restraints were removed and the MD simulation was performed in the NPT ensemble.

Generation of cell lines and FGFR signalling assay

HEK293T cells (verified by a morphology check under microscope, mycoplasma negative in 4′,6-diamidino-2-phenylindole (DAPI)) were used for lentiviral vector packaging and production of high-titre viral particles. An L6 myoblast cell line (no. GNR 4, National Collection of Authenticated Cell Cultures) was used as the host for stable expression of full-length (transmembrane) human FGFR1c, FGFR2c, FGFR3c, FGFR4 and mutants thereof. L6 cells endogenously express HSPGs but are devoid of FGFRs and αKlotho coreceptor and hence are naturally non-responsive to FGF23. However, via controlled ectopic co-expression of cognate FGFRs and αKlotho or exogenous supplementation with soluble αKlotho^ecto, these cells can respond to FGF23 stimulation. Accordingly, L6 cells are excellent hosts for reconstitution studies of FGF23 signalling in a physiological environment. Both HEK293T and L6 cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM, no. C11995500BT, Gibco) supplemented with 10% fetal bovine serum (FBS, no. FSD500, ExCell Bio), 100 U ml⁻¹ of Penicillin and 100 μg ml⁻¹ Streptomycin (no. P1400, Solarbio). Viral packaging and generation of the recombinant lentivirus particles in HEK293T cells were carried out using published protocols³³. For stable expression of individual wild-type or mutated FGFR cell lines, 2 × 10⁵ L6 cells were plated in six-well cell culture dishes and infected with lentivirus particles encoding given FGFR in the presence of polybrene (5 μg ml⁻¹; no. sc-134220, Santa Cruz Biotechnology). Stable transfectants were selected using G418 (0.5 mg ml⁻¹, no. HY-17561, MedChemExpress) or hygromycin (8 μg ml⁻¹, no. HY-B0490, MedChemExpress). For the FGFR1c+αKlotho^TM co-expressing cell line, L6 cells stably expressing FGFR1c^WT (resistant to G418) were infected with lentiviral particles encoding wild-type or mutated transmembrane αKlotho (αKlotho^TM) and the co-expressing cells were selected using hygromycin (80 μg ml⁻¹, no. HY-B0490, MedChemExpress).

For cell stimulation studies, parental and stably transfected L6 cells were seeded in 12-well cell culture plates at a density of 1 × 10⁵ cells per well and maintained for 24 h. On the next day, the cells were rinsed three times with phosphate buffered saline (PBS) and then serum starved for 12 h and costimulated with FGF23^WT and αKlotho^ecto for 5, 10, 20 and 40 min. For single time point stimulation, samples were harvested after 5 min.

After stimulation, cells were lysed and total lysates samples were analysed by western blotting, as previously described³³. The following antibodies were used: phosphorylated FGFR (1:1000, no. 3471S, Cell Signaling Technology), phosphorylated FRS2α (1:1000, no. 3864S, Cell Signaling Technology), phosphorylated PLCγ1 (1:1000, no. 2821S, Cell Signaling Technology), phosphorylated ERK1/2 (1:1000, no. 4370S, Cell Signaling Technology), α-tubulin (1:20000, no. 66031-1-Ig, Proteintech), total-FGFR1 (1:1000, no. 9740S, Cell Signaling Technology), total-FGFR2 (1:1000, no. 23328S, Cell Signaling Technology), total-FGFR3 (1:1000, no. ab133644, Abcam), total-FGFR4 (1:1000, no. 8562S, Cell Signaling Technology), HRP conjugated goat anti-mouse IgG (H + L) (1:5000, no. SA00001-1, Proteintech), HRP conjugated goat anti-rabbit IgG(H + L) (1:5000, no. SA00001-2, Proteintech). Blots were developed using enhanced chemiluminescence reagents (no. P10300, NCM Biotech Laboratories) by the ChemiDoc XRS+ system (Bio-Rad) or Amersham ImageQuant 800 (GE).

Determination of cell-surface expression of mutated FGFRs via endoglycosidase H sensitivity assay

Endoglycosidase H (Endo H) sensitivity was used to analyse potential impacts of ectodomain mutations on FGFR glycosylation/maturation and hence trafficking to the cell surface. In immunoblots, wild-type FGFRs migrate as a doublet of a major diffuse upper and a minor sharp lower band. The upper band represents the fully glycosylated mature FGFR, decorated with complex sugars that has passed the ER quality control and has been successfully trafficked to the cell surface. On the other hand, the faster migrating lower band is an incompletely processed high mannose form that is trapped in ER. Mutations affecting receptor maturation manifest in an increase in proportion of the faster migrating ER-resident band. The mannose-rich form is sensitive to Endo H which cleaves the bond between two N-acetylglucosamine (GlcNAc) subunits directly proximal to the asparagine residue. However, the fully glycosylated cell surface-resident band is resistant to Endo H. Accordingly, cell-surface abundance of FGFRs can be expressed as a ratio of Endo H-resistant fraction over the total receptor expression as determined by treating the receptor with PNGase F. This enzyme is an amidase that hydrolyzes the bond between the innermost GlcNAc and asparagine irrespective of complex sugar content and thus completely strips the FGFR from all its N-linked sugars. Accordingly, WT or mutant FGFR cell lines were lysed in an NP-40 lysis buffer (Biotime, no. P0013F, supplemented with 1 mM PMSF) for 15 min at 4 °C. First, 20 μg total protein (quantified by BCA assay) were denatured with glycoprotein denaturing buffer at 100 °C for 10 min and then treated with 500 units of Endo H (New England Biolabs, no. P0702S) or peptide-N-glycosidase F (PNGase F) (New England Biolabs, no. P0704S) following the manufacturer’s instructions. Endo H- and PNGase F-treated samples were immunoblotted with FGFR isoform-specific antibodies, as detailed above.

Size-exclusion chromatography–multi-angle dynamic light scattering

The molecular mass of the SEC-purified FGF23–FGFR1c-αKlotho–HS quaternary complex was determined by multi-angle light scattering following the established protocol⁶. Before the experiment, at least 60 ml of degassed running buffer (25 mM HEPES pH 7.5 containing 150 mM NaCl) were passed through the system to equilibrate the column and establish steady baselines for light scattering and refractive index detectors. Then, 50 μl of purified FGF23–FGFR1c-αKlotho–HS quaternary complex (1.5 mg ml⁻¹) was injected onto the Superdex 200 10/300 GL column and the eluent was continuously monitored at 280 nm absorbance, laser light scattering and refractive index at a flow rate of 0.5 ml min⁻¹. As a control, 50 μl of a purified FGF23–FGFR-αKlotho ternary complex (1.5 mg ml⁻¹) sample was analysed under the same condition. The experiments were performed at ambient temperature. Laser light scattering intensity and eluent refractive index values were used to derive molecular mass as implemented by the ASTRA software (Wyatt Technology Corp.).

Proximity ligation assay

Cells were seeded onto microscope cover glasses (no. WHB-12-CS-LC, WHB) placed inside 12-well cell culture dishes at 1 × 10⁵ cells per well and allowed to adhere for 24 h. On the next day, cells were washed three times with PBS and serum starved for 12 h. Following costimulation with FGF23^WT and αKlotho^ecto for 20 min, cells were washed three times with PBS and fixed with 4% paraformaldehyde for 30 min prior to PLA. The PLA reaction was performed using a Duolink PLA kit following the manufacturer’s instructions (no. DUO92101, Sigma-Aldrich) and visualized via fluorescence microscopy. Briefly, slides were treated with blocking solution for 60 min at 37 °C, rinsed three times with wash buffer A and then incubated overnight at 4 °C with two different primary antibodies raised in two different species for each FGFR isoform of interest (FGFR1 (1:20, no. PA5-25979, ThermoFisher) from rabbit, FGFR1 (1:100, no. ab824, Abcam) from mouse, FGFR2 (1:100, no. 23328S, Cell Signaling Technology) from rabbit, FGFR2 (1:50, no. sc-6930, Santa Cruz Biotechnology) from mouse, FGFR3 (1:100, no. MA5-32620, ThermoFisher) from rabbit, FGFR3 (1:100, no. sc-13121, Santa Cruz Biotechnology) from mouse, FGFR4 (1:100, no. 8562S, Cell Signaling Technology) from rabbit, FGFR4 (1:100, no. sc-136988, Santa Cruz Biotechnology) from mouse). Following three rinses with wash buffer A, slides were incubated with oligo-linked secondary antibodies (Duolink anti-mouse minus and anti-rabbit plus) for 1 h at 37 °C. Slides were rinsed again with wash buffer A and immersed in ligase solution for 30 min at 37 °C, to allow formation of circular DNA, followed by incubation with polymerase solution for 100 min at 37 °C for rolling circle amplification in a dark room. Slides were rinsed twice with wash buffer B for 10 min each, followed by rinsing with a 100-fold dilution of buffer B for 1 min. For imaging analysis, slides were covered by coverslips using a minimal volume of Duolink PLA mounting medium containing DAPI. Slides were examined using a confocal laser scanning microscope (C2si, Nikon).

Statistical analysis and reproducibility

All statistical analyses were carried out using GraphPad Prism 8.0. For statistical analysis of immunoblotting data, densitometric values (determined using ImageJ) from three independent experiments were used. For statistical analysis of PLA data, a number of fluorescent dots and cells (counted manually) was used from six randomly chosen microscope fields from two biologically independent experiments. Processing of the western blotting and PLA data in Figs. 2c,d, 3b–d and 4b,c was done using two-way ANOVA, followed by Tukey. One-way ANOVA, followed by Tukey was applied to process the western blotting data in Extended Data Fig. 3. Protein purifications were repeated at least eight times yielding samples with comparable purity/quantity. Western blotting experiments were carried out in biological triplicates with similar results. PLA assays were repeated at least twice independently, with analogous results.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.